Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Interventions for self‐harm in children and adolescents

Authors' declarations of interest

Version published: 07 March 2021 Version history

https://doi.org/10.1002/14651858.CD013667.pub2
Collapse all Expand all

Background

Self‐harm (SH; intentional self‐poisoning or self‐injury regardless of degree of suicidal intent or other types of motivation) is a growing problem in most countries, often repeated, and associated with suicide. Evidence assessing the effectiveness of interventions in the treatment of SH in children and adolescents is lacking, especially when compared with the evidence for psychosocial interventions in adults. This review therefore updates a previous Cochrane Review (last published in 2015) on the role of interventions for SH in children and adolescents.

Objectives

To assess the effects of psychosocial interventions or pharmacological agents or natural products for SH compared to comparison types of care (e.g. treatment‐as‐usual, routine psychiatric care, enhanced usual care, active comparator, placebo, alternative pharmacological treatment, or a combination of these) for children and adolescents (up to 18 years of age) who engage in SH.

Search methods

We searched the Cochrane Common Mental Disorders Specialized Register, the Cochrane Library (Central Register of Controlled Trials [CENTRAL] and Cochrane Database of Systematic Reviews [CDSR]), together with MEDLINE, Ovid Embase, and PsycINFO (to 4 July 2020).

Selection criteria

We included all randomised controlled trials (RCTs) comparing specific psychosocial interventions or pharmacological agents or natural products with treatment‐as‐usual (TAU), routine psychiatric care, enhanced usual care (EUC), active comparator, placebo, alternative pharmacological treatment, or a combination of these, in children and adolescents with a recent (within six months of trial entry) episode of SH resulting in presentation to hospital or clinical services. The primary outcome was the occurrence of a repeated episode of SH over a maximum follow‐up period of two years. Secondary outcomes included treatment adherence, depression, hopelessness, general functioning, social functioning, suicidal ideation, and suicide.

Data collection and analysis

We independently selected trials, extracted data, and appraised trial quality. For binary outcomes, we calculated odds ratios (ORs) and their 95% confidence internals (CIs). For continuous outcomes, we calculated the mean difference (MD) or standardised mean difference (SMD) and 95% CIs. The overall quality of evidence for the primary outcome (i.e. repetition of SH at post‐intervention) was appraised for each intervention using the GRADE approach.

Main results

We included data from 17 trials with a total of 2280 participants. Participants in these trials were predominately female (87.6%) with a mean age of 14.7 years (standard deviation (SD) 1.5 years). The trials included in this review investigated the effectiveness of various forms of psychosocial interventions. None of the included trials evaluated the effectiveness of pharmacological agents in this clinical population. There was a lower rate of SH repetition for DBT‐A (30%) as compared to TAU, EUC, or alternative psychotherapy (43%) on repetition of SH at post‐intervention in four trials (OR 0.46, 95% CI 0.26 to 0.82; N = 270; k = 4; high‐certainty evidence). There may be no evidence of a difference for individual cognitive behavioural therapy (CBT)‐based psychotherapy and TAU for repetition of SH at post‐intervention (OR 0.93, 95% CI 0.12 to 7.24; N = 51; k = 2; low‐certainty evidence). We are uncertain whether mentalisation based therapy for adolescents (MBT‐A) reduces repetition of SH at post‐intervention as compared to TAU (OR 0.70, 95% CI 0.06 to 8.46; N = 85; k = 2; very low‐certainty evidence). Heterogeneity for this outcome was substantial ( I² = 68%). There is probably no evidence of a difference between family therapy and either TAU or EUC on repetition of SH at post‐intervention (OR 1.00, 95% CI 0.49 to 2.07; N = 191; k = 2; moderate‐certainty evidence). However, there was no evidence of a difference for compliance enhancement approaches on repetition of SH by the six‐month follow‐up assessment, for group‐based psychotherapy at the six‐ or 12‐month follow‐up assessments, for a remote contact intervention (emergency cards) at the 12‐month assessment, or for therapeutic assessment at the 12‐ or 24‐month follow‐up assessments.

Authors' conclusions

Given the moderate or very low quality of the available evidence, and the small number of trials identified, there is only uncertain evidence regarding a number of psychosocial interventions in children and adolescents who engage in SH. Further evaluation of DBT‐A is warranted. Given the evidence for its benefit in adults who engage in SH, individual CBT‐based psychotherapy should also be further developed and evaluated in children and adolescents.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Interventions for children and adolescents who self‐harm

We have reviewed the international literature regarding psychosocial interventions, pharmacological (drug), and natural product (dietary supplementation) treatment trials in the field. A total of 17 trials meeting our inclusion criteria were identified. There is little evidence of beneficial effects for individual cognitive behavioural therapy (CBT)‐based psychotherapy, mentalisation‐based therapy for adolescents (MBT‐A), group‐based psychotherapy, enhanced assessment approaches, compliance enhancement approaches, family interventions, or remote contact interventions. There is some evidence of effectiveness for dialectical behaviour therapy (DBT‐A) for adolescents. However, few trials have been conducted and those that have are generally small, meaning that possible beneficial effects of some of these therapies cannot be ruled out.

Why is this review important?

Self‐harm (SH), which includes intentional self‐poisoning/overdose and self‐injury, is a major problem in many countries and is strongly linked with suicide. It is therefore important that effective treatments for SH patients are developed. There has been an increase in the use of interventions for SH in children and adolescents. It is therefore important to assess the evidence for their effectiveness.

Who will be interested in this review?

Hospital administrators (e.g. service providers), health policy officers and third party payers (e.g. health insurers), clinicians working with patients who engage in SH, patients themselves, and their relatives.

What questions does this review aim to answer?

This review is an update of a previous Cochrane Review from 2015 which found little evidence of beneficial effects of interventions for SH in children and adolescents. This updated review aims to further evaluate the evidence for effectiveness of interventions for children and adolescents with SH with a broader range of outcomes.

Which studies were included in the review?

To be included in the review, studies had to be randomised controlled trials of either psychosocial or drug treatments for children and adolescents up to 18 years of age who had recently engaged in SH.

What does the evidence from the review tell us?

There have been surprisingly few investigations of treatments for SH in children and adolescents, despite the size of this problem in many countries. We found positive effects of DBT‐A on repetition of SH. There is currently no clear evidence for the effectiveness of individual CBT‐based psychotherapy, MBT‐A, group‐based psychotherapy, enhanced assessment approaches, compliance enhancement approaches, family interventions, or remote contact interventions in preventing repetition of SH.

What should happen next?

We recommend further trials of DBT‐A. Given the evidence for its benefit for adults who engage in SH, individual CBT‐based psychotherapy should also be further developed and evaluated in children and adolescents. Given the extent of SH in children and adolescents, greater attention should be paid to the development and evaluation of specific therapies for this population.

Authors' conclusions

Implications for practice

Presentations of children and adolescents to clinical services following SH are common, yet there have been relatively few systematic investigations into interventions that may prevent recurrence. We found only 17 trials of psychosocial interventions in this clinical population. We found none of pharmacological treatments.

Given the evidence for its benefit for adults who engage in SH, individual CBT‐based psychotherapy needs to be further developed and evaluated in children and adolescents. However, treatment adherence with this approach may be challenging in this age group. One small feasibility study included in this review, for example, found that few adolescents completed the acute treatment phase, and even fewer attended any of the booster sessions offered (Sinyor 2020). We also found some positive effects of DBT‐A, but methodological factors limit confidence in the generalisability of the results.  We recommend further evaluation of these approaches to assess the impact of these interventions in different samples and settings. Given the multi‐component nature of these intervention approaches, greater use of head‐to‐head trials, which allow for dismantling of the effect size(s) between one or more component(s), should be considered.

Given the high incidence of family problems in children and adolescents who engage in SH, the lack of evidence for efficacy of family therapy approaches found in this review is both surprising and disappointing. On the strength of the evidence from three trials, including one recent larger trial, there is also little support for group‐based therapy for adolescents with a history of multiple episodes of SH. Additionally, it is notable that the authors of one of these trials reports an incident in which one participant posted confidential information about another participant on an online blog (Hazell 2009), highlighting the potential risks associated with group‐based therapy in this clinical population

An argument for intervention following an episode of SH is that it may improve other outcomes even if it does not reduce SH. Secondary outcomes were examined variably across the included trials. There was only limited evidence that experimental interventions might lead to better outcomes in these other domains. DBT‐A improved depression, hopelessness, and suicidal ideation in the short term. Both DBT‐A and therapeutic assessment appeared to improve treatment adherence (at least for individual therapy). Conspicuously, other treatments that might have been expected to improve depression, such as individual CBT‐based psychotherapy and MBT‐A did not perform any better than TAU. 

Indeed, few psychosocial interventions appear to perform better than TAU. However, TAU was not well described in most clinical trials we examined. TAU also varies greatly across clinical settings (Witt 2018b). A positive step forward would be the operationalisation of TAU to inform both clinical practice and research. It is possible that TAU has an advantage over some of these specific interventions because it offers more flexibility to tailor treatment to the specific needs of the patients.

Results of this review would also suggest that a comprehensive therapeutic assessment may increase engagement with subsequent treatment whilst, additionally, enabling the identification of psychosocial needs that should be addressed during treatment. Although this finding is based on a single cluster‐randomised trial which may overestimate the effectiveness of the intervention, this result suggests that a comprehensive therapeutic psychosocial assessment might be a useful part of a clinical intervention. This is in keeping with official guidance (NICE 2011).

Implications for research

Given that SH results from a complex interplay between genetic, biological, psychiatric, psychosocial, social, cultural, and other factors, the development of interventions for SH in children and adolescents could benefit from being based on detailed investigation of these factors, including those that might reduce the risk of further SH, as well as having benefits for other outcomes. Ideally, the development of new treatments should ensure their feasibility and suitability for the young people for whom they are designed; children and adolescents with experience of SH and their carers should be involved in this process.

Additionally, trials of interventions for children and adolescents who engage in SH should include a range of outcome measures, not just SH and suicide, but also acceptability, adherence, and attitudes to treatment by young people, their caregivers, and service providers, as these may help to identify contributors to any apparent benefit or lack of impact. In particular, the inclusion of outcomes that matter to those who engage in SH is required to further inform intervention development (Owens 2020). It is also important that adverse effects of treatment, both short‐ and long‐term, are carefully evaluated, such as the release of confidential information described in the Hazell 2009 trial. Use of an agreed set of outcome measures would also assist in evaluation across trials. Investigation of the mechanisms through which treatments might work is also desirable to assist with the identification of clinically relevant subgroups of patients who may benefit from certain, more intensive, forms of intervention.

Heed should also be paid to the principles of development and evaluation of treatments as laid out in the UK Medical Research Council guidance regarding complex interventions. Additionally, from a service planning perspective, future trials should also include economic evaluations in order to determine which interventions may be most feasible to routinely implement throughout a health service (Bustamante‐Madsen 2018).

Summary of findings

Open in table viewer
Summary of findings 1. Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents

Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: Individual CBT‐based psychotherapy
Comparison: TAU or other comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or other comparator

Risk with Comparison 1: Individual CBT‐based psychotherapy

Repetition of SH by post‐intervention

Study population

OR 0.93
(0.12 to 7.24)

51
(2 RCTs)

⊕⊕⊝⊝
LOW 1 2

Our confidence in the effect estimate of individual CBT‐based psychotherapy on repetition of SH at post‐intervention is limited. The true effect may be substantially different from the estimate of the effect.

160 per 1,000

150 per 1,000
(22 to 580)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level as we rated any of the sources of risk of bias (as described in Assessment of risk of bias in included studies) at high risk for one of the studies included in the pooled estimate.

2 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Open in table viewer
Summary of findings 2. Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: DBT‐A
Comparison: TAU or another comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or another comparator

Risk with Comparison 2: DBT‐A

Repetition of SH at post‐intervention

Study population

OR 0.46
(0.26 to 0.82)

270
(4 RCTs)

⊕⊕⊕⊕
HIGH

We are very confident that the true effect lies close to that of the estimate of the effect estimate of DBT‐A on repetition of SH at post‐intervention.

432 per 1,000

259 per 1,000
(165 to 384)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Open in table viewer
Summary of findings 3. Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: MBT‐A
Comparison: TAU or another comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or another comparator

Risk with Comparison 3: MBT‐A

Repetition of SH at post‐intervention

Study population

OR 0.70
(0.06 to 8.46)

85
(2 RCTs)

⊕⊝⊝⊝
VERY LOW 1,2,3

The evidence is very uncertain about the effect of MBT‐A on repetition of self‐harm by post‐intervention.

805 per 1,000

743 per 1,000
(198 to 972)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level as we rated any of the sources of risk of bias (as described in Assessment of risk of bias in included studies) at high risk for one of the studies included in the pooled estimate.

2 We downgraded this domain by one level as one of the studies included in the pooled estimate used a proxy measure (i.e. cut‐scores on the Risk Taking and Self‐Harm Inventory) to ascertain repetition of SH. It is unclear how scores on this measure may relate to actual SH behaviour.

3 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Open in table viewer
Summary of findings 4. Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents

Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: Family therapy
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with placebo

Risk with Comparison 7: Family therapy

Repetition of SH at post‐intervention

Study population

OR 1.00
(0.49 to 2.07)

191
(2 RCTs)

⊕⊕⊕⊝
MODERATE 1

We are moderately confident in the effect estimate of family therapy on repetition of SH at post‐intervention. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

216 per 1,000

216 per 1,000
(119 to 364)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Background

Description of the condition

Self‐harm (SH), which includes all intentional acts of self‐poisoning (such as intentional drug overdoses) or self‐injury (such as self‐cutting), regardless of degree of suicidal intent or other types of motivation (Hawton 2003), has been a growing problem in children and adolescents up to 18 years of age (Hawton 2012b). Rates of SH in children and adolescents have been increasing over recent decades across a number of comparable countries, according to the number of presentations to general hospitals and primary care (Cairns 2019; Griffin 2018; Morgan 2017). This increase may be attributable to a number of factors, including: younger age of onset of SH behaviours (Gardner 2019; Griffin 2018; Jung 2018; Perera 2018), changes in the potential lethality of methods of SH used by children and adolescents (Griffin 2018), increased risk of SH repetition in children and adolescents, relative to young adults (i.e. 20 to 24‐year‐olds; Bennardi 2016), changes in clinical documentation and improved administrative coding of cases of SH in children and adolescents, resulting in a higher detection rate of young people who engage in SH (McGill 2018).

In contrast to suicide rates, rates of hospital‐presenting SH are higher in young females than males in most countries. The female to male ratio peaks at approximately five to six times in 12 to 14‐year‐olds, before decreasing with age (Diggins 2017; Griffin 2018; McMahon 2014). While reasons for the differential prevalence of SH behaviours in young females compared to young males are complex, an earlier age of onset of psychiatric disorders in young females may represent an important factor (Rhodes 2014). However, only about 1:28 young males, and 1:18 young females who SH ever present to hospital (Geulayov 2018). Therefore, it is apparent that SH in children and adolescents in the community (i.e. without hospital presentation) is very common, although less is known about the treatment needs of these youth (Hawton 2012b; Madge 2008; McMahon 2014).

For those who present to hospital, the most common method of SH is self‐poisoning. Overdoses of analgesics and psychotropics, especially paracetamol or acetaminophen, are common in some countries, particularly high‐income countries (Cairns 2019; Hawton 2012c; Sheen 2002). Self‐cutting is the next most frequent method used by those who present to hospital. However, in the community, self‐cutting and other forms of self‐injury are far more frequent than self‐poisoning (Geulayov 2018; Madge 2008; Müller 2016).

SH is associated with increased risk of future suicide. While suicide is relatively uncommon in younger children, rates have been increasing across a number of countries in recent years, particularly among young females (Bould 2019; Lahti 2011; Roh 2018; Stefanac 2019; Skinner 2012; Sullivan 2015). For example, recent data from the UK showed that children and adolescents who presented to hospital on at least one occasion following an episode of SH were 30 times more likely to die by suicide within a year (Hawton 2020a). A history of SH, particularly with frequent repetition, is the strongest risk factor for suicide across a range of psychiatric disorders (Zahl 2004).

SH and suicide in children and adolescents are the result of a complex interplay between genetic, biological, psychiatric, psychosocial, social, cultural, and other factors (Hawton 2012b). Psychiatric disorders, particularly mood disorders, are associated with the largest population attributable risk for SH in children and adolescents. While personality disorders should not be diagnosed in younger children, emergent traits consistent with those in adult borderline personality disorder have also been found to be associated with a significant population attributable risk for SH in this population (Witt 2019a), particularly those who engage in frequent repetition of SH (Crowell 2012). Alcohol and illicit drug misuse may also play an important role.

Both psychological and biological factors appear to further increase vulnerability to SH. Psychological factors may include poor emotion regulation abilities, or poor emotional intelligence may also contribute to the risk of SH in this population (Brausch 2019; Mikolajczak 2009). Psychological influences on children and adolescents who engage in SH include feelings of entrapment, lack of belonging, and perceiving oneself as a burden (O'Connor 2012). Other contributors include perfectionism, low self‐esteem, social isolation, impulsivity, hopelessness, and poor parent‐child attachment (Hawton 2012b). 

Relationship problems are common in children and adolescents who engage in SH, especially problems with family members (Fortune 2016). Relationship problems with partners are more common in older adolescents (i.e. 15 to 18‐year‐olds) than in younger children (i.e. 10 to 14‐year‐olds; Hawton 2012c). A history of emotional, physical, or sexual abuse has been associated with a significant population attributable risk of SH (Liu 2018; Madge 2011; Witt 2019a). Bullying and victimisation (Heerde 2019), including cyber‐bullying (Heerde 2019; John 2018), can also increase the risk of SH. Exposure to suicidal behaviour in others, either through personal contact, or through portrayals in traditional (e.g. films or television dramas) or new media (including social media), may also be an important factor as SH in children and adolescents often has a 'contagious' quality (Hawton 2020b; McMahon 2013). Biological factors include disturbances in the serotonergic and stress response systems (Van Heeringen 2014).

Description of the intervention

Treatment for SH in children and adolescents may involve psychosocial interventions, pharmacological interventions, or a combination of the two approaches.

Psychosocial interventions

Psychological approaches used to treat children and adolescents who engage in SH typically involve brief individual‐ or group‐based psychological therapy. Treatment may vary in initial management, location of treatment, continuity, intensity, and frequency of contact with therapists. There is also considerable variation among countries in the availability of services to provide such interventions. Consequently, there is no standard psychosocial treatment of SH in children and adolescents. However, in high‐income countries, treatment generally consists of a combination of assessment, support, involvement of parents, family, and caregivers, and individual psychological therapies.

Pharmacological interventions

Given the prevalence of psychiatric disorders in children and adolescents who engage in SH (Hawton 2013), pharmacological treatments may include antidepressants, antipsychotics, and mood stabilisers (including anticonvulsants and lithium). SH also arises in the context of anxiety and general distress and thus anxiolytics (including both benzodiazepines and non‐benzodiazepine anxiolytics) may be trialled. Other pharmacological agents may also be trialled. However, treatment with pharmacological agents is generally less common than treatment with psychosocial interventions in this population, partly due to concerns about the risk of exacerbating SH (Miller 2014).

Combined psychosocial and pharmacological interventions

Treatment may also involve a combination of both psychosocial and pharmacological approaches, such as cognitive behavioural therapy combined with fluoxetine (Gilbert 2020).

How the intervention might work

Psychosocial interventions

Mood disorders, in particular, have been identified as key modifiable risk factors for children and adolescents who engage in SH (Witt 2019a). Psychosocial interventions may address some of the underlying psychological risk factors associated with SH. The mechanisms of action of these interventions might help children and adolescents improve their coping skills and tackle specific problems, manage psychiatric disorders, improve self‐esteem, increase a sense of social connectedness, and reduce impulsivity and harmful reactions to distressing situations. What follows, is a description of the psychosocial interventions that are typically available for children and adolescents who engage in SH.

Cognitive behavioural therapy‐based psychotherapy

Cognitive behavioural therapy (CBT)‐based psychotherapy helps people to identify and critically evaluate the ways in which they interpret and evaluate disturbing emotional experiences and events, and aims to help them change the ways in which they deal with problems (Westbrook 2008). This is achieved in three steps: first, people are helped to change the ways in which they interpret and evaluate distressing emotions; second, they learn strategies to help them change the way in which they think about the meanings and consequences of these emotions; finally, with the benefit of modified interpretation of emotions and events, they are helped to change their behaviour and develop positive functional behaviour (Jones 2012).

