Types of studies

We included randomised controlled trials (RCTs) in our review as they are well understood to be the ideal method for evaluating the effectiveness of healthcare interventions (National Forum on Early Childhood 2007; O'Connor 2011). Importantly, RCTs are considered less vulnerable to bias compared with other study designs (Lewis 2004); however, there are sometimes ethical and feasibility issues that limit the utility of RCTs (Bonnie 1998). It is unlikely that RCTs will be the sole research study design used to evaluate the effectiveness of educational support interventions for children and adolescents with chronic health conditions. As such, we considered controlled before‐and‐after studies for inclusion in this review. Controlled before‐and‐after studies incorporate a control group with similar characteristics to the intervention group; however, the group allocation is based on a non‐random method. These studies nonetheless need to have the intervention and control groups assessed before and after the intervention. Finally, our review included studies that may have used an interrupted time series (ITS) design. This study design involves measuring outcomes in people on multiple occasions before the intervention to identify underlying trends and on multiple occasions after the intervention (in total at least 7 occasions). The observations collected after the intervention provide information as to whether the observed trend noted post‐intervention is different to that of the underlying trend. ITS may or may not have a control group but the value of ITS analysis is that the subjects are collectively themselves the control group.

We expected to find other study designs in our search (e.g. observational studies and qualitative studies); however, due to the potential high risk of bias associated with these studies in relation to evidence of effectiveness (Ebrahim 2005), we did not include them in our review.

Nonetheless, we acknowledge that both RCTs and non‐randomised studies may be susceptible to bias, and as such, we examined and reported the risk of bias for all included studies. For studies involving a control group, we also accepted studies where group allocation was conducted at the cluster level.

We also aimed to include qualitative studies if they were conducted within the context of an included quantitative intervention study, particularly if the qualitative information was about the effectiveness of the intervention or factors influencing intervention effectiveness (e.g. enablers or inhibitors). We excluded standalone qualitative studies from our review.

Types of participants

Participants included children and adolescents aged 4 to 18 years who have a chronic illness or chronic health condition as defined in Description of the condition. We noted that in the United States of America (US) the term 'special healthcare needs' is often used either or instead of people with chronic health conditions. While not widely used outside of the US, we included studies with children and adolescents with special healthcare needs on the condition that the population in the study was described as having a 'chronic' health condition. We justified this as the definition of special healthcare needs is broader than that of chronic illness. Children with special healthcare needs are "those who have or are at increased risk for a chronic physical, developmental, behavioral, or emotional condition and who also require health and related services of a type or amount beyond that required by children generally" (McPherson 1998, p. 138). In this definition, the presence of a chronic health condition is considered together with the degree of support required to aid healthy functioning. As such, children and adolescents with a chronic health condition, disability, mental disorder (or a combination of these) could be captured under the broad umbrella term of special healthcare needs should they meet the service usage criteria of the special healthcare needs definition. In some circumstances, people may be captured under the definition of special healthcare needs when they have not yet experienced a chronic health condition, as the definition includes those who are at risk of developing a chronic illness due to biological and environmental risks. We did not include people at risk of developing a chronic health condition in our review.

We included studies with children and adolescents who have any chronic health condition regardless of severity according to the definition of a childhood chronic health condition by the Centers for Disease Control. It is understood that not all children and adolescents with a chronic health condition experience negative educational outcomes due to school absenteeism caused by their illness. However, children and adolescents with more severe chronic health conditions miss substantial amounts of school because of their illness (Moonie 2006; Newacheck 1992) and, for these pupils, school functioning may be impacted. Therefore, where possible, we aimed to collect information about severity of condition and number of days of school absenteeism so that we could perform a subgroup analysis to examine the potential impact of severity of disease on learning and developmental outcomes. We included studies of children and adolescents with a chronic health condition who have been hospitalised or who have had regular specialist medical intervention. We aimed to include studies of children and adolescents with a chronic health condition who were not necessarily hospitalised but who may be absent from school due to their illness. This is because there are certain conditions (e.g. chronic fatigue syndrome) that may not be severe enough to require hospitalisation or intense medical attention but have associated functional impairments that may lead to school absences and may place pupils at risk of school disengagement, poor quality of life and poor educational outcomes.

