1.1 Aims

Therefore, this systematic review aimed to (1) provide a comprehensive and critical summary of the effects of CBIs on the psychological distress (primary outcome) and physical function (secondary outcome) of adolescents with physical disabilities compared to active or inactive controls and (2) investigate the optimal characteristics of the effective CBIs found.

2.1 Search Strategy

Nine online databases were systematically searched from inception to October 2023, including seven English databases (Embase, PubMed, CINAHL, PsycINFO, Scopus, The Cochrane Library and Web of Science) and two Chinese databases (China National Knowledge Infrastructure and Wanfang Data). Search terms were structured according to the PICOS framework (Table S2). The target population consisted of adolescents aged 10–19 years with physical disabilities. Various search terms related to ‘physical disability’, such as ‘disabled persons’, ‘mobility limitation’, ‘handicap’ and ‘physically challenged’ were used to identify the population. Since studies on individuals with physical disabilities often use condition-specific keywords rather than general disability terms, common paediatric health conditions associated with impairments in body structure and function were included in the search (World Health Organization 2001). This list encompassed conditions such as ‘cerebral palsy’, ‘spinal dysraphism’, ‘spinal cord injury’, ‘scoliosis’, ‘arthritis’, ‘stroke’ and ‘multiple sclerosis’. Search terms such as ‘cognitive therapy’, ‘behavior therapy’, ‘cognitive behavioral therapy’, ‘mindfulness’ and ‘acceptance and commitment therapy’ were utilised to identify relevant interventions. The study design was limited to randomised controlled trials (RCTs). No restrictions were placed on the search terms used for controls and outcomes to ensure a comprehensive range of relevant articles. Tables S3–S11 outlined the search strategies and findings across all databases. Furthermore, a manual search of the reference lists of the included publications was conducted to identify any additional relevant articles.

2.2 Eligibility Criteria

Studies were included when they met the following criteria: (1) Populations: Adolescents between 10 and 19 years of age with congenital or acquired physical disabilities. (2) Interventions: The CBIs were categorised based on either explicit statements from the original article or a joint decision by the assessing authors regarding the treatment's components (i.e. cognitive and behavioural elements). (3) Controls: The CBIs were compared to control groups receiving either active interventions (e.g. attention control) or inactive interventions (e.g. usual care, standard practices, waitlist control and placebo). (4) Outcomes: Psychological distress, including symptoms associated with mood fluctuations and clinical mental health conditions (e.g. depression, anxiety, stress and distress; American Psychological Association 2018), should be assessed based on the study's outcome measures. (5) Study design: Only RCTs were included. (6) Only peer-reviewed studies published in English or Chinese were included as eligible for this review. Studies were excluded if they met any of the following criteria: (1) Functional limitations in daily activities due to intellectual/sensory/mental/mixed impairments. (2) Treatment group sample size smaller than five. (3) Unavailability of full-text or final results.

2.3 Study Selection

To identify duplicate publications, all retrieved records were imported into Endnote X9 (Clarivate Analytics, 2018). After removing duplicates, two reviewers (JYL and XXM) independently assessed the abstracts and titles using Endnote, following the predetermined eligibility criteria. Eligible records were then exported to Excel (Microsoft, 2019), and articles were identified through a comprehensive examination of the full text of the records.

2.4 Data Extraction

Data extraction was conducted using a data extraction form adapted from the Cochrane Handbook for Systematic Reviews of Interventions (Li, Higgins, and Deeks 2023). This form was initially tested on five randomly selected full-text studies and subsequently revised based on the pilot results. The following information was collected: (1) general information (author, publication year, country, setting, study design and measurements); (2) participants' characteristics (disease type, sample size, age range and gender distribution); (3) details of the interventions (the content of CBIs and comparators, theoretical basis, intervention aims, recipients, dosage, interventionist, delivery mode, feasibility and acceptability); (4) outcomes and measurement tools and (5) main findings. Two reviewers (JYL and XXM) independently extracted the data, and any disagreements were resolved through discussion or consultation with a third reviewer (YL). Corresponding authors were contacted for missing data or additional unpublished information. In cases of unresolved missing data, calculations were made using available values and established formulas to estimate the effect (Higgins et al. 2023).