Problem‐solving therapy (PST) is an integral part of CBT, although it can be delivered as a therapy in and of itself. PST assumes that ineffective and maladaptive coping behaviours that drive SH might be overcome by helping the person to learn skills to actively, constructively, and effectively solve the problems he or she faces in their daily lives (Nezu 2010). PST typically involves identification of the problem, generation of a range of solutions, implementation of chosen solutions based on appraisal, and the evaluation of these solutions (D'Zurilla 2010). Treatment goals include helping people to develop a positive problem‐solving orientation, use rational problem‐solving strategies, reduce the tendency to avoid problem‐solving, and reduce the use of impulsive problem‐solving strategies (Washburn 2012).

Dialectical behaviour therapy

In contrast to CBT and PST, which focus on changing behaviour and cognitive patterns, the focus of dialectical behavioural therapy (DBT) is to provide people with the skills to develop an awareness and acceptance of thoughts and emotions, including painful or distressing internal experiences, without judgement or attempts to alter, suppress, avoid, or otherwise change these experiences (Lynch 2006). The primary treatment goals of DBT are three‐fold: to reduce SH, reduce behaviours that interfere with the success of treatment, such as treatment non‐adherence, and reduce any other factors that may adversely affect the person's quality of life (e.g. frequency or duration of psychiatric hospitalisations) (Linehan 1993).

Miller 2007 adapted dialectical behaviour therapy for adolescents (DBT‐A) from Linehan’s initial conceptualisation of DBT, which was developed for adults diagnosed with borderline personality disorder. DBT‐A typically includes a combination of weekly individual and family therapy sessions, and telephone support as needed. As the aim of DBT‐A is to help children and adolescents adjust to maladaptive personality characteristics, the treatment is intensive and relatively prolonged, although usually less so than in adults (James 2008; Miller 2007).

Mentalisation‐based therapy

Mentalisation refers to the ability to understand the behaviour of both one's self and others in terms of motivational and emotional states (Allen 2008). Maladaptive and impulsive coping behaviours, including SH, are presumed to arise from a disrupted ability to engage in these processes. In mentalisation‐based therapy (MBT), the goal is to help people understand their emotions and behaviours, and develop strategies to regulate them to minimise the risk that they will engage in SH during times of distress (Rossouw 2018).

Mentalisation‐based therapy for adolescents (MBT‐A) is a relatively prolonged (one year) treatment which typically includes weekly individual sessions, and monthly family sessions (Fonagy 2019).

Group‐based psychotherapy

Group‐based psychotherapy treatment of children and adolescents who have self‐harmed integrates techniques from several therapies, including CBT, DBT‐A, MBT‐A, and specific group techniques. Group‐based psychotherapy may be more effective for children and adolescents than individual psychotherapy, as it provides them with a chance to work on skills related to developing interpersonal relationships and problem‐solving, which have been identified as important modifiable proximal risk factors for SH behaviours in this age group (Kaess 2020b).

Enhanced assessment approaches

Enhanced therapeutic assessment approaches combine standard psychosocial history and risk assessment techniques with brief cognitive analytic therapy and PST. Children and adolescents learn to identify sources of psychological pain and their connection to problem behaviours, such as SH, and identify ways to break this cycle (Ougrin 2012). The aim is to enhance adherence with subsequent treatment, and the potential benefit from it.

Compliance enhancement approaches

Of particular concern regarding after‐care of children and adolescents who present to hospital following an episode of SH, is the fact that adherence to recommended treatment tends to be relatively poor; between 25% and 50% of children and adolescents will not attend any follow‐up outpatient treatment sessions (Granboulan 2001; Taylor 1984). Efforts to maintain contact with children and adolescents, such as following up with them in the community, as well as efforts to address factors likely to impede attendance at treatment sessions, may be effective in improving treatment engagement and adherence in this population (Yuan 2019).

Family interventions

Family interventions typically involve conjoint therapy sessions with the child or adolescent and family members. It includes negotiation of goals, exploration of the episode of SH, communication between family members, problem‐solving, and discussion of developmental issues and their impact on the family. The basis of this therapy is that SH in children and adolescents may relate to family dysfunction, and therefore, efforts to improve family cohesion, attachment, adaptability, support, and parental warmth could help families function better and hence, reduce the risk of SH (Fortune 2016).

Remote contact interventions

Remote contact interventions, which may include letters, brief text messages delivered by telephone, telephone calls, and postcards, are low resource and non‐intrusive interventions that seek to maintain long‐term contact with children and adolescents. These interventions provide a sense of ongoing concern, and may mitigate the sense of social isolation reported by many children and adolescents who engage in SH. They may also help to improve their knowledge about triggers and warning signs for SH, provide them with information on alternative coping behaviours to SH, and where they can access help (Milner 2016).

These interventions may also be combined with emergency card interventions, which encourage children and adolescents to seek help when they feel distressed, and offer on‐demand emergency contact with psychiatric services or inpatient care. The aim is to reduce the risk of SH by facilitating rapid access to care.

Pharmacological interventions

Antidepressants

In relation to the prevention of SH and suicidal behaviour, the primary mechanism would be the effect of antidepressants on depression. However, there might also be other relevant specific effects, such as with drugs acting on the serotonin system, it having been suggested that serotonin levels are relevant to impulsivity, which is a feature sometimes associated with suicidal behaviour (Van Heeringen 2014). 

While different classifications of antidepressants have been suggested, a currently accepted classification is non‐selective monomamine inhibitors (e.g. amitriptyline, imipramine, dosulepin), selective serotonin reuptake inhibitors, subgrouped as non‐selective monomaine oxidase inhibitors (e.g., phenelzine) and monoamine oxidase A inhibitors (e.g., moclobemide), and other antidepressants (e.g., venlafaxine, mirtazapine, trazadone) (WHO 2014b).

An earlier approach was to group antidepressants as tricyclics, newer generation antidepressants (NGAs) (while recognising that many specific drugs in this category were introduced many years ago), and other antidepressants. This approach was used in the previous version of this review (Hawton 2015). For pragmatic reasons, we have therefore continued to use this categorisation in this update.

Antidepressants are often prescribed in the same dose range used to treat major depression. However, owing to the increased risk of overdose in this population, including the likelihood that children and adolescents who engage in self‐poisoning may use their own medication (Gjelsvik 2014), antidepressants associated with lower case fatality indices (e.g. SSRIs) are generally preferred (Hawton 2010), especially in people thought to be at risk of suicide.

In children and adolescents, there have been significant concerns that certain classes of antidepressants, particularly SSRIs, may increase suicidal ideation (Healy 2003). As a result, regulatory agencies in the UK (Medicines and Healthcare products Regulatory Agency; MHRA 2003), the US (US Food and Drug Administration; FDA 2004), and Europe (the European Medicines Agency; EMA 2005) have cautioned practitioners on the use of SSRIs in children and adolescents. More recently, review evidence suggests that risks may be elevated regardless of antidepressant class (Hetrick 2012). However, warnings from regulatory agencies may have had unintended consequences (Gibbons 2007; Lu 2014), although the evidence is mixed (Plöderl 2019; Whitely 2020). 

Antipsychotics

In people with a history of repeat SH, treatment with antipsychotics may be used to reduce heightened levels of arousal often experienced by them, especially in relation to stressful life events. By reducing this arousal, the urge to engage in SH may be reduced. Low potency second generation antipsychotics may reduce SH in children and adolescents diagnosed with major depression (Good 2006), and schizophrenia (Ma 2018). Lower doses may be prescribed to obtain this effect than is generally used in the treatment of psychotic disorders.

Anxiolytics, including both benzodiazepines and non‐benzodiazepine anxiolytics

Given that this population experiences a high prevalence of anxiety disorders (Hawton 2013), anxiolytics, including benzodiazepines and non‐benzodiazepine anxiolytics,  may be used to reduce suicidal behaviour through their specific effects on anxiety (Tyrer 2012). However, because of their GABAminergic effects, benzodiazepines may increase aggression and disinhibition (Albrecht 2014). In children and adolescents, current evidence from case series is that benzodiazepines are associated with an increased risk of suicidal ideation and SH (Kandemir 2008). Therefore, it is usually recommended that benzodiazepines are used very cautiously, if at all, in children and adolescents at risk of SH.

Mood stabilisers (including antiepileptics)

Mood stabilisers may have a role for children and adolescents diagnosed with bipolar disorder or unipolar depression, especially to prevent the recurrence of episodes of mood disorder (Cipriani 2013b). Therefore, these drugs may reduce the risk of SH. However, to date, this effect has only been found for lithium in adults (Cipriani 2013a). Lithium may reduce the risk of SH via a serotonin‐mediated reduction in impulsivity and aggression. It is also possible that the long‐term clinical monitoring, which all persons prescribed lithium treatment must undergo, might contribute to a reduction in SH (Cipriani 2013a).

Other pharmacological agents

Other pharmacological agents, particularly the N‐Methyl‐D‐aspartate receptor antagonist, ketamine, may also be trialed. Ketamine has been shown to have an antisuicidal effect, independent of its antidepressant effects (Sanacora 2017). As a result, the FDA has recently granted approval for the use of both ketamine and esketamine as adjunctive treatments to antidepressant therapy (FDA 2019). Ketamine has been associated with reduced suicidal ideation severity in the short term in adults with treatment‐resistant mood disorders (Wilkinson 2018; Witt 2020a). However, few trials have investigated the effect of ketamine over longer time periods. The effectiveness of ketamine on SH, and potential adverse effects of ketamine administration, such as dissociation, emergence psychosis, and rebound suicidal ideation, or behaviour, or both, remain under‐studied (Witt 2020a).

Natural products

In adults, there is some interest in the use of natural products, for example dietary supplementation of omega‐3 fatty acids (fish oils; Tanskanen 2001). Omega‐3 fatty acids have been implicated in the neural network, which is shown to correlate with the lethality of recent SH (Mann 2013). Blood plasma polyunsaturated fatty acid levels have also been implicated in the serotonin‐mediated link between low cholesterol and SH, suggesting that low omega‐3 fatty acid levels may have a negative impact on serotonin function (Sublette 2006). For those in whom SH is impulsive, omega‐3 supplementation may stimulate serotonin activity, thereby reducing the likelihood of engaging in SH (Brunner 2002).

Combined psychosocial and pharmacological interventions

A growing number of trials have investigated the effectiveness of combined psychosocial and pharmacological interventions, particularly in children and adolescents diagnosed with major depression. Given that achieving treatment response for psychosocial therapy alone may take up to four weeks or longer, combined approaches may provide a faster treatment response, and may have a superior effect to psychosocial intervention alone (Cox 2014). However, the effect of combined approaches on SH remains unclear (Cox 2014).

Why it is important to do this review

SH in children and adolescents is a major social and healthcare problem. It represents significant morbidity, is often repeated, and is linked with suicide. Many countries now have suicide prevention strategies, all of which include a focus on improved management of children and adolescents who engage in SH (WHO 2014a). SH is also associated with substantial healthcare costs (Kinchin 2017; Sinclair 2011).

In the UK, the National Collaborating Centre for Mental Health (NCCMH) produced the first guideline on the treatment of SH behaviours in 2004 (NCCMH 2004). This guideline focused on the short‐term physical and psychological management of SH. They updated this guidance in 2011, using interim data from a previous version of this review as the evidence‐base, and focused on the longer‐term psychological management of SH (NICE 2011). Subsequently, similar guidelines have been published by the Royal College of Psychiatrists (Royal College of Psychiatrists 2014), the Royal Australian and New Zealand College of Psychiatrists (Carter 2016), and German Professional Associations and Societies (Plener 2016), amongst others (Courtney 2019).

In 2021, the guidance contained in the 2011 NICE guidelines for the longer‐term management of SH will be due for updating. Therefore, we are updating our review (Hawton 2015), in order to provide contemporary evidence to guide clinical policy and practice.

Objectives

To assess the effects of psychosocial or pharmacological interventions for self‐harm (SH) compared to comparison types of care (e.g. treatment‐as‐usual, routine psychiatric care, enhanced usual care, active comparator, placebo, alternative pharmacological treatment, or a combination of these) for children and adolescents (up to 18 years of age) who engage in SH.

Methods

Criteria for considering studies for this review

Types of studies

We considered all randomised controlled trials (RCTs) of specific psychosocial or pharmacological treatments versus treatment‐as‐usual, routine psychiatric care, enhanced usual care, active comparator, placebo, alternative pharmacological treatment, or a combination of these, in the treatment of children and adolescents with a recent (within six months of trial entry) presentation for SH. All RCTs (including cluster‐RCTs and cross‐over trials) were eligible for inclusion regardless of publication type or language; however, we excluded quasi‐randomised trials.

Types of participants

While exact eligibility criteria often differ both within and between regions and countries (Witt 2020b), we included participants of both sexes and all ethnicities, who were up to 18 years of age, with a recent (i.e. within six months of trial entry) presentation to hospital or clinical services for SH.

We defined SH as all intentional acts of self‐poisoning (such as intentional drug overdoses) or self‐injury (such as self‐cutting), regardless of degree of suicidal intent or other types of motivation (Hawton 2003). This definition includes acts intended to result in death ('attempted suicide'), those without suicidal intent (e.g. to communicate distress, to temporarily reduce unpleasant feelings; sometimes termed 'non‐suicidal self‐injury'), and those with mixed motivation. We did not distinguish between attempted suicide and non‐suicidal self‐injury in this review, because there is a high level of co‐occurrence between them, particularly in children and adolescents (Andover 2012). Attempted suicide and non‐suicidal self‐injury cannot be distinguished in any reliable way, including on levels of suicidal intent (Klonsky 2011). Lastly, the motivations for SH are complex and can change, even within a single episode (De Beurs 2018).

We excluded trials in which participants were hospitalised for suicidal ideation only (i.e. without evidence of SH).

Types of interventions

Psychosocial interventions

These included:

  1. Individual CBT‐based psychotherapy;

  2. Dialectical behavioural therapy;

  3. Mentalisation therapy;

  4. Group‐based psychotherapy;

  5. Enhanced assessment approaches;

  6. Compliance enhancement approaches;

  7. Family interventions;

  8. Remote contact interventions.

Comparators

Treatment‐as‐usual (TAU) is likely to vary widely both between settings and between trials conducted over different time periods (Witt 2018). Following previous work, we defined TAU as routine clinical service provision that children and adolescents would receive had they not been included in the trial (i.e. routine care or 'standard disposition'; Hunt 2013). Other comparators could include no specific treatment or enhanced usual care, which refers to TAU that has in some way been supplemented, such as providing psychoeducation, assertive outreach, or more regular contact with case managers, and standard assessment approaches.

Pharmacological interventions

These included:

  1. Tricyclic antidepressants (TADs, e.g. amitriptyline);

  2. Newer generation antidepressants (NGAs), such as selective serotonin reuptake inhibitor (SSRIs, e.g. fluoxetine), serotonin and noradrenaline reuptake inhibitors (SNRIs, e.g. venlafaxine), norepinephrine reuptake inhibitors (NRIs, e.g. reboxetine), norepinephrine‐dopamine reuptake inhibitors (NDRIs, e.g. bupropion),  tetracyclic antidepressants (e.g. maprotiline), noradrenergic specific serotonergic antidepressants (NaSSAs, e.g. mirtazapine), serotonin antagonist or reuptake inhibitors (SARIs, e.g. trazodone), or reversible inhibitors of monoamine oxidase type A (RIMAs, e.g. moclobemide);

  3. Other antidepressants, such as irreversible monoamine oxidase inhibitors (MAOIs, e.g. phenelzine);

  4. Antipsychotics (e.g. quetiapine);

  5. Anxiolytics, including both benzodiazepines (e.g. diazepam), and non‐benzodiazepine anxiolytics (e.g. buspirone);

  6. Mood stabilisers, including antiepileptics (e.g. sodium valporate) and lithium;

  7. Other pharmacological agents (e.g. ketamine);

  8. Natural products (e.g. omega‐3 essential fatty acid supplementation).

Comparators

In pharmacological trials, where a comparison with the specific effects of a drug is being made, the comparator is typically placebo, which consists of any pharmacologically inactive treatment, such as sugar pills or injections with saline. We also included trials in which another pharmacological intervention (such as another standard pharmacological agent, reduced dose of the intervention agent, or active comparator) was used.

Types of outcome measures

For all outcomes, we were primarily interested in quantifying the effect of treatment assignment to the intervention at baseline, regardless of whether the intervention was received as intended (i.e. the intention‐to‐treat effect).

Primary outcomes

The primary outcome measure in this review was the occurrence of repeated SH over a maximum follow‐up period of two years. Repetition of SH was identified through self‐report, collateral report, clinical records, or research monitoring systems. As we wished to incorporate the maximum data from each trial, we included both self‐reported and hospital records of SH, where available. Preference was given to clinical records over self‐report where a study reported both measures. We also reported proportions of participants repeating SH, frequency of repeat episodes, and time to SH repetition (where available).

Secondary outcomes

Given increasing interest in the measurement of outcomes of importance to those who engage in SH (Owens 2020), we analysed data for the following secondary outcomes (where available) over a maximum follow‐up period of two years.

Treatment adherence

This was assessed using a range of measures of adherence, including: pill counts, changes in blood measures, and the proportion of participants that both started and completed treatment.

Depression

This was assessed as either continuous data, by scores on psychometric measures of depression symptoms, for example, total scores on the Beck Depression Inventory (BDI; Beck 1961), or scores on the depression subscale of the Hospital Anxiety and Depression Scale (HADS; Zigmond 1983), or as dichotomous data as the proportion of children and adolescents who met defined diagnostic criteria for depression.

Hopelessness

This was assessed as either continuous data, by scores on psychometric measures of hopelessness, for example, total scores on the Beck Hopelessness Scale (BHS; Beck 1974), or as dichotomous data as the proportion of children and adolescents reporting hopelessness.

General functioning

This was assessed as either continuous data, by scores on psychometric measures of general functioning, for example, total scores on the Global Assessment of Functioning (GAF; APA 2000), or as dichotomous data as the proportion of children and adolescents reporting improved general functioning.

Social functioning

This was assessed as either continuous data, by scores on psychometric measures of social functioning, for example, total scores on the Social Adjustment Scale (SAS; Weissman 1999), or as dichotomous data as the proportion of children and adolescents reporting improved social functioning.

Suicidal ideation

This was assessed as either continuous data, by scores on psychometric measures of suicidal ideation, for example, total scores on the Beck Scale for Suicidal Ideation (BSSI; Beck 1988), or as dichotomous data as the proportion of children and adolescents reaching a defined cut‐off for ideation.

Suicide

This included register‐recorded deaths, or reports from collateral informants, such as family members or neighbours.

Other

We remain open to including additional secondary outcomes, based on current outcome prioritisation work being undertaken by the author team.

Search methods for identification of studies

Electronic searches

An information specialist searched the following databases (to 4 July 2020), using relevant subject headings (controlled vocabularies) and search syntax as appropriate for each resource: Cochrane Common Mental Disorders Specialised Register (Appendix 1), Cochrane Library (Cochrane Central Register of Controlled Trials; CENTRAL), Cochrane Database of Systematic Reviews (CDSR), MEDLINE Ovid, Embase Ovid, and PsycINFO Ovid (Appendix 2).

A date restriction was applied as the search was to update an earlier version of this review (Hawton 2015). However, we did not apply any further restrictions on language or publication status to the searches.

We searched for retraction statements and errata once the included studies were selected.

We also searched the World Health Organization International Clinical Trials Registry Platform, and the US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov to identify ongoing trials.

The search was based on population only, participants who self‐harm (all ages). Records were screened to identify trials which were relevant to this review and two others (Witt 2021; Witt 2020c).

Searching other resources

Conference abstracts

In addition to conference abstracts retrieved via the main electronic search, we also screened the proceedings of recent (last five years) conferences organised by the largest scientific committees in the field:

  1. International Association for Suicide Prevention (both global congresses and regional conferences), and;

  2. Joint International Academy of Suicide Research and American Foundation for Suicide Prevention International Summits on Suicide Research.

Reference lists

We also checked the reference lists of all relevant RCTs, and the reference lists of major reviews that included a focus on interventions for SH in children and adolescents (Asarnow 2019; Berk 2016; Brent 2019; Busby 2020; Calear 2016; Cox 2017; Davasaambuu 2019; Devenish 2016; Flaherty 2018; Glenn 2019; Hawton 2015; Iyengar 2018; Joe 2018; Kothgassner 2020; Labelle 2015; Morken 2020; Robinson 2018; Yuan 2019).