We did not place any restrictions on studies and participants in our review by gender, developed or developing country status, or socioeconomic status. We did not apply any study design filters in the search strategies.

During the review process, we came across some health conditions that were difficult to determine if they were indeed chronic illnesses or chronic health conditions and relevant for this review (e.g. traumatic brain injury and malaria). In such cases, we referred to the Word Health Organisation Pediatric International Classification of Disease (ICD‐10) and consulted with co‐authors.

Types of interventions

Types of outcome measures

This review included studies that reported on any of the primary outcomes (school engagement, academic achievement) or any of the secondary outcomes (quality of life, transition to school, mental health and/or adverse outcomes). We considered short‐term and chronic outcomes for inclusion in the review (i.e. we did not place a restriction based on timing of outcomes).

Electronic searches

We searched the following electronic databases between January 18 and 25, 2021 for the period up to 31 December 2020.

• Cochrane Central Register of Controlled Trials (CENTRAL) (Including Public Health Group).

• MEDLINE (Ovid).

• Embase (Ovid).

• PsycINFO (Ovid).

• CINAHL (EBSCO).

• ERIC (Education Resources Information Center) (EBSCO).

• Applied Social Sciences Index and Abstracts: ASSIA (ProQuest).

• PubMed (from 2018).

We developed a preliminary search strategy in MEDLINE in consultation with a librarian at the Royal Children’s Hospital (Melbourne, Australia) and the Cochrane Public Health Group Information Specialist. The search strategy involved using Boolean operators to combine search terms (both text words and controlled vocabulary) relating to the population (i.e. hospitalised children/adolescents or children/adolescents with a chronic illness) and the intervention (i.e. educational support). As part of our search of the population, we also included specific chronic illness disease search terms based on their relative prevalence and known impact of causing significant absence from schooling and education (e.g. cancer, cystic fibrosis, eating disorders and scoliosis).

While our population of interest was children and adolescents with chronic health conditions, we decided to also include hospitalised children/adolescents in the electronic search strategy and manually reviewed titles and abstracts for the selection of studies that were based on children/adolescents with a chronic health condition. We chose to do so as there is little consistency in the use of either the term or definition of chronic illness or chronic health conditions. For example, as some chronic health conditions (but not all) are also associated with a disability or mental illness, published potentially relevant studies that refer to disability or mental health problems would be missed if we were to use solely chronic illness search terms in our electronic search strategy. This is also true for studies on chronic illness that are classified as special healthcare needs or described by a specific illness type. Thus, we believed our approach would better capture potentially relevant studies on chronic health conditions regardless of whether they were described in terms of disability, mental illness or special healthcare needs. Furthermore, we included in our electronic search the link to hospital as a proxy for severity of chronic illness.

In addition, we did not apply any study design filters in the search strategies other than a date limit (from 2019) to the PubMed search. The reason for this was to identify records that were available in PubMed, up to two years prior to the searches end period date (i.e. 31 December 2020), but were not available in Ovid Medline.

The full search strategies and terms can be found at Appendix 1.

Searching other resources

We handsearched the top five journals in the field for the 12 months prior to our electronic search date on February 8 and 9, 2021. These included the journals: Pediatrics, JAMA Pediatrics, Journal of School Health, Journal of Educational Psychology and the Journal of Pediatric Psychology. We also searched the bibliographies of included studies. We searched the following five grey literature registries and databases between February 8 and 12, 2021 to identify additional published and unpublished studies theses and conference abstracts. These included:

• ClinicalTrials.gov (clinicaltrials.gov/);

• WHO International Clinical Trials Registry platform (apps.who.int/trialsearch/);

• Australian New Zealand Clinical Trials Registry (www.anzctr.org.au/);

• OpenGrey (www.greynet.org/opengreyrepository.html);

• ProQuest Dissertation and Theses database.

We also contacted organisations and individual experts in the field via email and requested that they identify potentially relevant studies that may have been published or studies that may be ongoing.

Selection of studies

We imported search results (including author information, year of publication and titles and abstracts) into the reference management software Endnote X7. We removed duplicate records from Endnote. Two reviewers (TB plus one of either MT, SR or one of three research assistants) (see Acknowledgments) independently screened the remaining citations and assessed them for inclusion. An Excel workbook containing a checklist of the inclusion criteria was used to assess the eligibility of each study for inclusion in this review.