2.5 Quality Appraisal

The risk of bias was evaluated using the revised Cochrane risk-of-bias tool for randomised trials (RoB 2; Sterne et al. 2019). Both reviewers (JYL and XXM) independently assessed the risk of bias in each specific outcome of a trial according to these domains: randomisation process, deviations from intended interventions, missing outcome data, outcome measurement and selective reporting. Each domain required the reviewers to respond with ‘Yes’, ‘Probably yes’, ‘No’, ‘Probably no’ or ‘No information’ for specific questions. An overall level of risk of bias (‘High’, ‘Low’ or ‘Some concerns’) was assigned to each included outcome. A pilot test with five studies was conducted to ensure assessment consistency. Discrepancies between the reviewers were resolved through discussions with a third reviewer (YL).

2.6 Certainty of Evidence

The quality of the evidence (i.e. meta-analysis results) was presented using the Grading of Recommendation, Assessment, Development and Evaluation (Guyatt et al. 2011). Two independent assessors (JYL and XXM) evaluated the level of certainty of the evidence using five domains: risk of bias, indirectness, inconsistency, imprecision and publication bias. The evidence was then classified into four categories: ‘high’, ‘moderate’, ‘low’ and ‘very low’.

2.7 Data Synthesis

The data analysis was conducted using R software (Version 4.3.2), which involved computing the Hedges' g standard mean differences along with 95% confidence intervals (CIs) for each study. Due to variations in study populations and methods, a random-effects model was employed for all analyses. Effect sizes were calculated at postintervention (less than 8 weeks after the intervention), medium-term (3–5 months) and long-term (6 months or more) follow-ups, and categorised as small (g = 0.2), medium (g = 0.5) or large (g = 0.8; Cohen 1988). In the meta-analysis, only outcomes with data from two or more studies were included and pooled in the model. To ensure clarity, certain change scores were reversed, ensuring that negative effect sizes consistently represented relief from psychological distress. Three predefined subgroup analyses were conducted to explore potential variations in efficacy and heterogeneity: recipients (child-only/parent–child dyad), intervention durations (1–8/over 8 weeks) and delivery modes (face-to-face/online/video).

Heterogeneity was assessed using I² and Cochrane's Q statistics. I² values were categorised as ‘might not be important’ (0%–40%), ‘moderate’ (30%–60%), ‘substantial’ (50%–90%) or ‘considerable’ (75%–100%) heterogeneity (Deeks, Higgins, and Altman 2023). Sensitivity analyses were conducted using the leave-one-out method to examine the impact of excluding individual trials and identify potential sources of heterogeneity. The publication bias test was not performed as there were fewer than 10 studies in the meta-analysis for each outcome. Narrative synthesis was performed for studies that were not pooled for meta-analysis.

3.1 Study Selection

A total of 3507 records (3470 from nine main databases, 37 from reference lists) were identified. Following the removal of duplicate records (n = 1242) and the screening of titles and abstracts (n = 2078), a total of 187 relevant articles were identified for full-text screening. Ultimately, 12 RCTs were deemed eligible for inclusion in this review. Figure 1 presents the PRISMA flow chart illustrating the search and selection process. Table S12 lists all excluded articles during the full-text screening stage.

3.2 Study Characteristics

The characteristics of the included 12 RCTs are summarised in Table 1. Studies were published from 2003 to 2023, with most of them conducted in the United States (n = 5, 41.6%; Betz, Smith, and Macias 2010; Connelly et al. 2019; Kashikar-Zuck et al. 2005, 2012; LaMontagne et al. 2003). Other research locations included Canada (n = 3, 25.0%; Charette et al. 2015; Stinson et al. 2020, 2010), the United Kingdom (n = 2, 16.7%; Chalder et al. 2010; Crawley et al. 2018) and China (n = 2, 16.7%; Li et al. 2023; Xu 2007). Only one study was reported in Chinese (Xu 2007), and the remaining studies were reported in English. Most studies were two-arm parallel-design RCTs, while one study utilised a four-arm trial design (LaMontagne et al. 2003) and another study employed a crossover design (Kashikar-Zuck et al. 2005). Except for three pilot studies (Charette et al. 2015; Kashikar-Zuck et al. 2005; Stinson et al. 2010) and one unspecified (Xu 2007), the remaining eight trials were full-scale RCTs.