Correspondence

We consulted the corresponding authors of trials, and other experts in the field to find out if they were aware of any ongoing or unpublished RCTs on the treatment of children and adolescents who engage in SH that were not identified by the electronic searches.

Data collection and analysis

Selection of studies

Review authors KW, KH, and one of either SH, GR, TTS, ET, or PH, independently assessed the titles of reports identified by the electronic search for eligibility. We distinguished between:

  1. eligible or potentially eligible trials for retrieval, in which any psychosocial or psychopharmacological treatment was compared with a comparator (e.g. treatment‐as‐usual, routine psychiatric care, enhanced usual care, active comparator, placebo, alternative pharmacological treatment, or a combination of these);

  2. ineligible general treatment trials, not for retrieval (i.e. where there was no control treatment).

All trials identified as potentially eligible for inclusion then underwent a second screening. Pairs of review authors, working independently from one another, screened the full text of eligible or potentially eligible trials to identify whether the trial met our inclusion criteria. We resolved disagreements in consultation with the senior review author (KH). Where disagreements could not be resolved from the information reported in the trial, or where it was unclear whether the trial satisfied our inclusion criteria, we contacted corresponding trial authors for additional clarification.

We identified and excluded duplicate records, and collated multiple reports of the same trial, so that each trial, rather than each report, represented the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram, and completed a 'Characteristics of excluded studies' table (Liberati 2009).

Data extraction and management

Review author KW and one of either SH or GR independently extracted data from the included trials, using a standardised extraction form. Where there were any disagreements, they were resolved in consensus discussions with KH.

Data extracted from each eligible trial included:

  1. Participant information: number randomised, number lost to follow‐up or withdrawn, number analysed, mean or median age, sex composition, diagnoses, diagnostic criteria, inclusion criteria, and exclusion criteria.

  2. Methods: trial design, total duration of the trial, details of any 'run in' period (if applicable), number of trial centres and their location, setting, and date.

  3. Intervention(s): details of the intervention, including dose, duration, route of administration, whether concomitant treatments were permitted and details of these treatments, and any excluded treatments.

  4. Comparators(s): details on the comparator, including dose, duration, route of administration, whether concomitant treatments were permitted and details of these treatments, and any excluded treatments.

  5. Outcomes: raw data for each eligible outcome (see Types of outcome measures), details of other outcomes specified and reported, and time points at which outcomes were reported.

  6. Notes: source of trial funding, and any notable conflicts of interest of trial authors.

We extracted both dichotomous and continuous outcomes data from eligible trials. As the use of non‐validated psychometric scales is associated with bias, we extracted continuous data only if the psychometric scale used to measure the outcome of interest had been previously published in a peer‐reviewed journal, and was not subjected to item, scoring, or other modification by the trial authors (Marshall 2000).

We planned the following main comparisons:

  1. Individual CBT‐based psychotherapy (e.g. CBT, PST) versus treatment as usual (TAU) or other comparator;

  2. Dialectical behaviour therapy (DBT) versus TAU or other comparator;

  3. Mentalisation‐based therapy versus TAU or other comparator;

  4. Group‐based psychotherapy versus TAU or other comparator;

  5. Enhanced assessment approaches versus TAU or other comparator;

  6. Compliance enhancement approaches versus TAU or other comparator;

  7. Family interventions versus TAU or other comparator;

  8. Remote contact interventions versus TAU or other comparator;

  9. Tricyclic antidepressants versus placebo or other comparator drug or dose;

  10. Newer generation antidepressants versus placebo or other comparator drug or dose;

  11. Any other antidepressants versus placebo or other comparator drug or dose;

  12. Antipsychotics versus placebo or other comparator drug or dose;

  13. Anxiolytics, including both benzodiazepines and non‐benzodiazepine anxiolytics, versus placebo or other comparator drug or dose;

  14. Mood stabilisers, including antiepileptics and lithium, versus placebo or other comparator drug or dose;

  15. Other pharmacological agents versus placebo or other comparator drug or dose;

  16. Natural products versus placebo or other comparator drug or dose.

Assessment of risk of bias in included studies

Highly biased studies are more likely to overestimate treatment effectiveness (Moher 1998). Review author KW and one of either SH or GR independently evaluated the risk of bias for the primary outcome (i.e., repetition of SH at post‐intervention) by using version 2 of the Cochrane Risk of Bias tool, RoB 2 (Sterne 2019). This tool encourages consideration of the following domains:

  1. Bias in the randomisation process.

  2. Deviations from the intended intervention (assignment to intervention).

  3. Missing outcome data.

  4. Bias in the measurement of the outcome.

  5. Bias in the selection of the reported result.

For cluster‐RCTs, we also evaluated the following:

  1. Bias arising from the timing of identification and recruitment of participants.

Signalling questions in the RoB 2 tool provided the basis for the tool’s domain‐level judgements about the risk of bias. Two review authors independently judged each source of potential bias as low risk, high risk, or some concerns. An overall 'Risk of bias' judgement was then made for each study by combining ratings across these domains. Specifically, if any of the above domains were rated at high risk, the overall 'Risk of bias' judgement was rated as high risk. We reported this overall judgement, which could be low risk, high risk, or some concerns, in the text of the review, and in the 'Risk of bias' tables.

Where inadequate details were provided in the original report, we contacted corresponding trial authors to provide clarification. We resolved disagreements through discussions with KH.

We entered and organised our RoB 2 assessments on an Excel spreadsheet (Microsoft Excel RoB2 Macro), and made them available as electronic supplements.

Measures of treatment effect

Dichotomous outcomes

We summarised dichotomous outcomes, such as the number of participants engaging in a repeat SH episode, or number of deaths by suicide, using the summary odds ratio (OR) and the accompanying 95% confidence interval (CI), as the OR is the most appropriate effect size statistic for summarising associations between two dichotomous groups (Fleiss 1994).

Continuous outcomes

For outcomes measured on a continuous scale, we used mean differences (MD) and accompanying 95% CI where the same outcome measure was used. Where different outcome measures were used, we used the standardised mean difference (SMD) and its accompanying 95% CI.

We aggregated trials in a meta‐analysis only where treatments were sufficiently similar. For trials that could not be included in a meta‐analysis, we provided narrative descriptions of the results.

Hierarchy of outcomes

Where a trial measured the same outcome (for example, depression) in two or more ways, we planned to use the most common measure across trials in any meta‐analysis. We also planned to report scores from other measures in a supplementary table.

Timing of outcome assessment

The primary end point for this review was post‐intervention (i.e. at the conclusion of the treatment period). We also reported outcomes for the following secondary end points (where data were available):

  1. Between zero and six months after the conclusion of the treatment period.

  2. Between six and 12 months after the conclusion of the treatment period.

  3. Between 12 and 24 months after the conclusion of the treatment period.

Where there was more than one outcome assessment within a time period, we used data from the last assessment in the time period, unless different outcomes were assessed at different points. For treatment adherence, we also planned to use within‐treatment period results.

Unit of analysis issues

Zelen design trials

Trials in this area are increasingly using Zelen’s method, in which consent is obtained subsequent to randomisation and treatment allocation (Witt 2020b). This design may lead to bias if, for example, participants allocated to one particular arm of the trial disproportionally refuse to provide consent for participation or, alternatively, if participants only provide consent if they are allowed to cross over to the other treatment arm (Torgerson 2004).

Although no trial included in this review used Zelen's design, should we identify a trial using Zelen's method in future updates of this review, we plan to extract data for all randomised participants as this is consistent with Zelen's original intention (Zelen 1979), and preserves randomisation. This will typically be possible for our primary outcome, repetition of SH, as this will generally be ascertained from clinical, hospital, and/or medical records. However, for certain self‐reported outcome measures, data may only be reported on the basis of those who consented to participation. We therefore also plan to conduct sensitivity analyses to investigate what impact, if any, the inclusion of these trials may have on the pooled estimate of treatment effectiveness.

Cluster‐randomised trials

Cluster randomisation, for example by clinician or general practice, can lead to overestimation of the significance of a treatment effect, resulting in an inflation of the nominal type I error rate, unless appropriate adjustment is made for the effects of clustering (Donner 2002; Kerry 1998).

One trial included in this review used cluster randomisation (Ougrin 2011). We had planned to follow the guidance outlined in Higgins 2019a. Specifically, where possible, we planned to analyse data using measures that statistically accounted for the cluster design. Where this is was not possible, we planned to analyse data using the effective sample size. However, the trial authors were unable to provide values for either the inter‐cluster correlation coefficient or the design effect, and further, there was no similar cluster RCT of this intervention approach from which these values could be approximated. We were therefore unable to statistically account for the effects of clustering for this trial.

In future updates of this review, should we be able to obtain information on either the inter‐cluster correlation coefficient or the design effect, we will follow the guidance outlined in Higgins 2019a.

Cross‐over trials

A primary concern with cross‐over trials is the carry‐over effect, in which the effect of the intervention treatment (e.g. pharmacological, physiological, or psychological) influences the participant’s response to the subsequent control condition (Elbourne 2002). As a consequence, on entry to the second phase of the trial, participants may differ systematically from their initial state, despite a wash‐out phase. In turn, this may result in a concomitant underestimation of the effectiveness of the treatment intervention (Curtin 2002a; Curtin 2002b). No trial included in this review used cross‐over methodology. However, should we identify any cross‐over trials in future updates of this review, we will only extract data from the first phase of the trial, prior to cross‐over, to protect against the carry‐over effect.

Studies with multiple treatment arms

No trial included multiple treatment arms. Should any trial include multiple treatment groups where the intervention arms are sufficiently similar, for example, where comparison is made between two interventions of the same type, we will combine dichotomous data. For outcomes reported on a continuous scale, we will combine data using the formula in Higgins 2011.

Where the interventions are not sufficiently similar, we will split the comparator arm data following the advice in Higgins 2011.

Studies with adjusted effect sizes

Where trials reported both unadjusted and adjusted effect sizes, we included only observed, unadjusted effect sizes.

Dealing with missing data

We did not impute missing data, as we considered that the bias that would be introduced by doing this would outweigh any benefit of increased statistical power that may have been gained by including imputed data. However, where authors omitted standard deviations (SD) for continuous measures, we contacted corresponding authors to request missing data. Where missing data could not be provided, we calculated missing SDs using other data from the trial, such as CIs, based on methods outlined in Higgins 2019b.

Assessment of heterogeneity

Between‐study heterogeneity can be assessed using either the Chi² or I² statistics. However, in this review, we used only the I² statistic to quantify inconsistency, as this is considered to be more reliable (Deeks 2019). The I² statistic indicates the percentage of between‐study variation due to chance, and can take any value from 0% to 100% (Deeks 2019).

We used the following values to denote relative importance of heterogeneity, as per Deeks 2019:

  1. unimportant: 0% to 40%;

  2. moderate: 30% to 60%;

  3. substantial: 50% to 90%;

  4. considerable: 75% to 100%.

We also took the magnitude and direction of effects and strength of evidence for heterogeneity into account (e.g. the CI for I²).

Where substantial levels of heterogeneity were found, we explored reasons for this heterogeneity (see Subgroup analysis and investigation of heterogeneity for details).

Assessment of reporting biases

Reporting bias occurs when the decision to publish a particular trial is influenced by the direction and significance of the results (Egger 1997). Research suggests, for example, that trials with statistically significant findings are more likely to be submitted for publication, and subsequently, be accepted for publication, leading to possible overestimation of the true treatment effect (Hopewell 2009).

To assess whether trials included in any meta‐analysis were affected by reporting bias, we planned to enter data into a funnel plot when a meta‐analysis included results of at least 10 trials. Should evidence of any small study effects be identified, we planned to explore reasons for funnel plot asymmetry, including the presence of possible publication bias (Egger 1997).

Data synthesis

For the purposes of this review, we calculated the pooled OR and accompanying 95% CI using the random‐effects model, as this is the most appropriate model for incorporating heterogeneity between studies (Deeks 2019). We used the Mantel‐Haenszel method for dichotomous data, and the inverse variance method for continuous data. We conducted all analyses in Review Manager 5.4 (Review Manager 2020).

Subgroup analysis and investigation of heterogeneity

Subgroup analyses

We planned to undertake the following subgroup analyses where there were sufficient data to do so:

  1. sex (males versus females);

  2. repeater status (first SH episode versus repeat SH episode).

Given the increasing use of enhanced usual care rather than TAU in trials in the field (Witt 2020b), we also planned to undertake subgroup analyses to determine whether comparator choice influenced the pattern of results observed.

Formal tests for subgroup differences were undertaken in Review Manager 5.4 (Review Manager 2020). However, it is only possible to undertake these subgroup analyses if randomisation was stratified by these factors, otherwise, there is the risk that doing so could lead to confounding. Randomisation was stratified by sex in one trial (Cottrell 2018). We therefore requested data for the primary outcome of interest, repetition of SH, disaggregated by sex from the authors for this trial. No included trial stratified randomisation by repeater status or comparator choice.

Investigation of heterogeneity

Several meta‐analyses were associated with substantial levels of heterogeneity (i.e. I² ≥ 75%). For these analyses, KW and KH firstly independently triple‐checked data to ensure these were correctly entered. Next, we investigated the source of this heterogeneity using a formal statistical approach as outlined in Viechtbauer 2020.

Sensitivity analysis

We planned to undertake the following sensitivity analyses, where appropriate, to test whether key methodological factors or decisions may have influenced the main result:

  1. Where a trial made use of Zelen's method of randomisation (see Unit of analysis issues);

  2. Where a trial contributed to substantial between‐study heterogeneity (see Subgroup analysis and investigation of heterogeneity).

No included trial made use of Zelen's method of randomisation. We were therefore unable to undertake sensitivity analyses to investigate what impact, if any, Zelen's design had on the pooled estimate of treatment effectiveness.

However, several meta‐analyses were associated with substantial levels of between‐study heterogeneity. We therefore reported results of these sensitivity analyses in the text, alongside discussion of the likely causes of these differences.

Summary of findings and assessment of the certainty of the evidence

For each comparison, we constructed a 'Summary of findings' table for our primary outcome measure, repetition of SH at post‐intervention, following the recommendations outlined in Schünemann 2019. These tables provide information concerning the overall quality of the evidence from all included trials that measured the outcome. We assessed the quality of evidence across the following domains:

  1. 'Risk of bias' assessment.

  2. Indirectness of evidence.

  3. Unexplained heterogeneity or inconsistency of results.

  4. Imprecision of effect estimates.

  5. Potential publication bias.

For each of these domains, we downgraded the evidence from high certainty by one level (for serious) or by two levels (for very serious). For risk of bias, we downgraded this domain by one level when we rated any of the sources of risk of bias (as described in Assessment of risk of bias in included studies) at high risk for any of the studies included in the pooled estimate, or by two levels when we rated multiple studies at high risk for any of these sources. For indirectness of evidence, we considered the extent to which trials included in any meta‐analysis used proxy measures to ascertain repetition of SH; we downgraded this domain by one level if one study used proxy measures, and by two levels if multiple studies used proxy measures. For unexplained heterogeneity or inconsistency of results, we downgraded this domain by one level where the I² value indicated substantial levels of heterogeneity, or by two levels where the I² value indicated considerable levels of heterogeneity. For imprecision, we downgraded this domain by one level where the 95% CI for the pooled effect included the null value. Finally, for the potential publication bias domain, we considered any evidence of funnel plot asymmetry (if available), as well as other evidence such as suspected selective availability of data, and downgraded by one or more levels where publication bias was suspected.

We then used these domains to rate the overall certainty of evidence for the primary outcome according to the following:

  1. High certainty: further research is very unlikely to change our confidence in the estimate of effect;

  2. Moderate certainty: further research is likely to have an important impact on our confidence in the estimate of effect, and may change the estimate;

  3. Low certainty: further research is very likely to have an important impact on our confidence in the estimate of effect, and may change the estimate;

  4. Very low certainty: we are very uncertain about the estimate.

We constructed 'Summary of findings' tables using GRADEpro GDT software (GRADEpro GDT 2015).

Reaching conclusions

We based our conclusions only on findings from the quantitative or narrative synthesis of the studies included in this review. Our recommendations for practice and research suggest priorities for future research, and outline the remaining uncertainties in the area.

Results

Description of studies

Results of the search

For this update, a total of 7186 records were found using the search strategy as outlined in Appendix 1 and Appendix 2. Five further records were identified following correspondence and discussion with researchers in the field. After deduplication, the initial number was reduced to 4678. Of these, 4454 were excluded following title/abstract screening, whilst a further 157 were excluded after reviewing the full texts (Figure 1). There were 23 new studies (representing nine independent trials) of interventions for SH in children and adolescents identified by this update.

Figure 1
Study Flow Diagram

Study Flow Diagram

Included studies

In the previous version of this review (Hawton 2015), 11 trials of interventions for SH in children and adolescents were included. The present update located six new trials of interventions for SH in children and adolescents. The present review therefore includes 17 non‐overlapping trials (Asarnow 2017; Cooney 2010; Cotgrove 1995; Cottrell 2018; Donaldson 2005; Green 2011; Griffiths 2019; Harrington 1998; Hazell 2009; McCauley 2018; Mehlum 2014; Ougrin 2011; Rossouw 2012; Santamarina‐Pérez 2020; Sinyor 2020; Spirito 2002; Wood 2001a).

All trials have been published. Unpublished data were obtained from the study authors for 13 of these trials (Asarnow 2017; Cooney 2010; Cotgrove 1995; Donaldson 2005; Green 2011; Griffiths 2019; McCauley 2018; Ougrin 2011; Rossouw 2012; Santamarina‐Pérez 2020; Sinyor 2020; Spirito 2002; Wood 2001a).

Design

Most trials (94.1%) were randomised at the individual level employing either simple randomisation (Cooney 2010; Cotgrove 1995; Donaldson 2005; Harrington 1998; Santamarina‐Pérez 2020; Sinyor 2020; Spirito 2002; Wood 2001a), or a restricted randomisation scheme, such as a blocking (Griffiths 2019; Hazell 2009; Mehlum 2014) or minimisation (Green 2011; McCauley 2018; Rossouw 2012) procedure. Two trials used stratification (Asarnow 2017; Cottrell 2018). In one trial, cluster randomisation was used (Ougrin 2011).

Setting

Of the 17 independent RCTs included in this review, eight were from the UK (Cotgrove 1995; Cottrell 2018; Green 2011; Griffiths 2019; Harrington 1998; Ougrin 2011; Rossouw 2012; Wood 2001a), four were from the USA (Asarnow 2017; Donaldson 2005; McCauley 2018; Spirito 2002), and one was from each of Australia (Hazell 2009), Canada (Sinyor 2020), New Zealand (Cooney 2010), Norway (Mehlum 2014), and Spain (Santamarina‐Pérez 2020).

In the majority of trials, participants were recruited following a clinical presentation for SH. In one trial (Asarnow 2017), a minority (< 5%) of participants were recruited from schools; however, all participants in this trial had a history of multiple episodes of SH resulting in presentation to clinical services within three months preceding trial entry.

For most trials, treatment was delivered in an outpatient setting or the participants' home environment. In the remaining two trials, one of compliance enhancement (Spirito 2002) and one of enhanced therapeutic assessment (Ougrin 2011), the intervention was delivered whilst the adolescent was receiving treatment in hospital and/or the emergency department.

Participants and participant characteristics

The included trials comprised a total of 2280 participants. All had engaged in at least one episode of SH prior to trial entry. A history of SH prior to the index episode (i.e. a history of multiple episodes of SH) was a requirement for participation in nine trials (Asarnow 2017; Cooney 2010; Cottrell 2018; Green 2011; Hazell 2009; McCauley 2018; Mehlum 2014; Santamarina‐Pérez 2020; Wood 2001a). In two further trials, around half the sample had a history of multiple episodes of SH (Donaldson 2005; Ougrin 2011). For the remaining six trials, information on the proportion of participants with a history of multiple episodes of SH prior to the index episode was not reported (Cotgrove 1995; Griffiths 2019; Harrington 1998; Rossouw 2012; Sinyor 2020; Spirito 2002).

Information on the methods of SH for the index episode was not reported in the majority of trials. In one trial, only those who had engaged in self‐poisoning were eligible to participate (Harrington 1998), whilst in three further trials, the majority of participants had engaged in self‐poisoning (Cotgrove 1995; Donaldson 2005; Spirito 2002). Full details on the methods used at the index episode is provided in Table 1. Whilst the predominance of participants engaging in self‐poisoning in the majority of these trials reflects the typical pattern observed in those who present to hospital, SH in the community more often involves self‐cutting and other forms of self‐injury (Geulayov 2018; Müller 2016).

