Mindfulness-based interventions and cognitive behavioral therapy for depression, anxiety, and stress in adults with chronic conditions: a stratified subgroup meta-analysis Authors: Kenni Wojujutari Ajele, Botho Nanvula Ramonkga, Matthew O. Olasupo, Runcie C. W. Chidebe, David Musyimi Ndetei, Bashir A. Lwaleed, Christopher J. Armitage Published: 2026 ABSTRACT Anxiety, depression, and stress are common in individuals with diabetes and cancer and are associated with poor self-management and well-being. Cognitive behavioral therapy (CBT) and mindfulness-based interventions (MBIs) are widely used in nonpharmacological treatments, but their comparative effectiveness remains unclear. We conducted a stratified subgroup meta-analysis to indirectly compare CBT and MBIs in adults with diabetes or cancer, examining moderators including condition, delivery modality, and dose. Following PRISMA guidelines, we searched Scopus, Web of Science, EBSCOhost, and ScienceDirect through October 2025. Randomized trials (N = 107; n = 23,585) reporting validated post-intervention outcomes were pooled using random-effects models, with subgroup and meta-regression analyses performed in RStudio. Overall, CBT and MBIs significantly improved outcomes (SMD = −0.78). CBT showed larger effects, particularly for depression (SMD = −0.95) and diabetes (SMD = −1.21), although comparisons were indirect and heterogeneity was substantial. Stronger effects were observed in group-based interventions lasting ≥8 weeks, providing ≥8 contact hours, and including homework or booster components; higher session frequency was also associated with greater improvement (β = −0.086; p < 0.01). MBIs may offer scalable benefits, particularly in cancer care. These findings support condition-tailored psychosocial strategies, although results should be interpreted cautiously. Introduction Diabetes and cancer represent major global health burdens1,2, often accompanied by depression, anxiety, and distress3–6. Specifically, the prevalence of depression and anxiety in individuals with diabetes is nearly twice that of the general population7, affecting approximately 24% of patients8. Similarly, psychological distress affects about 30% of patients with cancer during or after treatment4,9–12. These symptoms correlate with diminished self-management, inadequate glycemic control, adverse cancer outcomes, and a lower quality of life13–15. Despite these risks, many patients remain untreated or undertreated; for example, more than 70% of cancer patients with major depressive disorder do not receive adequate treatment13–15. Current guidelines recommend routine screening and delivery of evidence-based psychosocial care across both cancer and diabetes continuums16–19. However, comparative evidence on which psychological interventions most effectively alleviate depression, anxiety, and distress in chronic illness populations remains limited. Cognitive behavioral therapy (CBT) and mindfulness-based interventions (MBIs), including mindfulness-based stress reduction (MBSR) and mindfulness-based cognitive therapy (MBCT), are well-established nonpharmacological treatments for populations with cancer and diabetes20–26. CBT targets maladaptive cognitions and behaviors, whereas MBIs cultivate present-moment awareness and acceptance to reduce emotional reactivity27–35. Both methods work for depression, anxiety, and stress-related disorders, but it is not clear which one works better for people with long-term medical conditions. In individuals with diabetes, meta-analyses have shown that cognitive-behavioral therapy (CBT) and mindfulness-based interventions (MBIs) are effective in reducing depressive symptoms and diabetes-related distress and improving glycemic control as measured by glycated hemoglobin (HbA1c) levels36–40. In cancer, both CBT and MBIs reduce depression and anxiety across various cancer types and delivery formats31,41–48. However, prior reviews typically examine only a single intervention or condition, often focusing on global distress rather than specific symptoms. Furthermore, they rarely include MBSR or tend to restrict their analyses to specific types of cancer or diabetes31,37,44,48–51. Consequently, comparative effectiveness across diabetes and cancer remains uncertain36,38,47,50,51, highlighting the need for a synthesis that informs condition-tailored psychosocial strategies. Furthermore, most evidence on CBT and MBIs originates from high-income Western settings, and their applicability and effectiveness in low- and middle-income countries (LMICs), where the burden of diabetes and cancer is rising rapidly, but mental health resources are scarce and remain poorly understood1. We examine regional variations to explore potential differences in intervention effects across contexts; however, as our findings are derived from indirect, study-level comparisons, they provide only provisional insights and should be interpreted cautiously when informing implementation strategies. Although both CBT and MBIs are effective across chronic illness settings, direct comparative trials are scarce, leaving uncertainty about their relative benefits across populations and program structures. Consequently, the present analysis relies on indirect, study-level comparisons rather than head-to-head evidence; such comparisons are inherently limited by study heterogeneity and potential confounding and, therefore, require cautious interpretation. We conducted a stratified subgroup meta-analysis of randomized controlled trials to address uncertainties about comparative benefits across conditions and program characteristics. Our primary objective was to compare CBT with MBIs in reducing depression, anxiety, and stress among adults with diabetes or cancer. Our secondary objectives were to test whether benefits differ by condition and assess variation by delivery modality (group, individual, digital, telephone, or hybrid) and program dose (weeks and total hours). We also evaluated study-level characteristics, including comparator type, risk of bias, and region or income level, using meta-regression and generated decision-relevant outputs to inform implementation and stepped-care pathways. Methods Design and protocol We conducted a stratified subgroup meta-analysis to compare the effectiveness of CBT and MBIs on depression, anxiety, and stress in adults with diabetes or cancer. In this approach, studies are grouped a priori into clinically meaningful strata defined by prespecified moderators, pooled within each stratum using random-effects models, and then compared across strata using omnibus tests and interaction terms to evaluate effect modification. This approach not only explains between-study heterogeneity while limiting post hoc, data-driven splitting but also permits indirect comparison between CBT and MBIs. The protocol prespecified the objectives, eligibility criteria, outcomes, moderators, and sensitivity analyses. Reporting was conducted in accordance with PRISMA guidelines52, and the protocol was prospectively registered with PROSPERO (CRD420251243457). Search strategy We searched Scopus, Web of Science Core Collection, EBSCOhost, and ScienceDirect from inception to October 2025. Search strings combined controlled vocabulary and free-text terms related to interventions, populations, study design, and outcomes. We used Boolean operators (e.g., AND, OR) to connect concepts and truncation (e.g., “therapy,” “mindful”) to capture word variants. For example, the PubMed strategy included MeSH terms and keywords such as “Cognitive Behavioral Therapy,” “Mindfulness-Based Stress Reduction,” “mindfulness,” “diabetes mellitus,” “cancer,” “randomized controlled trial,” “depression,” “anxiety,” and “stress.” Syntax was adapted for each database using thesauri (e.g., MeSH in PubMed, Emtree in Embase, APA Thesaurus in PsycINFO). Searches were limited to English-language publications. Full, line-by-line search strategies for all databases are presented in Supplementary Table 4. Records were