Shared decision‐making interventions for people with mental health conditions

See: Summary of findings 1 Shared decision‐making interventions compared with usual care for people with mental health conditions

Summary of findings 1. Shared decision‐making interventions compared with usual care for people with mental health conditions

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Shared decision‐making interventions compared with usual care for people with mental health conditions
Patient or population: people with mental health conditions Setting: various Intervention: shared decision‐making Comparison: usual care, cognitive training, placebo session
Outcomes	Risk with usual care	Risk with SDM intervention	Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Psychiatric symptoms Brief Psychiatric Rating Scale (BPRS; Overall 1988): 16 items with 7‐item Likert scale ('not present' to 'extremely severe') measured 6 months after intervention (Yamaguchi 2017).		MD ‐1.10 lower (‐5.54 lower to 3.34 higher)	‐	53 (1 RCT)	⊕⊝⊝⊝ Very low^a,b	Higher scores indicate more severe psychiatric symptoms; the results indicate little or no difference between groups. One further study that could not be pooled reported no statistically significant difference in PANSS scores between the groups when they were discharged from hospital (Hamann 2006).
Depression (1 to 6 months) Montgomery–Åsberg Depression Rating Scale (MADRS; Montgomery 1979): measured 3 months after intervention (Aljumah 2015). Patient Health Questionnaire‐9 (PHQ‐9; Kroenke 2001): measured 6 to 8 weeks after intervention (Loh 2007) or 3 months after intervention (LeBlanc 2015). Quick Inventory of Depressive Symptomatology Self‐Report (QIDS‐J; Rush 2003): measured 3 months after intervention (Aoki 2019a).		SMD ‐0.03 lower (‐0.17 lower to 0.12 higher)	‐	717 (4 RCTs)	⊕⊕⊝⊝ Low^c,d	Higher scores indicate more severe depression symptoms; the results indicate little or no difference between groups.
Depression (6 months or more) MADRS measured 6 months after intervention (Aljumah 2015). PHQ‐9 measured 6 months after intervention (LeBlanc 2015). QIDS‐J measured 6 months after intervention (Aoki 2019a). Hospital Anxiety and Depression Scale (HADS‐D; Zigmond 1983) measured 6 months after intervention (Lovell 2018).		SMD 0.03 higher (‐0.10 lower to 0.17 higher)	‐	898 (4 RCTs)	⊕⊕⊝⊝ Low^c,d	Higher scores indicate more severe depression symptoms; the results indicate little or no difference between groups.
Readmission (6 months or more) Rehospitalisation at 8 months after discharge (Hamann 2006) or 12 months after discharge (Hamann 2017).	Study population		RR 1.06 (0.77 to 1.46)	249 (2 RCTs)	⊕⊝⊝⊝ Very low^c‐e	‐
	362 per 1000ⁱ	384 per 1000 (279 to 529)	RR 1.06 (0.77 to 1.46)	249 (2 RCTs)	⊕⊝⊝⊝ Very low^c‐e	‐
Participation (observations on the process of SDM) Observing Patient Involvement in shared decisiON‐making (OPTION; Elwyn 2005) assessed from video recording on the encounter (LeBlanc 2015). Core components of SDM: scoring the transcripts of conversations between participants and doctors during consultation (SDM‐18; Salyers 2012) during consultation (Yamaguchi 2017).		SMD 1.14 higher (0.63 higher to 1.66 higher)	‐	133 (2 RCTs)	⊕⊝⊝⊝ Very low^c,f,g	Higher scores indicate more involvement in decision‐making; the results indicate an increase in involvement for the SDM group.
Participation (SDM‐specific‐reported outcomes, immediately after intervention) Combined Outcome Measure for Risk Communication and Treatment Decision‐making Effectiveness (COMRADE; Edwards 2003) measured immediately after decision‐making (Aoki 2019a). Decisional Conflict Scale (DCS; O'Connor 1995a) measured immediately after the clinical encounter (LeBlanc 2015). Man‐Song‐Hing Scale (Man‐Song‐Hing 1999) measured after intervention (Loh 2007).		SMD 0.63 higher (0.26 higher to 1.01 higher)	‐	534 (3 RCTs)	⊕⊕⊝⊝ Low^c,h	COMRADE, Man‐Song‐Hing Scale: higher scores indicate more involvement in decision‐making; DCS: lower scores indicate less decisional conflict; the results indicate an increase in involvement for the SDM group. In one further study that could not be pooled, participants in the intervention group reported significantly greater perceived involvement than those in the control group (Hamann 2006).
Adverse events ‐ not reported	There were no adverse effects reported.		‐	‐	‐	‐
*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; PANSS: Positive and Negative Syndrome Scale; RR: risk ratio; SDM: shared decision‐making
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.
^aDowngraded by one level: for risk of bias (high risk of bias for blinding of participants and outcome assessment) ^bDowngraded by two levels: for imprecision (insufficient number of participants for one study and large confidence interval) ^cDowngraded by one level: for indirectness (the outcome was measured using various approaches) ^dDowngraded by one level: for imprecision (large confidence interval) ^eDowngraded by two levels: for risk of bias (1 of 2 studies were at high risk of bias for randomisation and allocation concealment, and 2 of 2 studies were at high risk of bias for blinding of participants and outcome assessment) ^fDowngraded by two levels: for imprecision (small sample size) ^gDowngraded by one level: for risk of bias (2 of 2 studies were at high risk for blinding of participants and outcome assessment) ^hDowngraded by one level: for inconsistency (I² ≥ 50%; P value for heterogeneity ≤ 0.05) ⁱControl event rate calculated from means of usual care groups used in this analysis (Hamann 2006; Hamann 2017)

Background

Description of the condition

Mental health conditions

Mental health conditions comprise a wide range of problems, with different symptoms, including: neurodevelopmental disorders, schizophrenia spectrum and other psychotic disorders, bipolar and related disorders, depressive disorders, anxiety disorders, obsessive‐compulsive and related disorders, trauma and stressor‐related disorders, eating disorders, and personality disorders (APA 2013). A World Health Organization (WHO) survey estimated that 18.1% to 36.1% of the world’s population suffer from a diagnosable mental health condition during their life course (Kessler 2007). Hence, mental health conditions are globally prevalent and have a devastating impact on the lives of the people who experience them, their families, and communities (Whiteford 2015; WHO 2017). Mental health conditions affect people’s health and productivity both at home and in the workplace. This can lead to economic burdens for the person and their families. As more people experience mental health conditions, the more societal costs also rise. In Japan, the economic cost due to schizophrenia in 2008 has been estimated as 2.8 trillion Japanese yen (JPY) (USD 25.7 billion or USD 25,700 million) (Sado 2013), and depression in 2005 was JPY 2 trillion (USD 18 billion or USD 18,000 million) (Sado 2011). In Europe, the societal costs due to brain disorders, including mental health conditions, was estimated to be 800 billion euros (EUR) (USD 1 trillion or USD 1,000,000 million) a year, more than cancer, cardiovascular disease, and diabetes put together (Smith 2011). Moreover, the relative share and impact of the health burden caused by mental health conditions is increased due to stigma and lack of adequate support for treatment or care (Rehm 2019). Most notably, many low‐income countries invest less than 1% of their health budgets in mental health services (WHO 2015). Given this situation, WHO has a comprehensive mental health action plan (2013‐2020) under the catchphrase “no health without mental health” (Saxena 2013). Thus, mental health conditions are a worldwide health priority topic.

The recovery movement

The treatment of people with mental health conditions has evolved towards more comprehensive care in this century, where people are recognised as being at the centre, and health improvement is viewed in terms of recovery, rather than simply symptom relief. Recovery is a way of living a meaningful life even with limitations caused by mental illness, accepting and overcoming the challenge of the disability (Deegan 1988; Anthony 1993). Recovery is not a uniform process but varies from person to person. Recently, the term ‘personal recovery’ has been widely used to describe patient‐based recovery, which consists of elements such as re‐establishment of identity, finding meaning in life, empowerment, and taking responsibility for recovery (Van Eck 2018). The recovery process emphasises control being placed in the hands of the individual and not the professional (Jacobson 2001). In this regard, recovery‐oriented mental health care requires greater emphasis on the collaborative nature of interactions among health professionals, people with mental health conditions and their families (Duncan 2010). This concept has now been adopted at a national policy level in many Western countries (Perkin 2012; Van Hoof 2015; National Alliance on Mental Illness 2016), and has extended to Asian and African countries (Stein 2014; Singh 2015).

Shared decision‐making

The dominant paradigms in modern health care today are those of evidence‐based and person‐centred care. Thus, medical decision‐making has moved away from traditional, paternalistic approaches, where physicians drive the decision‐making process (Charles 1997). Increasingly, shared decision‐making (SDM) is advocated as an ideal model of treatment decision‐making in various medical fields, including mental health (Storm 2013). Charles and colleagues proposed an SDM model that encapsulates the most widely recognised core features (Charles 1997):

at least two participants ‐ physician and patient ‐ are involved;
both parties share information;
both parties take steps to build a consensus about the preferred treatment (that is, both participate in the decision‐making process); and
an agreement is reached on the treatment to implement (that is, a decision is made or is actively deferred).

Shared decision‐making emphasises the involvement of both parties in the collaborative process of understanding and deliberating the best available evidence about the risks and benefits across all available options, while ensuring that the patient’s values and preferences are fully clarified (Charles 1997; Elwyn 1999; Towle 1999). Shared decision‐making is an ethical imperative (Drake 2009; Elwyn 2017), and has been gaining support as a key principle of the delivery of person‐centred care (Barry 2012). Especially in the mental health field, it is a central part of the recovery paradigm described above, which derives from the patient’s right to autonomy and self‐determination (Storm 2013; Slade 2017).

Makoul and Clayman conducted a systematic review of 161 articles that specifically addressed SDM in health care to determine the range of conceptual definitions (Makoul 2006). They proposed nine essential elements as an integrative model of SDM during consultations with patients (Makoul 2006):

define and explain the healthcare problem;
present options;
discuss pros and cons (benefits, risks, costs);
clarify patient values and preferences;
discuss patient ability and self‐efficacy;
present what is known and make recommendations;
check and clarify the patient’s understanding;
make or explicitly defer a decision; and
arrange follow‐up.

Accordingly, for a decision to be a truly ‘shared’ decision it must have certain characteristics. It must involve at least two participants, and the sharing of information. The decision (which may be to do nothing) must be made and agreed upon by all parties (Charles 1997). Once a decision is made, there must be opportunities to review the decision (Edwards 2005).

Unsurprisingly, SDM does not mean the same thing in all cases. Trevena and Barratt proposed that the suitability of a decision for SDM depends upon the clinical context, patient preferences, and practitioner responsibilities (Trevena 2003). Kon suggested that SDM can best be understood as a continuum, at one end of which is patient‐driven decision‐making, at the opposite is physician‐driven decision‐making, and in the middle are many possible SDM approaches (Kon 2010).

Montori and colleagues examined the Charles 1997 SDM model in relation to long‐term conditions (Montori 2017). They concluded that for SDM to work in these conditions, it was necessary to add another component to the model: ongoing partnership between the clinical team (not just the clinician) and the patient (Montori 2017). SDM often evolves over multiple encounters because decision‐making is never just a single event or activity but rather is distributed over a range of people and times or episodes (Rapley 2008). Furthermore, especially in the public sector, SDM requires the active involvement of other parties, such as family members (Aoki 2019b) or peer‐support staff (Goscha 2015). In the case of long‐term conditions, 'planning' may be as much a feature as actually making decisions (Joseph‐Williams 2019).

Description of the intervention

This review is an update of an existing review of SDM for people with diagnosable mental health conditions published in 2010 (Duncan 2010), which included two cluster‐randomised trials. One was an SDM intervention for inpatients with schizophrenia, which consisted of a decision aid, decision support by nurses, and planning talk with their physicians (Hamann 2006). The other SDM intervention was for primary care patients with depression using a decision board during consultation with physicians (Loh 2007). To our knowledge, several articles about SDM interventions in mental health settings which could be incorporated into the updated version of the review had been reported since 2010 (Hamann 2011; Aljumah 2015; LeBlanc 2015). We have therefore identified an increasing number of recent trials in this area and have used these to inform the features of the interventions to be included in this review update.

We also recognise in this update that chronic or long‐term conditions require treatment decision‐making but that collaborative goal setting (between individual patients and clinicians) and action planning are also important (Coulter 2015).

Interventions eligible for inclusion in this review therefore include:

psychiatric ward‐based interventions for inpatients with mental health conditions, such as sharing treatment decision‐making between patient and clinician, perhaps also using decision support tools, or sharing ‘care planning’ between the patient and the interprofessional team as a role in decision coaching;
primary care‐based interventions for newly diagnosed or regular outpatients with mental disorders, such as sharing treatment decision‐making during initial or routine consultations, perhaps also using decision support tools, or sharing care planning between service users and their interprofessional team and/or peer support staff as a role in decision coaching; and
community‐based interventions, such as sharing care planning using telecommunication tools, web‐based tools, or home‐visiting care services.

In addition, studies of SDM educational or training programmes have been reported in recent years (Hamann 2011; Hamann 2017). Therefore, eligible interventions also include:

SDM educational or training programmes targeting patients or healthcare professionals, or both, in psychiatric ward‐based, primary‐care based, or community‐care based settings.

In this review update, we identify three overarching categories of intervention implementation in the context of both SDM interventions and interventions based on SDM educational or training programmes:

interventions targeting patients or carers such as family members, or both;
interventions targeting healthcare professionals; and
interventions targeting both.

Regarding patients, although the original review included only people with severe mental illnesses (schizophrenia and depression), despite no diagnostic restrictions (Duncan 2010), recent trials seem to include not only people with severe mental illnesses, such as schizophrenia, bipolar disorder, and depression, but also other mental illnesses such as neurodevelopmental disorder, anxiety disorder, personality disorder, and post‐traumatic stress disorder (Woltmann 2011; Westermann 2013; Metz 2018).

In short, as described above, there are several types of environmental settings, various types of interventions, and different diagnoses in this area. This situation allowed us to plan to conduct subgroup analyses for environmental setting, intervention types, participant diagnosis, and intervention elements as potential effect modifiers.

The main comparison sought for this review was between SDM interventions and usual care or control groups, which do not explicitly intend to involve patients. So far, we have found no adverse effects of SDM interventions in the mental health field, but the review sought to identify any harms reported by included studies.

How the intervention might work

Shared decision‐making includes collaboration and deliberation between patients and healthcare professionals, leading to well‐informed, preference‐based patient decisions and more cost‐effective health care. This, in turn, results in improved health outcomes (Elwyn 2016).

Information exchange during consultation is a central element of the identified SDM studies in various medical fields (Légaré 2018). There are also additional important elements, other than information exchange, such as sufficient rapport and trusting relationships between patient and healthcare professional in interventions of SDM in mental health (Zisman‐Ilani 2017; Aoki 2020). Shared decision‐making in mental health can foster and strengthen a therapeutic relationship between patient and healthcare professional, developing empathy, genuineness, trust, and mutual understanding between two parties (Corrigan 2012; Aoki 2020).

Moreover, in the SDM model, people can challenge existing barriers associated with mental health conditions, such as abuse of power, power asymmetry, assumptions of decisional incapacity, and stigma, while empowering patients through the decision‐making process (James 2017; Aoki 2020). Contrary to conventional assumptions of cognitive dysfunction, there is evidence that most people who access mental health services want to be involved in decisions about their care (Patel 2010; De las Cuevas 2012; Park 2014).

Shared decision‐making interventions in mental health seem to have an effect on both patients and healthcare professionals. For example, for patient outcomes, some trials indicate that SDM can have an impact on affective‐cognitive outcomes, such as knowledge, satisfaction, decision conflict, responsibility in decision‐making, beliefs towards treatment (Woltmann 2011; LeBlanc 2015; Ishii 2017), and behavioural outcomes, such as participation in decision‐making and communication improvements (Yamaguchi 2017; Aoki 2019a).

Shared decision‐making interventions can also change the behaviour of healthcare professionals to facilitate a patient’s active engagement in decision‐making during the consultation (Hamann 2011; Aoki 2020).

From a few trials, limited evidence suggests that SDM for people with mental health conditions may be able to improve long‐term outcomes, such as medication adherence, treatment adherence, and clinical symptoms (Aljumah 2015). This is largely centred around engagement with psychopharmacology, and the idea that SDM may be able to improve concordance between the expectations of patients and healthcare professionals (McKinnon 2014).

Why it is important to do this review

The original review included only two eligible studies: one undertaken in inpatients with schizophrenia and the other in outpatients newly diagnosed with depression. No definite conclusions could be drawn (Duncan 2010). However, recent further empirical evidence about SDM interventions in mental health warranted an update of the original review.

Shared decision‐making as a high‐priority interest in mental health care

Patient participation in treatment decision‐making has become a high priority in mental health care systems in recent years around the world (Thompson 2007). In many countries, mental health clinicians and policy makers advocate shared decision‐making as an important component of national mental health policies (Härter 2017).

Shared decision‐making is not only an ideal; it is also essential to focus on its effectiveness and implementation in practice (Adshead 2018). For example, guidance from the National Institute for Health and Care Excellence (NICE, UK) recommends that healthcare professionals should involve people with mental health conditions in decisions about prescribed medicines (NICE 2009; NICE 2015). World Federation of Societies of Biological Psychiatry (WFSBP) Guidelines for biological treatment of schizophrenia also specify that the goals of long‐term treatment have to be discussed with the patient and, if she/he agrees, with family members, relatives, carers and, in some cases, advocates, with the aims of providing adequate information and understanding the patient's personal goals (Hasan 2013).

Moreover, the cost‐effectiveness of SDM in mental health care has been drawing increasing attention. Cosh and colleagues have argued that SDM could reduce medical care costs by reducing admissions of inpatients with severe mental illness (Cosh 2017). This approach may also decrease the costs associated with the use of unnecessary, or unwanted, prescriptions (Latimer 2011; De las Cuevas 2012).

Originality of this review update

There are some systematic and narrative reviews related to this topic. A review and synthesis by James and Quirk identified a "rationale for SDM" as any argument or reason for SDM in mental health care outlined by authors in a journal paper and which described the rationales of SDM (James 2017). Their review excluded raw data or findings from experimental trials. Therefore, they did not draw conclusions on the effects of SDM interventions in mental health settings. A review by Zisman‐Ilani and colleagues indicated unique elements of SDM in mental health, such as facilitating patient motivation and providing patient communication skills training, which were rarely seen in other medical fields (Zisman‐Ilani 2017). However, this review was also descriptive and did not attempt statistical synthesis of the outcomes nor draw conclusions about effectiveness.