Open in table viewer
Table 1. Methods used at the index episode of self‐harm

Reference

Method

Self‐poisoning

n (%)

Self‐injury

n (%)

Combined self‐poisoning
and self‐injury

n (%)

Asarnow 2017

‐

‐

‐

Cooney 2010

‐

‐

‐

Cotgrove 19951

101 (96.2)

2 (1.9)

‐

Cottrell 2018

184 (22.1)

594 (71.4)

54 (6.5)

Donaldson 20052

33 (84.6)

‐

‐

Green 2011

5 (2.7)

67 (36.6)

111 (60.7)

Griffiths 2019

‐

‐

‐

Harrington 1998

162 (100.0)3

‐

‐

Hazell 20094

‐

‐

‐

McCauley 2018

‐

‐

‐

Mehlum 2014

‐

‐

‐

Ougrin 2011

28 (40.0)

37 (52.8)

5 (7.2)

Rossouw 2012

‐

‐

‐

Santamarina‐Pérez 2020

‐

‐

‐

Sinyor 2020

‐

‐

‐

Spirito 20025

54 (85.7)

‐

‐

Wood 2001a6

‐

‐

‐

n: number; %: percentage.

1The method used by the remaining two (1.9%) participants was not reported.

2 The method used by the remaining six (15.4%) participants was not reported.

3Over half (n = 92; 56.8%) used paracetamol/acetaminophen.

4Participants engaged in multiple forms of SH: cutting (97%); head banging (71%); intentional drug overdose (57%); smothering (36%); strangling (25%); other self‐poisoning (19%); attempted drowning (19%); jumping from a height (17%); and other self‐harm (35%).

5The method used by the remaining nine (14.3%) participants was not reported.

6Data on the proportion with a lifetime history of self‐poisoning or self‐injury were reported; however, data on the proportion using these methods at the index episode were not clearly reported.

Whilst all trials included both male and female participants, the majority of participants was female (87.6%), reflecting the typical pattern for SH (Hawton 2008). Of the 15 trials that reported information on age, the weighted mean age of participants at trial entry was 14.7 years (SD 1.5 years). In the 15 trials that reported information on psychiatric diagnoses (Asarnow 2017; Cooney 2010; Donaldson 2005; Green 2011; Griffiths 2019; Harrington 1998; Hazell 2009; McCauley 2018; Mehlum 2014; Ougrin 2011; Rossouw 2012; Santamarina‐Pérez 2020; Sinyor 2020; Spirito 2002; Wood 2001a), participants were most commonly diagnosed with major depression (64.3%), followed by any anxiety disorder (54.4%), any mood disorder (49.2%), substance use disorder (33.9%), and bipolar disorder (25.4%). Around half (49.1%) were diagnosed with borderline personality disorder. Only two studies reported information on the proportion of participants without psychiatric diagnoses at trial entry (Ougrin 2011; Spirito 2002); in these two trials, just over one‐third (34.5%) were not diagnosed with a major psychiatric disorder.

Information on comorbid diagnoses was reported in two trials (Cooney 2010; Donaldson 2005). Around two‐thirds (67.6%) were diagnosed with comorbid psychiatric diagnoses; however, the nature of these co‐morbidities was not clearly reported in either trial.

Interventions

The trials included in this review investigated the effectiveness of various forms of psychosocial interventions:

  1. Individual CBT‐based psychotherapy (e.g. CBT, PST) versus treatment‐as‐usual (TAU) or other comparator (Donaldson 2005; Sinyor 2020).

  2. Dialectical behaviour therapy (DBT) versus TAU or other comparator (Cooney 2010; McCauley 2018; Mehlum 2014; Santamarina‐Pérez 2020).

  3. Mentalisation‐based therapy versus TAU or other comparator (Griffiths 2019; Rossouw 2012).

  4. Group‐based psychotherapy versus TAU or other comparator (Green 2011; Hazell 2009; Wood 2001a).

  5. Enhanced assessment approaches versus TAU or other comparator (Ougrin 2011).

  6. Compliance enhancement approaches versus TAU or other comparator (Spirito 2002).

  7. Family interventions versus TAU or other comparator (Asarnow 2017; Cottrell 2018; Harrington 1998).

  8. Remote contact interventions versus TAU or other comparator (Cotgrove 1995).

There were no eligible trials of pharmacological interventions for SH in children and adolescents.

Comparators

Of the 17 RCTs included in this review, the majority (64.7%) compared the intervention to TAU (Cooney 2010; Cotgrove 1995; Cottrell 2018; Green 2011; Griffiths 2019; Harrington 1998; Hazell 2009; Ougrin 2011; Rossouw 2012; Spirito 2002; Wood 2001a). The remaining trials compared the effectiveness of the intervention to enhanced usual care (EUC; Asarnow 2017; Mehlum 2014; Santamarina‐Pérez 2020), or to alternative forms of psychotherapy (Donaldson 2005; McCauley 2018; Sinyor 2020).

Outcomes
Primary outcome

All trials reported data on the primary outcome of this review, repetition of SH. In the majority of these trials this was based on self‐reported information (Asarnow 2017; Cooney 2010; Donaldson 2005; Harrington 1998; Hazell 2009; McCauley 2018; Mehlum 2014; Rossouw 2012; Sinyor 2020; Spirito 2002; Wood 2001a), self‐reported information supplemented with information from a collateral informant such as a parent (Green 2011), or self‐reported information supplemented by clinical records (Griffiths 2019; Santamarina‐Pérez 2020). For the remaining three trials, information on repetition of SH was obtained from clinical or hospital records supplemented with information from general practitioners, social workers, and psychologists, where relevant (Cotgrove 1995), or on re‐presentation to hospital (Cottrell 2018) or emergency departments (Ougrin 2011).

Secondary outcomes

Treatment adherence

Treatment adherence was assessed as the proportion of participants that completed the full course of treatment (Donaldson 2005; Harrington 1998; Ougrin 2011; Rossouw 2012; Hazell 2009; Sinyor 2020), or the total number of treatment sessions attended (Cooney 2010; Mehlum 2014; Spirito 2002).

Depression

Depression was assessed using the Mood and Feelings Questionnaire (MFQ; Angold 1995) in the majority of trials (Green 2011; Hazell 2009; Mehlum 2014; Rossouw 2012; Wood 2001a), followed by the Montgomery‐Åsberg Depression Rating Scale (MADRS; Montgomery 1979) (Mehlum 2014; Sinyor 2020), the BDI (Santamarina‐Pérez 2020; Sinyor 2020), the Children's Depression Rating Scale‐Revised (CDRS‐R; Poznanski 1985) (Cottrell 2018), the Center for Epidemiologic Studies Depression Scale (CES‐D; Radloff 1991) (Donaldson 2005), and the depression subscale of the Revised Child Anxiety and Depression Scale (RCADS; Chorpita 2000) (Griffiths 2019).

Hopelessness

Hopelessness was assessed using the Beck Hopelessness Scale (BHS) in two trials (Harrington 1998; Mehlum 2014), followed by the Hopelessness Scale for Children (HSC; Kazdin 1983) (Cottrell 2018), and by the future optimism subscale score on the Reasons for Living Inventory‐Adolescent (RFL‐A; Osman 1998), which was reverse coded in the present review to indicate a perceived lack of optimism about the future (Cooney 2010).

General functioning

General functioning was assessed using the Children's Global Assessment Scale (C‐GAS; Shaffer 1985) in three trials (Hazell 2009; Ougrin 2011; Santamarina‐Pérez 2020), and by the Health of the Nation Outcome Scales for Children and Adolescents (HoNOSCA; Gowers 1999) in one trial (Green 2011).

Social functioning

No included trial reported data on social functioning.

Suicidal ideation

Suicidal ideation was assessed using the Suicidal Ideation Questionnaire‐Junior (SIQ‐JR; Reynolds 1985; Reynolds 1988) in the majority of trials (Donaldson 2005; Green 2011; Hazell 2009; McCauley 2018; Mehlum 2014; Santamarina‐Pérez 2020; Wood 2001a), followed by the BSSI (Cooney 2010; Cottrell 2018; Harrington 1998; Sinyor 2020).

Suicide

Suicide was assessed using medical or health service records (Cottrell 2018; Green 2011; Hazell 2009), or via interviews with collateral informants, like parents (Donaldson 2005). In the majority of trials, however, it was unclear how suicide was assessed (Cooney 2010; Harrington 1998; McCauley 2018; Mehlum 2014; Ougrin 2011; Rossouw 2012; Spirito 2002; Wood 2001a).

Excluded studies

A total of 157 studies were excluded from this update. The most common reason for exclusion was that not all trial participants had engaged in SH within six month of trial entry (90 studies). Reasons for exclusion for the remaining studies are reported in Figure 1.

Details on the reasons for exclusion for the 28 trials related to interventions in children and adolescents identified by this update are reported in the Characteristics of excluded studies section.

Ongoing studies

Of the five ongoing trials identified in the previous version of this review (Hawton 2015), three were included in this update (Asarnow 2017; Cottrell 2018; McCauley 2018). Two were excluded: one was subsequently published as a case report, and one recruited participants from non‐clinical settings.

Thirteen ongoing studies were identified in this update (see Characteristics of ongoing studies section for further information).

Studies awaiting classification

There were no potentially eligible studies which have not been included in this review.

Risk of bias in included studies

Risk of bias was evaluated for the primary outcome repetition of SH at post‐intervention. The results of the 'Risk of bias' assessments can be seen in Figure 2. Full 'Risk of bias' assessments, including the evidence we used to justify our ratings, are available here: doi.org/10.6084/m9.figshare.14152364.

Figure 2
Results of 'Risk of bias' assessments for each study

Results of 'Risk of bias' assessments for each study

Bias arising from randomisation process

All trials used random allocation to assign participants to the intervention and comparator arms. We therefore rated the majority (76.5%) as having a low risk of bias for this domain. Most trials (94.1%) randomised at the individual level. In one trial, cluster randomisation was used (Ougrin 2011). Three trials were rated as having some concerns for this domain. For two older trials (Cotgrove 1995; Donaldson 2005), insufficient detail on allocation concealment was reported. For the third, baseline differences between the intervention and comparator arms suggested there may have been a problem with the randomisation process. Specifically, over half (55.0%) of those allocated to the intervention arm were diagnosed with probable borderline personality disorder, compared to 15.0% of those allocated to the comparator arm (Griffiths 2019). One trial was rated as having high risk of bias for this domain as those allocated to the intervention arm had very significantly higher hopelessness scores at baseline compared to those in the comparator arm (Spirito 2002), suggesting there may have been a problem with the randomisation process. Additionally, no information on allocation concealment was reported in this trial.

Bias due to deviations from intended interventions

Whilst participants and clinical personnel were, typically, not blind to allocation owing to likely differences in treatment intensity between the intervention and control arms, most trials (82.3%) were nonetheless rated as at low risk of bias for this domain as no deviations from the intended intervention were apparent and analyses were conducted on an intention‐to‐treat (ITT) basis, although the statistical method(s) used to undertake these analyses was not always clearly reported. Two trials were rated as having some concerns for this domain. For one of these (Cooney 2010), per protocol analyses were undertaken and, additionally, some minor departures from the intended intervention occurred as a result of the experimental context. In the second, insufficient information was reported on the analysis method(s) used (Spirito 2002). One trial was rated as having high risk of bias for this domain (Cotgrove 1995). In this trial, some participants randomised to the control group mistakenly received the intervention and it is unclear how these cases were assessed in subsequent analyses. Additionally, the trial authors claimed the intervention was effective in preventing repeat SH even though comparison of repetition rates did not show a difference between arms, suggesting that selective reporting may have been apparent.

Bias due to missing outcome data

The majority of trials (88.2%) were at low risk of bias for this domain as fewer than 5% of the data were missing at the post‐intervention assessment, or the proportion of missing data was balanced between the intervention and control arms at post‐intervention. However, there were some concerns with respect to this domain for two trials. For one of these, there was evidence of a slightly larger proportion of missing data for the comparator arm as compared to the intervention arm (Hazell 2009), whilst in the second, the proportion of missing data in the intervention arm was over double that of the proportion missing from the comparator arm (Spirito 2002). Neither of these trials undertook sensitivity analyses to investigate the impact that missing data may have had on the estimate of treatment effectiveness.

Bias in measurement of the outcome

There were some concerns regarding bias in the measurement of the outcome for around half (52.9%) of the trials included in this review. Typically, this was because repetition of SH was ascertained from self‐reported information alone and participants were either not blind to treatment allocation and/or participant blinding was unlikely to have been possible given the differences in therapeutic intensity between the intervention and control arms (Asarnow 2017; Cooney 2010; Green 2011; Harrington 1998; Hazell 2009; McCauley 2018; Mehlum 2014; Sinyor 2020; Spirito 2002). Two trials were rated as having high risk of bias for this domain (Rossouw 2012; Wood 2001a). In the first trial, repetition of SH was determined from cut‐scores on the RTSHI and it is unclear how this scale may relate to actual self‐harming behaviour. In the second, the definition of repetition of SH was based on there being two or more further episodes, whilst in the two remaining trials of this intervention approach (i.e. group‐based psychotherapy; Green 2011; Hazell 2009), repetition was based on there being any further episodes of SH.

Bias in selection of the reported result

Only two trials (11.8%) were rated as being at low risk of bias for this domain. Of these, only one trial clearly reported that data had been analysed in accordance with a prespecified analysis plan that had been finalised before unblinded outcome data had been made available for analysis (Cottrell 2018), whilst for the second, there had been no major departures from the analysis plan as outlined in either a published trial protocol or clinical trials register (Asarnow 2017).

Instead, the majority (70.6%) of the trials included in this review were rated as having some concerns for this domain. In the majority of cases, this was because trials were published prior to the International Committee of Medical Journal Editors' (ICMJE) requirement in 2015 that all trials be preregistered in a publicly available clinical trials registry. It was, therefore, difficult to determine whether data had been analysed according to a prespecified plan, although there were no apparent departures from the analyses outlined in the methods section of these trials (Cooney 2010; Cotgrove 1995; Donaldson 2005; Green 2011; Harrington 1998; Mehlum 2014; Rossouw 2012). For two trials published subsequent to the ICMJE requirement, this domain was also rated as having some concerns, as the information provided within the clinical trials record was not sufficiently detailed to determine whether there had been departures from the proposed analysis plan (Griffiths 2019; Santamarina‐Pérez 2020). For three further trials, there were some concerns for this domain as data on repetition of SH for one or more eligible time point(s) was not reported (Hazell 2009; McCauley 2018; Wood 2001a); however, in all three of these trials, it was unlikely that the results were selectively reported for favourability.

Three trials were rated as being at high risk of bias for this domain as, although repetition of SH was a prespecified outcome, data had to be requested from the trial authors (Ougrin 2011; Spirito 2002) and one trials had not been preregistered with a clinical trials register despite being published subsequent to 2015 (Sinyor 2020).

Overall bias

As a consequence, most trials (94.1%) were rated as either having some concerns (k = 10; 58.8%) or were at high risk of bias ( k = 6; 35.3%).

Effects of interventions

See: Summary of findings 1 Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents; Summary of findings 2 Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents; Summary of findings 3 Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents; Summary of findings 4 Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents

Comparison 1: Individual CBT‐based psychotherapy (e.g. CBT, PST) versus treatment‐as‐usual (TAU) or other comparator

The effectiveness of CBT‐based psychotherapy (i.e. up to 10 treatment sessions) versus alternative psychotherapy was assessed in two trials of children and adolescents (weighted mean age: 16.1 ±  2.7 years; 77.8% female) presenting to clinical services following an episode of SH. In the first trial, the comparator was supportive relationship therapy, which was designed to be as close as possible to TAU for this population (Donaldson 2005, N = 39). In the second, the comparator was minimally‐directive supportive psychotherapy (Sinyor 2020, N = 24).

Primary outcome
1.1 Repetition of SH

Data from two trials did not show that CBT‐based psychotherapy reduces repetition of SH by post‐intervention (i.e. conclusion of the acute phase) compared to alternative psychotherapy (OR 0.93, 95% CI 0.12 to 7.24; N = 51, k = 2; I2 = 29%; Analysis 1.1). The overall risk of bias was high for one trial (Sinyor 2020) and there were some concerns for the other trial (Donaldson 2005). According to GRADE criteria, we judged the evidence to be of low certainty. 

For the second of these trials (Sinyor 2020), whilst time to SH repetition was also recorded, correspondence with trial authors revealed that so few participants remained in the study until the conclusion of the booster phase that the data for this outcome were determined to be too unreliable.

Secondary outcomes
1.2 Treatment adherence

There was no evidence of an effect for CBT‐based psychotherapy on the proportion of participants who completed the acute treatment phase (Analysis 1.2).

One trial also reported information on the proportion of participants who completed both the acute and booster phases of treatment; however, there was no evidence of an effect for CBT‐based psychotherapy (1/12 versus 4/12; OR 0.18, 95% CI 0.02 to 1.95; N = 24; k = 1; I² = not applicable; Sinyor 2020).

There was no evidence of an effect on the number of sessions attended (Analysis 1.3). However, data on the number of treatment sessions attended were only available for those who completed the three‐ and six‐month follow‐up assessments in one trial (Donaldson 2005), whilst only data on the number of treatment sessions attended during the acute phase of treatment were reported for the second trial as so few participants attended any sessions during the booster phase in this trial (Sinyor 2020).

1.3 Depression

There was no evidence of an effect of CBT‐based psychotherapy on depression scores at post‐intervention (Analysis 1.4), or at 12 months in one of these trials (mean 10.33, SD 11.45, n = 15 versus mean 13.89, SD 8.28, n = 15; MD ‐3.56, 95% CI ‐10.71 to 3.59; N = 30; k = 1; I² = not applicable; Donaldson 2005).

For one of these trials (Sinyor 2020), depression was also measured using the MADRS. Using these values did not materially affect this result (mean 17.36, SD 13.12, n = 11 versus mean 22.30, SD 12.55, n = 10; MD ‐5.90, 95% CI ‐16.57 to 4.77; N = 21; k = 1; I² = not applicable).

1.4 Hopelessness

No data available.

1.5 General functioning

No data available.

1.6 Social functioning

No data available.

1.7 Suicidal ideation

There was also no evidence of an effect of CBT‐based psychotherapy on suicidal ideation scores at post‐intervention (Analysis 1.5), or at 12 months in one trial (mean 24.89, SD 28.52, n = 15 versus mean 33.33, SD 30.42, n = 15; MD ‐8.44, 95% CI ‐29.54 to 12.66; N = 30; k = 1; I² = not applicable; Donaldson 2005).

1.8 Suicide

No participants died by suicide in either of these trials, including by the 12‐month follow‐up assessment in one of them (Donaldson 2005).

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

Not applicable.

Comparison 2: Dialectical behaviour therapy (DBT) versus TAU or another comparator

Four trials evaluated the effectiveness of a DBT program specially adapted for adolescents (i.e. DBT‐A), comprising individual, group‐based, and family therapy sessions, in children and adolescents (weighted mean age: 15.2 ± 1.5 years; 90.4% female) with a history of multiple episodes of SH compared to either TAU (Cooney 2010, N = 29), EUC (Mehlum 2014, N = 77; Santamarina‐Pérez 2020, N = 35), or alternative psychotherapy (McCauley 2018, N = 173).

Primary outcome
2.1 Repetition of SH

Data from these four trials showed there was evidence of an effect of DBT‐A on repetition of SH at post‐intervention (OR 0.46, 95% CI 0.26 to 0.82; N = 270; k = 4; I² = 0%; Analysis 2.1). Although there were some concerns with regards to the overall risk of bias for all four trials, according to the GRADE criteria, we judged the evidence to be of high certainty.  There was no evidence of a difference by comparator (i.e. TAU versus EUC versus alternative psychotherapy). 

However, there was no longer evidence of an effect for DBT‐A, as compared to alternative psychotherapy, on repetition of SH by 12 months in one of these trials (27/71 versus 30/58; OR 0.57, 95% CI 0.28 to 1.16; N = 129; k = 1; I² = not applicable; McCauley 2018).

With respect to frequency of repeat SH episodes, there was no evidence of an effect of DBT‐A at the post‐intervention assessment (Analysis 2.2). Once again, there was no evidence of a difference based on comparator.