imported to Covidence (http://covidence.org) for deduplication and screening. Eligibility criteria We included randomized or quasi-experimental trials enrolling adults aged 18 years or older with a medically confirmed diagnosis of diabetes (type 1, type 2, or mixed) or cancer, either during active treatment or in survivorship. The focus on adults ensured greater homogeneity in psychological treatment response, as pediatric populations differ developmentally and typically require distinct therapeutic approaches. Eligible studies had to evaluate a psychological intervention described by the authors as either cognitive behavioral therapy (CBT) or a mindfulness-based intervention (MBI) or meet predefined criteria based on established treatment frameworks. To ensure theoretical and clinical consistency, CBT programs were required to include core cognitive or behavioral components such as cognitive restructuring, behavioral activation, exposure, or problem-solving. MBIs were defined as interventions that include structured mindfulness training, typically involving practices such as meditation, body scanning, or mindful movement, and that are aligned with manualized protocols such as Mindfulness-Based Stress Reduction (MBSR) or Mindfulness-Based Cognitive Therapy (MBCT). Studies were required to include a comparator condition, such as standard care, waitlist, passive control, attention control, or active control, to allow estimation of treatment effects. In subgroup analyses, the category “single group” refers to within-group (pre–post) comparisons and was not treated as a comparator condition. Eligible studies also had to report post-intervention outcomes for depression, anxiety, or stress using validated instruments with established reliability and validity. Accepted measures included the Patient Health Questionnaire-9 (PHQ-9), Beck Depression Inventory (BDI), Generalized Anxiety Disorder 7-item scale (GAD-7), Hospital Anxiety and Depression Scale (HADS), and Depression Anxiety and Stress Scales-21 items (DASS-21). Sufficient statistical data, such as means and standard deviations, or equivalent effect size estimates, were required to calculate standardized mean differences (SMDs). We excluded non-randomized studies without a comparator group to reduce bias and improve internal validity. Studies focused on pediatric or adolescent populations were excluded due to developmental differences in symptom presentation and treatment response. Trials involving mixed medical populations were excluded unless the results for participants with diabetes or cancer could be reported separately, to ensure clinical specificity. Only interventions primarily grounded in CBT or mindfulness-based frameworks were included to maintain fidelity to the target treatment modalities. Studies were also excluded if they lacked extractable post-intervention data, were available only as conference abstracts or protocols without outcomes, or were duplicate publications without unique data. Study selection Records from all sources were imported into Covidence for deduplication and screening. Two reviewers independently screened titles and abstracts using a piloted form. Full texts were retrieved for potentially eligible records and assessed in duplicate. Disagreements were resolved by discussion, with a third reviewer available for arbitration. When multiple reports described the same trial, the information was reconciled, and the report with the most complete and least biased data was retained for quantitative synthesis. A PRISMA flow diagram documents record counts, reasons for exclusion, and the final study set. Data extraction Two reviewers independently extracted study-level and arm-level data into a structured Microsoft Excel workbook with predefined fields and validation rules. Variables included country, region, national income classification, setting, design, sample size, age, sex distribution, clinical condition and subtype, comparator type, attrition, and risk-of-bias domains. Intervention details captured include delivery format (group, individual, digital, telephone, or hybrid); number of sessions; session frequency; duration in weeks; total contact hours; inclusion of homework and booster sessions; adaptation type; and therapist qualifications, where reported. For outcomes, we recorded instrument names, assessment timing, means, standard deviations, and sample sizes for depression, anxiety, and stress at pre- and post-intervention, prioritizing intention-to-treat estimates when available. When multiple eligible measures were available for the same construction at post-intervention (e.g., HADS and PHQ-9 for depression), outcomes were first classified into predefined domains (stress, anxiety, depression) using a structured coding framework (Stress = 1, Anxiety = 2, Depression = 3). Each study contributed a single effect size per outcome domain. When more than one instrument assessed the same domain within a study, a predefined hierarchy was used to select a single measure based on instrument validity, frequency of use across studies, and proximity to the post-intervention point. This approach avoided double-counting and ensured independence of effect sizes. For multi-arm trials, all relevant arms were extracted, and shared comparators were adjusted to avoid double-counting. Cluster randomized trials were adjusted for clustering using effective sample sizes when intra-cluster correlation coefficients were available or could be imputed. Standard errors, confidence intervals, or p-values were converted to standard deviations using standard formulas as needed. Differences were settled by agreement, and a final inspection confirmed that the moderator coding was consistent for the grouped analyses and meta-regression. Risk of bias assessment Risk of bias was assessed independently by two reviewers using study design-specific tools. Randomized controlled trials were evaluated using the Cochrane Risk of Bias 2 (RoB 2) tool, while quasi-experimental studies were assessed using the Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool53,54. RoB 2 evaluates bias across five domains: randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result. Each domain is judged as “low risk,” “some concerns,” or “high risk of bias.” ROBINS-I assesses bias across seven domains, including confounding, participant selection, classification of interventions, deviations from intended interventions, missing data, outcome measurement, and selection of reported results. Overall risk is classified as low, moderate, serious, or critical. Discrepancies between reviewers were resolved through discussion or consultation with a third reviewer. Risk-of-bias visualizations were generated using the robvis tool (https://mcguinlu.shinyapps.io/robvis/). Risk-of-bias assessments are presented in Supplementary Fig. 1 (A: RoB 2; B: ROBINS-I). Most randomized trials demonstrated low risk or some concerns across domains, although some concerns were identified in outcome measurement and reporting bias, and a small subset of studies showed high risk in specific domains. Quasi-experimental studies were generally rated as moderate risk, though some showed serious risk in domains related to confounding and participant selection. Statistical analysis Analyses were conducted in R (version 4.3) using the metafor package. Standardized mean differences (SMDs) and corresponding sampling variances were computed using the escalc function. Separate random-effects models, estimated using restricted maximum likelihood, were fitted for depression, anxiety, and stress outcomes. SMDs were used because the selected studies employed a range of validated instruments to assess similar constructs, allowing for standardization and comparison of effect sizes across different scales. A random-effects model was applied to account for expected between-study heterogeneity arising from differences in study populations, intervention characteristics, and methodological features. This approach assumes