Stovell and colleagues conducted a systematic review of shared treatment decision‐making for psychosis, which identified 11 RCTs and showed small beneficial effects on indices of treatment‐related empowerment (Stovell 2016). However, given its focus on treatment decisions concerning psychosis, this review did not consider other mental health conditions and rehabilitation or care plan decisions beyond medical treatments.

As described above, an increasing number of systematic reviews of studies of SDM interventions have been reported in the mental health field thus far. However, there are few reviews examining the effects of SDM interventions on different types of decisions regarding treatment and care and addressing the broad range of mental health conditions since the original review conducted in 2010 (Duncan 2010).

In 2010, Légaré and colleagues first published a Cochrane Review entitled “Interventions for increasing the use of shared decision‐making by healthcare professionals” and have updated it periodically (Légaré 2018). The review has not restricted the 'types of participants’ to any specific health condition; instead, it has targeted a broad range of conditions. As the review title indicates, Légaré and colleagues have focused on interventions aimed at improving uptake of SDM by healthcare professionals, with a primary focus on how well this is adopted in practice (Légaré 2018). Légaré and colleagues have focused on healthcare professionals, whereas this review update focuses on service users and their carers, as well as healthcare professionals. Thus, this review has updated the available evidence for SDM in mental health settings and allowed us to conduct wider searches of the psychiatric and mental health literature to find additional studies and to focus on crucial psychiatric‐specific outcomes, which are not covered in the Légaré 2018 review. It is particularly important to focus on clinical symptoms (e.g. severity of psychotic symptoms), stages of recovery, and treatment/medication continuation because such outcomes form a link with personal recovery for people with mental health conditions.

For this review update, we also carefully considered recent research on outcome measurement tools used for assessing the effects of SDM interventions in the mental health field. Perestelo‐Perez and colleagues reviewed existing instruments used in SDM‐related studies in mental health (Perestelo‐Perez 2017b). The review revealed that there are three types of measurements: SDM antecedent measurements, such as the Autonomy Preference Index and Control Preference Scale (API, Ende 1989; CPS, Denger 1992); SDM process measurements, such as Observation Patient Involvement in Decision‐Making (OPTION, Edwards 2003; Edwards 2005; Elwyn 2013) and Shared Decision‐Making Questionnaire‐9 (SDM‐Q‐9, Kriston 2010); and SDM outcome measurements, such as the Decisional Conflict Scale and the Decision Regret Scale (DCS, O'Connor 1995a; DRS, Brehaut 2003). Information about the measurements used in previous trials informed our planning and decisions about which outcomes to assess in this review update.

It is over 10 years since the original Duncan 2010 review was published. Due to the growth of trials in this area, and a lack of systematic reviews with the same focus, an update of the systematic review is timely.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) and cluster‐RCTs (trials in which groups of participants are randomised). Although the original review's inclusion criteria allowed for controlled before‐and‐after (CBA) studies, interrupted time series (ITS) studies, and quasi‐RCTs, these study designs are at greater risk of bias than RCTs or cluster‐RCTs in evaluating the effectiveness of the interventions. Furthermore, we were aware that more literature in this area existed for this update, rather than only the two eligible studies identified in the original 2010 review. Therefore, we excluded study designs other than RCTs and cluster‐RCTs.

Types of participants

The people receiving the healthcare service within studies were those diagnosed with a mental health condition by any defined criteria, such as the International Classification of Diseases (ICD) (WHO 1992; WHO 2018), or the Diagnostic and Statistical Manual of Mental Disorders (DSM) (APA 2000; APA 2013). We included studies enrolling individuals of all ages. We included both public and private healthcare patients. We excluded simulated patients and those without any psychiatric diagnosis. We also excluded people who were making hypothetical decisions or advanced directive decisions.

The participants (those receiving the intervention) were people with mental health conditions or service users, informal carers such as family members, or healthcare professionals for people with mental health conditions (including general practitioners, psychiatrists, psychologists, nurses, social workers, occupational therapists, other allied health professionals, and lay support staff including peer support staff).

Types of interventions

The descriptions of the interventions were consistent with the SDM definition articulated by Charles 1997:

at least two participants, patient and physician, should be involved;
both parties share information;
both parties take steps to build a consensus about the preferred treatment; and
an agreement is reached on the treatment to implement.

Therefore, we included any intervention which:

met the four criteria identified by Charles 1997, above; and/or
consisted of SDM educational or training programmes targeted at people with mental health conditions (such as training in asking questions, discussion, clarifying own preferences, and reaching a decision) or healthcare professionals (such as training in problem definition, presenting options, communication skills, providing recommendations based on their expertise and previous experiences, and reaching a consensus), or both.

We excluded any intervention which:

did not meet the Charles 1997 criteria;
made the SDM element a secondary focus of the intervention (e.g. anxiety management);
consisted solely of information provided to people with mental health conditions about a condition (e.g. patient education without the two‐way sharing of information necessary for SDM);
aimed at enhancing communication between people with mental health conditions and healthcare professional, without focusing on a particular choice or decision;
targeted future care; that is, advanced directives ‐ also known as Ulysses contracts ‐ that set out how a person who is periodically mentally unwell wishes to be treated at those times; or
consisted exclusively of decision support interventions, such as decision coaching, patient decision aids, and question prompt sheets, and did not meet the Charles 1997 criteria.

We included interventions with decision support, such as decision coaching, patient decision aids, and question prompt sheets, if this formed a part of SDM.

Interventions took place in any care setting and were not restricted by the mode or intensity of delivery.

Included studies assessed a single intervention or a combination of interventions, and compared them with another type of intervention, with usual or standard care, or with no intervention.

Types of outcome measures

We have made changes from the original review for several outcomes. First, for the primary outcome, the level of patient involvement replaced satisfaction because patient involvement in decision‐making is crucial in the SDM process. In addition, since the concept of recovery has been gaining attention in this area in recent years, we decided to adopt recovery as one of the secondary outcomes in this review. We describe the primary and secondary outcomes below.

Primary outcomes

Clinical outcomes assessed using measurement tools such as psychiatric scales, depression scales, and anxiety scales

Psychiatric symptoms (for severe mental health conditions; e.g. Brief Psychiatric Rating Scale (BPRS; Overall 1988); Positive and Negative Syndrome Scale (PANSS; Kay 1988); or 48‐item Symptom Questionnaire (SQ‐48; Carlier 2012))
Depression (e.g. the Hospital Anxiety and Depression Scale (HADS; Zigmond 1983); Montgomery–Åsberg Depression Rating Scale (MADRS; Montgomery 1979); Beck Depression Inventory (BDI II; Beck 1996); or Patient Health Questionnaire‐9 (PHQ‐9; Kroenke 2001))
Anxiety (e.g. State‐Trait Anxiety Inventory (STAI; Spielberger 1983); or Hospital Anxiety and Depression Scale (HADS‐A; Zigmond 1983))
Readmission rates

Participation (by the person with the mental health condition) or level of involvement in the decision‐making process

Observations on the process of SDM (e.g. Observing Patient Involvement in Decision‐Making Scale for measuring patient involvement: OPTION (Elwyn 2003; Elwyn 2005); OPTION 5 Item (Elwyn 2013); Coding System to Measure Elements of Shared Decision‐Making During Psychiatric Visits (Salyers 2012))
Shared decision‐making‐specific user‐reported outcomes from encounters (e.g. Client Decision Conflict Scale (DCS; O'Connor 1995a); decision regret scale (DRS; Brehaut 2003); the 9‐item Shared Decision‐Making Questionnaire (SDMQ‐9; Kriston 2010); Combined Outcome Measure for Risk Communication and Treatment Decision‐Making (COMRADE; Edwards 2003); Clinical Decision‐Making Involvement and Satisfaction (CDIS‐P Involvement subscale; Slade 2014); Control Preferences Scale (CPS; Denger 1992); or Evaluating and Quantifying User and Carer Involvement in Mental Health Care Planning Patient‐Reported Outcome Measure (EQUIP ROPM; Bee 2016))

Assessing effects on clinical outcomes can be challenging because the effects of SDM interventions can depend on which treatment option is chosen. However, individuals wish not only to be involved in decision‐making but for their symptoms to improve, and this review therefore regards clinical outcomes as key outcomes alongside those measuring the degree of involvement in decision‐making.

Secondary outcomes

Recovery

e.g. Recovery Assessment Scale (Corrigan 2004); Developing Recovery Enhancing Environments Measure (DREEM; Ridgway 2004); Stages of Recovery Instrument (STORI; Andresen 2006); or Self‐Identified Stage of Recovery Parts A and B (SISR‐A and B; Andresen 2006)

Satisfaction

Overall satisfaction (with care) of person with mental health condition (e.g. Client Satisfaction Questionnaire‐8 (CSQ‐8; Attkisson 1982); Verona Service Satisfaction Scale (VSSS; Ruggeri 1993))
Users’ satisfaction concerning decision‐making (e.g. the satisfaction with decision scale (Holmes‐Rovner 1996); or Clinical Decision‐Making Involvement and Satisfaction (CDIS‐P Satisfaction subscale; Slade 2014))
Users' satisfaction with received information
Carer satisfaction (e.g. Carer satisfaction measured via the Carers and Users’ Expectations of Services—carer version (CUES‐C; Lelliott 2003))
Healthcare professional satisfaction

Knowledge

e.g. Patient/carer knowledge about disease, condition, or treatment options, provider knowledge

Treatment or medication continuation

e.g. Morisky Medication Adherence (MMAS; Morisky 1986; Morisky 2008)

Relationship or interaction between service users and health professionals

e.g. therapeutic alliance, concordance

Health service use outcomes

e.g. resource use; length of consultation; costs

Adverse outcomes

Any potential harms associated with interventions, including potential worsening of mental health condition

Timing of outcome assessment

We included all time points of outcome assessment in this review. We prespecified three categories: short‐term (until one month after decision‐making), medium‐term (one to six months after), and long‐term time points (six months or more), if applicable.

Main outcomes for the summary of findings tables

We reported the following outcomes in the summary of findings tables:

clinical outcomes, such as psychiatric scales and depression scales;
participation or involvement during the SDM process; and
adverse outcomes associated with interventions.

Search methods for identification of studies

We:

searched electronic bibliographic databases for published work;
searched trial registers and contacted authors for information on ongoing and recently completed studies;
searched the reference lists of relevant published studies; and
contacted authors of relevant studies to check for additional studies.

There were no language restrictions.

Electronic searches

We used an explicit search strategy, developed in collaboration with the Cochrane Consumer and Communication Group, to search the following bibliographic databases:

Cochrane Central Register of Controlled Trials (CENTRAL, the latest issue) in the Cochrane Library (2009 to 13 January 2020);
MEDLINE (OvidSP) (2009 to 14 January 2020);
Embase (OvidSP) (2009 to 14 January 2020); and
PsycINFO (OvidSP) (2009 to 14 January 2020).

We structured the search strategy according to a study design filter, mental illness search terms (based on advice from the Cochrane Common Mental Disorders Group), and shared decision‐making terms (Makoul 2006).

We updated and re‐ran the searches in February 2022.

We present the search strategy for CENTRAL in Appendix 1; Embase in Appendix 2; MEDLINE in Appendix 3; and PsycINFO in Appendix 4.

Searching other resources

We searched online trial registers for ongoing and recently completed studies using the following databases with the terms (shared decision‐making) OR SDM | psychiatry OR psychiatric OR psychology OR psychologic:

ClinicalTrials.gov, US National Institutes of Health (NIH) at clinicaltrials.gov/ (all dates);
World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) at apps.who.int/trialsearch/ (all dates); and
Web of Science (all dates).

We also searched reference lists of included studies and relevant systematic reviews, and primary studies.

Data collection and analysis

Selection of studies

We used the Cochrane RCT Classifier, which classifies records into two groups: 1) records with a low probability of being RCTs; and 2) records that have a high probability of being RCTs.

For the records with a low probability of being RCTs, one review author (TU) screened all titles and abstracts and confirmed that there were no RCTs.

For the records with a high probability of being RCTs, two review authors (YA, YY or MS, LS) independently screened all titles and abstracts to determine which studies met the inclusion criteria.

We retrieved in full text any papers identified as potentially relevant by at least one review author. Two review authors (YA, YY or MS, LS) independently screened full‐text articles for inclusion or exclusion, with discrepancies resolved by discussion and by consulting a third review author (TU or AE), if necessary, to reach consensus. We listed as excluded studies all the potentially relevant papers excluded from the review at this stage, with reasons provided.

We also provided citation details and any available information about ongoing studies. We collated and reported details of duplicate publications, so that each study (rather than each report) is the unit of interest in the review.

On 8 February 2022, we updated the searches. We searched references since January 2020 using the same bibliographic databases and resources described above. We also used the Cochrane RCT Classifier.

One review author (YY) screened all titles and abstracts for the records with a low probability of being RCTs, according to the RCT Classifier, and confirmed that there were no RCTs.

Two review authors (YA, UT) screened all titles and abstracts for the records that had a high probability of being RCTs. We retrieved in full text any papers identified as potentially relevant by at least one review author. Two review authors (YA, UT) independently screened full‐text articles for inclusion or exclusion, with discrepancies resolved by discussion and by consulting a third author (YY) if necessary, to reach consensus. All potentially‐relevant papers excluded from the review at this stage were listed as excluded studies, with reasons provided.

We then listed the studies which met the inclusion criteria as studies awaiting classification.

We reported the screening and selection process in an adapted PRISMA flow chart (Liberati 2009).

Data extraction and management

Two review author pairs (YA, YY or MS, LS) independently extracted data from included studies. For any studies involving the review authors, different members of the review author team assessed and extracted data from those studies. We resolved any discrepancies through discussion until consensus was reached, or through consultation with a third review author (TU or AE), where necessary. We developed and piloted a data extraction form using the Cochrane Consumers and Communication Review Group Data Extraction Template (available at: cccrg.cochrane.org/author-resources). We pilot tested the data extraction form with the first five included studies and refined it as necessary.

We extracted the following study data.

General information: title, source, publication date, country, language, author contact details, study design, aim, number of arms, consumer involvement, if informed consent was obtained, whether ethical approval was obtained.
Characteristics of participants: description of participants, geographic location, setting, methods of recruitment of participants, inclusion/exclusion criteria for participation in study, age (range, mean (standard deviation)), gender, ethnicity, principal diagnosis, other health problem/s, severity of illness, treatment receiving.
Characteristics of interventions: intervention description, whether SDM criteria were completely met (Charles 1997), aims of intervention, what was done, who delivered intervention, where/when/how often/how much was intervention provided, how people with mental health conditions accessed the intervention; whether the intervention was tailored/modified, how well the intervention was delivered.
Characteristics of outcomes and comparison groups: method of assessing outcome measures, method of follow‐up of non‐respondents, and timing of outcome assessment; loss‐to‐follow‐up rates, and characteristics of those lost to follow‐up. When two or more relevant measures were reported for each outcome, the scale of the validated assessment tool was chosen in the pooled statistical analysis.
Data and results: timing of outcome assessment, observed/total number (for dichotomous outcomes), mean change/standard deviation/number (for continuous outcomes), whether validated assessment tools were used.
Assessment of risk of bias: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, other sources of bias (e.g. baseline differences).
Funding source: details of the funding source, declaration of interests for the primary investigators.

One review author (YA) entered all extracted data into Review Manager 5 (RevMan 5) (Review Manager 2014). A second review (YY) author, working independently, checked for accuracy against the data extraction sheets. We contacted authors of individual studies to ask for additional information if required.

Assessment of risk of bias in included studies

We assessed and reported on the methodological risk of bias of included studies in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), and the guidelines of the Cochrane Consumers and Communication Review Group (Ryan 2019), which recommends the explicit reporting of the following: individual elements for RCTs: random sequence generation; allocation sequence concealment; blinding (participants, personnel); blinding (outcome assessment); completeness of outcome data, selective outcome reporting; and other sources of bias, such as baseline comparability.

We considered blinding separately for different outcomes where appropriate (for example, blinding may have the potential to differently affect subjective versus objective outcome measures).

For cluster‐RCTs, we assessed the adequacy of adjustment for clustering, and assessed and reported the risk of bias associated with an additional domain: selective recruitment of cluster participants.

We judged each item as being at high, low or unclear risk of bias as set out in the criteria provided by Higgins 2011, and provided a quote from the study report and a justification for our judgement for each item in the risk of bias table.

We deemed studies to be at the highest risk of bias if we scored them as at high or unclear risk of bias for either the sequence generation or allocation concealment domains, based on growing empirical evidence that these factors are particularly important potential sources of bias (Higgins 2011).

In all cases, two review authors independently assessed the risk of bias of included studies (YA, YY or MS, LS), with any disagreements resolved by discussion to reach consensus. We contacted study authors for additional information about the included studies, or for clarification of the study methods, as required.

We incorporated the results of the risk of bias assessment into the review through standard tables, and systematic narrative description and commentary about each of the domains, leading to an overall assessment of the risk of bias of included studies and a judgment about the internal validity of the review’s results.

Measures of treatment effect

For dichotomous outcomes, we analysed data based on the number of events and the number of people assessed in the intervention and comparison groups. We used these to calculate the risk ratio (RR) and 95% confidence interval (CI). For continuous measures, we analysed data based on the mean, standard deviation (SD) and number of people assessed for both the intervention and comparison groups to calculate mean difference (MD) and 95% CI. If the MD was reported without individual group data, we used this to report the study results. If more than one study measured the same outcome using different tools, we calculated the standardised mean difference (SMD) and 95% CI using the inverse variance method in Review Manager 5.

Unit of analysis issues

For cluster‐RCTs, we checked for unit‐of‐analysis errors. If errors were found, and sufficient information was available, we reanalysed the data using the appropriate unit of analysis, by taking account of the intra‐cluster correlation (ICC). We obtained estimates of the ICC by contacting authors of included studies or imputed them using estimates from external sources. If it was not possible to obtain sufficient information to reanalyse the data, we reported only the point estimate, and identified those studies as being high risk for 'other' bias based on the ‘unit‐of‐analysis error’.

Dealing with missing data

We contacted study authors to obtain missing data (participant, outcome, or summary data). For participant data, where possible, we conducted analysis on an intention‐to‐treat basis; otherwise, we analysed data as reported. We reported on the levels of loss to follow‐up and assessed this as a source of potential bias. For missing outcome or summary data, we imputed missing data where possible and reported any assumptions in the review.

Assessment of heterogeneity

Where studies were considered similar enough (based on consideration of populations, interventions, and other factors) to allow pooling of data using meta‐analysis, we assessed the degree of heterogeneity by visual inspection of forest plots and by examining the Chi² test for heterogeneity. A significance level of P = 0.1 was used in view of the low power of such tests.