There was also no evidence of an effect for DBT‐A, as compared to alternative psychotherapy, on frequency of repeated SH by the 12‐month follow‐up assessment in one of these trials (mean 2.54, SD 11.92, n = 71 versus mean 4.53, SD 18.30, n = 58; MD ‐1.99, 95% CI ‐7.46 to 3.48; N = 129; k = 1; I² = not applicable; McCauley 2018).

Secondary outcomes
2.2 Treatment adherence

There was evidence of an effect for DBT‐A when compared with alternative psychotherapy on the proportion of children and adolescents who completed treatment in one trial (39/86 versus 14/87; OR 4.33, 95% CI 2.12 to 8.82; N = 173; k = 1; I² = not applicable; McCauley 2018).

There was no evidence of an effect for DBT‐A on the number of individual therapy sessions attended (Analysis 2.3); however, there was evidence of an effect by comparator (test for subgroup differences: Chi² = 36.7, df = 2, P < 0.001, I² = 94.5%). Specifically, when compared to either TAU or alternative psychotherapy, participants randomised to DBT‐A attended a greater number of individual therapy sessions.

There was no evidence of an effect for DBT‐A on the number of group therapy sessions attended (Analysis 2.4). There was no evidence of a difference based on comparator. There was also no evidence that children and adolescents randomised to DBT‐A attended a greater number of family therapy sessions (Analysis 2.5); however, there was evidence of a difference by comparator for this outcome (test for subgroup differences: Chi² = 5.4, df = 1, P = 0.02, I² = 81.5%).

Compared with TAU, participants randomised to DBT‐A attended a greater number of family therapy sessions. Finally, there was no evidence of an effect for DBT‐A on the number of telephone therapy sessions in two trials (Analysis 2.6), or on the number of medication review meetings attended in one trial (mean 2.40, SD 2.20, n = 14 versus mean 1.60, SD 2.90, n = 15; MD 0.80, 95% CI ‐1.07 to 2.67; N = 29; k = 1; Cooney 2010).

2.3 Depression

There was evidence of an effect for DBT‐A as compared to EUC on depression scores at post‐intervention in two trials (SMD ‐0.42, 95% CI ‐0.81 to ‐0.03; N = 103; k = 2; I² = 0%; Analysis 2.7). Data on depression was also measured as scores on the depression subscale of the MFQ in one of these trials (Mehlum 2014); however, there was no evidence of an effect for DBT‐A according to this measure (mean 10.2, SD 5.0, n = 39 versus mean 12.6, SD 6.6, n = 38; MD ‐2.39, 95% CI ‐5.02 to 0.24; N = 77; k = 1; I² = not applicable).

Data on depression scores, measured using both the MADRS and the depression subscale of the MFQ were also available for one of these trials by the 12‐month assessment (Mehlum 2014); however, there was no evidence of an effect for DBT‐A according to either measure by this time point in this trial (MADRS: mean 15.09, SD 8.08, n = 38 versus mean 15.73, SD 9.06, n = 37; MD ‐0.64, 95% CI ‐4.53 to 3.25; N = 75; k = 1; I² = not applicable; SMFQ: mean 9.88, SD 5.53, n = 38 versus mean 9.19, SD 6.57, n = 37; MD 0.69, 95% CI ‐2.06 to 3.44; N = 75; k = 1; I² = not applicable).

2.4 Hopelessness

There was evidence of an effect for DBT‐A on hopelessness scores at the post‐intervention assessment (SMD ‐0.62, 95% CI ‐1.07 to ‐0.16; N = 100; k = 2; I² =  13%; Analysis 2.8). There was no evidence of a difference by comparator.

However, there was no longer evidence of an effect for DBT‐A, as compared to EUC, on hopelessness scores by the 12‐month assessment in one of these trials (mean 6.97, SD 5.66, n = 38 versus mean 7.26, SD 6.57, n = 37; MD ‐0.29, 95% CI ‐3.07 to 2.49; N = 75; k = 1; I² = not applicable; Mehlum 2014).

2.5 General functioning

There was no evidence of an effect for DBT‐A, as compared to EUC, on general functioning scores at post‐intervention in two trials (Analysis 2.9). The means obtained by correspondence for Santamarina‐Pérez 2020 differ modestly from those published (i.e. 65.00 versus 64.60 for the DBT‐A arm and 54.29 versus 54.60 for the comparator arm).

There was also no evidence of an effect for DBT‐A on general functioning scores by the 12‐month follow‐up assessment in one of these trials (mean 65.68, SD 11.81, n = 38 versus mean 64.22, SD 14.13, n = 37; MD 1.46, 95% CI ‐4.44 to 7.36; N = 75; k = 1; I² = not applicable; Mehlum 2014).

2.6 Social functioning

No data available.

2.7 Suicidal ideation

There was evidence of an effect of DBT‐A on suicidal ideation at the post‐intervention assessment in four trials (SMD ‐0.43, 95% CI ‐0.68 to ‐0.18; N = 256; k = 4; I² = 0%; Analysis 2.10). There was no evidence of a difference based on comparator.

However, there was no longer evidence of an effect for DBT‐A (as compared to either EUC or alterative psychotherapy) on suicidal ideation scores by the 12‐month follow‐up assessment in two of these trials (Analysis 2.11). Once again, there was no evidence of a difference based on comparator.

2.8 Suicide

Data obtained by correspondence with trial authors indicated there were no suicides in either arm either at post‐intervention or by the 12‐month follow‐up assessment in Cooney 2010, Mehlum 2014, or Santamarina‐Pérez 2020. In McCauley 2018, one participant who had been assigned to the alternative comparator group died by suicide by the 12‐month follow‐up assessment.

Correspondence with trial authors for Mehlum 2014 further indicated there had been no suicide deaths in either arm by the 24‐month follow‐up assessment in this trial.

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

Four analyses within this comparison were associated with substantial levels of heterogeneity (Analysis 2.3, I² = 92%; Analysis 2.4, I² = 94%; Analysis 2.5, I² = 88%; Analysis 2.9, I² = 85%); however, analyses did not indicate any individual study was associated with excessive influence for any of these analyses.

Comparison 3: Mentalisation‐based therapy versus TAU or other comparator

Two trials investigated the effectiveness of mentalisation‐based therapy for adolescents (MBT‐A) in children and adolescents (weighted mean age: 15.3 ± 1.3 years; 82.8% female) presenting to clinical services following SH (Griffiths 2019, N = 48; Rossouw 2012; N = 80).

Primary outcome
3.1 Repetition of SH

Data from two trials did not show that MBT‐A reduced repetition of SH compared with TAU at post‐intervention (OR 0.70, 95% CI 0.06 to 8.46; N = 85; k = 2; I2 = 68%; Analysis 3.1). There was substantial heterogeneity when comparing MBT‐A and TAU for repetition of SH at post‐intervention (I2 = 68%), and the pooled estimate was imprecise. As a consequence, according to the GRADE criteria, we judged the evidence to be of very low certainty. A sensitivity analysis using data for the proportion of participants who scored above the cut‐point suggesting likely SH on the Risk‐Taking and Self‐Harm Inventory (RTSHI; Vrouva 2010) for both trials did not materially affect this result (Analysis 3.2).

There was also no evidence of an effect for MBT‐A on repetition of SH by the six‐month follow‐up assessment in one of these trials (2/2 versus 5/9; OR 4.09, 95% CI 0.15 to 108.94; N = 11; k = 1; I² = not applicable; Griffiths 2019).

Once again, a post‐hoc sensitivity analysis using data for the proportion of participants who scored above the cut‐point on the RTSHI did not materially affect this result (17/22 versus 26/26; OR 0.06, 95% CI 0.00 to 1.16; N = 48; k = 1; I² = not applicable; Griffiths 2019).

Secondary outcomes
3.2 Treatment adherence

There was no evidence of an effect for MBT‐A on treatment adherence, as measured by the number of adolescents who completed all 12 months of treatment in one of these trials (20/40 versus 17/40; OR 1.35, 95% CI 0.56 to 3.27; N = 80; k = 1; I² = not applicable; Rossouw 2012).

One trial reported data on treatment adherence for the intervention arm only (Griffiths 2019). In this trial, correspondence with trial authors indicated: "50% [of] young people attended 50% or more sessions. Additionally, there were six (27.3%) young people who did not attend any sessions; a further five (22.7%) who attended at least one but less than half of the sessions; four (18.2%) who attended between 50‐75% of sessions; and seven (31.8%) who attended 75% or more."

3.3 Depression

There was no evidence of an effect for MBT‐A on depression scores at the post‐intervention assessment in two trials (Analysis 3.3), or by the six‐month assessment in one of these trials (mean 20.1, SD 5.7, n = 22 versus mean 18.5, SD 7.0, n = 26; MD 1.60, 95% CI ‐1.99 to 5.19; N = 48; k = 1; I² = not applicable; Griffiths 2019).

Depression was also measured dichotomously, as the proportion of participants scoring above the cut‐point on the depression subscale of the MFQ in one trial (Rossouw 2012). Although fewer adolescents in the intervention arm scored above this cut‐point, there was no evidence of an effect for MBT‐A by the post‐intervention assessment (19/39 versus 25/37; OR 0.46, 95% CI 0.18 to 1.16; N = 76; k = 1; I² = not applicable).

3.4 Hopelessness

No data available.

3.5 General functioning

No data available.

3.6 Social functioning

No data available.

3.7 Suicidal ideation

No data available.

3.8 Suicide

Correspondence with trial authors indicated that no participant died by suicide by the 6‐month follow‐up assessment in one of these trials (Griffiths 2019), or by the 12‐month follow‐up assessment in the second of these trials (Rossouw 2012).

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

One analysis within this comparison was associated with substantial levels of heterogeneity (Analysis 3.3, I² = 97%); however, analyses did not indicate any individual study was associated with excessive influence.

Comparison 4: Group‐based psychotherapy versus TAU or other comparator

The effectiveness of group therapy was assessed in three trials in children and adolescents (weighted mean age: 14.4 ±  1.4 years; 87.4% female) presenting to clinical services following SH (Green 2011, N = 366; Hazell 2009, N = 68; Wood 2001a, N = 63).

Given that the Green 2011 and Hazell 2009 trials were based in a large part on Wood 2001a, employed the same treatment manual (Wood 2001b), and involved authors of the earlier trial in the design of the intervention, we grouped these trials within a single analysis.

In all three trials, group therapy involved a variety of techniques, including CBT, PST, DBT, and group psychodynamic psychotherapy. The intervention consisted of six weekly acute group sessions, followed by weekly or bi‐weekly group therapy sessions continuing until the adolescent felt ready to leave the service.

Primary outcome
4.1 Repetition of SH

There was no evidence of an effect for group‐based psychotherapy on repetition of SH by the six‐month (Analysis 4.1) or 12‐month (Analysis 4.2) assessments.

As no trial reported information on repetition of SH by the post‐intervention assessment (i.e. the primary outcome of this review) we were unable to determine the quality of evidence for this outcome according to the GRADE criteria.

Secondary outcomes
4.2 Treatment adherence

No data available.

4.3 Depression

There was no evidence of an effect of group‐based psychotherapy on depression scores at either the six‐month (Analysis 4.3) or 12‐month (Analysis 4.4) assessments.

4.4 Hopelessness

No data available.

4.5 General functioning

There was no evidence of an effect of group‐based psychotherapy on general functioning scores by either the six‐month (Analysis 4.5) or 12‐month (Analysis 4.6) assessments in two trials.

4.6 Social functioning

No data available.

4.7 Suicidal ideation

There was no evidence of an effect of group‐based psychotherapy on suicidal ideation scores at either the six‐month (Analysis 4.7) or 12‐month (Analysis 4.8) assessments.

4.8 Suicide

There were no suicide deaths in either arm in any of the three trials of group‐based psychotherapy by the final follow‐up assessment.

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

One analysis within this comparison was associated with substantial levels of heterogeneity (Analysis 4.2, I² = 77%); however, analyses did not indicate any individual study was associated with excessive influence.

Comparison 5: Enhanced assessment approaches versus TAU or other comparator

One cluster‐RCT examined the effectiveness of enhanced assessment for the treatment of SH in children and adolescents (mean age: 15.6 ± 1.3 years; 80.0% female) referred for a psychological assessment following SH (Ougrin 2011, N = 70).

As the trial authors were unable to provide values for either the inter‐cluster correlation coefficient or the design effect, and further, there was no similar cluster‐RCT of this intervention approach from which these values could be approximated, we were unable to statistically account for the effects of clustering. Results presented in this section may therefore overestimate the effectiveness of this intervention.

Primary outcome
5.1 Repetition of SH

There was no evidence of an effect for enhanced therapeutic assessment on repetition of SH by the 12‐month (4/35 versus 5/34; OR 0.75, 95% CI 0.18 to 3.06; N = 69; k = 1; I² = not applicable) or the 24‐month (7/35 versus 9/34; OR 0.69, 95% CI 0.23 to 2.14; N = 69; k = 1; I² = not applicable) assessments.

As this trial did not report information on repetition of SH by the post‐intervention assessment (i.e. the primary outcome of this review) we were unable to determine the quality of evidence for this outcome according to the GRADE criteria.

Secondary outcomes
5.2 Treatment adherence

There was an effect for enhanced therapeutic assessment on treatment adherence. Children and adolescents in the enhanced therapeutic assessment arm were more likely to attend their first aftercare appointment (29/35 versus 17/35; OR 5.12, 95% CI 1.70 to 15.39; N = 70; k = 1; I² = not applicable).

5.3 Depression

No data available.

5.4 Hopelessness

No data available.

5.5 General functioning

There was no evidence of an effect for enhanced therapeutic assessment on general functioning at post‐intervention (mean 64.6, SD 12.9, n = 35 versus mean 60.1, SD 9.9, n = 35; MD 4.50, 95% CI ‐0.89 to 9.89; N = 70; k = 1; I² = not applicable).

5.6 Social functioning

No data available.

5.7 Suicidal ideation

No data available.

5.8 Suicide

Correspondence with trial authors confirmed there were no suicide deaths in either arm by the final 24‐month follow‐up assessment.

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

Not applicable.

Comparison 6: Compliance enhancement approaches versus TAU or other comparator

One trial investigated the effectiveness of standard aftercare planning plus an added compliance enhancement approach (consisting of a series of four telephone calls at 1, 2, 4 and 8 weeks' post‐discharge) to standard aftercare planning alone in children and adolescents (mean age: not reported; 86.2% female) admitted to the emergency department of a general hospital following an episode of SH (Spirito 2002, N = 76).

Primary outcome
6.1 Repetition of SH

There was no evidence of an effect for compliance enhancement on repetition of SH by the six‐month follow‐up assessment (3/29 versus 5/34; OR 0.67, 95% CI 0.15 to 3.08; N = 63; k = 1; I² = not applicable). Participants in the compliance enhancement group did, however, “engage in fewer repeat SH episodes” over this period compared to participants in the control group (mean 0.10 versus 0.15; Spirito 2002), although insufficient information was available to allow formal testing of this.

As this trial did not report information on repetition of SH by the post‐intervention assessment (i.e. the primary outcome of this review), we were unable to determine the quality of evidence for this outcome according to the GRADE criteria.

Secondary outcomes
6.2 Treatment adherence

There was no evidence of an effect for compliance enhancement on the proportion of participants that completed the full course of treatment (17/29 versus 16/34; OR 1.59, 95% CI 0.59 to 4.33; N = 63; k = 1; I² = not applicable), or on the average number of therapy sessions attended (mean 7.70, SD 5.80, n = 29 versus mean 6.40, SD 4.40, n = 34; MD 1.30, 95% CI ‐1.28 to 3.88; N = 63; k = 1; I² = not applicable).

6.3 Depression

No data available.

6.4 Hopelessness

No data available.

6.5 General functioning

No data available.

6.6 Social functioning

No data available.

6.7 Suicidal ideation

No data available.

6.8 Suicide

There were no suicide deaths in either arm by the final follow‐up assessment.

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

Not applicable.

Comparison 7: Family interventions versus TAU or other comparator

Three trials compared the effectiveness of a family intervention to either TAU (Cottrell 2018, N = 832; Harrington 1998, N = 162) or EUC (Asarnow 2017, N = 42) in children and adolescents (weighted mean age: 14.3 ± 1.4 years; 88.7% female) presenting to clinical services following SH.

In one older trial, therapy was delivered in the adolescent's home (Harrington 1998) whilst in the remaining two trials, therapy was delivered in an outpatient setting (Asarnow 2017; Cottrell 2018).

Primary outcome
7.1 Repetition of SH

There was no evidence of an effect for family therapy interventions on repetition of SH at post‐intervention in two trials (OR 1.00, 95% CI 0.49 to 2.07; N = 191; k = 2; I2 = 0%; Analysis 5.1). There was no evidence of a difference based on comparator (i.e. TAU versus EUC) for this outcome. According to GRADE criteria, we judged the evidence to be of moderate certainty. 

There was also no evidence of an effect for family therapy on repetition of SH by the 18‐month follow‐up assessment in one further trial (118/415 versus 103/417; OR 1.21, 95% CI 0.89 to 1.65; N = 832; k = 1; I² = 0%; Cottrell 2018).

Two trials also reported data on time to SH repetition. There was no evidence of an effect for family therapy, as compared to EUC, on time to repetition by the post‐intervention assessment (Asarnow 2017). There was also no effect on time to SH repetition for family therapy as compared to TAU by the 18‐month assessment in one further trial (HR 1.14, 95% CI 0.87 to 1.49; P = 0.33; Cottrell 2018).

Secondary outcomes
7.2 Treatment adherence

Treatment adherence was measured as the proportion of children and adolescents who completed the full course of treatment in two trials (Cottrell 2018; Harrington 1998). There was evidence of an effect for family therapy on treatment adherence by six months (OR 1.99, 95% CI 1.55 to 2.57; N = 993; k = 2; I² = 0%; Analysis 5.2).

7.3 Depression

There was no evidence of an effect for family therapy on depression scores at either the 12‐month (mean 33.2, SD 22.9, n = 248 versus mean 33.9, SD 21.7, n = 189; MD ‐0.70, 95% CI ‐4.91 to 3.51; N = 437; k = 1; I² = not applicable), or 18‐month (mean 30.6, SD 21.9, n = 204 versus mean 31.6, SD 19.0, n = 165; MD ‐1.00, 95% CI ‐5.18 to 3.18; N = 369; k = 1; I² = not applicable) assessments in one trial (Cottrell 2018).

7.4 Hopelessness

There was no evidence of an effect for family therapy on hopelessness scores at either the six‐month (mean 4.40, SD 3.30, n = 74 versus mean 4.20, SD 3.60, n = 74; MD 0.20, 95% CI ‐0.91 to 1.32, N = 148; k = 1; I² = not applicable; Harrington 1998), 12‐month (mean 4.90, SD 4.14, n = 255 versus mean 5.20, SD 4.14, n = 201; MD ‐0.30, 95% CI ‐1.07 to 0.47; N = 456; k = 1; I² = not applicable; Cottrell 2018), or 18‐month (mean 4.60, SD 4.26, n = 213 versus mean 4.80, SD 4.00, n = 179; MD ‐0.20, 95% CI ‐1.02 to 0.62; N = 392; k = 1; I² = not applicable; Cottrell 2018).

7.5 General functioning

No data available.

7.6 Social functioning

No data available.

7.7 Suicidal ideation

There was no evidence of an effect for family therapy on suicidal ideation scores at the six‐month follow‐up assessment in one trial (mean 23.6, SD 40.0, n = 74 versus mean 28.7, SD 36.3, n = 75; MD ‐5.10, 95% CI ‐17.37 to 7.17; N = 149; k = 1; I² = not applicable; Harrington 1998).

In one further trial, information on the proportion of participants with clinically significant suicidal ideation, as determined from cut‐scores on the BSSI, was reported in a secondary publication (Cottrell 2018). There was no evidence of an effect for family therapy on the proportion of participants with suicidal ideation at the 12‐month (111/257 versus 98/202; OR 0.81, 95% CI 0.56 to 1.17; N = 459; k = 1; I² = not applicable) or 18‐month (85/212 versus 80/180; OR 0.84, 95% CI 0.56 to 1.25; N = 392; k = 1; I² = not applicable) assessments in this trial.

7.8 Suicide

There were no deaths, including by suicide, in either arm by the 18‐month follow‐up assessment in one trial (Cottrell 2018).

In the remaining trial, there was no evidence of a difference in effect for suicide by the six‐month follow‐up assessment. One patient in the family therapy arm died by suicide; none died by suicide in the control group (Harrington 1998). As the denominators for the intervention and comparator arms for this outcome are not known, however, we could not calculate ORs for this trial.