that true effect sizes vary across studies and yields more conservative, generalizable estimates. Between-study heterogeneity was quantified using τ², I², and Cochran Q55. To evaluate the stability of the results, additional models were estimated using the DerSimonian–Laird (DL) and Sidik–Jonkman (SJ) variance estimators, both combined with the Hartung–Knapp correction. The latter approach, commonly referred to as the Hartung–Knapp–Sidik–Jonkman (HKSJ) method, has been recommended for improved accuracy in small samples56. To account for correlated outcomes within studies, we fitted multivariate random-effects models (rma.mv) with random study-level intercepts and jointly modeled outcomes. Prespecified subgroup analyses were conducted by condition (diabetes vs cancer), delivery format (group, individual, digital, telephone, or hybrid), duration category (<8 vs ≥8 weeks), total contact hours (<8 vs ≥8 h), homework inclusion (none, homework only, or homework plus booster), comparator type, setting, region, and national income classification. Program duration and total contact hours were categorized using prespecified thresholds informed by established intervention protocols and prior literature. An 8-week cutoff reflects the standard duration of manualized programs such as CBT and mindfulness-based interventions (e.g., MBSR, MBCT), which are typically delivered over eight weekly sessions. Similarly, an 8-h threshold represents the minimum cumulative therapeutic exposure commonly observed in structured interventions (e.g., 1 h per session across an 8-session program). To ensure consistency and comparability across literature, we established a priori thresholds to differentiate between low- and high-intensity intervention formats. Between-stratum differences were evaluated using omnibus Q tests for subgroup differences and by including interaction terms in mixed-effects meta-regression. Dose–response relations for session frequency and program duration were examined using meta-regression, along with study-level moderators including risk of bias and dropout category. We report two-sided α = 0.05, 95% confidence intervals, and 95% prediction intervals to support decision-making. Funnel plots, Egger regression tests, and trim-and-fill57 were used to check for publication bias. Sensitivity analyses included leave-one-out checks and influence diagnostics using Cook distance, DFFITS, studentized residuals, and hat values58,59. Robustness was evaluated by refitting primary models after excluding high-influence studies and by conducting sensitivity analyses separately for CBT and MBI studies, along with stratification by prespecified moderators. Subgroups with fewer than five studies were interpreted cautiously due to instability and increased susceptibility to small-study bias. We also considered comparator type (passive vs. active), as this may influence observed effect sizes. Results PRISMA flow diagram Figure 1 presents the PRISMA 2020 flow diagram summarizing the identification, screening, eligibility assessment, and inclusion processes. A total of 18,187 records were identified through database and register searches. After removing 4044 duplicates, 14,143 titles and abstracts were screened, and 13,548 were excluded for failing to meet the inclusion criteria. Full texts were sought and retrieved for 595 reports, which were all assessed for eligibility. Following full-text review, 488 reports were excluded due to non-English language (n = 37), wrong setting (n = 49), study protocol (n = 65), wrong outcomes (n = 170), wrong indication (n = 57), wrong intervention (n = 36), wrong study design (n = 58), or ineligible patient population (n = 16). Ultimately, 107 studies were included in the final review and meta-analysis.Fig. 1PRISMA 2020 flow diagram illustrating the study selection process, including identification, screening, eligibility assessment, and final inclusion of studies PRISMA 2020 flow diagram illustrating the study selection process, including identification, screening, eligibility assessment, and final inclusion of studies Characteristics of the reviewed studies As shown in Table 1, 107 trials (N = 23,585) varied in population, delivery, and intervention type.Table 1Summary of characteristics of included randomized trials (N = 107)DomainSummaryExample studiesScope and populationsWe synthesized 107 randomized trials among 23,585 adults with diabetes or cancer conducted across Asia, Europe, North America, Oceania, and Africa. Oncology samples included active treatment and survivorship. Diabetes samples covered type 1, type 2, and mixed cohorts.67–76Condition mixBreast and mixed solid tumors predominated in oncology, with frequent enrollment of women. Diabetes cohorts were mostly type 2 with mixed-gender community or clinic populations.77–88InterventionsTrials evaluated CBT variants such as behavioral activation, cognitive behavioral stress management, CBT for insomnia, and adherence-focused CBT, and MBI programs, including MBSR, MBCT, and mindfulness-integrated protocols.20–26,89–91ComparatorsComparators most often included standard care, waitlist, or passive control, with a smaller subset using active controls like education, attention control, health promotion, or alternate psychosocial programs.44,92–104Delivery formats and settingsDelivery spanned group, individual, digital, telephone, and blended formats across hospitals, specialty clinics, community or charity venues, and fully online platforms.105–118Remote modelsGuided internet programs and hybrid models were used in both conditions, reflecting an expansion in remote delivery.119–128Program dosePrograms typically included 4–24 sessions over 2–48 weeks. Many met thresholds of at least eight sessions or eight total contact hours. Homework was common, and booster contacts less frequent.129–138Outcomes and measuresCommon outcomes were depression, anxiety, and stress at post-intervention, measured with validated instruments such as HADS, PHQ-9, GAD-7, CES-D, PSS, DASS-21, STAI, and PROMIS short forms. Oncology studies often added distress thermometers and fear-of-recurrence measures.139–152Study quality and attritionTrial sizes were typically small to moderate, with frequent reliance on self-report outcomes and limited blinding. Treatment fidelity reporting varied, and attrition ranged widely.68,73,86,105,153Characteristics of randomized trials comparing cognitive behavioral therapy and mindfulness-based interventions in adults with diabetes or cancer across regions, delivery formats, doses, comparators, and outcome measures.CBT Cognitive Behavioral Therapy, MBIs Mindfulness-based Interventions, MBSR Mindfulness-based Stress Reduction, MBCT Mindfulness-Based Cognitive Therapy, HADS Hospital Anxiety and Depression Scale, PHQ-9 Patient Health Questionnaire-9, GAD-7 Generalized Anxiety Disorder-7, CES-D Center for Epidemiologic Studies Depression Scale, PSS Perceived Stress Scale, DASS-21 Depression Anxiety Stress Scales-21, STAI State-Trait Anxiety Inventory, PROMIS Patient-Reported Outcomes Measurement Information System. Table 1Summary of characteristics of included randomized trials (N = 107)DomainSummaryExample studiesScope and populationsWe synthesized 107 randomized trials among 23,585 adults with diabetes or cancer conducted across Asia, Europe, North America, Oceania, and Africa. Oncology samples included active treatment and survivorship. Diabetes samples covered type 1, type 2, and mixed cohorts.67–76Condition mixBreast and mixed solid tumors predominated in oncology, with frequent enrollment of women. Diabetes cohorts were mostly type 2 with mixed-gender community or clinic populations.77–88InterventionsTrials evaluated CBT variants such as behavioral activation, cognitive behavioral stress management, CBT for insomnia, and adherence-focused CBT, and MBI programs, including MBSR, MBCT, and mindfulness-integrated protocols.20–26,89–91ComparatorsComparators most often included standard care, waitlist, or passive control, with a