We reported our reasons for deciding that studies were similar enough to pool statistically. We quantified heterogeneity using the I² statistic. We considered an I² value of 50% or more to represent substantial levels of heterogeneity, but we interpreted this value in light of the size and direction of effects and the strength of the evidence for heterogeneity, based on the P value from the Chi² test (Higgins 2011). Where heterogeneity was present in pooled effect estimates, we explored possible reasons for variability by conducting subgroup analyses.

Where we detected substantial clinical, methodological, or statistical heterogeneity across included studies, we did not report pooled results from meta‐analysis but instead tried to use a narrative approach to data synthesis. In this event, we reported our reasons for deciding that studies were too dissimilar to meta‐analyse. We also explored possible clinical or methodological reasons for this variation by grouping studies that were similar in terms of populations, intervention features, methodological features, or other factors, to explore differences in intervention effects.

Assessment of reporting biases

We assessed reporting bias qualitatively based on the characteristics of the included studies (e.g. if only small studies indicating positive findings were identified for inclusion), or if information that we obtained from contacting experts and authors of studies suggested that there were relevant unpublished studies. If we could identify sufficient studies for inclusion in the review, we planned to construct a funnel plot to investigate small study effects, which can indicate the presence of publication bias. We planned to test for funnel plot asymmetry, with the choice of test made based on advice in Higgins 2011, and bearing in mind that there may be several reasons for funnel plot asymmetry when interpreting the results. However, we did not include sufficient studies to construct a funnel plot.

Data synthesis

We assessed suitability for meta‐analysis based upon whether the interventions in the included trials were similar enough in terms of participants, settings, intervention, comparison, and outcome measures to ensure meaningful conclusions from a statistically pooled result (see Assessment of heterogeneity). Due to the anticipated variability in the populations and interventions, and possibly other factors, of included studies, we used a random‐effects model for meta‐analysis.

When we were unable to pool the data statistically using meta‐analysis, we provided clear reasons for this decision, and conducted a narrative synthesis of results.

We planned to explore the main comparisons of the review as follows: intervention versus no intervention; intervention versus usual care; and one form of intervention versus another. However, the majority of studies were 'intervention versus usual care'. The exceptions were two studies that compared SDM interventions with cognitive training (Hamann 2011; Hamann 2017), and one study that included a placebo attention control comparison group (Mott 2014). We therefore analysed studies together but conducted subgroup analysis of SDM versus usual care and SDM versus cognitive training for the primary outcomes.

We planned to analyse the effects of interventions from studies comparing more than one intervention separately against the control, but no studies compared multiple interventions.

Subgroup analysis and investigation of heterogeneity

The potential subgroups for analysis included:

care setting (inpatient, outpatient, primary care, community secure environment);
types of intervention target population (directed to people with mental health conditions (and children versus older persons with mental health conditions), health professional, carers such as family members);
mental illness severity of people with mental health conditions (severe mental illness including schizophrenia, bipolar disorder, and depression versus non‐severe mental illness);
intervention types (SDM with or without decision support tools versus none, SDM training to health professionals or people with mental health conditions /carers).

If substantial heterogeneity was found, we determined potential reasons for heterogeneity by examining individual study characteristics and those of subgroups of the main body of evidence.

Sensitivity analysis

We conducted a sensitivity analysis on the risk of bias assessment, comparing the results of studies at higher and lower risk of bias. In these cases, we removed lower‐certainty studies (high overall risk of bias) from the analysis and examined how robust the results were when based only on higher‐certainty studies (overall low risk of bias or some concerns).

Summary of findings and assessment of the certainty of the evidence

We developed a summary of findings table to present the results of the meta‐analysis or narrative synthesis, or both, for the primary outcomes, including potential harms. We provided a source and rationale for each assumed risk cited in the table, and used the GRADE criteria to rank the certainty of the evidence based on the methods described in Chapter 11 of the Cochrane Handbook for Systematic Reviews of Interventions, using the GRADEprofiler (GRADEpro) software (Schünemann 2011). We ranked the certainty of the evidence for each outcome, downgrading the rating if one or more of the following criteria were present: risk of bias, inconsistency, imprecision of the observed effect, indirect evidence, and publication bias. We used footnotes to justify our decisions to downgrade the certainty of the evidence to aid the reader's understanding of the review.

Two review authors independently assessed each outcome against the GRADE criteria.

Where meta‐analysis was not possible, or possible for only some data for an outcome, we presented these findings descriptively alongside any pooled effect estimates from meta‐analysis.

No review authors selected, extracted data, or appraised the risk of bias for the study on which they were an author (Aoki 2019a). Members of the review author team other than YA, TY, and KW (authors on the Aoki 2019a trial) led the GRADE ratings for assessment of overall certainty of evidence for each outcome.

Results

Description of studies

See: Included studies; Studies awaiting classification; Ongoing studies

Results of the search

We conducted the initial database searches on 14 January 2020.

We generated 10,118 references (6935 references from MEDLINE, Embase, PsychINFO, and Web of Science plus 3183 references from CENTRAL, ClinicalTrials.gov, and WHO ICTRP) after removing duplicates from 12,129 references identified through database searching.

For the 6935 references from MEDLINE, Embase, PsychINFO, and Web of Science, we used the Cochrane RCT Classifier and classified two groups: 2956 references with a low probability of being RCTs; and 3976 references that had a high probability of being RCTs. For the 2956 references screened out based on the low probability of their being RCTs by the RCT Classifier, TU screened all titles and abstracts and checked there had been no RCT among them.

For 7162 references (3183 from CENTRAL, ClinicalTrials.gov, and WHO ICTRP plus the 3976 classified as having a high possibility of being RCTs by the RCT Classifier), two review authors independently screened all titles and abstracts of these references to determine which met the inclusion criteria. YA and YY screened half of them and LS and MS screened the other half. We identified and retrieved a total of 250 articles for appraisal in full‐text screening. Of the 250 full‐text articles, YA and YY independently screened 103 full‐text articles for inclusion or exclusion, with discrepancies resolved by discussion and by consulting TU to reach consensus. LS and MS independently screened the remaining 147 full‐text articles, with discrepancies resolved by discussion and by consulting AE. We ultimately identified 13 studies (25 articles) that met the inclusion criteria.

Thus, in this update, we have included 15 studies (28 articles) in the review: two studies (3 articles) were previously included studies in the 2010 version of the review (Duncan 2010) (see Figure 1).

Figure 1

PRISMA study flow diagram for initial search in January 2020

Search update

We updated the searches on 16 February 2022. We generated 2427 references (1636 references from MEDLINE, Embase, PsychINFO, and Web of Science plus 791 references from CENTRAL and ClinicalTrials.gov) after removing duplicates from 2662 references identified through database searching.

For the 1636 references from MEDLINE, Embase, PsychINFO, and Web of Science, we used the Cochrane RCT Classifier and classified two groups: 708 references with a low probability of being RCTs; and 928 references that had a high probability of being RCTs. For the 708 references screened out based on the low probability of their being RCTs by the RCT Classifier, YY screened all titles and abstracts and confirmed there had been no RCT among them.

For 1719 references (791 from CENTRAL and ClinicalTrials.gov plus the 928 references classified as having a high probability of being RCTs by the RCT Classifier), YA and TU independently screened all titles and abstracts of these references to determine which met the inclusion criteria. We identified and retrieved a total of 35 articles for appraisal in full‐text screening. From these, we identified four studies which met the inclusion criteria; we have listed these as studies awaiting classification (see Figure 2).

Figure 2

PRISMA study flow diagram for update search in February 2022

Included studies

This update search added 13 new studies (Aljumah 2015; Aoki 2019a; Hamann 2011; Hamann 2017; Ishii 2017; LeBlanc 2015; Lovell 2018; Mariani 2018; Mott 2014; Raue 2019; Troquete 2013; Woltmann 2011; Yamaguchi 2017), to the two previously included studies (Hamann 2006; Loh 2007), for a total of 15 studies.

Unit of randomisation

All the studies were randomised controlled trials. Of these, eight studies randomised individual participants, and seven randomised clusters (Hamann 2006; Loh 2007; LeBlanc 2015; Lovell 2018; Mariani 2018; Troquete 2013; Woltmann 2011). Four of seven trials accounted for the cluster effect in the published outcome data (Loh 2007; LeBlanc 2015; Lovell 2018; Woltmann 2011), and our meta‐analysis used the published data. However, the three remaining studies did not account for the cluster effect in the published data (Hamann 2006; Mariani 2018; Troquete 2013). Troquete 2013 did not report any of the primary or secondary outcomes of this systematic review. For Hamann 2006 and Mariani 2018, we did not reanalyse the data and we report these studies separately.

Settings and participants

The 15 RCTs, involving 3141 adults with mental health conditions, presented results from seven countries: Germany (four studies: Hamann 2006; Hamann 2011; Hamann 2017; Loh 2007), Japan (three studies: Aoki 2019a; Ishii 2017; Yamaguchi 2017), the Netherlands (one study: Troquete 2013), the UK (one study: Lovell 2018), the USA (four studies: LeBlanc 2015; Mott 2014; Raue 2019; Woltmann 2011), Saudi Arabia (one study: Aljumah 2015), and the Netherlands and Italy (one study: Mariani 2018). We did not find any trials which studied children.

The level of care was primary care in two studies (LeBlanc 2015; Loh 2007); community mental health service in two studies (Lovell 2018; Woltmann 2011); outpatient psychiatric service in four studies (Aljumah 2015; Aoki 2019a; Raue 2019; Yamaguchi 2017); specialised outpatient service in two studies such as a PTSD clinic (Mott 2014) and a forensic psychiatric service (Troquete 2013); acute wards in psychiatric hospital in four studies (Hamann 2006; Hamann 2011; Hamann 2017; Ishii 2017); and nursing home wards in one study (Mariani 2018).

The mental health conditions studied were schizophrenia in four studies (Hamann 2006; Hamann 2011; Hamann 2017; Ishii 2017); depression in four studies (Aljumah 2015; LeBlanc 2015; Loh 2007; Raue 2019); post‐traumatic stress disorder (PTSD) in one study (Mott 2014); and dementia in one study (Mariani 2018). Five studies included multiple clinical conditions: Aoki 2019a, depression and bipolar disorder; Lovell 2018, severe mental illness such as schizophrenia and bipolar disorder; Troquete 2013, substance‐related disorder, personality disorder, psychotic disorder, and others; Woltmann 2011, schizophrenia, bipolar disorder, depression, PTSD, and others; Yamaguchi 2017, schizophrenia, bipolar disorder, depression, and developmental disorder.

The care providers were physicians in five studies (Hamann 2011; Hamann 2017; LeBlanc 2015; Loh 2007; Mott 2014); case managers in two studies (Troquete 2013; Woltmann 2011); nurses in one study (Raue 2019); and pharmacists in one study (Aljumah 2015). Three studies included an interprofessional approach: Aoki 2019a and Hamann 2006, nurses and clinicians; Lovell 2018, nurses, occupational therapists, social workers, and others. Mariani 2018 included family carers and professionals, and Yamaguchi 2017 included peer supporters and clinicians.

Interventions

Of the 15 studies, nine studies used a decision support tool during consultation or decision coaching sessions. Of the nine studies, seven studies used a printed decision aid (Aljumah 2015; Aoki 2019a; Hamann 2006; LeBlanc 2015; Loh 2007; Mott 2014; Raue 2019), and two used electronic decision support systems (Woltmann 2011; Yamaguchi 2017). Among those that did not use decision support tools, Ishii 2017 used a question prompt sheet about treatment and Troquete 2013 used the assessment tool for risks and treatability.

Nurse interventions with decision aids before or after physician consultation were provided in three studies (Aoki 2019a; Hamann 2006; Raue 2019). Interventions by other care providers before or after physician consultation were pharmacists in one study (Aljumah 2015), case managers in one study (Woltmann 2011), and peer support specialists in one study (Yamaguchi 2017).

Almost all studies provided SDM training sessions to healthcare providers, and two of the 15 studies provided SDM training sessions to participants (Hamann 2011; Hamann 2017).

For details, see Characteristics of included studies.

Comparisons

Of the 15 studies, 12 studies provided people in the control group with usual care, such as usual physician consultation (Aoki 2019a; Hamann 2006; Ishii 2017; LeBlanc 2015; Loh 2007; Lovell 2018; Raue 2019; Yamaguchi 2017), usual care planning (Mariani 2018; Troquete 2013; Woltmann 2011), or usual pharmacy services (Aljumah 2015). The usual care participants in those studies received no decision‐making tool such as a decision aid or electronic decision support system and no decision support by decision coaching. The healthcare providers did not receive any training in SDM.

In Hamann 2011 and Hamann 2017, control participants received cognitive training (Lutz 2005) as a comparison to the intervention group, where participants received an SDM training program.

In Mott 2014, to ensure that participants received equal attention from study staff, control participants attended a 30‐minute placebo session.

Conceptual framework of SDM

The authors of Hamann 2006 and Ishii 2017 cited Charles 1997 for their definition of SDM and developed an interprofessional SDM model for inpatients.

Aoki 2019a used the SDM framework developed by Hamann 2006, which included nurse decision coaching before decision‐making with a physician, and modified it to be suitable for outpatients.

In Aljumah 2015, the authors used the SDM competency framework, developed by Simmons 2010, which was designed specifically for patients with depression.

Both studies that used an electronic decision support tool ‐ Woltmann 2011 and Yamaguchi 2017 ‐ cited the CommonGround computerised decision support system, which was created in the USA by Patricia Deegan, focusing on facilitating recovery‐oriented, shared decision‐making (Deegan 2008; Deegan 2010).

The authors of Loh 2007 stated that their approach to training physicians was based on the work of Towle 1999, Elwyn 2000, Elwyn 2001, and Elwyn 2001b.

In Hamann 2011 and Hamann 2017, the content of the training programme for patients was derived from theoretical considerations about patients' contributions to the SDM process (Towle 1999), from an adaptation of related approaches on patient competences in the medical encounter (Cegala 2000; Farin 2014).

The authors of Lovell 2018 cited Coulter 2017 and Montori 2017 for their definition of SDM and created an SDM programme which included a decision aid designed to encourage and directly support the conversations between patients and physicians.

Mariani 2018 cited the SDM principles in dementia and active listening (Gordon 2000), to enhance both verbal and non‐verbal communication skills to be used to assess and meet the patients' needs and preferences during the SDM interview.

In Mott 2014, the intervention was based on an existing decision‐making model by Elwyn and colleagues (Elwyn 2010; Elwyn 2012), which identifies SDM components: "choice talk", "option talk", and "decision talk".

Although three of the 15 studies did not refer to any particular concept (LeBlanc 2015; Raue 2019; Troquete 2013), the SDM of Troquete 2013 used a method of periodically monitoring violence risks and treatment needs (Van den Brink 2010).

Excluded studies

After full‐text assessment of articles for eligibility, we excluded 174 articles from the original search in January 2020 and 21 articles from the update search in February 2022. The reasons for exclusion were related to the design of the study, the type of participants, and the content of the intervention. Regarding the content of the intervention, the most common reason for exclusion was that the SDM intervention was part of a complex intervention addressing many facets of patient care. In these studies, the effects of SDM intervention could not be isolated. For more details, see Excluded studies.

We also identified 16 ongoing RCTs during the initial search in January 2020 and four ongoing RCTs during the update search in February 2022. Of the 16 ongoing RCTs from the initial search, we identified that one study published and, thus, we moved it to the list of 'studies awaiting classification' in February 2022. Thus, we ultimately identified 19 ongoing studies. For more details, see Ongoing studies.

For studies awaiting classification, initially,we identified five studies (four studies from the update search and one study transferred from the ongoing studies). Ultimately, we identified four studies after removing one duplicate. For more details, see Studies awaiting classification.

Risk of bias in included studies

We report further information on the rating and rationale for risk of bias assessments of the included studies in the risk of bias tables in the Characteristics of included studies, and summarise these in Figure 3 and Figure 4. The risk of bias assessment reported was based on the primary outcomes.

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Allocation

We considered 10 of 15 studies (67%) to be at low risk of bias for random sequence generation because these trial authors described their randomisation list or computerised randomisation methods (Aljumah 2015; Aoki 2019a; Hamann 2011; Hamann 2017; Ishii 2017; LeBlanc 2015; Loh 2007; Lovell 2018; Mott 2014; Yamaguchi 2017). In one study (7%), sequence generation took place after wards had been paired based on their characteristics (Hamann 2006); we judged this study as having a high risk of bias. We rated four studies (27%) as having an unclear risk of bias because they lacked a specific description of random sequence generation (Mariani 2018; Raue 2019; Troquete 2013; Woltmann 2011).

We rated nine of 15 studies (60%) as being at low risk of bias for allocation concealment: in these, allocation was done by a person with no involvement in the study (Aljumah 2015; Aoki 2019a; LeBlanc 2015; Troquete 2013; Yamaguchi 2017); closed concealment envelopes or drawing blind lots were used (Hamann 2011; Hamann 2017; Loh 2007); or an external telephone randomisation service was used (Lovell 2018). We assessed two studies (13%) as having a high risk of bias for allocation concealment: one study used envelopes (Mott 2014), and one study randomised at the ward level (Hamann 2006). We rated the four remaining studies (27%) as having an unclear risk of bias because they gave no description of allocation concealment (Ishii 2017; Mariani 2018; Raue 2019; Woltmann 2011).

Blinding

Due to the nature of the interventions, blinding of the participants and the personnel delivering the intervention was not possible. Therefore, for the blinding of participants and personnel, we judged all 15 studies to be at high risk of bias in this domain.

For the blinding of outcome assessment, we rated four of 15 studies (27%) as having a low risk of bias (Aljumah 2015; Aoki 2019a; Lovell 2018; Troquete 2013), six (40%) as having an unclear risk of bias (Hamann 2011; Ishii 2017; Loh 2007; Mariani 2018; Mott 2014; Woltmann 2011), and five (33%) as having a high risk of bias (Hamann 2006; Hamann 2017; LeBlanc 2015; Raue 2019; Yamaguchi 2017).

Incomplete outcome data

We considered that six of the 15 studies (40%) were at low risk of attrition bias because a similar and low proportion of participants from either group could not be included in the final analyses (Aljumah 2015; Aoki 2019a; Ishii 2017; LeBlanc 2015; Woltmann 2011; Yamaguchi 2017). We considered nine studies (60%) to be at a high risk of attrition bias because of significant loss to follow‐up (Hamann 2006; Hamann 2011; Hamann 2017; Loh 2007; Lovell 2018; Mariani 2018; Mott 2014; Raue 2019; Troquete 2013).