To be considered for inclusion in the review, the study must have met the following criteria:

• participants: must include children or adolescents (aged four to 18 years) with a chronic health condition. We also included studies involving children or adolescents with special healthcare needs providing the study reported findings with a reference to chronic illness;

• intervention: must include educational support;

• outcomes: must report the primary outcomes (e.g. school engagement or academic achievement) or secondary outcomes (e.g. quality of life, transition to school/school re‐entry, mental health or adverse outcomes);

• type of study: must be an RCT, controlled before‐and‐after study or ITS.

We retrieved the full texts of potentially relevant studies identified through the screening process and studies where eligibility was unclear. Two review authors: TB and one of either MT or SR, or one of three research assistants (see Acknowledgements) independently reviewed each study and recorded details about final inclusion/exclusion decisions, including reasons supporting each decision in the ’Characteristics of included studies’ table. We have detailed information about a randomly selected sample (n = 35) of excluded studies in the ’Characteristics of excluded studies’ table.

Where there was a discrepancy between the two reviewers about a decision to include/exclude a particular study, the two reviewers discussed the matter to try to resolve the disagreement. In the event that consensus was not achieved, a third review author (MK) arbitrated. We did not at this stage contact study authors to request further information because we did not need further information about the study to clarify decisions in this process.

Data extraction and management

We retrieved the full texts of studies that met the inclusion criteria. Two reviewers (TB and one of either MK or FDLMG (see Acknowledgements)) independently extracted study characteristic data, risk of bias data and outcome data from the full texts of included studies using a data extraction form, before entering the information into Review Manager 5 (RevMan 2012). We developed the data extraction and risk of bias forms based on the Cochrane Public Health Group Methods Manual (Cochrane Public Health Group 2011), and we included a risk of bias assessment. We piloted the data extraction form to assess the consistency of the data extracted by two review authors. In the data extraction form we collected information about study context, including information about the study methods, participants, details of the intervention, outcomes explored, and analyses used. We also included information about study outcomes and potential moderators/confounders (e.g. Illness type, child age) of the study outcomes in the data extraction form. We recorded information about outcomes in the ’Characteristics of included studies’ table and the ’summary of findings Table 1. Where available, we extracted the key characteristics of included studies including:

• descriptive information:

  • author;

  • year of publication;

  • year of study.

• study methods:

  • research question(s);

  • research aim(s);

  • study design (including information about the timing of assessments);

  • consent rate.

• participants:

  • demographics, using the PROGRESS checklist;

  • age/year level;

  • chronic illness type;

  • severity of chronic illness;

  • physical disability or cognitive disability, or both;

  • mental health condition or physical health condition, or both;

  • number of school days missed because of chronic illness.

• location:

  • country.

• details/nature of the intervention:

  • name of the intervention;

  • description of the intervention, including who delivered the intervention, what was delivered, where it was delivered and to whom it was delivered, as well as information pertaining to the multicomponent nature of the intervention;

  • theoretical foundation of the intervention;

  • setting (site of delivery) of the intervention (e.g. home, hospital, regular school or community‐based);

  • principal provider or co‐ordinator of the intervention (e.g. home, hospital, regular school or community‐based);

  • target of the intervention (e.g. child, family, school, hospital, community);

  • staffing (e.g. roles and numbers);

  • pedagogy (e.g. personalised, individual or group);

  • technology;

  • duration and intensity of the intervention;

  • format of the intervention;

  • cost of the intervention;

  • details of control and comparison groups;

  • retention rates.

• outcomes:

  • primary outcomes (e.g. school engagement, academic achievement) and measures used to assess outcome;

  • secondary outcomes (e.g. quality of life, transition to school/school re‐entry, mental health, adverse outcomes) and measures used to assess outcome;

  • potential confounders/moderators (e.g. health/well‐being, illness severity, chronic illness type).

• analysis:

  • how the data were analysed.

• other:

  • risk of bias information;

  • source(s) of research funding;

  • potential conflicts of interest;

  • implications for replication;

  • qualitative information on effectiveness.

Where disagreements in data extraction arose between review authors, the two review authors involved discussed the discrepancy with reference to the original full‐text paper to try to reach consensus. When required, we conferred with a third review author (MK). We contacted authors of included studies if additional information was required (e.g. if data were missing or if raw data were required) but no further information was provided. We did not find multiple publications reporting on the same included study.