smaller subset using active controls like education, attention control, health promotion, or alternate psychosocial programs.44,92–104Delivery formats and settingsDelivery spanned group, individual, digital, telephone, and blended formats across hospitals, specialty clinics, community or charity venues, and fully online platforms.105–118Remote modelsGuided internet programs and hybrid models were used in both conditions, reflecting an expansion in remote delivery.119–128Program dosePrograms typically included 4–24 sessions over 2–48 weeks. Many met thresholds of at least eight sessions or eight total contact hours. Homework was common, and booster contacts less frequent.129–138Outcomes and measuresCommon outcomes were depression, anxiety, and stress at post-intervention, measured with validated instruments such as HADS, PHQ-9, GAD-7, CES-D, PSS, DASS-21, STAI, and PROMIS short forms. Oncology studies often added distress thermometers and fear-of-recurrence measures.139–152Study quality and attritionTrial sizes were typically small to moderate, with frequent reliance on self-report outcomes and limited blinding. Treatment fidelity reporting varied, and attrition ranged widely.68,73,86,105,153Characteristics of randomized trials comparing cognitive behavioral therapy and mindfulness-based interventions in adults with diabetes or cancer across regions, delivery formats, doses, comparators, and outcome measures.CBT Cognitive Behavioral Therapy, MBIs Mindfulness-based Interventions, MBSR Mindfulness-based Stress Reduction, MBCT Mindfulness-Based Cognitive Therapy, HADS Hospital Anxiety and Depression Scale, PHQ-9 Patient Health Questionnaire-9, GAD-7 Generalized Anxiety Disorder-7, CES-D Center for Epidemiologic Studies Depression Scale, PSS Perceived Stress Scale, DASS-21 Depression Anxiety Stress Scales-21, STAI State-Trait Anxiety Inventory, PROMIS Patient-Reported Outcomes Measurement Information System. Summary of characteristics of included randomized trials (N = 107) Characteristics of randomized trials comparing cognitive behavioral therapy and mindfulness-based interventions in adults with diabetes or cancer across regions, delivery formats, doses, comparators, and outcome measures. CBT Cognitive Behavioral Therapy, MBIs Mindfulness-based Interventions, MBSR Mindfulness-based Stress Reduction, MBCT Mindfulness-Based Cognitive Therapy, HADS Hospital Anxiety and Depression Scale, PHQ-9 Patient Health Questionnaire-9, GAD-7 Generalized Anxiety Disorder-7, CES-D Center for Epidemiologic Studies Depression Scale, PSS Perceived Stress Scale, DASS-21 Depression Anxiety Stress Scales-21, STAI State-Trait Anxiety Inventory, PROMIS Patient-Reported Outcomes Measurement Information System. Mental health outcomes CBT and MBIs were associated with significant overall improvements in psychological symptoms among individuals with physical health conditions. As shown in Supplementary Table 2, the pooled effect for depression was large (SMD = −0.95, 95% CI: −1.16 to −0.73, p < 0.01 [k = 103, I² = 96.6%]), with similarly significant reductions for anxiety (SMD = −0.66, 95% CI: −0.79 to −0.53, p < 0.01 [k = 71, I² = 86.1%]) and stress (SMD = −0.62, 95% CI: −0.88 to −0.36, p < 0.01 [k = 37, I² = 92.2%]). These findings indicate that CBT and MBIs provide meaningful psychological benefits; however, substantial residual heterogeneity suggests that important sources of variability may not be fully captured by the stratification approach. Physical health condition and treatment type Across all physical health conditions, psychological interventions significantly reduced symptoms of depression, anxiety, and stress (SMD = −0.78; 95% CI: −0.91, −0.65, p < 0.01 [k = 211, I² = 87.7%]) (Table 2, Supplementary Fig. 2A). When stratified by both conditions, CBT showed greater improvement than MBIs. In individuals with cancer, CBT showed a stronger benefit (SMD = −0.86, 95% CI: −1.06, −0.67 [k = 75, I² = 89.20%]) compared to MBIs (SMD = −0.64; 95% CI: −0.79, −0.48 [k = 86, I² = 81.80%]). Similarly, for those with diabetes, CBT showed the highest benefit (SMD = −1.21, 95% CI: −1.94, −0.48 [k = 38, I² = 92.50%]) compared to MBIs (SMD = −0.54; 95% CI: −0.85, −0.23 [k = 12, I² = 80.20%]) (Table 2, Supplementary Fig. 2B). Subgroup differences were statistically significant (Q (3) = 6.77, p = 0.008), indicating differential effects across treatment type and condition.Table 2Summary of stratified subgroup meta-analysis of interventions, population, delivery, and contextual moderatorsFactorsGroupKSMD95% CIτ²I² (%)Subgroup DifferencesTreatmentPhysical HealthQ (3) = 6.77, p = 0.008CombinedAll211−0.78[−0.91, −0.65]0.5987.7CBTCancer75−0.86[−1.06, −0.67]0.5789.2Diabetes38−1.21[−1.94, −0.48]3.6092.5MBIsCancer86−0.64[−0.79, −0.48]0.3881.8Diabetes12−0.54[−0.85, −0.23]0.2080.2TreatmentPopulation SubgroupQ (10) = 27.17, p = 0.002CBTActive Cancer42−0.76[−1.00, −0.52]0.4782.60Active Cancer + Survivors4−1.07[−1.80, −0.34]0.1361.30Cancer Survivors29−0.97[−1.34, −0.60]0.8092.70Type 1 Diabetes (T1D)8−0.95[−1.95, 0.06]1.2186.40T1D + Type 2 Diabetes (T2D)5−0.44[−1.07, 0.20]0.2078.40T2D25−1.49[−2.64, −0.34]6.3494.40MBIsActive Cancer49−0.59[−0.77, −0.42]0.2982.00Active Cancer + Survivors13−1.21[−2.12, −0.31]2.0392.30Cancer Survivors24−0.45[−0.55, −0.36]0.007.40T1D + T2D7−0.41[−0.65, −0.17]0.0230.10T2D5−0.67[−1.58, 0.25]0.4990.50TreatmentComparator TypeQ(8) = 93.97, p < 0.001CBTActive Control31−1.24[−2.15, −0.34]4.700.92Passive Control2−0.31[−0.55, −0.06]0.000.00Passive Control26−1.05[−1.48, −0.62]0.930.89Single-Group17−1.00[−1.28, −0.71]0.240.80Standard Care37−0.75[−1.04, −0.46]0.620.91MBIsActive Control14−1.10[−2.00, −0.20]2.220.92Passive Control42−0.63[−0.77, −0.49]0.150.78Single-Group12−0.72[−0.88, −0.55]0.020.25Standard Care30−0.41[−0.63, −0.19]0.260.79TreatmentDelivery FormatQ (7) = 21.50, p = 0.001CBTGroup31−1.63[−2.55, −0.71]4.960.94Telephone-based/Hybrid22−0.47[−0.61, −0.33]0.050.63Individual29−0.94[−1.29, −0.58]0.740.89Online/Digital31−0.76[−1.04, −0.48]0.450.89MBIsGroup57−0.63[−0.85, −0.41]0.530.84Individual17−0.81[−1.16, −0.46]0.400.88Online/Digital18−0.59[−0.73, −0.45]0.010.24Telephone-based/Hybrid6−0.45[−0.59, −0.31]0.000.00TreatmentRegionQ(10) = 42.48, p < 0.001CBTAfrica5−3.74[−12.01, 4.53]43.000.98Asia27−1.28[−1.70, −0.87]0.950.94Europe37−0.42[−0.60, −0.24]0.140.72North America18−0.84[−1.21, −0.47]0.480.89Northern America9−1.07[−1.79, −0.35]0.790.92Oceania14−1.17[−1.52, −0.83]0.270.89South America3−0.43[−1.11, 0.26]0.000.00MBIsAsia32−0.74[−1.01, −0.47]0.480.88Europe28−0.47[−0.58, −0.36]0.030.41Oceania4−0.78[−1.54, −0.03]0.150.68North America34−0.66[−0.99, −0.33]0.680.82TreatmentTreatment DurationQ(3) = 17.11, p = 0.001CBT<8 weeks39−0.82[−1.12, −0.51]0.7186.50≥8 weeks74−1.00[−1.32, −0.68]1.1191.90MBIs<8 weeks24−1.10[−1.60, −0.60]1.2392.30≥8 weeks74−0.49[−0.59, −0.40]0.1062.80Treatment GroupTotal Hours per SessionQ(3) = 16.07, p = 0.001CBT<8 h52−0.67[−0.85, −0.48]0.3585.2≥8 h61−1.21[−1.65, −0.77]1.9692.9MBIs<8 h25−1.06[−1.54, −0.58]1.1891.8TreatmentHomework/BoosterQ(5) = 23.76, p = 0.001CBTNon6−0.66[−1.31, −0.01]0.3180.2Homework only93−0.94[−1.19, −0.68]0.8590.7Homework + Booster14−0.99[−1.74, −0.23]1.5292.2MBIsNon4−1.80[−2.90, −0.69]0.4184.7Homework only77−0.60[−0.76, −0.43]0.3478.5Homework + Booster17−0.48[−0.61, −0.35]0.0235.7Treatment TypePhysical HealthQ (7) = 106.26, p < 0.001AdaptedCancer85−0.84[−1.00, −0.69]0.4286.20Cancer7−0.03[−0.16, 0.10]0.000.00Diabetes17−1.61[−3.44, 0.21]11.1093.20Diabetes6−1.32[−2.20, −0.43]0.6191.80IntegratedDiabetes13−0.78[−1.24, −0.32]0.5190.30Cancer28−1.22[−1.65, −0.80]1.0088.30UnmodifiedDiabetes14−0.66[−1.05, −0.27]0.3886.20Cancer41−0.37[−0.48, −0.25]0.0549.70Number of studies (k). Standardized mean differences (SMDs) reflect pooled intervention benefit relative to control. Positive values indicate greater symptoms; negative values reflect symptom improvement. τ² indicates between-study variance and I² reflects heterogeneity. Subgroup differences were tested using mixed-effects models. Table 2Summary of