Selective reporting

We rated three studies (20%) as being at a low risk of reporting bias (Ishii 2017; LeBlanc 2015; Lovell 2018); these studies' protocols were registered publicly and outcomes were reported as planned (see: Ishii 2017, secondary reference Ishii 2014; LeBlanc 2015, secondary reference LeBlanc 2013; Lovell 2018, secondary reference Bower 2015). We judged 11 studies (73%) to have an unclear risk of reporting bias because no protocol was publicly available (Aljumah 2015; Aoki 2019a; Hamann 2006; Hamann 2011; Hamann 2017; Loh 2007; Mott 2014; Raue 2019; Troquete 2013; Woltmann 2011; Yamaguchi 2017). We considered Mariani 2018 to be at a high risk of reporting bias because the authors stated in the protocol that the primary outcome was the proportion of dementia residents whose preferences, needs, and related actions were known, satisfied, and documented in their ‘life‐and‐care plan’. However, in the results paper, the primary outcome was the agreement of the residents’ ‘life‐and‐care plans’ with the five operationalised recommendations.

Other potential sources of bias

Of 15 studies, we rated 10 studies (67%) which discussed other potential sources of bias. We considered the five studies which lacked a sample size calculation and were of a relatively small sample size as being at high risk of bias (Hamann 2011; Mariani 2018; Mott 2014; Woltmann 2011; Yamaguchi 2017). In Ishii 2017 and Troquete 2013, although the sample size calculation was conducted, the participants were fewer than planned, leading to underpowered analyses. In Lovell 2018, missing baseline data for the cohort sample was cluster mean imputed. We considered the three studies in which participants were recruited to the trial after the clusters had been randomised as being at a high risk of bias (Hamann 2006; Loh 2007; Lovell 2018). In addition, three studies did not account for the cluster effect (Hamann 2006; Mariani 2018; Troquete 2013).

Effects of interventions

Primary outcomes

Clinical outcomes assessed using measurement tools such as psychiatric scales, depression scales, and anxiety scales

Psychiatric symptoms (for severe mental health conditions)

Two studies reported psychiatric symptom outcomes (Hamann 2006; Yamaguchi 2017; see Table 1). Data from Yamaguchi 2017, using the Brief Psychiatric Rating Scale (BPRS) at six months' follow‐up, were available for the analysis. The mean difference (MD) for this study was ‐1.10 (95% confidence interval (CI) ‐5.54 to 3.34; 1 study, 53 participants; very low‐certainty evidence; Analysis 1.1), indicating little or no difference between groups. One study not included in the analysis reported no statistically significant difference in Positive and Negative Syndrome Scale (PANSS) scores between the intervention and control groups when they were discharged from hospital (Hamann 2006). However, this study did not adequately adjust for clustering and so may have produced overly precise results. Therefore, we are uncertain whether SDM interventions improve psychiatric symptoms compared with control due to very low‐certainty evidence.

Table 1. Psychiatric symptoms

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Note
Hamann 2006	Positive and Negative Syndrome Scale, PANSS	At discharge	‐	58	‐	59.3	P > 0.05
Yamaguchi 2017	Brief Psychiatric Rating Scale, BPRS	After 6 months' follow‐up	26	34.0 (SD 7.9)	27	35.1 (SD 8.6)	P = 0.31

SDM: shared decision‐making

Depression

Six studies reported depression symptoms (Aljumah 2015; Aoki 2019a; LeBlanc 2015; Loh 2007; Lovell 2018; Raue 2019; see Table 2).

Table 2. Depression

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Aljumah 2015	Montgomery‐Åsberg Depression Rating Scale, MADRS	3 months	110	21.07 (SD 12.21)	110	21.01 (SD 12.63)	P = 0.971
Aljumah 2015	Montgomery‐Åsberg Depression Rating Scale, MADRS	6 months	110	20.65 (SD 11.97)	110	20.86 (SD 12.54)	P = 0.897
Aoki 2019a	Quick Inventory of Depressive Symptomatology, QIDS‐J	3 months	35	10.49 (SD 5.12)	53	9.57 (SD 5.80)	No difference
Aoki 2019a	Quick Inventory of Depressive Symptomatology, QIDS‐J	6 months	35	10.34 (SD 5.68)	53	10.36 (SD 6.17)	No difference
LeBlanc 2015	Patient Health Questionnaire‐9, PHQ‐9	3 months	114	9.0	101	9.2	P = 0.78
LeBlanc 2015	Patient Health Questionnaire‐9, PHQ‐9	6 months	109	8.9	101	9.3	P = 0.91
LeBlanc 2015	Remission rate, PHQ score < 5	3 months	114	19.6%	101	18.7%	P = 0.85
LeBlanc 2015	Remission rate, PHQ score < 5	6 months	109	21.5%	101	14.4%	P = 0.18
LeBlanc 2015	Responsiveness, > 50% PHQ‐9 improvement	3 months	114	33.5%	101	30.9%	P = 0.77
LeBlanc 2015	Responsiveness, > 50% PHQ‐9 improvement	6 months	109	34.8%	101	27.3%	P = 0.15
Loh 2007	Patient Health Questionnaire‐9, PHQ‐9	6‐8 weeks	128	13.7 (SD 5.8)	66	14.6 (SD 5.3)	P = 0.610
Lovell 2018	The Depression subscale of the Hospital Anxiety and Depression Scale, HADS‐D	6 months	208	9.8 (SD 5.5)	172	8.9 (SD 5.8)	P = 0.963
Raue 2019	Hamilton Depression Rating Scale, HDRS	8 weeks	114	12.4 (0.8)	88	11.7 (1.0)	No difference
Raue 2019	Hamilton Depression Rating Scale, HDRS	12 weeks	114	12.9 (0.8)	88	12.0 (0.9)	No difference

SDM: shared decision‐making

For the outcomes at one to six months' follow‐up, data from five studies (using the Montgomery‐Åsberg Depression Rating Scale (MADRS), Patient Health Questionnaire‐9 (PHQ‐9), Quick Inventory of Depressive Symptomatology (QIDS‐J), and Hamilton Depression Rating Scale (HDRS)) were available for statistical analysis. The pooled estimate standardised mean difference (SMD) was 0.14 (95% CI ‐0.19 to 0.47, a small effect; 5 studies, 919 participants; very low‐certainty evidence; Analysis 1.2), indicating little or no difference between groups.

For the sensitivity analysis of the outcomes at one to six months' follow‐up, removing the lower‐certainty study (Raue 2019), the pooled estimate SMD was ‐0.03 (95% CI ‐0.17 to 0.12; a small effect; 4 studies, 717 participants; low‐certainty evidence; Analysis 1.3), indicating little or no difference between groups. Sensitivity analysis showed that heterogeneity decreased; however, the overall effect did not change much.

For the outcomes at six months or more, data from five studies (using MADRS, PHQ‐9, QIDS‐J, and the Hospital Anxiety and Depression Scale ‐ Depression subscale (HADS‐D)) were available for statistical analysis. The pooled estimate SMD was 0.21 (95% CI ‐0.19 to 0.60, a small effect; 5 studies, 1100 participants; very low‐certainty evidence; Analysis 1.4), indicating little or no difference between groups.

For the sensitivity analysis of the outcomes at six months or more, removing the lower‐certainty study (Raue 2019), the pooled estimate SDM was 0.03 (95% CI ‐0.10 to 0.17, a small effect; 4 studies, 898 participants; low‐certainty evidence; Analysis 1.5), indicating little or no difference between groups. Sensitivity analysis found that heterogeneity decreased; however, the overall effect did not change much.

LeBlanc 2015 reported the remission rate and responsiveness of depression and found little or no difference between the two groups for: remission at three months (risk ratio (RR) 1.06, 95% CI 0.68 to 1.65; 215 participants; low‐certainty evidence; Analysis 1.6); remission at six months (RR 1.58, 95% CI 0.97 to 2.55; 210 participants; low‐certainty evidence; Analysis 1.7); response at three months (RR 1.09, 95% CI 0.81 to 1.47; 215 participants; low‐certainty evidence; Analysis 1.8); or response at six months (RR 1.34, 95% CI 0.98 to 1.83; 210 participants; low‐certainty evidence; Analysis 1.9).

Accordingly, we are uncertain if SDM interventions improve depression symptoms compared with control due to low‐ or very low‐certainty evidence.

Anxiety

One study (380 participants) reported on anxiety (Lovell 2018), using the Hospital Anxiety and Depression Scale ‐ Anxiety subscale (HADS‐A), at six months' follow‐up and found little or no difference between the intervention and control groups (Table 3).

Table 3. Anxiety

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Lovell 2018	The Anxiety subscale of the Hospital Anxiety and Depression Scale, HADS‐A	6 months	208	12.1 (SD 5.4)	172	10.9 (SD 5.9)	P = 0.339

SDM: shared decision‐making

Readmission rates

Three studies conducted in psychiatric wards reported readmission rates between two arms (Hamann 2006; Hamann 2011; Hamann 2017; see Table 4). For the outcomes at one to six months' follow‐up, data from two studies were available for the statistical analysis (Hamann 2006; Hamann 2011). The RR was 1.06 (95% CI 0.52 to 2.14; 128 participants; very low‐certainty evidence; Analysis 1.10), indicating little or no difference between groups. For six months or more, data from two studies were available for the statistical analysis (Hamann 2006; Hamann 2017). The RR was 1.06 (95% CI 0.77 to 1.46; 249 participants; very low‐certainty evidence; Analysis 1.11), indicating little or no difference between groups. In addition, Hamann 2006 did not account for the cluster effect and so may have produced overly precise results. Accordingly, we are uncertain if SDM interventions impact readmission compared with no intervention due to very low‐certainty evidence.

Table 4. Readmission rate

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Hamann 2006	Rehospitalisation rate	6 months	36	22%	37	22%	P > 0.05
Hamann 2006	Rehospitalisation rate	18 months	38	53%	41	46%	P > 0.05
Hamann 2011	Rehospitalisation rate	6 months	29	17%	26	15%	P = 0.57
Hamann 2017	Rehospitalisation rate	12 months	95	31%	75	31%	P = 0.98

SDM: shared decision‐making

Participation (by the person with the mental health condition) or level of involvement in the decision‐making process

Six studies reported participation or level of involvement in the decision‐making process (Aoki 2019a; Hamann 2006; LeBlanc 2015; Loh 2007; Lovell 2018; Yamaguchi 2017; see Table 5).

Table 5. Participation (by the person with mental health condition) or level of involvement in the decision‐making process

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Aoki 2019a	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE communication	After decision‐making	32	median 44	53	median 38	P < 0.001
Aoki 2019a	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE confidence	After decision‐making	32	median 41	53	median 37	P = 0.005
Hamann 2006	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE total	After the intervention		79.5		69.7	P = 0.03 n = 75 (Total number of participants)
LeBlanc 2015	Decisional Conflict Scale, DCS (0 = conflict, 100 = comfort)	After encounter	138	79.7	114	74.5	P = 0.01
LeBlanc 2015	Participation‐Involvement patient, OPTION (observation)	Assessed from video recording on the encounter	57	46.6	39	32.5	P = 0.01
Loh 2007	Man‐Song‐Hing Scale	After intervention	128	28.0 (SD 2.9)	66	25.5 (SD 3.0)	P = 0.003
Lovell 2018	Equip patient reported outcome measure, EQUIP PROM‐14	6 months	192	21.3 (SD 9.6)	153	21.6 (SD 11.2)	P = 0.715
Yamaguchi 2017	Core components of SDM, SDM‐18	During consultation	18	6.3 (SD 1.3)	19	4.1 (SD 1.3)	P < 0.001
Yamaguchi 2017	Patient activation measure, PAM	6 months	26	52.3 (SD 18.1)	27	45.8 (SD 10.8)

SDM: shared decision‐making

Observations on the process of SDM

Two studies assessed participation or level of involvement in the decision‐making process by observations on the process of SDM using Observing Patient Involvement in Decision‐Making Scale (OPTION) and SDM‐18 (LeBlanc 2015; Yamaguchi 2017). The pooled estimate of the SMD was 1.14 (95% CI 0.63 to 1.66, a large effect; 2 studies, 133 participants; very low‐certainty evidence; Analysis 1.12), indicating an increase in participation for the group that received the intervention. However, we are uncertain if SDM interventions improve observations on the process of SDM compared with control due to very low‐certainty evidence.

Shared decision‐making‐specific user‐reported outcomes from encounters: immediately after encounter

Four studies evaluated patient participation or level of involvement in the decision‐making process immediately after decision‐making using SDM‐specific user‐reported outcomes, such as Decision Conflict Scale (DCS), Combined Outcome Measure for Risk Communication and Treatment Decision‐Making (COMRADE), and the Man‐Song‐Hing Scale (Aoki 2019a; Hamann 2006; LeBlanc 2015; Loh 2007). Data from three studies were available for the statistical analysis (Aoki 2019a; LeBlanc 2015; Loh 2007). The pooled estimate of the SMD was 0.63 (95% CI 0.26 to 1.01, a moderate effect; 534 participants; low‐certainty evidence; Analysis 1.13), indicating an increase in participation for the group that received the intervention.

In the study not included in the meta‐analysis, participants in the intervention group reported significantly greater perceived involvement than those in the control group (Hamann 2006). However, Hamann 2006 failed to adequately adjust for clustering and so may have produced overly precise results.

Shared decision‐making‐specific user‐reported outcomes from encounters: at six months or more

Two studies assessed patient participation or level of involvement in the decision‐making process at six months' follow‐up, using the Evaluating and Quantifying User and Carer Involvement in Mental Health Care Planning Patient‐Reported Outcome Measure (EQUIP ROPM) and the Patient Activation Measure (PAM) (Lovell 2018; Yamaguchi 2017). The pooled estimate of the SMD was 0.13 (95% CI ‐0.30 to 0.56, a small effect; 398 participants; low‐certainty evidence; Analysis 1.14), indicating little or no difference between groups.

Accordingly, SDM interventions may improve SDM‐specific user‐reported participation outcomes immediately after the encounter compared with control, based on low‐certainty evidence. On the other hand, there may be little or no effect of SDM interventions on levels of user‐reported involvement compared to control at long‐term follow‐up, again based on low‐certainty evidence.

Secondary outcomes

Recovery

Two studies, Lovell 2018 and Yamaguchi 2017, reported on recovery, using Developing Recovery Enhancing Environments Measure (DREEM) and Self‐Identified Stage of Recovery (SISR), respectively (see Table 6). Data from both studies were available for the statistical analysis. The estimate of SMD was 0.10 (95% CI ‐0.13 to 0.32, a small effect; 313 participants; very low‐certainty evidence; Analysis 1.15), indicating little or no difference between the two groups. Thus, we are uncertain whether SDM interventions improve recovery compared with control due to very low‐certainty evidence.

Table 6. Recovery

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Lovell 2018	Developing Recovery Enhancing Environments Measure, DREEM	6 months	142	43.5 (SD 13.3)	118	42.6 (SD 12.5)	P = 0.990
Yamaguchi 2017	Self‐Identified Stage of Recovery, SISR Part A	6 months	26	3.19 (SD 1.2)	27	2.93 (SD 1.36)	P = 0.99
Yamaguchi 2017	Self‐Identified Stage of Recovery, SISR Part B	6 months	26	15.04 (SD 5.38)	27	14.04 (SD 4.15)	P = 0.40

SDM: shared decision‐making

Satisfaction

Ten studies reported service users' satisfaction (Aoki 2019a; Hamann 2006; Hamann 2011; Hamann 2017; Ishii 2017; LeBlanc 2015; Loh 2007; Lovell 2018; Woltmann 2011; Yamaguchi 2017; see Table 7).

Table 7. Service user satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	After decision making	32	24.31 (SD 2.90)	53	23.75 (SD 3.71)	No difference
Hamann 2006	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	At hospital discharge	‐	16.3 (SD 3.7)	‐	16.4 (SD 3.2)	P = 0.42
Hamann 2011	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	32	25.5 (SD 4.1)	29	26.7 (SD 3.2)	P = 0.23
Hamann 2017	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	‐	25.7 (SD 4.2)	‐	25.8 (SD 5.2)	P = 0.88
Ishii 2017	Overall satisfaction, Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	At hospital discharge	9	23.7 (SD 3.9)	13	22.1 (SD 3.7)	No difference
LeBlanc 2015	User satisfaction ‐ right amount of information	Immediately after encounter	132	92.5%	109	91.9%	P = 0.81
LeBlanc 2015	User satisfaction ‐ information given was clear	Immediately after encounter	132	68.7%	109	58.7%	P = 0.09
LeBlanc 2015	User satisfaction ‐ information given was helpful	Immediately after encounter	132	69.2%	109	52.8%	P = 0.01
LeBlanc 2015	User satisfaction ‐ strongly desire to receive information this way for other treatment decisions	Immediately after encounter	132	68.2%	109	50.5%	P = 0.005
Lovell 2018	User satisfaction ‐ strongly recommend the way information was shared to others	Immediately after encounter	132	77.6%	109	59.1%	P = 0.002
Loh 2007	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	128	29.8 (SD 2.7)	66	27.0 (SD 3.6)	P = 0.014
Lovell 2018	Overall satisfaction, Verona Service Satisfaction Scale ‐ European Version‐54, VSSS‐EU‐54	6 months	191	3.5 (SD 0.7)	156	3.5 (SD 0.8)	P = 0.045
Woltmann 2011	Overall satisfaction, Seven statements related to satisfaction, a 5‐point Likert scale	Post participation	40	3.88 (SD 0.54)	40	3.78 (SD 0.56)	No difference
Yamaguchi 2017	Overall satisfaction, Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	6 months	26	26.0 (4.4)	27	24.3 (4.8)	P = 0.21

SDM: shared decision‐making

Overall satisfaction (with care) of person with the mental health condition: immediately after the intervention

Five studies reported on overall satisfaction immediately after the intervention using the Client Satisfaction Questionnaire–8 Japanese version (CSQ‐8), German version of the CSQ (ZUF‐8), and a 5‐point Likert scale (Aoki 2019a; Hamann 2011; Hamann 2017; Loh 2007; Woltmann 2011). Data from four studies were available for the statistical analysis. The pooled estimate of SMD was 0.26 (95% CI ‐0.29 to 0.80, a small effect; 4 studies, 420 participants; very low‐certainty evidence; Analysis 1.16), indicating little or no difference between the two groups. One study not included in the meta‐analysis reported no statistically significant difference in ZUF‐8 scores between the intervention and control groups (Hamann 2017).