Subgroup analyses

Randomisation was stratified by sex in two of these trials; however, for one of these trials we were did not receive these data from the trial authors in time for publication of this review (Asarnow 2017).

For the second trial, data on repetition of SH disaggregated by sex were reported in a secondary publication (Cottrell 2018). There was no evidence of an effect for family therapy on repetition of SH by the 18‐month assessment in either males (11/47 versus 7/48; OR 1.79, 95% CI 0.63 to 5.10; N = 95; k = 1; I² = not applicable) or females (107/368 versus 96/369; OR 1.17, 95% CI 0.84 to 1.61; N = 737; k = 1; I² = not applicable) in this trial. There was also no evidence of a difference by sex for this outcome (test for subgroup differences: chi² = 0.59, df = 1, P = 0.44).

Sensitivity analyses

Not applicable.

Comparison 8: Remote contact interventions versus TAU or other comparator

One trial investigated the effectiveness of an emergency card enabling children and adolescents (mean age: 14.9 years, SD not reported; 84.8% female) who were admitted to hospital following an episode of SH to re‐admit themselves to the paediatric ward of the same hospital on demand if they felt suicidal over a 12‐month period in addition to TAU (Cotgrove 1995, N = 105).

Primary outcome
8.1 Repetition of SH

There was no evidence of an effect for emergency cards on repetition of SH by the 12‐month follow‐up assessment (3/47 versus 7/58; OR 0.50, 95% CI 0.12 to 2.04; N = 105; k = 1; I² = not applicable).

As this trial did not report information on repetition of SH by the post‐intervention assessment (i.e. the primary outcome of this review), we were unable to determine the quality of evidence for this outcome according to the GRADE criteria.

Secondary outcomes
8.2 Treatment adherence

No data available.

8.3 Depression

No data available.

8.4 Hopelessness

No data available.

8.5 General functioning

No data available.

8.6 Social functioning

No data available.

8.7 Suicidal ideation

No data available.

8.8 Suicide

No data available.

Subgroup analyses

No included trial stratified randomisation by sex or repeater status.

Sensitivity analyses

Not applicable.

Comparison 9: Tricyclic antidepressants versus placebo or other comparator drug or dose

There were no eligible trials in which tricyclic antidepressants were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 10: Newer generation antidepressants versus placebo or other comparator drug or dose

There were no eligible trials in which NGAs were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 11: Any other antidepressants versus placebo or other comparator drug or dose

There were no eligible trials in which any other antidepressants were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 12: Antipsychotics versus placebo or other comparator drug or dose

There were no eligible trials in which antipsychotics were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 13: Anxiolytics, including both benzodiazepines and non‐benzodiazepine anxiolytics, versus placebo or other comparator drug or dose

There were no eligible trials in which anxiolytics, including both benzodiazepines and non‐benzodiazepine anxiolytics, were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 14: Mood stabilisers, including antiepileptics and lithium, versus placebo or other comparator drug or dose

There were no eligible trials in which mood stabilisers, including antiepileptics and lithium, were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 15: Other pharmacological agents versus placebo or other comparator drug or dose

There were no eligible trials in which other pharmacological agents were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Comparison 16: Natural products versus placebo or other comparator drug or dose

There were no eligible trials in which natural products were compared either to placebo or to other comparator drug or dose for children and adolescents engaging in SH.

Discussion

This review included 17 trials, six of which have been completed since the previous version of this review (Hawton 2015).

Previously, we commented on the small number of trials in this population, especially given the size of the problem of SH and its increase in young people in many countries (Bould 2019; Lahti 2011; Roh 2018; Skinner 2012; Stefanac 2019; Sullivan 2015), as well as its association with suicide and other adverse outcomes (e.g. alcohol and other substance misuse, and psychiatric morbidity) in adulthood (Ohlis 2020). The reasons for this paucity of trials are unclear. One reason may be that conducting such trials in children and adolescents who are engaging in SH can be particularly challenging, especially in very young individuals for whom parental consent will be required. However, there is a clear need for more and better quality trials that can inform clinical practice.

Summary of main results

The trials included in this review investigated the effectiveness of various forms of psychosocial interventions. None of the included trials evaluated the effectiveness of pharmacological agents in this clinical population.

Individual CBT‐based psychotherapy

On the basis of data from two feasibility trials, there is probably little to no effect of individual CBT‐based psychotherapy (i.e. up to 10 acute sessions) compared with TAU on repetition of SH at post‐intervention.  There was also no evidence of effect for individual CBT‐based psychotherapy on any of the secondary outcomes.

Dialectical behaviour therapy (DBT)

On the basis of data from four trials, DBT for adolescents (DBT‐A) reduces repetition of SH at post‐intervention compared with TAU (Cooney 2010), EUC (Mehlum 2014; Santamarina‐Pérez 2020), and alternative psychotherapy (McCauley 2018).

Results for DBT‐A on frequency of repeated SH by this time point were less clear; although in contrast to our findings the trial authors for one of these trials analysed the longitudinal slope of decline in the mean number of SH episodes per participant based on information reported at the baseline, nine‐week, and 15‐week assessments, and reported an effect for DBT‐A (Mehlum 2014). It is difficult to rationalise these two sets of results. The results of our analyses would, however, suggest that any differences in frequency of SH repetition between the groups following treatment may not be marked.

There was some evidence that those allocated to DBT‐A attended a greater number of individual therapy sessions and were more likely to experience some improvement by the end of the treatment phase in some of the secondary outcomes included in this review, including depression, hopelessness, and suicidal ideation scores.

Mentalisation‐based therapy

We are uncertain whether MBT‐A reduced repetition of SH at the post‐intervention assessment as compared with TAU on the basis of data from two trials (Griffiths 2019; Rossouw 2012). However, this was based on fewer adolescents scoring above the cut‐point on the Risk Taking and Self‐Harm Inventory (RTSHI). It is unclear how this scale may relate to actual SH behaviour.

There was no apparent effect of MBT‐A on depression. However, in contrast to our findings, the authors of one trial reported that although "[t]he level of self‐rated depression decreased for participants in both groups...The linear rate of decrease was somewhat greater for the MBT‐A group (p<0.04) and the model yielded a significant difference at 12 months" (Rossouw 2012, p. 1308). These results suggest there might be some effect for MBT‐A over usual care in terms of this outcome, but this treatment approach requires evaluation in further trials before a stronger conclusion can be reached.

Group‐based psychotherapy

Three trials investigated group‐based therapy in adolescents with a history of multiple SH episodes (Green 2011; Hazell 2009; Wood 2001a). On the basis of data from these three trials, there is probably little to no effect of group‐based psychotherapy compared with TAU on repetition of SH. However, considerable heterogeneity was found, with the results of the earlier two studies showing effects in different directions (Hazell 2009; Wood 2001a), whilst the results of the third, much larger, trial indicated no superiority of group‐based psychotherapy compared with TAU (Green 2011). It is important to note, when interpreting this result, that the definition of repetition in Wood 2001a was based on there being two or more further episodes, whilst in the two remaining trials of this intervention approach (i.e. Green 2011; Hazell 2009), repetition was based on there being any further episodes of SH. 

Enhanced assessment approaches

Given the known poor treatment adherence of children and adolescents who engage in SH (Granboulan 2001; Taylor 1984), efforts have been made to increase adherence through therapeutic assessment following SH. Based on data from a single trial, there is probably little or no effect of an enhanced therapeutic assessment approach compared with TAU on repetition of SH at 12 or 24 months (Ougrin 2011). This approach may increase treatment adherence in terms of the number of participants who attended their first treatment session. However, these results were based on a single cluster‐RCT, which may have overestimated the effectiveness of this intervention. 

Compliance enhancement approaches

On the basis of data from a single trial, the evidence is uncertain as to whether compliance enhancement (consisting of standard discharged planning together with a series of four telephone calls at 1, 2, 4 and 8 weeks' post‐discharge) has any effect on repetition of SH when compared with TAU (Spirito 2002). However, this trial was likely underpowered to evaluate these outcomes. 

Family interventions

On the basis of data from three trials, there is probably little or no effect for either a home‐based (Harrington 1998) or clinic‐based (Asarnow 2017; Cottrell 2018) family intervention as compared with standard treatment on repetition of self‐harm either at post‐intervention or by the 18‐month follow‐up assessment. 

Remote contact interventions

On the basis of data from a single trial, there is probably little or no effect of an emergency card allowing patients to re‐admit themselves to hospital compared with TAU on repetition of SH (Cotgrove 1995). However, the trial was underpowered to evaluate this outcome. Few adolescents made use of the emergency card (10.6%); however, none of those who used the card engaged in repeated SH.

Overall completeness and applicability of evidence

Completeness of evidence

There have been relatively few trials of interventions for SH in children and adolescents, especially compared with the number of trials of psychosocial interventions for adults (Hawton 2016). Therefore, our conclusions are limited to a small range of interventions. Additionally, as there were no eligible trials of pharmacological interventions, perhaps due to concerns about safety in this clinical population, our findings are limited to psychosocial interventions.

Where it was unclear that the trials satisfied our inclusion criteria, we contacted corresponding trial authors for clarification. We also contacted corresponding authors where data were either not clearly reported, or where we required data reported in a different format to allow for their inclusion in a meta‐analysis. However, despite engaging in over 80 emails with corresponding authors we were not always able to obtain all relevant data. This was due to a combination of non‐response to our enquiries and to authors being unable to access relevant data, either due to moving on to later positions or as a result of working from home due to COVID‐19 pandemic quarantine orders. This is a common problem in meta‐analyses (Selph 2014).

Unfortunately, presence of publication bias could not be formally evaluated as no meta‐analysis in the present review included 10 or more trials. Therefore, we could not rule out the possibility that publication bias may have affected the studies within this review. This is a problem that commonly affects clinical data (Easterbrook 1991).

Whilst all of the included trials reported information on repetition of SH, publication bias may have been more common for the secondary outcomes assessed by this review. However, formal testing of publication bias was not possible due to the small number of trials.

Applicability of evidence

The majority of participants in these trials were female, reflecting the typical pattern for SH in hospital‐presenting populations (Hawton 2008). Only two trials stratified randomisation by sex (Asarnow 2017; Cottrell 2018). Given that there are some differences in the motivations for SH reported by males as compared to females (Claes 2007), further work is needed on the treatment needs and preferences of males who engage in SH, as well as their experiences of clinical services (Hassett 2017), and how these may differ from females who engage in SH.

The majority of trials included either children and adolescents who had all engaged in intentional drug overdoses or self‐poisoning, or samples where the majority had, again reflecting the typical pattern observed in patients who present to general hospitals following SH (Hawton 2007). However, there are other important patient subgroups, such as those who engage in self‐cutting, who may have different treatment needs (Hawton 2004). None of the trials included in this review specifically focused on these patients, although it should be noted that method switching is common in those who engage in repeat episodes of SH (Witt 2019b). Nine trials focused on those with a history of repeated SH, which is a particular issue in this clinical population given its association with subsequent SH repetition (Hawton 2012a) and suicide (Hawton 2020a). However, no trial investigated impacts of psychosocial interventions for those with an initial episode of SH versus those engaging in repeated SH. We were therefore unable to undertake subgroup analyses to investigate the impact of these interventions by repeater status.

This review focused exclusively on those who engaged in SH. As a result, we have excluded trials in which participants were diagnosed with conditions such as borderline personality disorder but where SH was not required for trial entry. We also excluded trials in which participants engaged in repetitive self‐injurious behaviour in the context of an intellectual disability or developmental disorder (e.g. an autism spectrum disorder). Readers interested in the use of psychosocial interventions for these patient groups are instead referred to the relevant reviews (Chezan 2017; Wong 2020).

Quality of the evidence

Certainty of evidence, as assessed using the GRADE approach, was generally moderate to very low suggesting that further research is likely to have an important impact on our confidence in the estimates of treatment effectiveness, and may in fact change the estimates. This is particularly likely to affect results for those interventions that so far have only been assessed in single trials.

Additionally, using the Cochrane 'Risk of bias' tool, version 2 (Sterne 2019), for almost all trials included in this review, there were some concerns of a high risk of bias in relation to at least one aspect of trial design, with weaknesses most commonly observed with respect to selection of the reported result and measurement of the outcome.

For most trials (70.6%), insufficient information was reported to determine whether data were analysed in accordance with a prespecified analysis plan. In 2015, the International Committee of Medical Journal Editors (ICMJE) recommended all clinical trials should be preregistered in a public trials registry (Witt 2020b). Whilst the majority (83.3%) of the six trials published subsequent to this recommendation were registered, in some cases insufficient detail was provided within the clinical trial register to determine how key outcome(s) were defined. This made it difficult to determine whether there had been any substantive changes to the proposed analysis plan and, if so, the reasons for any such departures. Future trials should provide sufficient detail within the clinical trial register to determine how key outcome(s) are defined and measured to aid in the determination as to whether there has been any substantive changes to the proposed analysis plan and, if so, the reasons for any such departures.

For around half of the trials (52.9%) included in this review, there were some concerns relating to bias in the measurement of repetition of SH. This was typically because repetition of SH was based on self‐reported information. Given that up to one‐fifth of SH episodes recorded in medical and clinical records are not reported by participants, prevalence estimates derived from self‐report alone may underestimate the true rate of SH (Mars 2016). By supplementing data on self‐reported SH with information from clinical or medical records, future trials could compare results based on self‐reported information with those obtained from objective sources to investigate what impact, if any this bias may have had on the estimate of treatment effectiveness.

Additionally, participants and clinical personnel were, typically, not blind to allocation owing to likely differences in treatment intensity between the intervention and control arms (Witt 2020b). Indeed, due to safety considerations, unbinding may be unavoidable in these types of trials. However, given that repetition of SH was based on self‐reported information in a number of the included trials (Asarnow 2017; Cooney 2010; Donaldson 2005; Harrington 1998; Hazell 2009; McCauley 2018; Mehlum 2014; Rossouw 2012; Sinyor 2020; Spirito 2002; Wood 2001a), this introduces potential bias. 

Lastly, the trials included in this review were, in general, relatively small to detect differences in proportions of patients who engage in a repeat episode of SH, although it is acknowledged that some of the trials were feasibility studies (e.g. Cooney 2010; Griffiths 2019; Sinyor 2020). Whilst sample sizes have increased over time, most trials in this field are still underpowered. We have previously calculated that trials in this field may need to recruit a minimum of 1862 participants per arm to detect an effect for repetition of SH with 80% power at the conventional alpha level (Witt 2020b). Future trials should therefore supply a priori power calculations to justify their sample size.

Potential biases in the review process

We are confident we have identified all relevant trials of psychosocial interventions for SH in children and adolescents. However, we cannot rule out the possibility that some relevant outcome data may be missing from this review. Although data on repetition of SH were available for all of the included trials, limited data were available on secondary outcomes. Nevertheless, by using the random‐effects model in all analyses, our results possess greater generalisability than if we had used the fixed‐effect model (Erez 1996).

Agreements and disagreements with other studies or reviews

This review is an update of the 2015 Cochrane Review on interventions for SH in children and adolescents (Hawton 2015). The previous review included 11 trials of eight different approaches, finding that whilst there had been relatively few trials on interventions for children and adolescents who have engaged in SH, certain approaches warranted further evaluation. The results in this update largely concur with the previous iteration. There appear to be limited positive findings regarding DBT‐A on repetition of SH by the end of the treatment phase. Results of this review would also suggest that a comprehensive therapeutic assessment may increase engagement with subsequent treatment.

We identified 18 further reviews that included a focus on interventions for SH in children and adolescents that have been completed since the previous version of this review was published. Four were systematic reviews (Calear 2016; Davasaambuu 2019; Devenish 2016; Iyengar 2018), four included meta‐analyses (Kothgassner 2020; Labelle 2015; Robinson 2018; Yuan 2019), and the remainder were narrative reviews (Asarnow 2019; Berk 2016; Brent 2019; Busby 2020; Cox 2017; Flaherty 2018; Glenn 2019; Joe 2018; Morken 2020).

However, whilst these reviews generally indicated that delivering some form of psychotherapy is likely to be more effective than nothing (e.g. Iyengar 2018; Kothgassner 2020; Robinson 2018; Yuan 2019), consistent with results found in adults (Hetrick 2016), these reviews have tended to statistically pool results from very different interventions together and so the results are largely meaningless for clinical practice as they provide little insight into which approach may be most beneficial for particular clinically relevant subgroups of patients.

Future reviews should investigate which component(s) of these typically multi‐component intervention approaches are most effective. Individual participant data meta‐analyses would also assist with the identification of clinically relevant subgroups of patients who may benefit from certain more intensive forms of intervention.

Study Flow Diagram

Figures and Tables -
Figure 1

Study Flow Diagram

Navigate to figure in ReviewOpen in new tab

Results of 'Risk of bias' assessments for each study

Figures and Tables -
Figure 2

Results of 'Risk of bias' assessments for each study

Navigate to figure in ReviewOpen in new tab

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 1: Repetition of SH by post‐intervention

Figures and Tables -
Analysis 1.1

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 1: Repetition of SH by post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 2: Treatment adherence: Proportion completing treatment

Figures and Tables -
Analysis 1.2

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 2: Treatment adherence: Proportion completing treatment

Navigate to figure in ReviewOpen in new tab

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 3: Treatment adherence: Number of treatment sessions attended

Figures and Tables -
Analysis 1.3

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 3: Treatment adherence: Number of treatment sessions attended

Navigate to figure in ReviewOpen in new tab

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 4: Depression scores at post‐intervention

Figures and Tables -
Analysis 1.4

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 4: Depression scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 5: Suicidal ideation scores at post‐intervention

Figures and Tables -
Analysis 1.5

Comparison 1: Individual CBT‐based psychotherapy versus TAU or other comparator, Outcome 5: Suicidal ideation scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 1: Repetition of SH at post‐intervention

Figures and Tables -
Analysis 2.1

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 1: Repetition of SH at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 2: Frequency of SH repetition at post‐intervention

Figures and Tables -
Analysis 2.2

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 2: Frequency of SH repetition at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 3: Treatment adherence: Number of individual therapy sessions attended

Figures and Tables -
Analysis 2.3

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 3: Treatment adherence: Number of individual therapy sessions attended

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 4: Treatment adherence: Number of group therapy sessions attended

Figures and Tables -
Analysis 2.4

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 4: Treatment adherence: Number of group therapy sessions attended

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 5: Treatment adherence: Number of family therapy sessions attended

Figures and Tables -
Analysis 2.5

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 5: Treatment adherence: Number of family therapy sessions attended

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 6: Treatment adherence: Number of telephone therapy sessions

Figures and Tables -
Analysis 2.6

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 6: Treatment adherence: Number of telephone therapy sessions

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 7: Depression scores at post‐intervention

Figures and Tables -
Analysis 2.7

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 7: Depression scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 8: Hopelessness scores at post‐intervention

Figures and Tables -
Analysis 2.8

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 8: Hopelessness scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 9: General functioning scores at post‐intervention

Figures and Tables -
Analysis 2.9

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 9: General functioning scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 10: Suicidal ideation scores at post‐intervention

Figures and Tables -
Analysis 2.10

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 10: Suicidal ideation scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 11: Suicidal ideation scores by 12‐months

Figures and Tables -
Analysis 2.11

Comparison 2: DBT‐A versus TAU or another comparator, Outcome 11: Suicidal ideation scores by 12‐months

Navigate to figure in ReviewOpen in new tab

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 1: Repetition of SH by post‐intervention

Figures and Tables -
Analysis 3.1

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 1: Repetition of SH by post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 2: Repetition of SH at post‐intervention (Risk‐Taking and Self‐Harm Inventory)

Figures and Tables -
Analysis 3.2

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 2: Repetition of SH at post‐intervention (Risk‐Taking and Self‐Harm Inventory)

Navigate to figure in ReviewOpen in new tab

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 3: Depression scores at post‐intervention

Figures and Tables -
Analysis 3.3

Comparison 3: MBT‐A versus TAU or another comparator, Outcome 3: Depression scores at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 1: Repetition of SH by six months

Figures and Tables -
Analysis 4.1

Comparison 4: Group‐based psychotherapy, Outcome 1: Repetition of SH by six months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 2: Repetition of SH by 12 months

Figures and Tables -
Analysis 4.2

Comparison 4: Group‐based psychotherapy, Outcome 2: Repetition of SH by 12 months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 3: Depression scores at six months

Figures and Tables -
Analysis 4.3

Comparison 4: Group‐based psychotherapy, Outcome 3: Depression scores at six months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 4: Depression scores at 12 months

Figures and Tables -
Analysis 4.4

Comparison 4: Group‐based psychotherapy, Outcome 4: Depression scores at 12 months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 5: General functioning scores at six months

Figures and Tables -
Analysis 4.5

Comparison 4: Group‐based psychotherapy, Outcome 5: General functioning scores at six months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 6: General functioning scores at 12 months

Figures and Tables -
Analysis 4.6

Comparison 4: Group‐based psychotherapy, Outcome 6: General functioning scores at 12 months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 7: Suicidal ideation scores at six months

Figures and Tables -
Analysis 4.7

Comparison 4: Group‐based psychotherapy, Outcome 7: Suicidal ideation scores at six months

Navigate to figure in ReviewOpen in new tab

Comparison 4: Group‐based psychotherapy, Outcome 8: Suicidal ideation scores at 12 months

Figures and Tables -
Analysis 4.8

Comparison 4: Group‐based psychotherapy, Outcome 8: Suicidal ideation scores at 12 months

Navigate to figure in ReviewOpen in new tab

Comparison 5: Family therapy, Outcome 1: Repetition of SH at post‐intervention

Figures and Tables -
Analysis 5.1

Comparison 5: Family therapy, Outcome 1: Repetition of SH at post‐intervention

Navigate to figure in ReviewOpen in new tab

Comparison 5: Family therapy, Outcome 2: Treatment adherence by six months

Figures and Tables -
Analysis 5.2

Comparison 5: Family therapy, Outcome 2: Treatment adherence by six months

Navigate to figure in ReviewOpen in new tab
Summary of findings 1. Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents

Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: Individual CBT‐based psychotherapy
Comparison: TAU or other comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or other comparator

Risk with Comparison 1: Individual CBT‐based psychotherapy

Repetition of SH by post‐intervention

Study population

OR 0.93
(0.12 to 7.24)

51
(2 RCTs)

⊕⊕⊝⊝
LOW 1 2

Our confidence in the effect estimate of individual CBT‐based psychotherapy on repetition of SH at post‐intervention is limited. The true effect may be substantially different from the estimate of the effect.