Assessment of risk of bias in included studies

We assessed the risk of bias of the included studies using version 1 of the Cochrane risk of bias tool for randomised trials (Higgins 2011). This tool assesses selection bias (i.e. random sequence generation and allocation concealment), performance bias (i.e. blinding of participants and personnel), detection bias (i.e. blinding of outcome assessment), attrition bias (i.e. incomplete outcome data), reporting bias (i.e. selective reporting), and other sources of bias. There were no controlled before‐and‐after studies that were included, but we are aware that for these studies the recommended tool is the Risk Of Bias In Non‐randomized Studies of Interventions (ROBINS‐I) as described in Chapter 25 of the Cochrane Handbook for Systematic Reviews of Interventions version 6.3 (Sterne 2022). Similarly, there were no ITS included studies so we did not need to assess risk of bias in ITS studies using the Effective Practice and the Organisation of Care Risk of bias tool for ITS study designs. We included additional criteria for cluster‐RCTs including 'recruitment to cluster', 'baseline imbalance', 'loss of clusters', 'incorrect analysis', 'contamination' and 'compatibility with individual RCTs'.

Measures of treatment effect

We prioritised the differences between intervention and control group follow‐up scores. However, due to poor reporting, this was not always possible and we were often required to use the estimate of treatment effect that was available and outlined in the original paper. Where available, we reported standard deviations and 95% confidence intervals. It was likely that studies would use different scales to measure the continuous outcomes explored in this review. In this circumstance, we aimed where possible to report standardised mean differences (SMD) with 95% confidence intervals.

No included studies reported ordinal, dichotomous or qualitative data, but the methods proposed to handle ordinal data can be found in the protocol (Tollit 2015).

Unit of analysis issues

It is possible that for cluster‐RCTs and cluster‐non‐RCTs, authors may have undertaken analysis at the individual level while not accounting for the effects of group clustering. As there was one included study that was a cluster‐RCT that met the inclusion criteria for this review and the analyses included adjustment for clustering appropriately using hierarchical linear models to compare outcomes, there was no need to adjust for clustering as described in the protocol (Tollit 2015). We made no adjustments or separate analysis for trials with multiple treatment groups, multicomponent interventions or non‐inferiority studies as published in the protocol (Tollit 2015) as these were not relevant to any of the included studies. The full protocol wording for this section can be found in Tollit 2015.

Dealing with missing data

Missing data can be related to publication bias and reporting bias. We extracted and reported missing data and retention data using the procedures described in Data extraction and management. We reported information about biases in the Risk of bias section in the ’Characteristics of included studies’ table. In addition, where possible and appropriate, we employed the following strategies to deal with different types of missing data in our review as published in the protocol (Tollit 2015).

• Where there was missing information on methods of the included studies, we contacted authors of trials to obtain information on the methods used or to clarify the methods when this was unclear in the paper. We contacted authors by email in the first instance, using contact details provided in the publication.

• We contacted authors of original papers to obtain missing data required for the analysis, including missing outcome data. We also contacted study authors to ascertain the reasons for these missing data.

• Whenever possible, we conducted our analyses using an intention‐to‐treat methodology, which involves using all participants who were initially randomised in the trial in the analyses. In the analyses, the group treatment that participants were initially allocated to (e.g. intervention group versus control group) were analysed accordingly, regardless of the subsequent treatment received or irrespective of whether participants were lost at follow‐up.

• Where summary data were missing and where possible, we calculated these data using existing information as provided in the paper (e.g. using confidence intervals, standard errors, t values, P values, F values). Where we could not calculate standard deviations and standard deviations of the change due to missing data of supplementary statistics, we were not able to impute the standard deviations and standard deviations of change statistics from other existing studies in this review due to a lack of information, or from studies from a different review as this is the first review on this topic.

Assessment of heterogeneity

As meta‐analysis was not possible due to the small number of studies for each outcome, we were unable to assess for statistical heterogeneity as outlined in the original protocol (Tollit 2015). Instead, we examined study heterogeneity by comparing the study characteristics of trials that met the inclusion criteria. Included in our assessment was an examination of similarities and differences in study design (methodological heterogeneity), and participants, interventions and outcome measures across the studies (clinical heterogeneity). In lieu of pooling studies in a meta‐analysis, we presented information for each individual study in the Characteristics of included studies table and provided an overview of the results based on the direction of effects as per McKenzie 2021; we interpreted the results with caution (see Table 1).