stratified subgroup meta-analysis of interventions, population, delivery, and contextual moderatorsFactorsGroupKSMD95% CIτ²I² (%)Subgroup DifferencesTreatmentPhysical HealthQ (3) = 6.77, p = 0.008CombinedAll211−0.78[−0.91, −0.65]0.5987.7CBTCancer75−0.86[−1.06, −0.67]0.5789.2Diabetes38−1.21[−1.94, −0.48]3.6092.5MBIsCancer86−0.64[−0.79, −0.48]0.3881.8Diabetes12−0.54[−0.85, −0.23]0.2080.2TreatmentPopulation SubgroupQ (10) = 27.17, p = 0.002CBTActive Cancer42−0.76[−1.00, −0.52]0.4782.60Active Cancer + Survivors4−1.07[−1.80, −0.34]0.1361.30Cancer Survivors29−0.97[−1.34, −0.60]0.8092.70Type 1 Diabetes (T1D)8−0.95[−1.95, 0.06]1.2186.40T1D + Type 2 Diabetes (T2D)5−0.44[−1.07, 0.20]0.2078.40T2D25−1.49[−2.64, −0.34]6.3494.40MBIsActive Cancer49−0.59[−0.77, −0.42]0.2982.00Active Cancer + Survivors13−1.21[−2.12, −0.31]2.0392.30Cancer Survivors24−0.45[−0.55, −0.36]0.007.40T1D + T2D7−0.41[−0.65, −0.17]0.0230.10T2D5−0.67[−1.58, 0.25]0.4990.50TreatmentComparator TypeQ(8) = 93.97, p < 0.001CBTActive Control31−1.24[−2.15, −0.34]4.700.92Passive Control2−0.31[−0.55, −0.06]0.000.00Passive Control26−1.05[−1.48, −0.62]0.930.89Single-Group17−1.00[−1.28, −0.71]0.240.80Standard Care37−0.75[−1.04, −0.46]0.620.91MBIsActive Control14−1.10[−2.00, −0.20]2.220.92Passive Control42−0.63[−0.77, −0.49]0.150.78Single-Group12−0.72[−0.88, −0.55]0.020.25Standard Care30−0.41[−0.63, −0.19]0.260.79TreatmentDelivery FormatQ (7) = 21.50, p = 0.001CBTGroup31−1.63[−2.55, −0.71]4.960.94Telephone-based/Hybrid22−0.47[−0.61, −0.33]0.050.63Individual29−0.94[−1.29, −0.58]0.740.89Online/Digital31−0.76[−1.04, −0.48]0.450.89MBIsGroup57−0.63[−0.85, −0.41]0.530.84Individual17−0.81[−1.16, −0.46]0.400.88Online/Digital18−0.59[−0.73, −0.45]0.010.24Telephone-based/Hybrid6−0.45[−0.59, −0.31]0.000.00TreatmentRegionQ(10) = 42.48, p < 0.001CBTAfrica5−3.74[−12.01, 4.53]43.000.98Asia27−1.28[−1.70, −0.87]0.950.94Europe37−0.42[−0.60, −0.24]0.140.72North America18−0.84[−1.21, −0.47]0.480.89Northern America9−1.07[−1.79, −0.35]0.790.92Oceania14−1.17[−1.52, −0.83]0.270.89South America3−0.43[−1.11, 0.26]0.000.00MBIsAsia32−0.74[−1.01, −0.47]0.480.88Europe28−0.47[−0.58, −0.36]0.030.41Oceania4−0.78[−1.54, −0.03]0.150.68North America34−0.66[−0.99, −0.33]0.680.82TreatmentTreatment DurationQ(3) = 17.11, p = 0.001CBT<8 weeks39−0.82[−1.12, −0.51]0.7186.50≥8 weeks74−1.00[−1.32, −0.68]1.1191.90MBIs<8 weeks24−1.10[−1.60, −0.60]1.2392.30≥8 weeks74−0.49[−0.59, −0.40]0.1062.80Treatment GroupTotal Hours per SessionQ(3) = 16.07, p = 0.001CBT<8 h52−0.67[−0.85, −0.48]0.3585.2≥8 h61−1.21[−1.65, −0.77]1.9692.9MBIs<8 h25−1.06[−1.54, −0.58]1.1891.8TreatmentHomework/BoosterQ(5) = 23.76, p = 0.001CBTNon6−0.66[−1.31, −0.01]0.3180.2Homework only93−0.94[−1.19, −0.68]0.8590.7Homework + Booster14−0.99[−1.74, −0.23]1.5292.2MBIsNon4−1.80[−2.90, −0.69]0.4184.7Homework only77−0.60[−0.76, −0.43]0.3478.5Homework + Booster17−0.48[−0.61, −0.35]0.0235.7Treatment TypePhysical HealthQ (7) = 106.26, p < 0.001AdaptedCancer85−0.84[−1.00, −0.69]0.4286.20Cancer7−0.03[−0.16, 0.10]0.000.00Diabetes17−1.61[−3.44, 0.21]11.1093.20Diabetes6−1.32[−2.20, −0.43]0.6191.80IntegratedDiabetes13−0.78[−1.24, −0.32]0.5190.30Cancer28−1.22[−1.65, −0.80]1.0088.30UnmodifiedDiabetes14−0.66[−1.05, −0.27]0.3886.20Cancer41−0.37[−0.48, −0.25]0.0549.70Number of studies (k). Standardized mean differences (SMDs) reflect pooled intervention benefit relative to control. Positive values indicate greater symptoms; negative values reflect symptom improvement. τ² indicates between-study variance and I² reflects heterogeneity. Subgroup differences were tested using mixed-effects models. Summary of stratified subgroup meta-analysis of interventions, population, delivery, and contextual moderators Number of studies (k). Standardized mean differences (SMDs) reflect pooled intervention benefit relative to control. Positive values indicate greater symptoms; negative values reflect symptom improvement. τ² indicates between-study variance and I² reflects heterogeneity. Subgroup differences were tested using mixed-effects models. As shown in Table 2 and Supplementary Fig. 2B, both CBT and MBIs improved psychological outcomes across cancer and diabetes subgroups. When stratified, CBT produced strong reductions in depression among cancer survivors (SMD = −0.97; 95% CI: −1.34, −0.60 [k = 29, I² = 93%]) and individuals with T2D (SMD = −1.49; 95% CI: −2.64, −0.34 [k = 25, I² = 94%]), while MBIs showed more modest benefits for cancer survivors (SMD = −0.45 [k = 24, I² = 7%]) and T2D (SMD = −0.67 [k = 5, I² = 90.50%]) (Table 2). Subgroup differences across population types were significant (Q (10) = 27.17, p < 0.01), indicating differential treatment benefits by clinical profile (Table 2, Supplementary Fig. 2C). Comparator type and delivery format Comparisons by control group in Table 2 and Supplementary Fig. 2D revealed that CBT demonstrated greater benefit when compared to active controls (SMD = −1.24, 95% CI: −2.15, −0.34 [k = 31, I² = 92%]) and standard care (SMD = −0.75 [k = 37, I² = 91%]). MBIs also showed improvement over standard care (SMD = −0.41 [k = 30, I² = 79%]) and passive controls (SMD = −0.63 [k = 42, I² = 78%]). Subgroup differences were significant (Q (8) = 93.97, p < 0.01), indicating that the comparator condition moderated the level of treatment benefit. Treatment delivery format also influenced outcomes (Table 2, Supplementary Fig. 2E). Group-based CBT interventions showed the strongest benefit (SMD = −1.63 [k = 31, I² = 94%]), followed by individual (SMD = −0.94 [k = 29, I² = 89%]) and online formats (SMD = −0.76 [k = 31, I² = 89%]). This pattern was consistent across both the diabetes and cancer populations. MBIs delivered in groups (SMD = −0.63 [k = 57, I² = 84%]) and individual sessions (SMD = −0.81 [k = 17, I² = 88%]) were more effective than telephone-based or online modalities (Table 2). These differences were significant (Q (7) = 21.50, p < 0.01), suggesting that format may influence intervention benefits. CBT trials used active comparators more frequently, which generally yielded smaller SMDs, whereas MBIs were often compared with passive or usual care controls. This structural difference may partly explain the larger pooled effects for CBT. Geographic region and program duration Regional differences were evident (Table 2). CBT delivered in Oceania (SMD = −1.17, 95% CI: −1.52 to −0.83 [k = 14, I² = 89%]) and Asia (SMD = −1.28, 95% CI: −1.70 to −0.87 [k = 27, I² = 94%]) demonstrated larger effect sizes than in Europe or South America. For MBIs, benefits were consistent across Asia, North America, and Oceania (Table 2, Supplementary Fig. 2F). Subgroup variation was significant (Q (10) = 42.48, p < 0.001), suggesting that geographic and cultural factors may influence treatment success. Longer treatment duration was associated with greater improvement. As shown in Table 2 and Supplementary Fig. 2G, both shorter (<8 weeks) and longer (≥8 weeks) psychological interventions were associated with significant mental health improvements, though the degree of benefit varied by duration. The 8-week threshold reflects the standard length of established CBT and MBI protocols, such as MBSR and MBCT, and is commonly used in clinical trials and prior reviews. CBT programs lasting 8 weeks or more yielded larger results (SMD = −1.00 [k = 74, I² = 91.9%]) compared to shorter programs (SMD = −0.82 [k = 39, I² = 86.5%]). MBIs showed a similar pattern, with ≥8-week programs producing smaller but significant improvements (SMD = −0.49 [k = 74, I² = 62.8%]) relative to shorter ones (SMD = −1.10 [k = 24, I² = 92.3%]) (Table 2, Supplementary Fig. 2G). Subgroup differences were significant (Q (3) = 17.11, p < 0.01), indicating that improvements varied by treatment duration. Treatment dose, format, and homework Total contact hours and the inclusion of homework or booster sessions moderated treatment effectiveness (Table 2, Supplementary Fig. 2I). For CBT, programs involving ≥8 h were significantly more beneficial (SMD = −1.21 [k = 61, I² = 92.9%]) than those with <8 h (SMD = −0.67 [k = 52, I² = 62.8%]). The 8-h threshold reflects the minimum total contact time typically found in structured psychological interventions (e.g., 1 h per session over an 8-session program), consistent with standard CBT and MBI protocols and previous reviews. MBIs followed a similar trend. Including homework and booster sessions produced stronger improvements across both interventions, particularly for MBIs with non-homework designs (SMD = −1.80 [k = 4, I² = 84.7%]) and CBT with both homework and booster (SMD = −0.99 [k = 14, I² = 92.2%]). Subgroup differences were statistically significant (Q (5) = 23.76, p < 0.01), indicating that structured reinforcement improved outcomes. Treatments across conditions and program features Across conditions and symptom domains, CBT generally showed larger effect sizes than MBIs, with the clearest differences observed for depression and diabetes populations; however, several subgroup estimates were based on small numbers of studies and should be interpreted cautiously (Table 3, Supplementary Fig. 2J, K–L). In cancer, CBT was associated with greater reductions in anxiety and depression than MBIs (for example, anxiety: SMD = −0.77, 95% CI: −1.03, −0.51 [k = 27, I² = 86.40%] vs −0.60, 95% CI: −0.76, −0.44 [k = 30, I² = 70.90%]; depression: SMD = −0.99, 95% CI: −1.31, −0.67 [k = 36, I² = 90.70%] vs −0.73, 95% CI: −1.03, −0.42 [k = 36, I² = 86.10%]) (Table 3). In diabetes, CBT showed larger average reductions than MBIs for both anxiety and depression; however, some MBI subgroup estimates were based on very small numbers of studies (e.g., k = 4–6) and should be interpreted with caution (for example, anxiety: SMD = −0.72, 95% CI: −1.28, −0.16 [k = 10, I² = 81.70%] vs −0.39, 95% CI: −0.68, −0.11 [k = 4, I² = 0%]; depression: SMD = −1.45, 95% CI: −2.62, −0.27 [k = 25, I² = 94.10%] vs −0.78, 95% CI: −1.41, −0.15 [k = 6, I² = 87.00%]). Subgroup differences in outcome and condition were statistically significant (Q(11) = 77.38, p < 0.01).Table 3Subgroup meta-analysis of psychological interventions by physical health status, mental health outcome, treatment duration, and homework or booster componentsPhysical HealthMental HealthTreatmentkSMD95% CIτ²I² (%)Subgroup differencesCancerAnxietyCBT27−0.77[−1.03, −0.51]0.3686.40Q (11) = 77.38, p < 0.0001MBIs30−0.60[−0.76, −0.44]0.1370.90DepressionCBT36−0.99[−1.31, −0.67]0.7590.70MBIs36−0.73[−1.03, −0.42]0.6086.10StressMBIs20−0.56[−0.96, −0.16]0.5882.90CBT12−0.71[−1.27, −0.15]0.6688.10DiabetesAnxietyMBIs4−0.39[−0.68, −0.11]0.000.00CBT10−0.72[−1.28, −0.16]0.4281.70DepressionCBT25−1.45[−2.62, −0.27]6.6694.10MBIs6−0.78[−1.41, −0.15]0.3287.00StressCBT3−1.04[−3.31, 1.23]0.6383.80MBIs2−0.11[−0.75, 0.53]0.000.00Physical HealthTreatment DurationTreatmentQ (6) = 21.92, p = 0.0013Cancer<8 weeksCBT29−0.74[−1.07, −0.41]0.5583.40≥8 weeksCBT46−0.94[−1.19, −0.70]0.5891.10<8 weeksMBIs24−1.10[−1.60, −0.60]1.2392.30≥8 weeksMBIs62−0.49[−0.59, −0.39]0.0756.60Diabetes<8 weeksCBT10−1.07[−1.89, −0.25]1.1791.50≥8 weeksCBT28−1.30[−2.31, −0.28]5.4493.00≥8 weeksMBIs12−0.54[−0.85, −0.23]0.2080.20Mental HealthTreatment DurationTreatmentQ (11) = 20.28, p = 0.041Anxiety<8 weeksCBT12−0.67[−1.04, −0.29]0.2273.40<8 weeksMBIs9−0.79[−1.19, −0.40]0.2184.00≥8 weeksCBT25−0.79[−1.09, −0.49]0.4389.10≥8 weeksMBIs25−0.50[−0.64, −0.36]0.0446.60Depression<8 weeksMBIs11−1.32[−2.31, −0.33]1.9694.40<8 weeksCBT21−0.94[−1.45, −0.43]1.1090.30≥8 weeksCBT40−1.26[−1.93, −0.59]3.0993.30≥8 weeksMBIs31−0.56[−0.72, −0.40]0.1369.60Stress<8 weeksCBT6−0.66[−1.57, 0.25]0.5481.70<8 weeksMBIs4−1.30[−3.95, 1.34]2.6295.50≥8 weeksCBT9−0.85[−1.54, −0.15]0.7191.00≥8 weeksMBIs18−0.36[−0.59, −0.14]0.1163.50Physical HealthHomework/BoosterTreatmentQ (10) = 57.02, p < 0.0001CancerNonCBT4−1.07[−1.32, −0.82]0.000.00NonMBIs4−1.80[−2.90, −0.69]0.4184.70Homework onlyCBT67−0.82[−1.01, −0.64]0.4688.80Homework onlyMBIs72−0.59[−0.76, −0.42]0.3377.10Homework + BoosterMBIs10−0.52[−0.69, −0.35]0.0137.10Homework + BoosterCBT4−1.47[−5.12, 2.18]5.0796.70DiabetesNonCBT2−0.02[−4.52, 4.47]0.2081.30Homework onlyCBT26−1.52[−2.63, −0.40]6.1993.60Homework onlyMBIs5−0.67[−1.58, 0.25]0.4990.50Homework + BoosterMBIs7−0.41[−0.65, −0.17]0.0230.10Homework + BoosterCBT10−0.82[−1.40, −0.23]0.5988.10This table summarizes standardized mean differences (SMDs) comparing cognitive Behavioral therapy (CBT) and mindfulness-based interventions (MBIs) across physical health conditions (cancer, diabetes), mental health outcomes (anxiety, depression, stress), treatment duration, and homework or booster components. Negative SMD values indicate symptom reduction. Number of studies (k). Heterogeneity is reported using τ² and I². Subgroup differences were examined using mixed-effects models, with significant moderation observed across several comparisons. Table 3Subgroup meta-analysis of psychological interventions by physical health status, mental health outcome, treatment duration, and homework or booster componentsPhysical HealthMental HealthTreatmentkSMD95% CIτ²I² (%)Subgroup differencesCancerAnxietyCBT27−0.77[−1.03, −0.51]0.3686.40Q (11) = 77.38, p < 0.0001MBIs30−0.60[−0.76, −0.44]0.1370.90DepressionCBT36−0.99[−1.31, −0.67]0.7590.70MBIs36−0.73[−1.03, −0.42]0.6086.10StressMBIs20−0.56[−0.96, −0.16]0.5882.90CBT12−0.71[−1.27, −0.15]0.6688.10DiabetesAnxietyMBIs4−0.39[−0.68, −0.11]0.000.00CBT10−0.72[−1.28, −0.16]0.4281.70DepressionCBT25−1.45[−2.62, −0.27]6.6694.10MBIs6−0.78[−1.41, −0.15]0.3287.00StressCBT3−1.04[−3.31, 1.23]0.6383.80MBIs2−0.11[−0.75, 0.53]0.000.00Physical HealthTreatment DurationTreatmentQ (6) = 21.92, p = 0.0013Cancer<8 weeksCBT29−0.74[−1.07, −0.41]0.5583.40≥8 weeksCBT46−0.94[−1.19, −0.70]0.5891.10<8 weeksMBIs24−1.10[−1.60, −0.60]1.2392.30≥8 weeksMBIs62−0.49[−0.59, −0.39]0.0756.60Diabetes<8 weeksCBT10−1.07[−1.89, −0.25]1.1791.50≥8 weeksCBT28−1.30[−2.31, −0.28]5.4493.00≥8 weeksMBIs12−0.54[−0.85, −0.23]0.2080.20Mental HealthTreatment DurationTreatmentQ (11) = 20.28, p = 0.041Anxiety<8 weeksCBT12−0.67[−1.04, −0.29]0.2273.40<8 weeksMBIs9−0.79[−1.19, −0.40]0.2184.00≥8 weeksCBT25−0.79[−1.09, −0.49]0.4389.10≥8 weeksMBIs25−0.50[−0.64, −0.36]0.0446.60Depression<8 weeksMBIs11−1.32[−2.31, −0.33]1.9694.40<8 weeksCBT21−0.94[−1.45, −0.43]1.1090.30≥8 weeksCBT40−1.26[−1.93, −0.59]3.0993.30≥8 weeksMBIs31−0.56[−0.72, −0.40]0.1369.60Stress<8 weeksCBT6−0.66[−1.57, 0.25]0.5481.70<8 weeksMBIs4−1.30[−3.95, 1.34]2.6295.50≥8 weeksCBT9−0.85[−1.54, −0.15]0.7191.00≥8 weeksMBIs18−0.36[−0.59, −0.14]0.1163.50Physical HealthHomework/BoosterTreatmentQ (10) = 57.02, p < 0.0001CancerNonCBT4−1.07[−1.32, −0.82]0.000.00NonMBIs4−1.80[−2.90, −0.69]0.4184.70Homework onlyCBT67−0.82[−1.01, −0.64]0.4688.80Homework onlyMBIs72−0.59[−0.76, −0.42]0.3377.10Homework + BoosterMBIs10−0.52[−0.69, −0.35]0.0137.10Homework + BoosterCBT4−1.47[−5.12, 2.18]5.0796.70DiabetesNonCBT2−0.02[−4.52, 4.47]0.2081.30Homework onlyCBT26−1.52[−2.63, −0.40]6.1993.60Homework onlyMBIs5−0.67[−1.58, 0.25]0.4990.50Homework + BoosterMBIs7−0.41[−0.65, −0.17]0.0230.10Homework + BoosterCBT10−0.82[−1.40, −0.23]0.5988.10This table summarizes standardized mean differences (SMDs) comparing cognitive Behavioral therapy (CBT) and mindfulness-based interventions (MBIs) across physical health conditions (cancer, diabetes), mental health outcomes (anxiety, depression, stress), treatment duration, and homework or booster components. Negative SMD values indicate symptom reduction. Number of studies (k). Heterogeneity is reported using τ² and I². Subgroup differences were examined using mixed-effects models, with significant moderation observed across several comparisons. Subgroup meta-analysis of psychological interventions by physical health status, mental health outcome, treatment duration, and homework or booster components This table summarizes standardized mean differences (SMDs) comparing cognitive Behavioral therapy (CBT) and mindfulness-based interventions (MBIs) across physical health conditions (cancer, diabetes), mental health outcomes (anxiety, depression, stress), treatment duration, and