Overall satisfaction at hospital discharge

Two studies conducted in psychiatric wards reported on overall satisfaction at hospital discharge, using CSQ‐8 and ZUF‐8 (Hamann 2006; Ishii 2017). Ishii 2017 (22 participants) reported on satisfaction at hospital discharge and found little or no difference between the two groups (MD 1.60, 95% CI ‐1.65 to 4.85; very low‐certainty evidence; Analysis 1.17).

Hamann 2006 also reported on satisfaction at discharge and found no statistically significant difference in ZUF‐8 scores between the intervention and control groups. However, this study did not adjust for clustering and so may have produced overly precise results.

Overall satisfaction at six months and more

Two studies reported on satisfaction at six months after the intervention using the Verona Service Satisfaction Scale ‐ European Version‐54 (VSSS‐EU‐54) and CSQ‐8 (Lovell 2018; Yamaguchi 2017). The pooled estimate of the SMD was 0.09 (95% CI ‐0.22 to 0.40, a small effect; 2 studies, 400 participants; very low‐certainty evidence; Analysis 1.18), indicating little or no difference between the two groups.

Accordingly, we are uncertain if SDM interventions improve the overall satisfaction of people with mental health conditions compared with no intervention due to very low‐certainty evidence.

Users’ satisfaction concerning decision‐making

None of the 15 studies examined the effect of shared decision‐making on users' satisfaction concerning decision‐making.

Users' satisfaction with received information

LeBlanc 2015 assessed satisfaction with received information immediately after the encounter. The assessment included these elements: right amount of information given; information given was clear; information given was helpful; strong desire to receive information this way for other treatment decisions; and strongly recommend the way information was shared to others, and the results were as follows:

Right amount of information given: RR 1.00 (95% CI 0.94 to 1.07; 241 participants; low‐certainty evidence; Analysis 1.19), indicating little or no difference between groups.
Information given was clear: RR 1.19 (95% CI 0.98 to 1.44; 241 participants; low‐certainty evidence; Analysis 1.20), indicating little or no difference between groups.
Information given was helpful: RR 1.33 (95% CI 1.08 to 1.65; 241 participants; moderate‐certainty evidence; Analysis 1.21), indicating an improvement in the group receiving an SDM intervention compared with control.
Strong desire to receive information this way for other treatment decisions: RR 1.35 (95% CI 1.08 to 1.68; 241 participants; moderate‐certainty evidence; Analysis 1.22), indicating an improvement in the group receiving an SDM intervention compared with control.
Strongly recommend the way information was shared to others: RR 1.32 (95% CI 1.11 to 1.58; 241 participants; moderate‐certainty evidence; Analysis 1.23), indicating an improvement in the group receiving an SDM intervention compared with control.

Therefore, SDM interventions probably improve users' satisfaction with received information (information given was helpful; strongly desire to receive information this way for other treatment decisions; and strongly recommend the way information was shared to others) compared with no intervention, based on moderate‐certainty evidence. However, there may be little or no effect on satisfaction with amount and clarity of information, based on low‐certainty evidence.

Carer satisfaction

One study (50 participants) reported on carer satisfaction using Carers and Users’ Expectations of Services—carer version (CUES‐C) and found little or no difference between groups (MD ‐1.40, 95% CI ‐6.69 to 3.89; low‐certainty evidence; Analysis 1.24) (Lovell 2018, see Table 8). Consequently, SDM interventions may have little or no effect on carer satisfaction compared with no intervention.

Table 8. Carer satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Lovell 2018	Carers and Users’ Expectations of Services ‐ carer version, CUES‐C	6 months	24	22.71	26	24.12	No difference

Healthcare professional satisfaction

Four studies examined professional satisfaction using a 5‐point Likert scale and a professional caregivers' job satisfaction questionnaire (JSQ) (Hamann 2006; LeBlanc 2015; Mariani 2018; Woltmann 2011; see Table 9). Two studies were included in the statistical analysis. On a continuous scale, the MD was 0.70 (95% CI 0.26 to 1.14; 1 study, 20 participants; Analysis 1.25), indicating an increase in professional satisfaction for the group that received SDM compared with control (Woltmann 2011). However, this is based on very low‐certainty evidence. On a categorical (original) scale, the RR was 1.35 (95% CI 1.16 to 1.58; 1 study, 256 participants; moderate‐certainty evidence; Analysis 1.26), indicating an increase in professional satisfaction for the group that received SDM compared with control (LeBlanc 2015).

Table 9. Healthcare provider satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Hamann 2006	5‐point Likert scale: overall satisfaction with what had been achieved during hospitalisation	At discharge	‐	3.8	‐	3.5	P = 0.02
LeBlanc 2015	Satisfied/extremely satisfied 1‐item, 5‐point Likert scale	Immediately after the clinical encounter	139	54%	117	76.3%	P = 0.02
Mariani 2018	Professional caregivers' job satisfaction questionnaire, JSQ	6 months	16	42.84 (SD 14.33)	18	43.33 (SD 10.97)	P = 0.576
Woltmann 2011	Case manager satisfaction, 6 statements related to satisfaction, a 5‐point Likert scale	After participation	10	4 (SD 0.5)	10	3.3 (SD 0.5)	P = 0.002

Hamann 2006 used a 5‐point Likert scale and reported that psychiatrists in the intervention group were more satisfied with what had been achieved during hospitalisation (SDM mean 3.8/comparison mean 3.5, P = 0.02). However, Hamann 2006 failed to adequately adjust for clustering and so may have produced overly precise results. Mariani 2018 used a professional caregivers' job satisfaction questionnaire and found no difference between the intervention and control groups (34 participants, SDM mean 42.8/comparison mean 43.3, P = 0.58). This study also failed to adequately adjust for clustering and so may have produced overly precise results.

Therefore, regarding healthcare professional satisfaction, the effects were mixed. Consequently, SDM interventions may have little or no effect on healthcare professional satisfaction measured continuously, compared with no intervention, due to low‐certainty evidence. On the other hand, SDM interventions probably improve healthcare professionals' satisfaction measured categorically, compared with no intervention, and based on moderate‐certainty evidence.

Knowledge

Three studies assessed patient knowledge regarding disease, condition, or treatment options (Hamann 2006; LeBlanc 2015; Woltmann 2011; see Table 10). Data from two studies were available for the statistical analysis (322 participants). The pooled estimate of SMD was 0.41 (95% CI 0.18 to 0.63, a moderate effect; very low‐certainty evidence; Analysis 1.27), indicating an improvement in the group receiving an SDM intervention compared with control.

Table 10. Knowledge

Study

Scale used

Timing

N SDM

SDM mean

N control

Control mean

Notes

Hamann 2006

Patient knowledge about their disease, original items with 7 multiple‐choice questions

At discharge

‐

15.0 (SD 4.4)

‐

10.9 (SD 5.4)

P = 0.01

n = 88 (total number of participants)

LeBlanc 2015

Overall knowledge including both tailored to information in the decision aid and generic information about depression

Immediately after the clinical encounter

138

63.5

116

56.3

P = 0.03

Woltmann 2011

Client knowledge of the care plan (plan goals recalled)

2 to 4 days after the care planning session

75%

57%

P = 0.02

SDM: shared decision‐making

In the study not pooled, the authors measured patient knowledge before discharge using an invalidated questionnaire with 7 multiple‐choice questions (Hamann 2006). Patients' knowledge in the intervention group as measured by this scale had improved at discharge (88 participants, SDM mean 15/comparison mean 10.9, P = 0.01) (Hamann 2006). However, Hamann 2006 failed to adequately adjust for clustering and so may have produced overly precise results.

Accordingly, we are uncertain if SDM interventions improve patient knowledge compared with no intervention due to very low‐certainty evidence.

Treatment or medication continuation

Clinic visits

Four studies reported on clinic visits using visit rate or visit frequency (Aoki 2019a; Hamann 2011; Ishii 2017; Mott 2014; see Table 11). Data from these four studies were used for the statistical analysis. For one to six months, Mott 2014 reported on psychotherapy visit rate at four months' follow‐up and found little or no difference between the groups (RR 0.98, 95% CI 0.37 to 2.59; 20 participants; very low‐certainty evidence; Analysis 1.28). For six months or more, three studies reported on clinic visit rate at six months' follow‐up (Aoki 2019a; Hamann 2011; Ishii 2017). They found little or no difference between the two groups (RR 1.07, 95% CI 0.93 to 1.23; 171 participants; very low‐certainty evidence; Analysis 1.29).

Table 11. Treatment continuation

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Adherence with outpatient visits	6 months	35	56%	53	51%	P = 0.656
Hamann 2011	“Has this patient shown up at your practice since being discharged from the hospital?” (Physicians answered yes/no)	6 months	32	94%	29	90%	P = 0.45
Hamann 2011	“Are you still in psychiatric treatment?” (Participants answered yes/no)	6 months	25	100%	23	91%	P = 0.22
Hamann 2011	"How much does this patient engage in planning for his or her therapy?" (Physicians answered)	6 months	25	3.5 (SD 0.9)	23	3.2 (SD 0.9)	P = 0.19
Ishii 2017	Whether a patient received outpatient psychiatric treatment within 30 days prior to follow‐up time on medical records	6 months	9	88.9%	13	69.2%	No difference
Loh 2007	Participant assessment of treatment adherence	6‐8 weeks	128	4.3 (SD 0.9)	66	3.9 (SD 1.0)	No difference
Loh 2007	Physician assessment of treatment adherence	6‐8 weeks	128	4.8 (SD 0.6)	66	4.3 (SD 1.1)	No difference
Mott 2014	Initiated psychotherapy visits 1 to 9	4 months	9	44%	11	45%	No difference

Treatment adherence by service users or healthcare providers

Two studies assessed treatment adherence by service users or healthcare providers (Loh 2007; Hamann 2011). Loh 2007 used two separate treatment adherence outcome measures at six to eight weeks after the intervention: a patient rating and a physician rating. Both were Likert‐type scales based on a single question. For the patient rating, there was little or no difference between the two groups (194 participants, SDM mean 4.3/comparison mean 3.9). For the physician rating, there was little or no difference between the two groups (194 participants, SDM mean 4.8/comparison mean 4.3). Hamann 2011 used two separate treatment adherence outcome measures at six months' follow‐up: a physician rating of adherence and a patient rating. The physician rating was a five‐point Likert scale based on a single question and the patient rating was based on categorical data (Yes/No data). For both patient and physician ratings, little or no differences were found between groups (Hamann 2011).

Accordingly, we are uncertain whether SDM interventions improve treatment continuation compared with no intervention due to very low‐certainty evidence.

Medication continuation from one to six months

Four studies examined medication continuation from one to six months (Aljumah 2015; Aoki 2019a; Hamann 2006; Hamann 2017; see Table 12), Two were included in the statistical analysis (Aljumah 2015; Aoki 2019a).

Table 12. Medication continuation

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aljumah 2015	Morisky Medication Adherence Scale, MMAS	3 months	110	5.79 (SD 1.89)	110	5.04 (SD 1.98)	P = 0.004
Aljumah 2015	Morisky Medication Adherence Scale, MMAS	6 months	110	5.99 (SD1.88)	110	4.94 (SD 1.94)	P < 0.0001
Aoki 2019a	Visual analogue scale, VAS	3 months	22	8.79 (SD 1.44)	44	8.57 (SD 1.60)	P = 0.910
Aoki 2019a	Visual analogue scale, VAS	6 months	22	8.58 (SD 1.44) )	44	8.44 (SD 1.62	P = 0.872
Hamann 2006	Estimated compliance from physician's point of view	At discharge	‐	1.7	‐	2.0	P > 0.05
Hamann 2006	Overall compliance determined by participant rated, physician rated, and plasma level	6 months after discharge	39	41%	47	55%	P > 0.05
Hamann 2006	Overall compliance determined by participant rated, physician rated, and plasma level	18 months after discharge	30	60%	38	58%	P > 0.05
Hamann 2011	“Are you still taking medication for your psychiatric condition?” (Participants answered yes/no)	6 months post hospital discharge	25	100%	23	87%	P = 0.10
Hamann 2011	"How do you estimate your patient’s compliance?" (Physician assessed)	6 months post hospital discharge	29	4.0 (SD 1.1)	26	4.2 (SD 0.9)	P = 0.78
Hamann 2017	Medication Adherence Rating Scale, MARS	6 months post hospital discharge	‐	2.6 (2.1)	‐	2.5 (2.2)	P = 0.72 n = 100 (total number of participants)
Hamann 2017	Medication Adherence Rating Scale, MARS	12 months post hospital discharge	‐	2.4 (2.1)	‐	2.8 (2.3)	P = 0.42 n = 85 (total number of participants)
LeBlanc 2015	Participants reported medication usage	After encounter	154	89.9%	134	79.1%	P = 0.15
LeBlanc 2015	Filled prescription within 30 days	For trial period	154	86.2%	134	93.2%	P = 0.19
LeBlanc 2015	% proportion of days covered (PDC) > 80% (of filled prescription)	For trial period	113	94.7%	93	97.8%	P = 0.67
Raue 2019	Initiation of antidepressant medication in the Cornel Service Index	12 weeks after intervention	103	23.3%	78	15.4%	P = 0.154
Yamaguchi 2017	Morisky Medication Adherence Scale, MMAS	6 months	26	5.7 (SD 1.5)	27	5.4 (SD 1.5)	P = 0.74

SDM: shared decision‐making

On a continuous scale, two measures were used (the Morisky Medication Adherence Scale (MMAS) and a visual analogue scale (VAS)). The SMD was 0.33 (95% CI 0.10 to 0.57, a small effect; 2 studies, 286 participants; Analysis 1.30), indicating an improvement in medication adherence for the group that received shared decision‐making compared with control. However, this is based on low‐certainty evidence. In a study not pooled in this meta‐analysis, Hamann 2017 (100 participants) reported no statistically significant difference in Medication Adherence Rating Scale (MARS) scores between the groups at six months' follow up (T = 0.36, P = 0.72).

On a categorical scale, one measure of adherence was used (overall adherence determined by patient rating with MARS, physician rating, and plasma level) (Hamann 2006). The RR was 0.74 (95% CI 0.47 to 1.17; one study, 86 participants; very low‐certainty evidence; Analysis 1.31), indicating little or no difference in medication continuation between groups.

Medication continuation at six months or more

Seven studies reported on medication continuation at six months or more (Aljumah 2015; Aoki 2019a; Hamann 2006; Hamann 2011; Hamann 2017; LeBlanc 2015; Yamaguchi 2017; see Table 12). Four studies were included in the statistical analysis. On a continuous scale, three measures were used (MMAS, VAS, service user estimated proportion of how much medicine taken). The SMD was 0.27 (95% CI ‐0.03 to 0.56, a small effect; 4 studies, 394 participants; low‐certainty evidence; Analysis 1.32), indicating little or no difference in medication continuation between groups. In a study not pooled in this meta‐analysis, Hamann 2017 (85 participants) reported no statistically significant difference in MARS scores between the groups at 12 months' follow‐up (T = ‐0.81, P = 0.42).

On a categorical scale, four measures were used (overall adherence determined by patient rating, physician rating using MARS scale, and plasma level; self‐reported if participant were taking medication for psychiatric condition, Yes/No; overall adherence determined by patient rating with Medication Adherence Questionnaire (MAQ), physician rating, and patient clinical visit; proportion of patients who filled their prescription within 30 days). The RR was 1.05 (95% CI 0.94 to 1.17; 4 studies, 577 participants; very low‐certainty evidence; Analysis 1.33), indicating little or no difference in medication continuation between groups.

Therefore, we are uncertain whether SDM interventions improve medication continuation compared with no intervention due to very low or low‐certainty evidence.

Treatment/medication continuation

Raue 2019 reported the combined proportion of possible antidepressant pills taken, and possible psychotherapy sessions attended over 12 weeks, with no difference reported between the groups (P = 0.154).

Carer participation or level of involvement in SDM process

One study reported on carer participation at six months using a patient‐reported outcome measure, PROM‐14 (Lovell 2018; see Table 13). The MD for this study was 3.60 (95% CI ‐0.99 to 8.19; 1 study; 68 participants; low‐certainty evidence; Analysis 1.34), indicating little or no difference between groups. Shared decision‐making interventions may therefore have little or no effect on carer participation or level of involvement compared with control.

Table 13. Carer participation in decision‐making

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Lovell 2018	User involvement in care planning (carers): Equip patient‐reported outcome measure, PROM‐14	6 months	22	20.1 (SD 8.0)	46	16.5 (SD 10.9)	P = 0.899

SDM: shared decision making

Relationship or interaction between service users and healthcare professionals

Relationship between service users and healthcare professionals, assessed by users

Four studies reported on the relationship between service users and healthcare professionals, as assessed by users, using the Trust in Physician Scale, California Psychotherapy Alliance Scale (CALPAS), and Scale to Assess Therapeutic Relationship (STAR) (Hamann 2011; Hamann 2017; Lovell 2018; Yamaguchi 2017; see Table 14). Data from three studies (457 participants) were available for the statistical analysis (Hamann 2011; Lovell 2018; Yamaguchi 2017).The pooled estimate of SMD was ‐0.13 (95% CI ‐0.54 to 0.28, a small effect; very low‐certainty evidence; Analysis 1.35), indicating little or no difference between two groups. One study reported on service user‐professional relationship assessed by users using the Trust in Physician Scale but no data were available to compute a standardised mean difference (Hamann 2017). This study reported that participants in the intervention group showed no decline in trust in their physicians compared with the control group (T = ‐1.15, P = 0.25) (Hamann 2017).