3.3 Methodological Quality of the Included Studies

Five of the 12 included RCTs (41.7%) were considered at ‘high’ risk of bias; the remaining seven studies were with ‘some concerns’. Regarding the randomisation process, three studies (25.0%) raised concerns about potential bias due to insufficient information on allocation concealment and/or sequence generation. Furthermore, two studies (16.7%) exhibited biased aspects related to deviations from the intended interventions, while the remaining 10 studies were considered to have a ‘low’ risk of bias. Except for one study, which indicated that the higher missing values in the intervention group could be attributed to the longer duration of CBIs, other studies were deemed ‘low’ risk (91.7%). Eleven (91.7%) studies were found to have biased results concerning outcome measurements. This bias arose from the use of self-reported instruments and the absence of blinding. In terms of the selection of the reported results, four studies (33.3%) raised ‘some concerns’ due to insufficient information about prespecified plans/protocols. The summary of risk-of-bias for the 12 included studies is presented in Figure 2.

3.4 Evidence Quality of Included Studies for Meta-Analysis

The certainty of evidence ranged from ‘moderate’ to ‘low’ (Table S14). The certainty of the effects of CBIs on depressive symptoms was downgraded by one level due to the risk of bias in the included studies. For the effects of CBIs on anxiety, stress and physical function, the evidence had a ‘low’ level of certainty. This was primarily due to the risk of bias, as well as either inconsistency (high heterogeneity) or imprecision (small sample sizes below 400).

3.5 Effectiveness of Cognitive-Based Interventions

3.5.1 Stress

Two RCTs involving 118 subjects investigated the impact of CBIs on post-intervention stress (Li et al. 2023; Stinson et al. 2010). Although participants receiving CBIs demonstrated improvements in stress, these enhancements were not statistically significant (g = −0.15, 95% CI [−0.52, 0.21], I² = 0%; Figure 3). Only one study evaluated the long-term effects of a mindfulness-based intervention on stress, revealing non-significant improvements in stress symptoms 6-month postintervention (g = 0.02, 95% CI [−0.44, 0.49]; Li et al. 2023).

3.5.3 Anxiety

3.5.3.1 Meta-Analysis Results of Anxiety at Postintervention

Five trials with 551 participants were included to assess the effectiveness of CBIs on anxiety at postintervention (Charette et al. 2015; Connelly et al. 2019; LaMontagne et al. 2003; Stinson et al. 2020; Xu 2007). The pooled results did not demonstrate significant improvements (g = −0.43, 95% CI [−0.93, 0.07], I² = 81%; Figure 4). Upon conducting sensitivity analyses, it was observed that the potential benefits of CBIs emerged after the removal of an outlier (g = −0.59, 95% CI [−1.14, −0.05]; Figure S1; Connelly et al. 2019). Nevertheless, persistent heterogeneity was observed after sensitivity analyses (I² = 58%–83%), prescribed subgroup analyses were therefore performed (shown in Table S15). The findings indicate that child-only CBIs were more effective in reducing anxiety (g = −0.72, 95% CI [−1.37, −0.07], n = 1) compared to parent–child dyad CBIs (g = −0.04, 95% CI [−0.27, 0.20], n = 3). Interventions conducted within 1–8 weeks (g = −0.80, 95% CI [−1.38, −0.22], n = 3) showed greater improvements in anxiety compared to interventions lasting over 8 weeks (g = 0.03, 95% CI [−0.17, 0.24], n = 2). Face-to-face (g = −1.39, 95% CI [−2.06, −0.71], n = 1) and video-based delivery modes (g = −0.52, 95% CI [−0.94, −0.10], n = 2) were more beneficial in improving anxiety.

3.5.3.2 Meta-Analysis Results of Anxiety at Follow-Up

When exploring the lasting effects of CBIs on anxiety, three RCTs with 443 subjects in the medium term and three RCTs with 441 subjects in the long term were included (Connelly et al. 2019; Crawley et al. 2018; Stinson et al. 2020). The impact of CBIs on anxiety symptoms was not significant in both the medium term (g = −0.14, 95% CI [−0.63, 0.34], I² = 77%; Figure S2) and long term (g = −0.37, 95% CI [−0.85, 0.11], I² = 75%; Figure S3). The sensitivity analyses proved the robustness of the findings (Figures S4 and S5). We did not perform a subgroup analysis for the follow-up anxiety outcome due to the limited number of studies (only three) that provided data.