Assessment of reporting biases

We extracted information on potential reporting biases as described in 'Data extraction and management'. Likely types of reporting biases on this topic are: publication bias, outcome reporting bias (selective reporting bias), citation bias and time lag bias. As we could only include a small number of studies (10 or fewer) in the review, we described reporting biases in the Risk of bias tables.

As the number of studies in this review did not exceed 10, we did not produce a funnel plot. Original protocol wording for this section, however, can be found in Tollit 2015

Data synthesis

Synthesis by meta‐analysis as described in the protocol (see Tollit 2015) was not able to be performed in this review due to the small number of studies for each outcome, which would preclude an accurate estimation of between study variability for random effects meta‐analysis. Instead, we synthesised the data grouped by outcome measure using a vote counting method in accordance with Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (version 6.2), 'Statistical synthesis when meta‐analysis of effect estimates is not possible' (McKenzie 2021), and presented the findings in a Direction of effects table. Vote counting can be used to synthesise results based on a reported direction of effect, and/or where there is inconsistency in the effect measures or data reported across studies. Wherever possible, the effect estimate comparing intervention and control groups was calculated and reported as Standard Mean Difference (SMD) together with Confidence Intervals (CIs) and P values, where available. We have used a standard metric of 1 for a result that favours the intervention, and 0 for a result that favours the control group.

The vote counting method requires that, in order to assess the effect of interventions on one of our outcomes, each study must contribute one result to the vote counting method. In some studies, study authors did not report an overall outcome measure but instead reported the results of multiple subscales of the outcome measure. In these instances, it was therefore necessary to select one subscale result for inclusion in the synthesis. We did this by developing a post hoc Decision Rule (DR) informed by the best available evidence as to the clinical importance of each item on children's learning outcomes. The following Decision Rule was developed.

Decision Rule: Overall measure statistic (behaviours > social skills/competence) > Subscale measure statistic: Working memory (verbal) > Working memory (visual) > Attention > Reading > Applied mathematics > Spelling > Calculation.

The DR therefore applied specifically to the two primary outcomes school engagement and academic achievement as follows. For the outcome school engagement, the following hierarchy was used to select one measure for inclusion in the synthesis: overall measure > working memory (verbal) > working memory (visual) > attention. For the outcome academic achievement, the following hierarchy was used: overall measure > reading > applied mathematics > spelling > calculation.

The original protocol wording for this section can be found in Tollit 2015.

Subgroup analysis and investigation of heterogeneity

Subgroup analysis and quantitative examination of heterogeneity, as described in the protocol (see Tollit 2015), was not able to be performed because of the small number of included studies.

Sensitivity analysis

Sensitivity analysis, as described in the protocol (see Tollit 2015), was not able to be performed because of the small number of included studies.

Summary of findings and assessment of the certainty of the evidence

We assessed the overall certainty of the evidence grouped by outcome measure using the GRADE approach and the GRADEpro software. Two people (TB and FDLMG) independently conducted the assessments and, where there were inconsistencies, they were resolved through discussion. We rated the certainty of evidence as high, moderate, low or very low. As all studies were RCTs, we started with high‐certainty of evidence and the following factors were considered for downgrading the certainty: overall risk of bias, consistency of effect, imprecision, indirectness and publication bias, based on the guidance of the Cochrane Handbook of Sytematic Reviews of Interventions chapter 14 (Schünemann 2011).

We prepared the summary of findings Table 1 based on the Cochrane Effective Practice and Organisation of Care (EPOC) worksheet for preparing a Summary of Findings (SoF) table using GRADE (Cochrane 2017). We used the template provided for the SoF table in RevMan 5.4.1. The table includes summary information about both the primary and secondary outcomes of this review: (primary) school engagement and academic achievement; (secondary) transition back to school, mental health and quality of life in addition to any adverse effects as an outcome. As we determined there were insufficient data that could be appropriately pooled, we used the 'Summary of findings' table format for narrative synthesis.