Table 14. Relationship between service users and healthcare providers

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Hamann 2006	Working Alliance Inventry, WAI (by physicians)	At discharge	‐	60.6	‐	69.0	P > 0.05
Hamann 2011	Difficult Doctor‐Patient Relationship Questionnaire, DDPRQ (by physician)	At discharge	32	40.4 (SD 7.6)	29	44.6 (SD 8.4)	P = 0.05
Hamann 2011	Trust in physician (by participant)	At discharge	32	41.8 (SD 7.4)	29	46.4 (SD 7.2)	P = 0.02
Hamann 2011	Therapeutic alliance (by physician)	Post intervention	32	23.8 (SD 4.7)	29	24.1 (SD 4.8)	P = 0.83
Hamann 2017	Difficult Doctor‐Patient Relationship Questionnaire, DDPRQ (by physician)	At discharge	‐	43.0 (SD 8.1)	‐	44 (SD 7.4)	P = 0.37
Hamann 2017	Trust in physician (by participant)	At discharge	‐	40.3 (SD 7.5)	‐	41.1 (SD 6.8)	P = 0.25
Lovell 2018	California Psychotherapy Alliance Scale, CALPAS	6 months	191	4.8 (SD 1.4)	152	4.9 (SD 1.5)	P = 0.949
Yamaguchi 2017	Relationship‐Scale to Assess Therapeutic Relationship, STAR. Positive collaboration (by clinician)	6 months	26	18.7 (SD 3.2)	27	17.7 (SD 3.9)	P = 0.07
Yamaguchi 2017	Relationship‐STAR, emotional difficulties (by clinician)	6 months	26	10.8 (SD 1.3)	27	10.5 (SD 1.2)	P = 0.59
Yamaguchi 2017	Relationship‐STAR, positive clinician input (by clinician)	6 months	26	9.9 (SD 1.7)	27	9.6 (SD 1.4)	P = 0.17
Yamaguchi 2017	Relationship‐STAR, positive collaboration (by participant)	6 months	26	19.4 (SD 5.6)	27	17.2 (SD 5.5)	P = 0.05
Yamaguchi 2017	Relationship‐STAR, positive clinician input (by participant)	6 months	26	9.0 (SD 2.2)	27	7.7 (SD 2.9)	P = 0.03
Yamaguchi 2017	Relationship‐STAR, non‐supportive clinician input (by participant)	6 months	26	10.4 (SD 2.4)	27	10 (SD 2.6)	P = 0.69

Relationship between service users and healthcare professionals, assessed by professionals

Four studies reported on the relationship between service users and healthcare professionals, as assessed by professionals, using the Working Alliance Inventory (WAI), Difficult Doctor‐Patient Relationship Questionnaire (DDPRQ), Therapeutic Alliance scale, and Scale to Assess Therapeutic Relationship (STAR) (Hamann 2006; Hamann 2011; Hamann 2017; Yamaguchi 2017; Table 14). Data from two studies (114 participants) were available for statistical analysis (Hamann 2011; Yamaguchi 2017). The pooled estimate of SMD was 0.17 (95% CI ‐0.31 to 0.65, a small effect; very low‐certainty evidence; Analysis 1.36), indicating little or no difference between two groups. Of the two studies not used for the statistical analysis, Hamann 2006 reported that participants in the intervention group did not differ from those in the control group in cooperation, as reflected in the WAI (therapist version) (mean 60.6/60.9, P > 0.05). This study did not adjust for clustering and so may have produced overly precise results. Hamann 2017 reported on therapeutic alliance and found no difference in DDPRQ scores between the intervention and control groups (T = ‐0.90, P = 0.37)

Accordingly, we are uncertain whether SDM interventions improve relationships or interactions between service users and healthcare professionals compared with control as results are based on very low‐certainty evidence.

Health service use outcomes

Length of consultation

Three studies reported on consultation duration or time spent in individual contacts between psychiatrist and patient (Aoki 2019a; Hamann 2006; Loh 2007; see Table 15). Data from two studies (282 participants) assessing consultation duration (minutes) were available for the statistical analysis (Aoki 2019a; Loh 2007). The pooled estimate of SMD was 0.09 (95% CI ‐0.24 to 0.41, a small effect; moderate‐certainty evidence; Analysis 1.37), indicating little or no difference between the two groups. Although the authors of the Aoki 2019a trial reported that the median duration of the SDM intervention was 26 minutes and 24 minutes for control, we used mean duration, which the authors provided us, for the statistical analysis in this review. In the study not pooled in the meta‐analysis, Hamann 2006 reported that the participants of the intervention group did not differ from those in the control group in the time spent in individual contact with psychiatrists, as reported by the participant (SDM mean 64 minutes/control mean 60 minutes, P > 0.05). However, Hamann 2006 failed to adequately adjust for clustering and so may have produced overly precise results.

Table 15. Health service use outcomes

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Consultation duration (minutes)	During initial consultation	35	28.71 (SD 12.66)	53	30.49 (SD 15.91)	P = 0.983
Hamann 2006	Rating of time spent per week with participant (minutes)	At discharge	‐	64.0	‐	60.0	P > 0.05
Ishii 2017	Length of stay (days)	At hospital discharge	9	66.7 (SD 40.4)	13	66.5 (SD 17.4)	No difference
Loh 2007	Consultation duration (minutes)	During consultation	128	29.2 (SD 10.7)	66	26.7 (SD 12.5)	No difference

SDM: shared decision‐making

Therefore, SDM interventions probably have little or no effect on length of consultation compared with no intervention, based on moderate‐certainty evidence.

Length of hospital stay

One study conducted in a psychiatric ward reported on length of hospital stay (days) (Ishii 2017; see Table 15). The MD for this study was 0.20 (95% CI ‐27.84 to 28.24; 22 participants; very low‐certainty evidence; Analysis 1.38), indicating little or no difference between groups. Accordingly, we are uncertain whether SDM interventions improve length of hospital stay compared with no intervention due to very low‐certainty evidence.

Adverse outcomes

There were no adverse effects on health outcomes and no other adverse events reported.

Subgroup analysis and investigation of heterogeneity

Insufficient studies prevented the planned subgroup analyses of care setting, types of intervention target population, mental illness severity of people with mental health conditions, and intervention types. However, two studies used cognitive training in the control group (Hamann 2011; Hamann 2017). The other 13 studies had usual care as the control group. Therefore, we conducted subgroup analysis according to control group (SDM versus usual care and SDM versus cognitive training) for the primary outcomes as follows.

Clinical outcomes: readmission (one to six months)

When considered separately by subgroups for SDM versus usual care and SDM versus cognitive training, readmission (one to six months) showed no difference between studies that used usual care versus those that used cognitive training as the control group (RR 1.03 versus RR 1.12; test for subgroup difference P = 0.91, I² = 0%; Analysis 1.10).

Clinical outcomes: readmission (six months or more)

When analysing SDM versus usual care and SDM versus cognitive training, readmission (six months or more) showed no difference between studies that used usual care versus those that used cognitive training as the control group (RR 1.14 versus RR 1.00; test for subgroup difference P = 0.69, I² = 0%; Analysis 1.11).

For other outcomes, we could not conduct formal subgroup analyses because there were too few studies in each subgroup.

Sensitivity analysis

As previously described, we conducted sensitivity analyses for depression symptoms, removing lower‐certainty studies (high overall risk of bias) for the analyses.

Discussion

Summary of main results

In this update review, we added 13 new studies to the two studies from the original Cochrane Review for a total of 15 studies comparing SDM for mental health conditions to usual care (12 studies), cognitive training (two studies), or 30‐minute placebo session (one study).

The 15 studies recruited a total of 3141 people with mental health conditions. The number of included studies has increased considerably in ten years (i.e. the time elapsed between the previous and present versions of this review). This suggests that this field has been garnering attention and rapidly expanding. Although the majority of the countries included were in Europe or the USA, three studies were in Japan. This shows that SDM in psychiatry has been attracting a lot of attention in middle‐ and upper‐income countries. We also observed one study by international collaborators. Regarding the setting, various fields – including primary care, outpatient services, community care, or psychiatric wards – were represented. The clinical conditions also covered many kinds of mental health conditions, such as schizophrenia, depression, bipolar disorder, dementia, and post‐traumatic stress disorder (PTSD). Although SDM is a common key concept for all included studies, the content of interventions, such as the duration and healthcare providers, varied.

Primary Outcomes

There were little or no differences in effects between groups receiving the intervention or control for clinical outcomes, such as psychiatric symptoms, depression, anxiety, and readmission.

We are uncertain if SDM interventions for people with mental health conditions improve observed level of participation in the SDM process, but this approach may increase the service users' self‐reported participation or level of involvement in the decision‐making process compared with usual care in the short term. There was insufficient evidence for sustained participation in the decision‐making process over the longer term.

Moreover, we graded the certainty of the evidence for most of the primary outcomes described above as low or very low, which means that results are likely to change with more research.

Secondary outcomes

We are uncertain about the effects of SDM interventions on recovery.

For service users' satisfaction, while one study which assessed some aspects of users' satisfaction with received information immediately after the encounter using categorical measurements found that those receiving the SDM intervention are probably more satisfied, we are uncertain about effects on overall satisfaction levels. No included studies reported users’ satisfaction with decision‐making.

Regarding carer satisfaction, one study reported there may be little or no difference between two groups.

Several studies assessed healthcare professionals' satisfaction and the results were mixed: some suggested there may be little or no effect on satisfaction levels, while one study using another measure found that SDM interventions probably improve healthcare professionals' satisfaction compared with no intervention.

Regarding patient knowledge, although three studies showed that participants' knowledge in SDM groups had improved compared with control, this is a small effect and very low‐certainty evidence. Therefore, we are uncertain about effects.

The results were mixed for treatment or medication adherence in the short term, and findings were based on low‐ or very low‐certainty evidence. Accordingly, we are uncertain about the effects. We are also uncertain about both treatment and medication adherence over the longer term.

There may be little or no difference in carer participation, and we are uncertain about the effects on the relationship between service users and healthcare professionals, and health service use outcomes, such as length of hospital stay. Shared decision‐making interventions probably have little or no effect on consultation length.

No adverse events were reported.

Overall completeness and applicability of evidence

This review appears to highlight two benefits for clinical settings. First, people with mental health conditions receiving the SDM intervention may be more involved in the decision‐making process, compared with usual care. Second, there was probably no difference between intervention and control groups with regard to the consultation duration. Shared decision‐making emphasises the process of conversation between the service user and healthcare provider. Concerns are then sometimes raised that SDM interventions may prolong the consultation duration. Accordingly, our results may help to address these concerns.

However, overall, for most outcomes of interest in this review, the effects were of small or negligible size, and the evidence was of low or very low certainty. Thus, there is a low level of certainty about the findings based on the studies assembled thus far. On the other hand, the benefits described above could be nonetheless clinically important. This is because greater levels of involvement in the decision‐making process appeared to be consistent with the concept of personal recovery in the mental health field, which consists of elements such as re‐establishment of identity, finding meaning in life, empowerment, and sense of responsibility (Van Eck 2018). The recovery process places control in the hands of the individual and not the professional (Jacobson 2001). Accordingly, it is worth mentioning that SDM interventions for people with mental health conditions are increasing, and greater emphasis is being placed on the collaborative nature of interactions among healthcare providers, people with mental health conditions, and their families. In addition, the findings of this update review also suggest that people receiving usual care may not be as involved in the decision‐making processes as they wish. This means that there is a need to continue exploring further interventions to promote service users' involvement and autonomy in this field.

We found that a variety of scales was used to measure service users' involvement in decision‐making processes. This indicates that there is not yet consensus on a standardised scale to measure the level of service user involvement in SDM interventions for people with mental illness, which may contribute to the variability in effects across studies.

The meta‐analyses showed considerable heterogeneity for almost all outcomes. The heterogeneity of reporting possibly may be due to the fact that SDM interventions are complex and that complexity is reflected in the range of scales and approaches to measurement in use. It is notable that there was also considerable diversity in the components of the SDM interventions adopted by the included studies. These variations included issues of timing (such as those that were implemented during consultation versus those that required the service user to prepare before consultation), use of tools (such as those that used decision support tools versus those that did not), and variability in whom the intended target of the SDM intervention was (those that involved only the physician versus those involved an interprofessional team). Considerable diversity was also found in follow‐up periods for outcome assessment. Many studies were also underpowered to detect important differences in outcomes. Heterogeneity in the various outcomes may also reflect the inclusion of clinically diverse studies in this review update. Therefore, it should be remembered that the pooled effect estimates may not be applicable across the board (e.g. to different persons, mental health conditions, or situations).

While there was diversity in the SDM interventions studied, most participants included were adults with severe mental illnesses, such as schizophrenia, depression, and bipolar disorder, in higher‐income countries (Europe, the USA, and Japan). We did not find any studies which included children or those in low‐ and middle‐income countries. People with dementia were included in only one trial (Mariani 2018), although the older population is increasing globally. Shared decision‐making interventions targeting children with mental health conditions have been reported (Brinkman 2011; Abrines‐Jaume 2016; Levy 2016; Liverpool 2021b), but there are as yet no available RCTs through which effects of these interventions in children might be determined. Autonomy and self‐determination of these vulnerable populations should be advocated from the standpoint of recovery. Thus, further research for these under‐researched populations, including children and people in low‐ and middle‐income countries, is also needed.

Implementation of SDM interventions in the clinical environment requires consideration in terms of healthcare costs, although there were no studies which evaluated cost‐effectiveness of the interventions in this update review. Cost‐effectiveness should be assessed and examined in future trials.

Whether the intervention was implemented with fidelity is an important consideration when assessing the outcomes of SDM interventions. However, this aspect of implementation was not clearly reported, except by one study (Yamaguchi 2017).

Furthermore, the number of identified studies was relatively small overall. This limited our ability to conduct further analyses, such as subgroup analyses, to further investigate potential modifiers of the effects of SDM interventions.

Certainty of the evidence

We assessed the methodological risk of bias of included studies in this update review in accordance with the Cochrane Handbook (Higgins 2011), and used the GRADE criteria to rank the certainty of the evidence (Schünemann 2011).

GRADE appraisal of the certainty of the evidence indicated low‐ or very low‐certainty evidence for almost all outcomes in this updated review. We provide a summary of the reasons for downgrading below.

We assessed the methodological limitations of included studies and rated several studies as having a high risk of bias. It should also be noted that we rated many of the studies as 'unclear risk' for several items as we did not have enough information to assess the risk of bias based on the information available to us. We rated five or more studies as at unclear or high risk of bias for the key items of sequence generation and allocation concealment. Regarding blinding of participants, none of the studies could be rated as 'low risk' of bias due to the nature of the intervention; we rated all as 'high risk'. For blinding of outcome assessors, we rated only four studies as having a low risk of bias, and the remainder as having a high or unclear risk of bias. In 12 of the 15 studies, the risk of selective outcome reporting was high or unclear (the majority of studies had no published protocol), indicating that bias may be present due to not reporting all findings. Because of the small number of studies that assessed common outcomes, it was not possible to analyse publication bias due to failure to publish negative studies.

For imprecision, several (six of 15 studies) lacked statistical power because of the small sample sizes studied. We also found statistically significant levels of heterogeneity in several outcomes and this was a common reason for downgrading the certainty of the evidence. For example, there was high heterogeneity in depressive symptoms and service user involvement, which were measured with different scales. Several knowledge scales that were not standardised were used. Moreover, only five of the 15 included studies assessed the primary outcome; namely, the extent to which interventions can engage service users in the decision‐making process. Furthermore, we found that these five studies did not use a common scale, but rather various kinds of scales. Outcomes measured in very different ways was a common reason for downgrading the evidence due to indirectness. Thus, we expect that the certainty of the evidence in this area will improve if researchers develop or recognise (or both), and use, validated common measurements to assess the impact of interventions.

Moreover, the 15 studies varied in the setting, the diseases targeted, the nature of the decision‐making about what to choose, the elements of decision support provided to service users including decision support tools, the type of comparison provided (the content of the intervention compared with usual care), and the targeted outcome measures. This too contributed to inconsistency across studies.

For the reasons described above, the overall evidence certainty of the results of this update review is low or very low, which limits our confidence in the results. In conclusion, more and better studies are needed to increase the certainty of the evidence in this field and to inform decisions about implementation of SDM interventions in mental health services.

Potential biases in the review process

Although we took every effort to minimise the potential for biases in the review process, three sources of potential bias may exist.

First, while our searches were comprehensive, there is a possibility that some relevant studies were missed for assessment by the review.

Second, a potential bias in reviews in this area is the adoption of clear criteria for what constitutes an SDM intervention for people with mental health conditions. We inherited the original review and clearly defined SDM based on Charles and colleagues' criteria (Charles 1997). This allowed us to establish a standard procedure for conducting this update review. Although SDM research has received a lot of attention, not only in this area, and the overall number of related publications has been increasing over the years, different researchers often use various definitions of SDM (Makoul 2006). For example, even when authors define SDM, there may be no choice or decision‐making involved, only the facilitation of communication to encourage service users to speak up in the consultation (Moncrieff 2016), or motivational interviewing, which aims to increase motivation for a particular treatment (Ludman 2002). Shared decision making requires equipoise in decision‐making (Elwyn 2006). That is, there is a range of possible and appropriate treatment options. In the process of choosing one of the options (including choosing none), the preferences and values of the service user and the health care provider regarding the options are clarified, and it is essential to make decisions based on those preferences and values. What SDM 'looks like' in mental health decision‐making may nonetheless be potentially somewhat different from how it is encountered in other healthcare areas (Zisman‐Ilani 2017). This may make selecting the studies for this review open to bias.

Third, the review included studies undertaken by some review authors. Assessment for inclusion, data extraction, and certainty assessment of these studies was undertaken by review authors not involved in the primary studies, in order to minimise any potential bias. On the other hand, this is also a strength of this update because the review team is composed of individuals with experience in SDM for those with mental health conditions.

Agreements and disagreements with other studies or reviews

There are some systematic and narrative reviews related to this topic. A review by James and Quirk identified a ‘rationale for SDM’ as any argument or reason for SDM in mental health care outlined by authors in a journal paper and which described the rationales (James 2017). Their review excluded raw data or outcome findings of experimental trials. A review by Zisman‐Ilani and colleagues indicated unique elements of SDM in mental health, such as facilitating patient motivation and providing patient communication skills training, which were rarely seen in other medical fields (Zisman‐Ilani 2017). However, this review was also descriptive and did not attempt statistical synthesis of the outcomes. Stovell and colleagues conducted a systematic review of shared treatment decision‐making for psychosis, which identified 11 RCTs and showed small beneficial effects on indices of treatment‐related empowerment (Stovell 2016). However, given its focus on treatment decisions concerning psychosis, this review did not consider other mental health conditions and rehabilitation or care plan decisions beyond medical treatments.

A Cochrane Review regarding interventions for increasing the use of SDM by healthcare professionals has been completed (Légaré 2018). The review suggests that interventions by health care providers to promote the use of SDM may slightly improve participants' mental health‐related quality of life compared with usual care, with little or no difference in physical health‐related quality of life (Légaré 2018). A Cochrane Review focusing on the effects of decision aids, a tool which may promote SDM, has also been conducted and updated periodically to reflect recent evidence (Stacey 2017). This review found that in a variety of decision‐making situations, people who received the decision support tool intervention gained more knowledge and were better able to clarify their values compared with those receiving usual care. Those who used the tools also took a more active role in decision‐making, a finding aligned with our results in this update review indicating that participants may perceive greater levels of involvement.

A Cochrane Review that aimed to determine the effects of decision coaching was published in 2021 (Jull 2021). This was the first version of the review and included 28 studies, which suggests that decision support interventions are recently gathering much attention. The review found that decision coaching did not indicate any adverse effects and may improve participants' knowledge.