3.5.4 Depressive Symptoms

3.5.4.1 Meta-Analysis Results of Depressive Symptoms at Postintervention

Four RCTs assessed depressive symptoms at postintervention and were included in the meta-analysis pooling 554 participants (Connelly et al. 2019; Kashikar-Zuck et al. 2005, 2012; Stinson et al. 2020). The pooled analysis yielded non-significant results (g = −0.05, 95% CI [−0.25, 0.16], I² = 14%; Figure 5) and sensitivity analyses provided further support for the reliability of these findings (Figure S6). Subgroup analyses were performed for different intervention durations and delivery modes, revealing no statistically significant reduction in depressive symptoms (Table S15).

3.5.4.2 Meta-Analysis Results of Depressive Symptoms at Follow-Up

Three RCTs with 443 subjects in medium term showed no significant improvement in depressive symptoms (g = −0.02, 95% CI [−0.21, 0.17], I² = 8%), which was confirmed by sensitivity analysis (Figures S7 and S8; Connelly et al. 2019; Crawley et al. 2018; Stinson et al. 2020). In the long-term analysis involving four RCTs with 555 subjects (Connelly et al. 2019; Crawley et al. 2018; Kashikar-Zuck et al. 2012; Stinson et al. 2020), Connelly's study significantly contributed to the heterogeneity in the meta-analysis of depressive symptoms (Connelly et al. 2019). This resulted in a change in the effect size estimate from −0.15 (95% CI [−0.36, 0.06], I² = 34%; Figure S9) to −0.24 (95% CI [−0.46, −0.03], I² = 0%; Figure S10). The subgroup analysis results showed that the child-only target CBIs were more beneficial in improving long-term depressive symptoms (g = −0.54, 95% CI [−1.06, −0.02], n = 1; Table S15). Subgroup differences in different durations and delivery modes were not significant.

4.1 Psychological Distress

CBIs did not show significant effectiveness in improving stress levels compared to active controls, whereas Dunning et al. reported small but significant benefits in reducing stress among children and adolescents (Dunning et al. 2019). A possible reason may be that our synthesis included only two studies (Li, Higgins, and Deeks 2023; Stinson et al. 2010), whereas the latter study encompassed 20 studies that focused primarily on typically developing youth. The small sample size in the synthesis may have hindered our ability to detect small effects favouring CBIs in the context of physical disabilities. In terms of distress, two studies yielded non-significant results, precluding a meta-analysis due to methodological heterogeneity. Moreover, the low certainty of evidence for stress outcomes and the high risk of bias in the study examining distress (Betz, Smith, and Macias 2010) underscore the need for further RCTs with robust designs in this field.

In contrast to the findings on cognitive behavioural therapy for children and adolescents (Vigerland et al. 2016), CBIs did not yield overall significant improvements in anxiety and depressive symptoms over time among adolescents with physical disabilities. This difference may be attributed to differences in eligibility criteria, as most participants in our analysis may not have had mood symptoms at the time of recruitment, whereas Vigerland et al.'s review specifically targeted reducing symptoms of established psychological conditions. Additionally, the limited effects could be attributed to only three studies focusing on psychological distress as a primary outcome (Kashikar-Zuck et al. 2005; LaMontagne et al. 2003; Xu 2007), suggesting that CBIs may indirectly impact these outcomes. Future research should target psychological issues in adolescents with physical disabilities and consider recruiting individuals with psychological symptoms for a more precise depiction of the effectiveness of CBIs.

Notably, it was observed that CBIs had small to medium effects on postintervention anxiety and long-term depressive symptoms after excluding one study (Connelly et al. 2019). The variation in controls among the included studies, with Connelly et al.'s study and its multisite trial (Stinson et al. 2020) employing a more active control combining guided disease education and regular therapist contact, contributed to the inconsistent findings. Thus, the exclusion of Connelly et al.'s study, which had the largest weight but showed non-significant effects on anxiety and depressive symptoms, had a considerable impact on the overall results. Nevertheless, the evidence supporting anxiety and depression outcomes is of low to moderate quality, primarily due to potential internal biases and heterogeneity, indicating that future studies have the potential to alter the conclusions or enhance our confidence in them.