160 per 1,000

150 per 1,000
(22 to 580)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level as we rated any of the sources of risk of bias (as described in Assessment of risk of bias in included studies) at high risk for one of the studies included in the pooled estimate.

2 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Figures and Tables -
Summary of findings 1. Comparison 1: Individual CBT‐based psychotherapy compared to TAU or other comparator for self‐harm in children and adolescents
Navigate to table in Review
Summary of findings 2. Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: DBT‐A
Comparison: TAU or another comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or another comparator

Risk with Comparison 2: DBT‐A

Repetition of SH at post‐intervention

Study population

OR 0.46
(0.26 to 0.82)

270
(4 RCTs)

⊕⊕⊕⊕
HIGH

We are very confident that the true effect lies close to that of the estimate of the effect estimate of DBT‐A on repetition of SH at post‐intervention.

432 per 1,000

259 per 1,000
(165 to 384)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Figures and Tables -
Summary of findings 2. Comparison 2: DBT‐A compared to TAU or another comparator for self‐harm in children and adolescents
Navigate to table in Review
Summary of findings 3. Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: MBT‐A
Comparison: TAU or another comparator

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with TAU or another comparator

Risk with Comparison 3: MBT‐A

Repetition of SH at post‐intervention

Study population

OR 0.70
(0.06 to 8.46)

85
(2 RCTs)

⊕⊝⊝⊝
VERY LOW 1,2,3

The evidence is very uncertain about the effect of MBT‐A on repetition of self‐harm by post‐intervention.

805 per 1,000

743 per 1,000
(198 to 972)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level as we rated any of the sources of risk of bias (as described in Assessment of risk of bias in included studies) at high risk for one of the studies included in the pooled estimate.

2 We downgraded this domain by one level as one of the studies included in the pooled estimate used a proxy measure (i.e. cut‐scores on the Risk Taking and Self‐Harm Inventory) to ascertain repetition of SH. It is unclear how scores on this measure may relate to actual SH behaviour.

3 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Figures and Tables -
Summary of findings 3. Comparison 3: MBT‐A compared to TAU or another comparator for self‐harm in children and adolescents
Navigate to table in Review
Summary of findings 4. Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents

Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents

Patient or population: self‐harm in children and adolescents  (up to 18 years or age)
Intervention: Family therapy
Comparison: placebo

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with placebo

Risk with Comparison 7: Family therapy

Repetition of SH at post‐intervention

Study population

OR 1.00
(0.49 to 2.07)

191
(2 RCTs)

⊕⊕⊕⊝
MODERATE 1

We are moderately confident in the effect estimate of family therapy on repetition of SH at post‐intervention. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

216 per 1,000

216 per 1,000
(119 to 364)

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio; OR: Odds ratio;

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

1 We downgraded this domain by one level where the 95% CI for the pooled effect included the null value.

Figures and Tables -
Summary of findings 4. Comparison 7: Family therapy compared to placebo for self‐harm in children and adolescents
Navigate to table in Review
Table 1. Methods used at the index episode of self‐harm

Reference

Method

Self‐poisoning

n (%)

Self‐injury

n (%)

Combined self‐poisoning
and self‐injury

n (%)

Asarnow 2017

‐

‐

‐

Cooney 2010

‐

‐

‐

Cotgrove 19951

101 (96.2)

2 (1.9)

‐

Cottrell 2018

184 (22.1)

594 (71.4)

54 (6.5)

Donaldson 20052

33 (84.6)

‐

‐

Green 2011

5 (2.7)

67 (36.6)

111 (60.7)

Griffiths 2019

‐

‐

‐

Harrington 1998

162 (100.0)3

‐

‐

Hazell 20094

‐

‐

‐

McCauley 2018

‐

‐

‐

Mehlum 2014

‐

‐

‐

Ougrin 2011

28 (40.0)

37 (52.8)

5 (7.2)

Rossouw 2012

‐

‐

‐

Santamarina‐Pérez 2020

‐

‐

‐

Sinyor 2020

‐

‐

‐

Spirito 20025

54 (85.7)

‐

‐

Wood 2001a6

‐

‐

‐

n: number; %: percentage.

1The method used by the remaining two (1.9%) participants was not reported.

2 The method used by the remaining six (15.4%) participants was not reported.

3Over half (n = 92; 56.8%) used paracetamol/acetaminophen.

4Participants engaged in multiple forms of SH: cutting (97%); head banging (71%); intentional drug overdose (57%); smothering (36%); strangling (25%); other self‐poisoning (19%); attempted drowning (19%); jumping from a height (17%); and other self‐harm (35%).

5The method used by the remaining nine (14.3%) participants was not reported.

6Data on the proportion with a lifetime history of self‐poisoning or self‐injury were reported; however, data on the proportion using these methods at the index episode were not clearly reported.

Figures and Tables -
Table 1. Methods used at the index episode of self‐harm
Navigate to table in Review
Comparison 1. Individual CBT‐based psychotherapy versus TAU or other comparator

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1.1 Repetition of SH by post‐intervention Show forest plot

2

51

Odds Ratio (M‐H, Random, 95% CI)

0.93 [0.12, 7.24]

1.2 Treatment adherence: Proportion completing treatment Show forest plot

2

63

Odds Ratio (M‐H, Random, 95% CI)

0.57 [0.20, 1.63]

1.3 Treatment adherence: Number of treatment sessions attended Show forest plot

2

55

Mean Difference (IV, Random, 95% CI)

‐0.23 [‐2.07, 1.60]

1.4 Depression scores at post‐intervention Show forest plot

2

52

Std. Mean Difference (IV, Random, 95% CI)

‐0.39 [‐0.95, 0.16]

1.5 Suicidal ideation scores at post‐intervention Show forest plot

2

51

Std. Mean Difference (IV, Random, 95% CI)

‐0.14 [‐0.69, 0.41]
Figures and Tables -
Comparison 1. Individual CBT‐based psychotherapy versus TAU or other comparator
Navigate to table in Review
Comparison 2. DBT‐A versus TAU or another comparator

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

2.1 Repetition of SH at post‐intervention Show forest plot

4

270

Odds Ratio (M‐H, Random, 95% CI)

0.46 [0.26, 0.82]

2.1.1 Comparator: TAU

1

28

Odds Ratio (M‐H, Random, 95% CI)

2.55 [0.20, 31.86]

2.1.2 Comparator: Enhanced usual care

2

105

Odds Ratio (M‐H, Random, 95% CI)

0.29 [0.10, 0.85]

2.1.3 Comparator: Alternative psychotherapy

1

137

Odds Ratio (M‐H, Random, 95% CI)

0.50 [0.25, 0.98]

2.2 Frequency of SH repetition at post‐intervention Show forest plot

4

271

Mean Difference (IV, Random, 95% CI)

‐0.71 [‐1.55, 0.14]

2.2.1 Comparator: TAU

1

27

Mean Difference (IV, Random, 95% CI)

0.00 [‐0.92, 0.92]

2.2.2 Comparator: Enhanced usual care

2

107

Mean Difference (IV, Random, 95% CI)

‐1.02 [‐1.84, ‐0.20]

2.2.3 Comparator: alternative psychotherapy

1

137

Mean Difference (IV, Random, 95% CI)

‐2.99 [‐8.40, 2.42]

2.3 Treatment adherence: Number of individual therapy sessions attended Show forest plot

4

267

Mean Difference (IV, Random, 95% CI)

5.95 [‐0.18, 12.07]

2.3.1 Comparator: TAU

1

29

Mean Difference (IV, Random, 95% CI)

16.10 [12.16, 20.04]

2.3.2 Comparator: Enhanced usual care

2

112

Mean Difference (IV, Random, 95% CI)

1.82 [‐0.63, 4.27]

2.3.3 Comparator: Alternative psychotherapy

1

126

Mean Difference (IV, Random, 95% CI)

4.70 [1.71, 7.69]

2.4 Treatment adherence: Number of group therapy sessions attended Show forest plot

3

285

Mean Difference (IV, Random, 95% CI)

5.39 [‐0.20, 10.98]

2.4.1 Comparator: Enhanced usual care

2

112

Mean Difference (IV, Random, 95% CI)

6.12 [‐3.09, 15.33]

2.4.2 Comparator: Alternative psychotherapy

1

173

Mean Difference (IV, Random, 95% CI)

3.80 [1.73, 5.87]

2.5 Treatment adherence: Number of family therapy sessions attended Show forest plot

3

141

Mean Difference (IV, Random, 95% CI)

1.09 [‐3.85, 6.02]

2.5.1 Comparator: TAU

1

29

Mean Difference (IV, Random, 95% CI)

4.90 [2.57, 7.23]

2.5.2 Comparator: Enhanced usual care

2

112

Mean Difference (IV, Random, 95% CI)

‐1.02 [‐5.43, 3.39]

2.6 Treatment adherence: Number of telephone therapy sessions Show forest plot

2

112

Mean Difference (IV, Random, 95% CI)

0.16 [‐1.23, 1.55]

2.6.1 Comparator: Enhanced usual care

2

112

Mean Difference (IV, Random, 95% CI)

0.16 [‐1.23, 1.55]

2.7 Depression scores at post‐intervention Show forest plot

2

103

Std. Mean Difference (IV, Random, 95% CI)

‐0.42 [‐0.81, ‐0.03]

2.7.1 Comparator: EUC

2

103

Std. Mean Difference (IV, Random, 95% CI)

‐0.42 [‐0.81, ‐0.03]

2.8 Hopelessness scores at post‐intervention Show forest plot

2

100

Std. Mean Difference (IV, Random, 95% CI)

‐0.62 [‐1.07, ‐0.16]

2.8.1 Comparator: TAU

1

23

Std. Mean Difference (IV, Random, 95% CI)

‐0.24 [‐1.07, 0.59]

2.8.2 Comparator: Enhanced usual care

1

77

Std. Mean Difference (IV, Random, 95% CI)

‐0.76 [‐1.22, ‐0.30]

2.9 General functioning scores at post‐intervention Show forest plot

2

102

Mean Difference (IV, Random, 95% CI)

5.19 [‐5.31, 15.69]

2.9.2 Comparator: Enhanced usual care

2

102

Mean Difference (IV, Random, 95% CI)

5.19 [‐5.31, 15.69]

2.10 Suicidal ideation scores at post‐intervention Show forest plot

4

256

Std. Mean Difference (IV, Random, 95% CI)

‐0.43 [‐0.68, ‐0.18]

2.10.1 Comparator: TAU

1

23

Std. Mean Difference (IV, Random, 95% CI)

‐0.24 [‐1.07, 0.59]

2.10.2 Comparator: Enhanced usual care

2

108

Std. Mean Difference (IV, Random, 95% CI)

‐0.66 [‐1.05, ‐0.27]

2.10.3 Comparator: Alternative psychotherapy

1

125

Std. Mean Difference (IV, Random, 95% CI)

‐0.27 [‐0.62, 0.08]

2.11 Suicidal ideation scores by 12‐months Show forest plot

2

205

Mean Difference (IV, Random, 95% CI)

‐0.78 [‐6.91, 5.35]

2.11.1 Comparator: Enhanced usual care

1

75

Mean Difference (IV, Random, 95% CI)

‐1.60 [‐10.91, 7.71]

2.11.2 Comparator: Alternative psychotherapy

1

130

Mean Difference (IV, Random, 95% CI)

‐0.15 [‐8.29, 7.99]
Figures and Tables -
Comparison 2. DBT‐A versus TAU or another comparator
Navigate to table in Review
Comparison 3. MBT‐A versus TAU or another comparator

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

3.1 Repetition of SH by post‐intervention Show forest plot

2

85

Odds Ratio (M‐H, Random, 95% CI)

0.70 [0.06, 8.46]

3.2 Repetition of SH at post‐intervention (Risk‐Taking and Self‐Harm Inventory) Show forest plot

2

119

Odds Ratio (M‐H, Random, 95% CI)

0.48 [0.10, 2.25]

3.3 Depression scores at post‐intervention Show forest plot

2

128

Std. Mean Difference (IV, Random, 95% CI)

‐0.72 [‐2.86, 1.42]
Figures and Tables -
Comparison 3. MBT‐A versus TAU or another comparator
Navigate to table in Review
Comparison 4. Group‐based psychotherapy

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

4.1 Repetition of SH by six months Show forest plot

2

430

Odds Ratio (M‐H, Random, 95% CI)

1.72 [0.56, 5.24]

4.2 Repetition of SH by 12 months Show forest plot

3

490

Odds Ratio (M‐H, Random, 95% CI)

0.80 [0.22, 2.97]

4.3 Depression scores at six months Show forest plot

2

420

Mean Difference (IV, Random, 95% CI)

0.39 [‐2.76, 3.54]

4.4 Depression scores at 12 months Show forest plot

3

473

Mean Difference (IV, Random, 95% CI)

‐0.94 [‐4.04, 2.16]

4.5 General functioning scores at six months Show forest plot

2

402

Std. Mean Difference (IV, Random, 95% CI)

‐0.05 [‐0.25, 0.15]

4.6 General functioning scores at 12 months Show forest plot

2

396

Std. Mean Difference (IV, Random, 95% CI)

‐0.11 [‐0.30, 0.09]

4.7 Suicidal ideation scores at six months Show forest plot

2

421

Mean Difference (IV, Random, 95% CI)

1.26 [‐7.75, 10.27]

4.8 Suicidal ideation scores at 12 months Show forest plot

3

471

Mean Difference (IV, Random, 95% CI)

‐1.51 [‐9.62, 6.59]
Figures and Tables -
Comparison 4. Group‐based psychotherapy
Navigate to table in Review
Comparison 5. Family therapy

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

5.1 Repetition of SH at post‐intervention Show forest plot

2

191

Odds Ratio (M‐H, Random, 95% CI)

1.00 [0.49, 2.07]

5.1.1 Comparator: TAU

1

149

Odds Ratio (M‐H, Random, 95% CI)

1.02 [0.41, 2.51]

5.1.2 Comparator: Enhanced usual care

1

42

Odds Ratio (M‐H, Random, 95% CI)

0.98 [0.29, 3.31]

5.2 Treatment adherence by six months Show forest plot

2

993

Odds Ratio (M‐H, Random, 95% CI)

1.99 [1.55, 2.57]
Figures and Tables -
Comparison 5. Family therapy
Navigate to table in Review
Risk of bias for analysis 1.1 Repetition of SH by post‐intervention

Bias

Study

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Donaldson 2005

 Some concerns Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns

"[Participants were r]andomized to one of two treatment conditions" (p.114). Correspondence with study authors confirmed that a random numbers table was used to generate the allocation sequence. However, no specific information on allocation concealment was reported. Additionally, "...analyses yielded no differences between...groups in age...gender...race...or socioeconomic status... The two groups were comparable on rates of previous suicide attempts and psychiatric diagnoses" (p.116).

No information on participant blinding was reported. However, both treatments were so similar that it is possible participants were unaware of which treatment they were receiving. However, for clinical personnel "[t]he same seven therapists provided treatment in both...conditions” (p.114). However, there were no apparent deviations from the intended intervention. Additionally, "[f]ollow‐up data from all 31 families who completed follow‐ups (regardless
of number of treatment sessions attended) were included in data analyses consistent with an intent to‐treat model" (p.115).

Data were available for 100% of those randomised.

"Structured adolescent and parent follow‐up interviews...assessed incidents of further suicidal behaviors" (p.115). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Additionally, whilst no information on outcome assessor blinding was reported, information on SH was ascertained from participant self‐report.

No information is available on whether the data that produced this result was analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. "Outcome measures were administered 3 months (end of active treatment) and 6 months (end of maintenance)..." (p.115). All eligible outcome measurements within the outcome domain were reported. Additionally, "...interviews...assessed incidents of further suicidal behaviors" (p.115). All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as no specific information on allocation concealment was reported.

Sinyor 2020

 Low risk of bias Low risk of bias Low risk of bias Some concerns High risk of bias High risk of bias

"A random number generator was used to create the treatment allocation sequence...A member of the principal investigator's lab who was otherwise not involved in the study generated the list and created individual electronic files with allocation by subject number, protected by unique passwords that were only sent to the PI following each subject's inclusion..." (p.688). Additionally, "[t]here was no difference in mean age, sex, baseline SSI, MADRS, CGI‐S or BDI scores [at baseline] between groups" (p.689).

"This study is a single‐site, single‐blind...RCT" (p.687). As this was a single, (assessor) blinded trial, participants and clinical personnel were not blinded to treatment allocation. However, there were no apparent deviations from the intended intervention. Additionally, "[t]his study used an intent‐to‐treat approach including all 24 subjects randomized into the trial each of whom had at least a baseline visit" (p.689).

Data were available for 100% of those randomised.

"...repeated SH (i.e. any episode of self‐injury regardless of intent occurring between research visits) as well as suicide attempts (i.e. any episode of self‐injury with at least some intent to die between research visits) derived from the Columbia Suicide Severity Rating Scale..." (p.688). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Additionally, whilst "a blinded research coordinator...obtained study measures" (p.688), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

No information is available as to whether the data that produced this result was analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. However,  correspondence with trial authors revealed data on all outcomes were collected at the conclusion of the acute treatment phase (10 weeks), and at 12 months (i.e., following the conclusion of the booster phase), although data on repetition of self‐harm for both these time points had to be requested from trial authors. Additionally,  data for all eligible analyses of the data were obtained following correspondence.

Overall, this trial was rated as at high risk of bias as, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation. Additionally, insufficient information was reported in the trial protocol to determine whether the data that produced this result analysed in accordance with a pre‐specified analysis plan that was finalized before unblinded outcome data were available for analysis.

Figures and Tables -
Risk of bias for analysis 1.1 Repetition of SH by post‐intervention
Navigate to table in Review
Risk of bias for analysis 2.1 Repetition of SH at post‐intervention

Bias

Study

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Subgroup 2.1.1 Comparator: TAU

Cooney 2010

 Low risk of bias Some concerns Low risk of bias Some concerns Some concerns Some concerns

"Participants were randomly assigned…via a computerised randomisation
procedure…In this procedure, the treatment assignment series was generated by the fifth author...in conjunction with a research assistant who was otherwise entirely independent from the study. Each treatment assignment was placed in a sealed numbered envelope...by the assistant, and once the participant had completed all pre‐treatment screening and
research assessments, he or she was given the next envelope in the series and opened it to discover what treatment he or she was to receive” (p.10). Additionally, there were no significant differences between the intervention and comparator arms for any of the demographic characteristics and clinical variables measured (Table 1, p.17).