homework or booster components. Negative SMD values indicate symptom reduction. Number of studies (k). Heterogeneity is reported using τ² and I². Subgroup differences were examined using mixed-effects models, with significant moderation observed across several comparisons. When stratified, longer CBT courses generally produced greater average reductions, especially in cancer and diabetes (for example, cancer CBT ≥ 8 weeks: SMD = −0.94, 95% CI: −1.19, −0.70 [k = 46, I² = 91.10%]), while MBIs often demonstrated stronger effects in shorter cancer programs (for example, <8 weeks: SMD = −1.10, 95% CI: −1.60, −0.60 [k = 24, I² = 92.30%]) (Table 3). Homework and boosters were associated with stronger improvements, notably for CBT in diabetes (homework only: SMD = −1.52, 95% CI: −2.63, −0.40 [k = 26, I² = 93.60%]) and for MBIs in cancer with homework only or homework plus booster (for example, homework plus booster: SMD = −0.52, 95% CI: −0.69, −0.35 [k = 10, I² = 37.10%]) (Table 3). However, some subgroup findings, particularly those with small k, may be unstable and should be interpreted cautiously. Session dose-response and multivariate modeling As shown in Supplementary Table 1, Supplementary Fig. 3C–I, multivariate meta-regression models confirmed significant overall benefits of CBT and MBIs across outcomes: depression (SMD = −0.89), anxiety (SMD = −0.82), and stress (SMD = −0.83). The overall session dose significantly predicted greater reductions (β = −0.086; p ≤ 0.01), whereas total weeks did not. Outcome-specific models showed stronger session-dose effects for depression (β = −0.091; p < 0.01) and anxiety (β = −0.074; p = 0.04) (Supplementary Table 1), suggesting that greater treatment frequency, rather than duration, played a key role in improving mental health outcomes. Improvements varied across both symptom domains and treatment intensities (Supplementary Fig. 3C–I). Session frequency showed a stronger association with improvement than total program duration, suggesting that intensive weekly sessions may be more impactful than extended low-frequency programs. Sensitivity analysis and publication bias Leave-one-out analyses (Supplementary Fig. 3A, B, Supplementary Table 3) demonstrated that no single study excessively influenced the overall results. The most influential study for depression shifted the pooled estimate by 8.3%, while for stress, the largest change was 14.8%. Overall, findings remained consistent across replications. Publication bias was detected for the outcomes in Supplementary Fig. 3A (Egger’s test: t = −6.39, p < 0.01). While the trim-and-fill method maintained a statistically significant adjusted pooled estimate (SMD = −0.39; 95% CI: −0.56 to −0.21; p < 0.01), attenuation of effect sizes and small study sizes suggest possible selective reporting, warranting cautious interpretation (Supplementary Fig. 3B). Excluding high-risk studies did not materially change estimates, supporting robustness. Moderation by Study Design and Treatment Type The multivariate meta-regression, including the interaction between study design and treatment type, revealed a statistically significant overall moderator effect, QM(6) = 12.73, p = 0.048 (see Supplementary Table 5). The pooled effect for the reference category (depression outcomes, randomized trials, CBT, cancer) was significant (b = 0.80, 95% CI [0.51, 1.10], p < 0.001). Study design emerged as a significant moderator, with quasi-experimental studies showing larger effect sizes than randomized trials (b = 1.18, p = 0.011), suggesting potential inflation of effects in non-randomized designs. The main effect of treatment type (MBIs vs CBT) was not statistically significant (b = −0.15, p = 0.468). However, a significant interaction between study design and treatment type (b = −1.36, p = 0.016) indicated that the relative effectiveness of MBIs compared to CBT varied by study design, with larger effects in quasi-experimental studies being attenuated for MBIs. Substantial residual heterogeneity remained (QE(209) = 1759.80, p < 0.001), indicating that additional sources of variability were not fully accounted for. Specifically, the larger effects observed in quasi-experimental studies were attenuated for MBIs. Outcome type and physical health condition were not significant moderators. Substantial residual heterogeneity remained (QE(209) = 1759.80, p < 0.001), suggesting that additional sources of variability were not accounted for in the model. Discussion In this stratified subgroup meta-analysis, we found that both CBT and MBIs reduce depression, anxiety, and stress in adults with diabetes or cancer. CBT tended to show larger reductions than MBIs, particularly for depressive symptoms and in people with Type 2 diabetes and cancer survivors; however, these findings are based on indirect comparisons across separate trials and should be interpreted with caution. Given evidence of publication bias and attenuation of effect sizes after adjustment, the magnitude of these differences may be overestimated. The differences between CBT and MBIs may be due to the trials’ specific features, not one method being better. An additional limitation is reliance on self-report outcome measures, such as the PHQ-9 and GAD-7, in trials where participants were not blinded to treatment allocation. This introduces a risk of detection bias, as participant expectations and perceived benefit may influence reported outcomes. Given the large effect sizes observed in some subgroups, these effects may in part reflect expectancy or reporting biases rather than true clinical differences. CBT studies more often used active comparators, which typically yield smaller effect sizes, whereas MBIs were more frequently compared to passive or usual-care controls, and these structural differences likely contributed to the observed effects. Reductions were greater in programs that used group delivery, lasted at least eight sessions, provided at least eight total contact hours, and included homework or booster sessions. Dose-response modeling indicated that higher session frequencies, rather than the overall program duration, were associated with greater symptom reduction. Meta-regression analyses identified several moderators of treatment effects. Importantly, study design significantly moderated effect sizes, with quasi-experimental studies yielding larger estimates than randomized trials. This pattern suggests that non-randomized designs may overestimate intervention effects, raising the risk for selection bias and reinforcing the need to interpret pooled estimates with caution. Our cross-condition design offers a critical advance over prior reviews that examined CBT or MBIs in isolation. Isolated syntheses offer advantageous information for individual treatments but do not help in comparing options or creating specific plans for different methods. For example, in diabetes, previous reviews report that CBT improves mood and occasionally glycemic control, whereas reviews of MBIs describe psychological benefits without comparing outcomes with CBT or examining treatment characteristics such as delivery format or dose36–40. Without a direct or stratified comparison, clinicians and policymakers lack evidence to determine which intervention may be more appropriate for specific patient populations, clinical settings, or resource levels. By pooling and stratifying data across both conditions and interventions, our analysis identifies not only whether CBT or MBIs are effective, but also under which circumstances and for whom they offer the greatest clinical benefit. In cancer, meta-analyses confirm that MBIs reduce anxiety and depression, consistent with our findings, while our stratified analyses indicate that CBT often yields larger average improvement in cancer, and especially in diabetes31,37,44,46,48–51. By identifying group delivery and higher session dose as key moderators for CBT and briefer formats for MBIs, we provide program design guidance that prior reviews rarely addressed in