4.2 Physical Function

CBIs did not elicit notable immediate or medium term changes in physical function, but a small positive effect was observed in the long term. These results concur with a large multicenter RCT of cognitive behavioural therapy for paediatric chronic pain, which indicated non-significant between-group differences in daily activity limitations immediately after treatment but significant differences at the 6-month follow-up (Palermo et al. 2016). Specifically, the enhancements in physical function may be attributed to improved self-management skills and behavioural changes. Drawing upon the generic cognitive model, the generation of beneficial behaviours stems from the adjustment of dysfunctional thoughts and emotional states (Beck and Haigh 2014). Consequently, alterations in physical function follow a progressive trajectory, potentially necessitating a duration to manifest notable effects.

4.3 Promising Intervention Components Affecting the Effectiveness of CBIs

The interventions in the included studies exhibited variations in characteristics, highlighting a potential knowledge gap in the optimal application of CBIs for adolescents with physical disabilities. Regarding parent involvement, our subgroup analyses indicated that CBIs targeting adolescents only were more effective in reducing anxiety and depression. This effectiveness may be attributed to adolescents' high value for autonomy and control over their experiences (National Academies of Sciences, Engineering and Medicine 2019), where providing opportunities for independence positively impacts their engagement. However, substantial evidence supports the importance and effectiveness of involving parents in adolescent mental health interventions (Dardas, van de Water, and Simmons 2018; Grist et al. 2019), which challenges this speculation. The controversy may stem from age differences among the studied adolescents and variations in parental involvement. Nonetheless, more quantitative or qualitative evidence is necessary to confirm the effectiveness of CBIs in different recipient groups due to the limited number of studies available within each subgroup.

In terms of intervention durations, the review results suggested that CBIs of 1–8 weeks had a better effect in reducing postintervention anxiety, which is further supported by higher attrition rates observed in studies with longer-term interventions (3 months; Connelly et al. 2019; Stinson et al. 2020). One plausible reason is that short-term interventions are more likely to capture the attention and engagement of adolescents who juggle various school, family and social activities. Importantly, research has shown that 8 weeks of mindfulness-based training can induce similar alterations in brain function and emotion regulation as extended periods of traditional mental practices (Gotink et al. 2016). Therefore, research focusing on shorter-term CBIs for adolescents holds considerable promise for further exploration.

Face-to-face and video modes of delivery were found to be beneficial in reducing post-intervention anxiety, while online self-guided interventions with limited healthcare provider contact showed no significant effects on psychological symptoms (Connelly et al. 2019; Stinson et al. 2020). With the characteristic of real-time conversational interaction, chatbots embedded with mental health promotion techniques have demonstrated the potential to improve stress, distress and depression (Abd-Alrazaq et al. 2020). Given the demands of physically disabled adolescents for telemedicine services that transcend time and distance constraints, real-time virtual interventions (e.g. chatbots) that provide accessible spoken, written and visual language via electronic devices warrant further investigation.

Certain content elements have been identified as potentially crucial for successful CBIs. Cognitive components, such as guided imagery, challenging negative thoughts and building confidence in disease management, were found to be significant. Behavioural strategies, including relaxation exercises, goal setting and coping skills training for disease and emotional symptoms, were also commonly incorporated into successful interventions. These findings align with a systematic review of psychological interventions for young people with chronic illnesses, emphasising the benefits of learning techniques to identify and modify maladaptive thought and behaviour patterns (Sansom-Daly et al. 2012).

4.4 Strengths and Limitations

This review contributes to current knowledge by summarising the effectiveness of CBIs in improving psychological outcomes among adolescents with physical disabilities. A comprehensive search across multiple databases in both English and Chinese languages was conducted to minimise selection bias. Additionally, the synthesis of outcomes at different time points aimed to enhance the comprehensiveness of understanding the durability effects of CBIs. However, some limitations in this review warrant caution. First, the heterogeneity regarding participants' specific diseases and intervention characteristics in the included studies can potentially impact the reliability of the pooled results. Second, the inclusion of studies with biases and the limited certainty of evidence may affect the validity of the findings. Third, the observational nature along with the small number of included studies limited confidence in the subgroup analyses. Finally, the insufficient number of related studies prevented subgroup analyses on other important potential moderators (e.g. CBIs with different theoretical models), thereby limiting the generalisability of the findings.