In addition, SDM interventions have attracted attention in various areas dealing with physical diseases. A Cochrane Review exploring whether SDM interventions reduce the use of antibiotics for acute respiratory infections in primary care has been published (Coxeter 2015). The review found that SDM interventions significantly reduced antibiotic prescriptions for acute respiratory infections, compared with usual care (Coxeter 2015). Another Cochrane Review evaluated the effects and harms of SDM interventions in asthma treatment. The number of studies included in the review was relatively small, and each study was different, so meta‐analysis was not possible (Kew 2017). However, evidence from individual studies indicated that SDM may improve quality of life and asthma control, and may reduce medical visits for asthma (Kew 2017).

Although this update review did not find the effects regarding improvement of clinical symptoms due to low‐ or very low‐certainty evidence, evaluations of these SDM interventions in the other somatic areas suggest that SDM interventions may potentially improve clinical outcomes. Efforts to promote consumer involvement in health and decision‐making and to enable the delivery of more person‐centred care should continue, and focusing on SDM interventions and their use in practice may be one means of promoting care that aligns with these principles.

There is a growing number of SDM interventions in various areas and reviews are being conducted. Overall, however, the certainty of the evidence seems to be moderate, low, or very low, and we are not yet convinced of the effectiveness of SDM interventions in any area for improving health outcomes. More and better studies, including agreed components of SDM and core outcomes and measures, are needed to increase the certainty of the evidence in this field. Shared decision‐making continues to be supported from values‐based healthcare and ethical perspectives, has gained policy prominence, and major guidelines have been published to promote its more routine use in clinical practice (NICE 2009; NICE 2015). We need to continue making efforts to implement the recommendations of these guidelines into clinical practice.

Figure 1

PRISMA study flow diagram for initial search in January 2020

Figure 2

PRISMA study flow diagram for update search in February 2022

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Analysis 1.1

Comparison 1: Shared decision‐making versus control, Outcome 1: Clinical outcomes ‐ psychiatric symptoms

Analysis 1.2

Comparison 1: Shared decision‐making versus control, Outcome 2: Clinical outcomes ‐ depression (1 to 6 months)

Analysis 1.3

Comparison 1: Shared decision‐making versus control, Outcome 3: Clinical outcomes ‐ depression (1 to 6 months) ‐ sensitivity analysis removing low‐quality studies

Analysis 1.4

Comparison 1: Shared decision‐making versus control, Outcome 4: Clinical outcomes ‐ depression (6 months or more)

Analysis 1.5

Comparison 1: Shared decision‐making versus control, Outcome 5: Clinical outcomes ‐ depression (6 months or more) ‐ sensitivity analysis removing low‐quality studies

Analysis 1.6

Comparison 1: Shared decision‐making versus control, Outcome 6: Clinical outcomes ‐ depression remission (1 to 6 months)

Analysis 1.7

Comparison 1: Shared decision‐making versus control, Outcome 7: Clinical outcomes ‐ depression remission (6 months or more)

Analysis 1.8

Comparison 1: Shared decision‐making versus control, Outcome 8: Clinical outcomes ‐ depression response (1 to 6 months)

Analysis 1.9

Comparison 1: Shared decision‐making versus control, Outcome 9: Clinical outcomes ‐ depression response (6 months or more)

Analysis 1.10

Comparison 1: Shared decision‐making versus control, Outcome 10: Clinical outcomes ‐ readmission rates (1 to 6 months)

Analysis 1.11

Comparison 1: Shared decision‐making versus control, Outcome 11: Clinical outcomes ‐ readmission rates (6 months or more)

Analysis 1.12

Comparison 1: Shared decision‐making versus control, Outcome 12: Participation ‐ observations on the process of SDM

Analysis 1.13

Comparison 1: Shared decision‐making versus control, Outcome 13: Participation ‐ SDM‐specific user‐reported outcomes from encounters (immediately after intervention)

Analysis 1.14

Comparison 1: Shared decision‐making versus control, Outcome 14: Participation ‐ SDM‐specific user‐reported outcomes from encounters (6 months or more)

Analysis 1.15

Comparison 1: Shared decision‐making versus control, Outcome 15: Recovery

Analysis 1.16

Comparison 1: Shared decision‐making versus control, Outcome 16: Satisfaction ‐ overall users' satisfaction immediately after intervention

Analysis 1.17

Comparison 1: Shared decision‐making versus control, Outcome 17: Satisfaction ‐ overall users' satisfaction at hospital discharge

Analysis 1.18

Comparison 1: Shared decision‐making versus control, Outcome 18: Satisfaction ‐ overall users' satisfaction in 6 months or more

Analysis 1.19

Comparison 1: Shared decision‐making versus control, Outcome 19: Satisfaction ‐ users' satisfaction with received information: right amount of information (categorical)

Analysis 1.20

Comparison 1: Shared decision‐making versus control, Outcome 20: Satisfaction ‐ users' satisfaction with received information: information given was clear (categorical)

Analysis 1.21

Comparison 1: Shared decision‐making versus control, Outcome 21: Satisfaction ‐ users' satisfaction with received information: information given was helpful (categorical)

Analysis 1.22

Comparison 1: Shared decision‐making versus control, Outcome 22: Satisfaction ‐ users' satisfaction with received information: strongly desire to receive information this way for other treatment decisions (categorical)

Analysis 1.23

Comparison 1: Shared decision‐making versus control, Outcome 23: Satisfaction ‐ users' satisfaction with received information: strongly recommend the way information was shared to others (categorical)

Analysis 1.24

Comparison 1: Shared decision‐making versus control, Outcome 24: Satisfaction ‐ carer satisfaction

Analysis 1.25

Comparison 1: Shared decision‐making versus control, Outcome 25: Satisfaction ‐ healthcare professional satisfaction

Analysis 1.26

Comparison 1: Shared decision‐making versus control, Outcome 26: Satisfaction ‐ healthcare professional satisfaction (categorical)

Analysis 1.27

Comparison 1: Shared decision‐making versus control, Outcome 27: Knowledge

Analysis 1.28

Comparison 1: Shared decision‐making versus control, Outcome 28: Treatment continuation ‐ clinic visits (1 to 6 months)

Analysis 1.29

Comparison 1: Shared decision‐making versus control, Outcome 29: Treatment continuation ‐ clinic visits (6 months or more)

Analysis 1.30

Comparison 1: Shared decision‐making versus control, Outcome 30: Medication continuation (1 to 6 months)

Analysis 1.31

Comparison 1: Shared decision‐making versus control, Outcome 31: Medication continuation (1 to 6 months) (categorical)

Analysis 1.32

Comparison 1: Shared decision‐making versus control, Outcome 32: Medication continuation (6 months or more)

Analysis 1.33

Comparison 1: Shared decision‐making versus control, Outcome 33: Medication continuation (6 months or more) (categorical)

Analysis 1.34

Comparison 1: Shared decision‐making versus control, Outcome 34: Carer participation

Analysis 1.35

Comparison 1: Shared decision‐making versus control, Outcome 35: Relationship between service users and healthcare professionals, assessed by users

Analysis 1.36

Comparison 1: Shared decision‐making versus control, Outcome 36: Relationship between service users and healthcare professionals, assessed by healthcare professionals

Analysis 1.37

Comparison 1: Shared decision‐making versus control, Outcome 37: Health service use outcomes ‐ length of consultation

Analysis 1.38

Comparison 1: Shared decision‐making versus control, Outcome 38: Health service use outcomes ‐ length of hospital stay

Summary of findings 1. Shared decision‐making interventions compared with usual care for people with mental health conditions

Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Shared decision‐making interventions compared with usual care for people with mental health conditions
Patient or population: people with mental health conditions Setting: various Intervention: shared decision‐making Comparison: usual care, cognitive training, placebo session
Outcomes	Risk with usual care	Risk with SDM intervention	Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Psychiatric symptoms Brief Psychiatric Rating Scale (BPRS; Overall 1988): 16 items with 7‐item Likert scale ('not present' to 'extremely severe') measured 6 months after intervention (Yamaguchi 2017).		MD ‐1.10 lower (‐5.54 lower to 3.34 higher)	‐	53 (1 RCT)	⊕⊝⊝⊝ Very low^a,b	Higher scores indicate more severe psychiatric symptoms; the results indicate little or no difference between groups. One further study that could not be pooled reported no statistically significant difference in PANSS scores between the groups when they were discharged from hospital (Hamann 2006).
Depression (1 to 6 months) Montgomery–Åsberg Depression Rating Scale (MADRS; Montgomery 1979): measured 3 months after intervention (Aljumah 2015). Patient Health Questionnaire‐9 (PHQ‐9; Kroenke 2001): measured 6 to 8 weeks after intervention (Loh 2007) or 3 months after intervention (LeBlanc 2015). Quick Inventory of Depressive Symptomatology Self‐Report (QIDS‐J; Rush 2003): measured 3 months after intervention (Aoki 2019a).		SMD ‐0.03 lower (‐0.17 lower to 0.12 higher)	‐	717 (4 RCTs)	⊕⊕⊝⊝ Low^c,d	Higher scores indicate more severe depression symptoms; the results indicate little or no difference between groups.
Depression (6 months or more) MADRS measured 6 months after intervention (Aljumah 2015). PHQ‐9 measured 6 months after intervention (LeBlanc 2015). QIDS‐J measured 6 months after intervention (Aoki 2019a). Hospital Anxiety and Depression Scale (HADS‐D; Zigmond 1983) measured 6 months after intervention (Lovell 2018).		SMD 0.03 higher (‐0.10 lower to 0.17 higher)	‐	898 (4 RCTs)	⊕⊕⊝⊝ Low^c,d	Higher scores indicate more severe depression symptoms; the results indicate little or no difference between groups.
Readmission (6 months or more) Rehospitalisation at 8 months after discharge (Hamann 2006) or 12 months after discharge (Hamann 2017).	Study population		RR 1.06 (0.77 to 1.46)	249 (2 RCTs)	⊕⊝⊝⊝ Very low^c‐e	‐
	362 per 1000ⁱ	384 per 1000 (279 to 529)	RR 1.06 (0.77 to 1.46)	249 (2 RCTs)	⊕⊝⊝⊝ Very low^c‐e	‐
Participation (observations on the process of SDM) Observing Patient Involvement in shared decisiON‐making (OPTION; Elwyn 2005) assessed from video recording on the encounter (LeBlanc 2015). Core components of SDM: scoring the transcripts of conversations between participants and doctors during consultation (SDM‐18; Salyers 2012) during consultation (Yamaguchi 2017).		SMD 1.14 higher (0.63 higher to 1.66 higher)	‐	133 (2 RCTs)	⊕⊝⊝⊝ Very low^c,f,g	Higher scores indicate more involvement in decision‐making; the results indicate an increase in involvement for the SDM group.
Participation (SDM‐specific‐reported outcomes, immediately after intervention) Combined Outcome Measure for Risk Communication and Treatment Decision‐making Effectiveness (COMRADE; Edwards 2003) measured immediately after decision‐making (Aoki 2019a). Decisional Conflict Scale (DCS; O'Connor 1995a) measured immediately after the clinical encounter (LeBlanc 2015). Man‐Song‐Hing Scale (Man‐Song‐Hing 1999) measured after intervention (Loh 2007).		SMD 0.63 higher (0.26 higher to 1.01 higher)	‐	534 (3 RCTs)	⊕⊕⊝⊝ Low^c,h	COMRADE, Man‐Song‐Hing Scale: higher scores indicate more involvement in decision‐making; DCS: lower scores indicate less decisional conflict; the results indicate an increase in involvement for the SDM group. In one further study that could not be pooled, participants in the intervention group reported significantly greater perceived involvement than those in the control group (Hamann 2006).
Adverse events ‐ not reported	There were no adverse effects reported.		‐	‐	‐	‐
*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; PANSS: Positive and Negative Syndrome Scale; RR: risk ratio; SDM: shared decision‐making
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.
^aDowngraded by one level: for risk of bias (high risk of bias for blinding of participants and outcome assessment) ^bDowngraded by two levels: for imprecision (insufficient number of participants for one study and large confidence interval) ^cDowngraded by one level: for indirectness (the outcome was measured using various approaches) ^dDowngraded by one level: for imprecision (large confidence interval) ^eDowngraded by two levels: for risk of bias (1 of 2 studies were at high risk of bias for randomisation and allocation concealment, and 2 of 2 studies were at high risk of bias for blinding of participants and outcome assessment) ^fDowngraded by two levels: for imprecision (small sample size) ^gDowngraded by one level: for risk of bias (2 of 2 studies were at high risk for blinding of participants and outcome assessment) ^hDowngraded by one level: for inconsistency (I² ≥ 50%; P value for heterogeneity ≤ 0.05) ⁱControl event rate calculated from means of usual care groups used in this analysis (Hamann 2006; Hamann 2017)

Summary of findings 1. Shared decision‐making interventions compared with usual care for people with mental health conditions

Table 1. Psychiatric symptoms

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Note
Hamann 2006	Positive and Negative Syndrome Scale, PANSS	At discharge	‐	58	‐	59.3	P > 0.05
Yamaguchi 2017	Brief Psychiatric Rating Scale, BPRS	After 6 months' follow‐up	26	34.0 (SD 7.9)	27	35.1 (SD 8.6)	P = 0.31
SDM: shared decision‐making

Table 1. Psychiatric symptoms

Table 2. Depression

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Aljumah 2015	Montgomery‐Åsberg Depression Rating Scale, MADRS	3 months	110	21.07 (SD 12.21)	110	21.01 (SD 12.63)	P = 0.971
Aljumah 2015	Montgomery‐Åsberg Depression Rating Scale, MADRS	6 months	110	20.65 (SD 11.97)	110	20.86 (SD 12.54)	P = 0.897
Aoki 2019a	Quick Inventory of Depressive Symptomatology, QIDS‐J	3 months	35	10.49 (SD 5.12)	53	9.57 (SD 5.80)	No difference
Aoki 2019a	Quick Inventory of Depressive Symptomatology, QIDS‐J	6 months	35	10.34 (SD 5.68)	53	10.36 (SD 6.17)	No difference
LeBlanc 2015	Patient Health Questionnaire‐9, PHQ‐9	3 months	114	9.0	101	9.2	P = 0.78
LeBlanc 2015	Patient Health Questionnaire‐9, PHQ‐9	6 months	109	8.9	101	9.3	P = 0.91
LeBlanc 2015	Remission rate, PHQ score < 5	3 months	114	19.6%	101	18.7%	P = 0.85
LeBlanc 2015	Remission rate, PHQ score < 5	6 months	109	21.5%	101	14.4%	P = 0.18
LeBlanc 2015	Responsiveness, > 50% PHQ‐9 improvement	3 months	114	33.5%	101	30.9%	P = 0.77
LeBlanc 2015	Responsiveness, > 50% PHQ‐9 improvement	6 months	109	34.8%	101	27.3%	P = 0.15
Loh 2007	Patient Health Questionnaire‐9, PHQ‐9	6‐8 weeks	128	13.7 (SD 5.8)	66	14.6 (SD 5.3)	P = 0.610
Lovell 2018	The Depression subscale of the Hospital Anxiety and Depression Scale, HADS‐D	6 months	208	9.8 (SD 5.5)	172	8.9 (SD 5.8)	P = 0.963
Raue 2019	Hamilton Depression Rating Scale, HDRS	8 weeks	114	12.4 (0.8)	88	11.7 (1.0)	No difference
Raue 2019	Hamilton Depression Rating Scale, HDRS	12 weeks	114	12.9 (0.8)	88	12.0 (0.9)	No difference
SDM: shared decision‐making

Table 2. Depression

Table 3. Anxiety

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Lovell 2018	The Anxiety subscale of the Hospital Anxiety and Depression Scale, HADS‐A	6 months	208	12.1 (SD 5.4)	172	10.9 (SD 5.9)	P = 0.339
SDM: shared decision‐making

Table 3. Anxiety

Table 4. Readmission rate

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Hamann 2006	Rehospitalisation rate	6 months	36	22%	37	22%	P > 0.05
Hamann 2006	Rehospitalisation rate	18 months	38	53%	41	46%	P > 0.05
Hamann 2011	Rehospitalisation rate	6 months	29	17%	26	15%	P = 0.57
Hamann 2017	Rehospitalisation rate	12 months	95	31%	75	31%	P = 0.98
SDM: shared decision‐making

Table 4. Readmission rate

Table 5. Participation (by the person with mental health condition) or level of involvement in the decision‐making process

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Aoki 2019a	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE communication	After decision‐making	32	median 44	53	median 38	P < 0.001
Aoki 2019a	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE confidence	After decision‐making	32	median 41	53	median 37	P = 0.005
Hamann 2006	Combined Outcome Measure for Risk communication And treatment Decision making Effectiveness, COMRADE total	After the intervention		79.5		69.7	P = 0.03 n = 75 (Total number of participants)
LeBlanc 2015	Decisional Conflict Scale, DCS (0 = conflict, 100 = comfort)	After encounter	138	79.7	114	74.5	P = 0.01
LeBlanc 2015	Participation‐Involvement patient, OPTION (observation)	Assessed from video recording on the encounter	57	46.6	39	32.5	P = 0.01
Loh 2007	Man‐Song‐Hing Scale	After intervention	128	28.0 (SD 2.9)	66	25.5 (SD 3.0)	P = 0.003
Lovell 2018	Equip patient reported outcome measure, EQUIP PROM‐14	6 months	192	21.3 (SD 9.6)	153	21.6 (SD 11.2)	P = 0.715
Yamaguchi 2017	Core components of SDM, SDM‐18	During consultation	18	6.3 (SD 1.3)	19	4.1 (SD 1.3)	P < 0.001
Yamaguchi 2017	Patient activation measure, PAM	6 months	26	52.3 (SD 18.1)	27	45.8 (SD 10.8)
SDM: shared decision‐making

Table 5. Participation (by the person with mental health condition) or level of involvement in the decision‐making process

Table 6. Recovery

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Lovell 2018	Developing Recovery Enhancing Environments Measure, DREEM	6 months	142	43.5 (SD 13.3)	118	42.6 (SD 12.5)	P = 0.990
Yamaguchi 2017	Self‐Identified Stage of Recovery, SISR Part A	6 months	26	3.19 (SD 1.2)	27	2.93 (SD 1.36)	P = 0.99
Yamaguchi 2017	Self‐Identified Stage of Recovery, SISR Part B	6 months	26	15.04 (SD 5.38)	27	14.04 (SD 4.15)	P = 0.40
SDM: shared decision‐making