As this was a single (assessor) blinded trial, participants and clinical personnel were not blinded to treatment allocation. Additionally, "[i]Initially, owing to funding restrictions, we had planned to limit the number of participants in group sessions to two [persons] per family (i.e. the young person and one parent or caregiver). Approximately 12 weeks into group sessions, we extended this to three family members per group session..." (pp.11‐13).  However, It is unlikely that the addition of a third family member would have significantly affected the outcome. Additionally, given that both the intervention and comparator arms comprised family therapy, it is likely these deviations could have been balanced between groups. However, although data for all 29 participants randomised to the intervention or comparator arms is reported at the mid‐treatment assessment, at longer follow‐up periods only data on those available for assessment were presented.

Data were available for 93.3% of those randomised.  However, analysis method did not correct for bias, nor were sensitivity analyses undertaken to investigate the potential effect of missing data. Despite this, data were available for 92.8% of those randomised to the intervention arm, and 93.7% of those randomised to the comparator arm.

"The primary outcome variable was the occurrence and frequency of self‐harm (i.e. suicide attempts and non‐suicidal self‐injuries)... To measure this, we used the Suicide Attempt‐Self‐Injury Interview...which we adapted for use with adolescents" (p.12). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. However, whilst "[a]ll subsequent assessments were administered by an assessor...who was also blind to treatment condition" (p.10), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

No information is available as to whether the data that produced this result were analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. However, "[o]utcome assessments were conducted before treatment assignment, and scheduled to occur at approximately 3 [i.e., mid treatment] and 6‐month [i.e., post‐intervention] intervals... (p.10), all eligible outcome measurements within the outcome domain were reported. "The primary outcome variable was the occurrence and frequency of self‐harm (i.e. suicide attempts and non‐suicidal self‐injuries)... To measure this, we used the Suicide Attempt‐Self‐Injury Interview...which we adapted for use with adolescents" (p.12), all eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation. Additionally, there were some minor deviations from the intended intervention that may have arisen as a result of the trial context.

Subgroup 2.1.2 Comparator: Enhanced usual care

Santamarina‐Pérez 2020

 Low risk of bias Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns

"...participants were then randomly assigned...using a computer‐generated list of random numbers and following a simple randomization procedure in a 1:1 ratio" (p.7). Additionally, "an independent and off‐site statistician generated the random allocation sequence" (p.7). There were no significant differences between the intervention and comparator arms for most of the demographic characteristics and clinical variables measured (Table 1, p.6). However, those assigned to the intervention arm had lower depression scores (MD 7.10, 95% CI 1.46 to 12.74, p=0.019). This result may have been compatible with chance.

"Given the characteristics of the treatment arms, therapists and patients were not blind to the type of treatment" (p.7). Additionally, there were no apparent deviations from the intended intervention. "An intention‐to‐treat analysis was performed and involved all patients who were randomly assigned" (p.7).

Data were available for 85.7% of those randomised. However, Analysis method did not correct for bias, nor were sensitivity analyses undertaken to investigate the potential effect of missing data. Despite this, data were available for 83.3% of those randomised to the intervention arm, and 88.2% of those randomised to the comparator arm.

"Frequency of NSSIs and the number of SAs... were collected based on the information reported during individual sessions and using medical records. These variables were collected by the patient’s clinician  using a register..." (p.5). Repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Additionally, "[c]linical assessments of the participants were performed by an independent psychologist...who was blind to treatment assignment at baseline and at the end of the treatment after 16 weeks" (p.3).

Referring back to the trial protocol (NCT02406625), no specific information was reported. However, "...variables were collected..at week 4, week 8, week 12, and week 16 of treatment" (p.5). All eligible outcome measurements within the outcome domain were reported. Additionally, "[f]requency of NSSIs and the number of SAs...were collected by the patient’s clinician using a register..." (p.5). All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as insufficient information was reported in the trial protocol to determine whether the data that produced this result analysed in accordance with a pre‐specified analysis plan that was finalized before unblinded outcome data were available for analysis.

Mehlum 2014

 Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns Some concerns

"Treatment allocation of participants...was based on a permuted block randomisation procedure with an undisclosed and variable blocking factor...management of the randomisation procedures was performed by an external group..." (p.1083). Additionally, there were "[n]o significant differences between the treatment groups were found on any baseline characteristics" (p.1087).

As this was a single (assessor) blinded trial, participants and clinical personnel were not blinded to treatment allocation. Additionally, there were no apparent deviations from the intended intervention. "Data analysis was by intention to treat" (p.1086).

Data were available for 100% of those randomised.

"The primary outcomes were as follows: the number of self‐reported self‐harm episodes (suicide attempts and non‐suicidal self‐harm episodes combined)..." (p.1085). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). Additionally, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. However, whilst "[t]en independent assessors, blind to treatment allocation and to results from baseline interviews, conducted interviews at trial completion…" (p.4), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

Referring back to the trial protocol (NCT00675129), no specific information was reported. However, "...self‐harm was measured at 9 weeks (covering the first 9 weeks) and 15 weeks (covering the next 6 weeks)" (p.1085). All eligible outcome measurements within the outcome domain were reported. Additionally, "[t]he primary outcomes were as follows: the number of self‐reported self‐harm episodes (suicide attempts and non‐suicidal self‐harm episodes combined)..." (p.1085). All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as insufficient information was reported in the trial protocol to determine whether the data that produced this result analysed in accordance with a pre‐specified analysis plan that was finalized before unblinded outcome data were available for analysis.

Subgroup 2.1.3 Comparator: Alternative psychotherapy

McCauley 2018

 Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns Some concerns

"Participants were randomized to treatment condition groups using a computerized adaptive minimization randomization procedure that matched participants across conditions within sites on age, number of suicide attempts, number of previous self‐injuries, and psychotropic medication use..." (p.2‐3). Additionally, there were no significant differences between the intervention and comparator arms for any of the demographic characteristics and clinical variables measured (Table 1, p.4).

"Participants learned their treatment assignment at the first therapy session." (p.3). However, '[t]herapists provided treatment in only 1 study arm..." (p.3). Given the similarity in treatment intensity between the intervention and comparator arms in this trial, it is possible that clinical personnel could have remained blind to treatment allocation. However, there were no apparent deviations from the intended intervention. Additionally, "[t]reatment effects were evaluated using intention‐to‐treat analyses" (p.3). Furthermore, "[a]ll dropouts and missing data were replaced using multiple imputation based on averaging 10 iterative Markov‐Chain
Monte Carlo imputations to complete the missing data, providing a full intention‐to‐treat analysis" (p.4).

Data were available for 100% of those randomised.

"Suicide attempts, NSSI, and self‐harm were measured using the SASII [Suicide Attempt Self‐Injury Interview]" (p.3). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Additionally, whilst "[a]ssessors naive to treatment group" (p.3), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

Referring back to the trial protocol (NCT01528020), no specific information was reported. "Assessments occurred at baseline (before treatment), 3 months (middle of treatment), 6 months (end of treatment), 9 months, and 12 months" (p.3). Data on repetition of SH at the 3 month (middle of treatment) and 9 months assessment were not reported, however. Whilst the 9 month assessment represents a post‐intervention assessment time point, the authors do report data for other post‐intervention assessment time points for which a favourable effect for the intervention was not apparent. It is therefore unlikely trial authors selectively reported results for favourability. All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation.

Figures and Tables -
Risk of bias for analysis 2.1 Repetition of SH at post‐intervention
Navigate to table in Review
Risk of bias for analysis 3.1 Repetition of SH by post‐intervention

Bias

Study

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Griffiths 2019

 Some concerns Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns

"Randomisation (at the individual level) was independent and concealed, using randomised‐permuted blocks adjusted to permit access to group treatment without undue delay. Group allocation remained concealed until completion of self‐report ratings" (p.3).  Whilst, "[a]t entry to the study, there were no significant between group differences on demographic or service usage variables..." (p.3), the trial authors also report that "[u]sing the recommended cut‐off [for the 11‐item Borderline Personality Features Scale for Children] 55% of participants in the treatment arm warranted further assessment for borderline personality disorder compared to only 15% of the control arm" (p.6) (OR 6.60, 95% CI 1.70 to 25.62).

As this was a single (assessor) blinded trial, participants and clinical personnel were not blind to treatment allocation. Referring back to the trial protocol (NCT02771691), little specific information was provided on the content of the intervention. However, there was no apparent deviation from the intended intervention. Additionally, "...we also carried out intention‐to‐treat analyses to determine indicative treatment effects on the primary and secondary outcome measures, adjusting for pre‐specified baseline covariates" (p.4).

Data were available for 100% of those randomised.

"Our primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by emergency department presentation in...electronic records" (p.4). Additionally, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. "All measures were administered blind by the research assistants...For self‐harm related emergency department use, the research assistants reviewed...electronic records after the last face to face assessment had been conducted in order to maintain blinding" (p.4).

Referring back to the trial protocol (NCT02771691), no specific information was reported. "Our primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by emergency department presentation in...electronic records…[all measures were administered…at baseline, 12, 24 and 36 weeks" (p.4). All eligible outcome measurements within the outcome domain were reported. "Our primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by
emergency department presentation in...electronic records" (p.4). All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as just over half of the participants assigned to the intervention arm warranted further assessment for borderline personality disorder at baseline compared to 15% of those assigned to the control arm, suggesting there may have been a problem with the randomisation process.

Rossouw 2012

 Low risk of bias Low risk of bias Low risk of bias High risk of bias Some concerns High risk of bias

"Allocation was by minimization, controlling for past hospital admissions, gender, and age" (p.1305). Additionally, "...participants were randomized by an independent statistician working offsite..." (p.1306). Additionally,  there were no significant differences between the intervention and comparator arms for most of the demographic characteristics and clinical variables measured (Table 1, p.1308). However, whilst those assigned to the intervention arm were significantly younger (MD ‐0.60, 95% CI ‐1.15 to ‐0.05; p=0.04), this result may have been compatible with chance.

"...participants were...blinded to assignment" (p.1305). Given the difference in therapeutic intensity between the intervention and comparator arms, however, it is unlikely that clinical personnel blinding could have been convincingly achieved. Additionally, there were no apparent deviations from the intended intervention. "Data analysis was by intention to treat" (p.1306).

Data were available for 88.7% of those randomised. However, analysis method did not correct for bias, nor were sensitivity analyses undertaken to investigate the potential effect of missing data. Despite this, data were available for 90.0% of those randomised to the intervention arm, and 87.5% of those randomised to the comparator arm.

"The primary outcome was self‐harm assessed by self‐report...using the self‐harm scale of the Risk‐Taking and Self‐Harm Inventory (RTSHI)" (p.1306). However, this was based on fewer adolescents scoring above the cut‐point on the Risk Taking and Self‐Harm Inventory (RTSHI) and it is unclear how this scale may relate to actual SH behaviour. Additionally, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). Additionally, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. However, whilst "assessors…were…blinded to assignment" (p.1305), information on SH was ascertained from participant self‐report.

No information is available as to whether the data that produced this result analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. However,  "[t]he primary outcome was self‐harm assessed by self‐report at baseline and every 3 months until 12 months after randomization" (p.1306). All eligible outcome measurements within the outcome domain were reported. Additionally, "[t]he primary outcome was self‐harm assessed by self‐report..." (p.1306). All eligible analyses of the data were reported.

Overall, this trial was rated as at high risk of bias as, repetition of self‐harm assessed by self‐report using the self‐harm scale of the Risk‐Taking and Self‐Harm Inventory (RTSHI) and it is unclear how this scale may relate to actual SH behaviour. Additionally, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation.

Figures and Tables -
Risk of bias for analysis 3.1 Repetition of SH by post‐intervention
Navigate to table in Review
Risk of bias for analysis 3.2 Repetition of SH at post‐intervention (Risk‐Taking and Self‐Harm Inventory)

Bias

Study

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Griffiths 2019

 Some concerns Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns

"Randomisation (at the individual level) was independent and concealed, using randomised‐permuted blocks adjusted to permit access to group treatment without undue delay. Group allocation remained concealed until completion of self‐report ratings" (p.3).  Whilst, "[a]t entry to the study, there were no significant between group differences on demographic or service usage variables..." (p.3), the trial authors also report that "[u]sing the recommended cut‐off [for the 11‐item Borderline Personality Features Scale for Children] 55% of participants in the treatment arm warranted further assessment for borderline personality disorder compared to only 15% of the control arm" (p.6) (OR 6.60, 95% CI 1.70 to 25.62).

As this was a single (assessor) blinded trial, participants and clinical personnel were not blind to treatment allocation. Referring back to the trial protocol (NCT02771691), little specific information was provided on the content of the intervention. However, there was no apparent deviation from the intended intervention. Additionally,  "...we also carried out intention‐to‐treat analyses to determine indicative treatment effects on the primary and secondary outcome measures, adjusting for pre‐specified baseline covariates" (p.4).

Data were available for 100% of those randomised.

"Our primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by emergency department presentation in...electronic records" (p.4). Additionally, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. "All measures were administered blind by the research assistants...For self‐harm related emergency department use, the research assistants reviewed...electronic records after the last face to face assessment had been conducted in order to maintain blinding" (p.4).

Referring back to the trial protocol (NCT02771691), no specific information was reported. However, "[o]ur primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by emergency department presentation in...electronic records…[all measures were administered…at baseline, 12, 24 and 36 weeks" (p.4). All eligible outcome measurements within the outcome domain were reported. Additionally, "[o]ur primary outcome was self‐harm post‐treatment as assessed by...[t]he self‐harm subscale of the Risk‐Taking and Self‐Harm Inventory for Adolescents (RTSHI)...[and s]elf‐harm related hospital use as reported by emergency department presentation in...electronic records" (p.4). All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as just over half of the participants assigned to the intervention arm warranted further assessment for borderline personality disorder at baseline compared to 15% of those assigned to the control arm, suggesting there may have been a problem with the randomisation process.

Rossouw 2012

 Low risk of bias Low risk of bias Low risk of bias High risk of bias Some concerns High risk of bias

"Allocation was by minimization, controlling for past hospital admissions, gender, and age" (p.1305). Additionally, "...participants were randomized by an independent statistician working offsite..." (p.1306). Additionally,  there were no significant differences between the intervention and comparator arms for most of the demographic characteristics and clinical variables measured (Table 1, p.1308). However, whilst those assigned to the intervention arm were significantly younger (MD ‐0.60, 95% CI ‐1.15 to ‐0.05; p=0.04), this result may have been compatible with chance.

"...participants were...blinded to assignment" (p.1305). Given the difference in therapeutic intensity between the intervention and comparator arms, however, it is unlikely that clinical personnel blinding could have been convincingly achieved. However, there were no apparent deviations from the intended intervention. Additionally, "[d]ata analysis was by intention to treat" (p.1306).

Data were available for 88.7% of those randomised. However, analysis method did not correct for bias, nor were sensitivity analyses undertaken to investigate the potential effect of missing data. Despite this, data were available for 90.0% of those randomised to the intervention arm, and 87.5% of those randomised to the comparator arm.

"The primary outcome was self‐harm assessed by self‐report...using the self‐harm scale of the Risk‐Taking and Self‐Harm Inventory (RTSHI)" (p.1306). However, this was based on fewer adolescents scoring above the cut‐point on the Risk Taking and Self‐Harm Inventory (RTSHI) and it is unclear how this scale may relate to actual SH behaviour. Additionally, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Although, whilst "assessors…were…blinded to assignment" (p.1305), information on SH was ascertained from participant self‐report.

No information is available as to whether the data that produced this result analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. However, "[t]he primary outcome was self‐harm assessed by self‐report at baseline and every 3 months until 12 months after randomization" (p.1306). All eligible outcome measurements within the outcome domain were reported. Additionally, "[t]he primary outcome was self‐harm assessed by self‐report..." (p.1306). All eligible analyses of the data were reported.

Overall, this trial was rated as at high risk of bias as, repetition of self‐harm assessed by self‐report using the self‐harm scale of the Risk‐Taking and Self‐Harm Inventory (RTSHI) and it is unclear how this scale may relate to actual SH behaviour. Additionally, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation.

Figures and Tables -
Risk of bias for analysis 3.2 Repetition of SH at post‐intervention (Risk‐Taking and Self‐Harm Inventory)
Navigate to table in Review
Risk of bias for analysis 5.1 Repetition of SH at post‐intervention

Bias

Study

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Subgroup 5.1.1 Comparator: TAU

Harrington 1998

 Low risk of bias Low risk of bias Low risk of bias Some concerns Some concerns Some concerns

"...a series of opaque and sealed envelopes containing either a blank sheet or the letter F were prepared and randomly assorted by an assistant" (p.2).  Additionally, "[t]he groups were well matched on clinical characteristics" (p.514), including age, sex, history of SH, method of SH, family history of SH, and other clinical and functional characteristics (Table 1, p.514).

Given the difference in therapeutic intensity between the intervention and comparator arms, it is unlikely that participant blinding could have been convincingly achieved. Furthermore, "[the envelopes] were opened by the social worker at the time of the assessment, who then assigned the case to the family intervention plus routine care or routine care alone" (p.513), suggesting that clinical personnel were not blind to treatment assignment. However, there were no apparent deviations from the intended intervention. Additionally, "[a]ll the analyses were conducted ‘intent to treat’"(p.514).

Data were available for 97.4% of those randomised.

Correspondence with trial authors revealed repetition of SH was ascertained from self‐report. However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, It is likely that repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Additionally, whilst "[t]reatment assignment was entered on a register and concealed from the outcome assessors" (p.2), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

No information is available as to whether the data that produced this result were analysed in accordance with a pre‐specified analysis plan that was finalised before unblinded outcome data were available for analysis. However, "[o]utcome assessments were conducted...2 months and...at 6 months" (p.514). All eligible outcome measurements within the outcome domain were reported. Additionally, correspondence with trial authors revealed repetition of SH was ascertained from self‐report. All eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation.

Subgroup 5.1.2 Comparator: Enhanced usual care

Asarnow 2017

 Low risk of bias Low risk of bias Low risk of bias Some concerns Low risk of bias Some concerns

"Eligible participants were randomly assigned in a 1:1 ratio to [intervention] or [comparator] using a computerized algorithm. To improve balance across conditions, the randomization procedure was stratified on gender and SA versus NSSI‐only status at presentation. Enrollment...staff were masked to randomization status." (p.507). Additionally, "[intervention] and [comparator] youths were similar in regard to demographic characteristics and clinical variables, with no statistically significant between group differences" (p.509; Table 1, p.511).

Given the difference in therapeutic intensity between the intervention and comparator arms, it is unlikely that participant or clinical personnel blinding could have been convincingly achieved. However, there were no apparent deviations from the intended intervention. Additionally, "[p]rimary analyses were intent to treat, incorporating data from all
participants regardless of degree of participation" (p.509).

"Although data were available for 100% of the sample using any report [i.e., youth self‐report supplemented by parental report, where necessary], youth‐report data were available for 55% of E‐TAU youths versus 100% of SAFETY youths" (p.510). However, "[s]ensitivity analyses using parent report when youth report was unavailable yielded similar results...Additional sensitivity analyses assuming that all E‐TAU youths who dropped out early reached the 3‐month time point SA‐free also yielded comparable results" (p.511).

"...youth self‐report was used in primary analyses of SA and SH outcomes" (p.509). However, previous work has demonstrated that self‐harm prevalence estimates derived from self‐report may be underestimated, and supplementing prevalence estimates with medical or clinical record information is advisable (Mars 2016). However, repetition of self‐harm was ascertained in the same way for both the intervention and comparator arms. Although, whilst "[a]ssessment staff were naive to treatment condition" (p.508), information on SH was ascertained from participant self‐report. Self‐reported self‐harm represents an participant‐reported outcome.

Referring back to the trial protocol (NCT00692302), there were no apparent departures from the analysis plan. Referring back to the trial protocol (NCT00692302), all eligible outcome measurements within the outcome domain were reported. Additionally, referring back to the trial protocol (NCT00692302), all eligible analyses of the data were reported.

Overall, this trial was rated as at some concerns for risk of bias as, although outcome assessors were blinded, repetition of self‐harm was ascertained from self‐reported information and participants were likely not blind to treatment allocation.

Figures and Tables -
Risk of bias for analysis 5.1 Repetition of SH at post‐intervention
Navigate to table in Review