Table 6. Recovery

Table 7. Service user satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	After decision making	32	24.31 (SD 2.90)	53	23.75 (SD 3.71)	No difference
Hamann 2006	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	At hospital discharge	‐	16.3 (SD 3.7)	‐	16.4 (SD 3.2)	P = 0.42
Hamann 2011	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	32	25.5 (SD 4.1)	29	26.7 (SD 3.2)	P = 0.23
Hamann 2017	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	‐	25.7 (SD 4.2)	‐	25.8 (SD 5.2)	P = 0.88
Ishii 2017	Overall satisfaction, Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	At hospital discharge	9	23.7 (SD 3.9)	13	22.1 (SD 3.7)	No difference
LeBlanc 2015	User satisfaction ‐ right amount of information	Immediately after encounter	132	92.5%	109	91.9%	P = 0.81
LeBlanc 2015	User satisfaction ‐ information given was clear	Immediately after encounter	132	68.7%	109	58.7%	P = 0.09
LeBlanc 2015	User satisfaction ‐ information given was helpful	Immediately after encounter	132	69.2%	109	52.8%	P = 0.01
LeBlanc 2015	User satisfaction ‐ strongly desire to receive information this way for other treatment decisions	Immediately after encounter	132	68.2%	109	50.5%	P = 0.005
Lovell 2018	User satisfaction ‐ strongly recommend the way information was shared to others	Immediately after encounter	132	77.6%	109	59.1%	P = 0.002
Loh 2007	Overall satisfaction, German version of the Client Satisfaction Questionnaire, ZUF‐8	Post intervention	128	29.8 (SD 2.7)	66	27.0 (SD 3.6)	P = 0.014
Lovell 2018	Overall satisfaction, Verona Service Satisfaction Scale ‐ European Version‐54, VSSS‐EU‐54	6 months	191	3.5 (SD 0.7)	156	3.5 (SD 0.8)	P = 0.045
Woltmann 2011	Overall satisfaction, Seven statements related to satisfaction, a 5‐point Likert scale	Post participation	40	3.88 (SD 0.54)	40	3.78 (SD 0.56)	No difference
Yamaguchi 2017	Overall satisfaction, Client Satisfaction Questionnaire–8 Japanese version, CSQ‐8	6 months	26	26.0 (4.4)	27	24.3 (4.8)	P = 0.21
SDM: shared decision‐making

Table 7. Service user satisfaction

Table 8. Carer satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Lovell 2018	Carers and Users’ Expectations of Services ‐ carer version, CUES‐C	6 months	24	22.71	26	24.12	No difference

Table 8. Carer satisfaction

Table 9. Healthcare provider satisfaction

Study	Scale used	Timing	N SDM	SDM mean	N comparison	Comparison mean	Notes
Hamann 2006	5‐point Likert scale: overall satisfaction with what had been achieved during hospitalisation	At discharge	‐	3.8	‐	3.5	P = 0.02
LeBlanc 2015	Satisfied/extremely satisfied 1‐item, 5‐point Likert scale	Immediately after the clinical encounter	139	54%	117	76.3%	P = 0.02
Mariani 2018	Professional caregivers' job satisfaction questionnaire, JSQ	6 months	16	42.84 (SD 14.33)	18	43.33 (SD 10.97)	P = 0.576
Woltmann 2011	Case manager satisfaction, 6 statements related to satisfaction, a 5‐point Likert scale	After participation	10	4 (SD 0.5)	10	3.3 (SD 0.5)	P = 0.002

Table 9. Healthcare provider satisfaction

Table 10. Knowledge

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Hamann 2006	Patient knowledge about their disease, original items with 7 multiple‐choice questions	At discharge	‐	15.0 (SD 4.4)	‐	10.9 (SD 5.4)	P = 0.01 n = 88 (total number of participants)
LeBlanc 2015	Overall knowledge including both tailored to information in the decision aid and generic information about depression	Immediately after the clinical encounter	138	63.5	116	56.3	P = 0.03
Woltmann 2011	Client knowledge of the care plan (plan goals recalled)	2 to 4 days after the care planning session	36	75%	33	57%	P = 0.02
SDM: shared decision‐making

Table 10. Knowledge

Table 11. Treatment continuation

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Adherence with outpatient visits	6 months	35	56%	53	51%	P = 0.656
Hamann 2011	“Has this patient shown up at your practice since being discharged from the hospital?” (Physicians answered yes/no)	6 months	32	94%	29	90%	P = 0.45
Hamann 2011	“Are you still in psychiatric treatment?” (Participants answered yes/no)	6 months	25	100%	23	91%	P = 0.22
Hamann 2011	"How much does this patient engage in planning for his or her therapy?" (Physicians answered)	6 months	25	3.5 (SD 0.9)	23	3.2 (SD 0.9)	P = 0.19
Ishii 2017	Whether a patient received outpatient psychiatric treatment within 30 days prior to follow‐up time on medical records	6 months	9	88.9%	13	69.2%	No difference
Loh 2007	Participant assessment of treatment adherence	6‐8 weeks	128	4.3 (SD 0.9)	66	3.9 (SD 1.0)	No difference
Loh 2007	Physician assessment of treatment adherence	6‐8 weeks	128	4.8 (SD 0.6)	66	4.3 (SD 1.1)	No difference
Mott 2014	Initiated psychotherapy visits 1 to 9	4 months	9	44%	11	45%	No difference

Table 11. Treatment continuation

Table 12. Medication continuation

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aljumah 2015	Morisky Medication Adherence Scale, MMAS	3 months	110	5.79 (SD 1.89)	110	5.04 (SD 1.98)	P = 0.004
Aljumah 2015	Morisky Medication Adherence Scale, MMAS	6 months	110	5.99 (SD1.88)	110	4.94 (SD 1.94)	P < 0.0001
Aoki 2019a	Visual analogue scale, VAS	3 months	22	8.79 (SD 1.44)	44	8.57 (SD 1.60)	P = 0.910
Aoki 2019a	Visual analogue scale, VAS	6 months	22	8.58 (SD 1.44) )	44	8.44 (SD 1.62	P = 0.872
Hamann 2006	Estimated compliance from physician's point of view	At discharge	‐	1.7	‐	2.0	P > 0.05
Hamann 2006	Overall compliance determined by participant rated, physician rated, and plasma level	6 months after discharge	39	41%	47	55%	P > 0.05
Hamann 2006	Overall compliance determined by participant rated, physician rated, and plasma level	18 months after discharge	30	60%	38	58%	P > 0.05
Hamann 2011	“Are you still taking medication for your psychiatric condition?” (Participants answered yes/no)	6 months post hospital discharge	25	100%	23	87%	P = 0.10
Hamann 2011	"How do you estimate your patient’s compliance?" (Physician assessed)	6 months post hospital discharge	29	4.0 (SD 1.1)	26	4.2 (SD 0.9)	P = 0.78
Hamann 2017	Medication Adherence Rating Scale, MARS	6 months post hospital discharge	‐	2.6 (2.1)	‐	2.5 (2.2)	P = 0.72 n = 100 (total number of participants)
Hamann 2017	Medication Adherence Rating Scale, MARS	12 months post hospital discharge	‐	2.4 (2.1)	‐	2.8 (2.3)	P = 0.42 n = 85 (total number of participants)
LeBlanc 2015	Participants reported medication usage	After encounter	154	89.9%	134	79.1%	P = 0.15
LeBlanc 2015	Filled prescription within 30 days	For trial period	154	86.2%	134	93.2%	P = 0.19
LeBlanc 2015	% proportion of days covered (PDC) > 80% (of filled prescription)	For trial period	113	94.7%	93	97.8%	P = 0.67
Raue 2019	Initiation of antidepressant medication in the Cornel Service Index	12 weeks after intervention	103	23.3%	78	15.4%	P = 0.154
Yamaguchi 2017	Morisky Medication Adherence Scale, MMAS	6 months	26	5.7 (SD 1.5)	27	5.4 (SD 1.5)	P = 0.74
SDM: shared decision‐making

Table 12. Medication continuation

Table 13. Carer participation in decision‐making

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Lovell 2018	User involvement in care planning (carers): Equip patient‐reported outcome measure, PROM‐14	6 months	22	20.1 (SD 8.0)	46	16.5 (SD 10.9)	P = 0.899
SDM: shared decision making

Table 13. Carer participation in decision‐making

Table 14. Relationship between service users and healthcare providers

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Hamann 2006	Working Alliance Inventry, WAI (by physicians)	At discharge	‐	60.6	‐	69.0	P > 0.05
Hamann 2011	Difficult Doctor‐Patient Relationship Questionnaire, DDPRQ (by physician)	At discharge	32	40.4 (SD 7.6)	29	44.6 (SD 8.4)	P = 0.05
Hamann 2011	Trust in physician (by participant)	At discharge	32	41.8 (SD 7.4)	29	46.4 (SD 7.2)	P = 0.02
Hamann 2011	Therapeutic alliance (by physician)	Post intervention	32	23.8 (SD 4.7)	29	24.1 (SD 4.8)	P = 0.83
Hamann 2017	Difficult Doctor‐Patient Relationship Questionnaire, DDPRQ (by physician)	At discharge	‐	43.0 (SD 8.1)	‐	44 (SD 7.4)	P = 0.37
Hamann 2017	Trust in physician (by participant)	At discharge	‐	40.3 (SD 7.5)	‐	41.1 (SD 6.8)	P = 0.25
Lovell 2018	California Psychotherapy Alliance Scale, CALPAS	6 months	191	4.8 (SD 1.4)	152	4.9 (SD 1.5)	P = 0.949
Yamaguchi 2017	Relationship‐Scale to Assess Therapeutic Relationship, STAR. Positive collaboration (by clinician)	6 months	26	18.7 (SD 3.2)	27	17.7 (SD 3.9)	P = 0.07
Yamaguchi 2017	Relationship‐STAR, emotional difficulties (by clinician)	6 months	26	10.8 (SD 1.3)	27	10.5 (SD 1.2)	P = 0.59
Yamaguchi 2017	Relationship‐STAR, positive clinician input (by clinician)	6 months	26	9.9 (SD 1.7)	27	9.6 (SD 1.4)	P = 0.17
Yamaguchi 2017	Relationship‐STAR, positive collaboration (by participant)	6 months	26	19.4 (SD 5.6)	27	17.2 (SD 5.5)	P = 0.05
Yamaguchi 2017	Relationship‐STAR, positive clinician input (by participant)	6 months	26	9.0 (SD 2.2)	27	7.7 (SD 2.9)	P = 0.03
Yamaguchi 2017	Relationship‐STAR, non‐supportive clinician input (by participant)	6 months	26	10.4 (SD 2.4)	27	10 (SD 2.6)	P = 0.69

Table 14. Relationship between service users and healthcare providers

Table 15. Health service use outcomes

Study	Scale used	Timing	N SDM	SDM mean	N control	Control mean	Notes
Aoki 2019a	Consultation duration (minutes)	During initial consultation	35	28.71 (SD 12.66)	53	30.49 (SD 15.91)	P = 0.983
Hamann 2006	Rating of time spent per week with participant (minutes)	At discharge	‐	64.0	‐	60.0	P > 0.05
Ishii 2017	Length of stay (days)	At hospital discharge	9	66.7 (SD 40.4)	13	66.5 (SD 17.4)	No difference
Loh 2007	Consultation duration (minutes)	During consultation	128	29.2 (SD 10.7)	66	26.7 (SD 12.5)	No difference
SDM: shared decision‐making

Table 15. Health service use outcomes

Comparison 1. Shared decision‐making versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Clinical outcomes ‐ psychiatric symptoms Show forest plot	1	53	Mean Difference (IV, Random, 95% CI)	‐1.10 [‐5.54, 3.34]

1.2 Clinical outcomes ‐ depression (1 to 6 months) Show forest plot	5	919	Std. Mean Difference (IV, Random, 95% CI)	0.14 [‐0.19, 0.47]

1.3 Clinical outcomes ‐ depression (1 to 6 months) ‐ sensitivity analysis removing low‐quality studies Show forest plot	4	717	Std. Mean Difference (IV, Random, 95% CI)	‐0.03 [‐0.17, 0.12]

1.4 Clinical outcomes ‐ depression (6 months or more) Show forest plot	5	1100	Std. Mean Difference (IV, Random, 95% CI)	0.21 [‐0.19, 0.60]

1.5 Clinical outcomes ‐ depression (6 months or more) ‐ sensitivity analysis removing low‐quality studies Show forest plot	4	898	Std. Mean Difference (IV, Random, 95% CI)	0.03 [‐0.10, 0.17]

1.6 Clinical outcomes ‐ depression remission (1 to 6 months) Show forest plot	1	215	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.68, 1.65]

1.7 Clinical outcomes ‐ depression remission (6 months or more) Show forest plot	1	210	Risk Ratio (M‐H, Random, 95% CI)	1.58 [0.97, 2.55]

1.8 Clinical outcomes ‐ depression response (1 to 6 months) Show forest plot	1	215	Risk Ratio (M‐H, Random, 95% CI)	1.09 [0.81, 1.47]

1.9 Clinical outcomes ‐ depression response (6 months or more) Show forest plot	1	210	Risk Ratio (M‐H, Random, 95% CI)	1.34 [0.98, 1.83]

1.10 Clinical outcomes ‐ readmission rates (1 to 6 months) Show forest plot	2	128	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.52, 2.14]

1.10.1 SDM versus usual care	1	73	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.43, 2.44]
1.10.2 SDM versus cognitive training	1	55	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.34, 3.73]
1.11 Clinical outcomes ‐ readmission rates (6 months or more) Show forest plot	2	249	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.77, 1.46]

1.11.1 SDM versus usual care	1	79	Risk Ratio (M‐H, Random, 95% CI)	1.14 [0.73, 1.78]
1.11.2 SDM versus cognitive training	1	170	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.63, 1.57]
1.12 Participation ‐ observations on the process of SDM Show forest plot	2	133	Std. Mean Difference (IV, Random, 95% CI)	1.14 [0.63, 1.66]

1.13 Participation ‐ SDM‐specific user‐reported outcomes from encounters (immediately after intervention) Show forest plot	3	534	Std. Mean Difference (IV, Random, 95% CI)	0.63 [0.26, 1.01]

1.14 Participation ‐ SDM‐specific user‐reported outcomes from encounters (6 months or more) Show forest plot	2	398	Std. Mean Difference (IV, Random, 95% CI)	0.13 [‐0.30, 0.56]

1.15 Recovery Show forest plot	2	313	Std. Mean Difference (IV, Random, 95% CI)	0.10 [‐0.13, 0.32]

1.16 Satisfaction ‐ overall users' satisfaction immediately after intervention Show forest plot	4	420	Std. Mean Difference (IV, Random, 95% CI)	0.26 [‐0.29, 0.80]

1.17 Satisfaction ‐ overall users' satisfaction at hospital discharge Show forest plot	1	22	Mean Difference (IV, Random, 95% CI)	1.60 [‐1.65, 4.85]

1.18 Satisfaction ‐ overall users' satisfaction in 6 months or more Show forest plot	2	400	Std. Mean Difference (IV, Random, 95% CI)	0.09 [‐0.22, 0.40]

1.19 Satisfaction ‐ users' satisfaction with received information: right amount of information (categorical) Show forest plot	1	241	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.94, 1.07]

1.20 Satisfaction ‐ users' satisfaction with received information: information given was clear (categorical) Show forest plot	1	241	Risk Ratio (M‐H, Random, 95% CI)	1.19 [0.98, 1.44]

1.21 Satisfaction ‐ users' satisfaction with received information: information given was helpful (categorical) Show forest plot	1	241	Risk Ratio (M‐H, Random, 95% CI)	1.33 [1.08, 1.65]

1.22 Satisfaction ‐ users' satisfaction with received information: strongly desire to receive information this way for other treatment decisions (categorical) Show forest plot	1	241	Risk Ratio (M‐H, Random, 95% CI)	1.35 [1.08, 1.68]

1.23 Satisfaction ‐ users' satisfaction with received information: strongly recommend the way information was shared to others (categorical) Show forest plot	1	241	Risk Ratio (M‐H, Random, 95% CI)	1.32 [1.11, 1.58]

1.24 Satisfaction ‐ carer satisfaction Show forest plot	1	50	Mean Difference (IV, Random, 95% CI)	‐1.40 [‐6.69, 3.89]

1.25 Satisfaction ‐ healthcare professional satisfaction Show forest plot	1	20	Mean Difference (IV, Random, 95% CI)	0.70 [0.26, 1.14]

1.26 Satisfaction ‐ healthcare professional satisfaction (categorical) Show forest plot	1	256	Risk Ratio (M‐H, Random, 95% CI)	1.35 [1.16, 1.58]

1.27 Knowledge Show forest plot	2	322	Std. Mean Difference (IV, Random, 95% CI)	0.41 [0.18, 0.63]

1.28 Treatment continuation ‐ clinic visits (1 to 6 months) Show forest plot	1	20	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.37, 2.59]

1.29 Treatment continuation ‐ clinic visits (6 months or more) Show forest plot	3	171	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.93, 1.23]

1.30 Medication continuation (1 to 6 months) Show forest plot	2	286	Std. Mean Difference (IV, Random, 95% CI)	0.33 [0.10, 0.57]

1.31 Medication continuation (1 to 6 months) (categorical) Show forest plot	1	86	Risk Ratio (M‐H, Random, 95% CI)	0.74 [0.47, 1.17]

1.32 Medication continuation (6 months or more) Show forest plot	4	394	Std. Mean Difference (IV, Random, 95% CI)	0.27 [‐0.03, 0.56]

1.33 Medication continuation (6 months or more) (categorical) Show forest plot	4	577	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.94, 1.17]

1.34 Carer participation Show forest plot	1	68	Mean Difference (IV, Random, 95% CI)	3.60 [‐0.99, 8.19]

1.35 Relationship between service users and healthcare professionals, assessed by users Show forest plot	3	457	Std. Mean Difference (IV, Random, 95% CI)	‐0.13 [‐0.54, 0.28]

1.36 Relationship between service users and healthcare professionals, assessed by healthcare professionals Show forest plot	2	114	Std. Mean Difference (IV, Random, 95% CI)	0.17 [‐0.31, 0.65]

1.37 Health service use outcomes ‐ length of consultation Show forest plot	2	282	Std. Mean Difference (IV, Random, 95% CI)	0.09 [‐0.24, 0.41]

1.38 Health service use outcomes ‐ length of hospital stay Show forest plot	1	22	Mean Difference (IV, Random, 95% CI)	0.20 [‐27.84, 28.24]

Comparison 1. Shared decision‐making versus control