depth42,44,60. The efficacy of CBT in addressing diabetes and depression likely stems from its concentrated emphasis on illness-specific maladaptive cognitions, behavioral avoidance, and self-management obstacles34,35,61,62. Techniques such as behavioral activation, cognitive restructuring, and problem solving directly address regimen burden and catastrophic thinking27,30,34,35. Mindfulness-based interventions (MBIs) seem to be a good fit for cancer patients, who often deal with uncertainty, fear of recurrence, fatigue, and pain31,42,43,46. By fostering nonjudgmental awareness and acceptance rather than cognitive change, MBIs align with the stronger performance we observed for briefer, less structured programs in cancer settings30,33. Between-study heterogeneity remained high despite stratification, indicating that the prespecified moderators did not fully account for variability across studies. This residual heterogeneity may reflect unmeasured factors such as differences in patient subgroups, clinical characteristics, intervention fidelity, and condition-specific features. Further subgroup and sensitivity analyses examining specific patient groups, such as cancer type or diabetes subtype, were limited by inconsistent reporting and small numbers of studies within categories, which restricts the reliability of these estimates. These unmeasured factors may have influenced outcomes even after stratification. Larger gains with group-based, higher-dose CBT are consistent with skill acquisition and practice opportunities36,40,62. Stronger performance of shorter MBIs suggests early, efficient uptake of mindfulness skills with diminishing returns at longer durations36,38,40. Regional differences, including greater reductions in some Asian settings, may reflect cultural receptivity, therapist training, implementation fidelity, or differences in usual care36,44,48. Regional subgroup analyses suggested variation in intervention effects; however, these categories were broad and should be interpreted cautiously. Regions such as Asia and Oceania encompass diverse health systems, resource levels, and implementation contexts, limiting the precision of these comparisons. For low- and middle-income countries (LMICs), structural factors such as limited mental health workforce capacity, infrastructure constraints, and variability in digital access may influence the feasibility of delivering higher-intensity interventions, including frequent group sessions or technology-supported formats. These contextual factors may moderate both the scalability and effectiveness of CBT and MBIs and should be considered when interpreting regional differences and translating findings into practice. Although most trials were small to moderate in size and small-study bias was present, findings were generally stable across influence diagnostics; however, bias-adjusted analyses showed attenuation of effect sizes, suggesting that the magnitude of observed benefits may be influenced by publication bias and should be interpreted with caution. Inconsistent reporting of intervention fidelity and therapist competence limited mechanistic precision but did not alter the overall pattern41,48. For health systems in resource-limited regions, scalability is a paramount concern. Our findings suggest that brief, group-based MBIs may represent a scalable option in cancer care settings, while group-based CBT delivered at higher doses may be beneficial for adults with diabetes or prominent depressive symptoms63,64. This highlights an important trade-off between effectiveness and scalability: the higher-intensity formats that offer greater benefit may be more resource-intensive and harder to implement on a scale. These observations are based on indirect comparisons with substantial heterogeneity, potential publication bias, and some small subgroup sizes and should therefore be interpreted cautiously when informing policy or clinical pathways. For diabetes populations with significant depressive symptoms, investing in training non-specialist providers (e.g., nurses, diabetes educators) to deliver structured, group-based CBT could maximize reach and impact65. Digital and telephone-based formats also offer promise for overcoming geographic and workforce barriers, provided implementation support is in place. Two complementary pathways emerge. Adults with diabetes or prominent depressive symptoms may benefit from group-based CBT delivered at higher doses, while cancer services may consider brief MBI programs as scalable options. Homework and booster elements warrant deliberate planning. Across both modalities, adding structured homework and follow-on booster contacts was associated with stronger improvement. Implementation decisions should also consider that these conclusions arise from indirect comparisons rather than direct head-to-head trials. Thus, patient preference, resource constraints, and cultural adaptation remain essential determinants of modality choice. MBIs frequently excelled when delivered in less structured formats without formal homework, whereas CBT thrived with explicit homework and booster sessions. Some isolated reviews have informed recommendations by national organizations such as the UK National Institute for Health and Care Excellence (NICE)66, particularly for cancer-related distress and psychological support in diabetes care. However, most of these guidelines are based on single-condition, single-intervention syntheses and lack detailed guidance on optimal modality selection, delivery format, or dose tailoring. Our findings build on that foundation by offering comparative data across conditions, identifying actionable moderators, and highlighting how delivery and dose affect benefit. Health systems should invest in workforce training for group CBT, develop brief MBI curricula suited to low-intensity rollouts, embed homework and booster strategies in CBT pathways, and implement routine outcome monitoring to ensure programs deliver expected benefits at scale16–19. This study has several limitations. Several subgroup estimates were based on fewer than five studies and should therefore be interpreted with caution due to instability and susceptibility to small-study effects. Despite extensive stratification, heterogeneity remained substantial, and some subgroup cells, particularly for diabetes-related stress in certain regions, were small, increasing uncertainty. A major limitation is that comparisons between CBT and MBIs are indirect, as they are drawn from separate trials rather than direct head-to-head studies. In addition, outcome measures and follow-up periods varied across studies, limiting conclusions regarding the durability of effects. CBT tended to show larger effect sizes than MBIs; however, these differences are based on indirect comparisons and may be confounded by systematic differences in study design, including comparator type. Notably, CBT trials more frequently used active comparators, whereas MBIs were predominantly compared to passive or usual care controls. This imbalance in comparator types represents a key structural limitation that may inflate estimates of relative effectiveness and preclude definitive conclusions regarding comparative superiority. Furthermore, reliance on study-level moderators rather than patient-level data leaves room for residual confounding related to baseline severity and comorbidity. Although study design-specific tools (RoB 2 and ROBINS-I) were used, the inclusion of a small number of quasi-experimental studies may still introduce residual bias, and their estimates should be interpreted with caution. Publication and small-study biases were detected, and adjustment using trim-and-fill resulted in attenuated effect sizes, suggesting that the magnitude of observed benefits may be overestimated despite the persistence of the overall pattern. Even so, the consistency of findings across robustness checks, along with clear moderation by delivery format and intervention dose, supports cautious, condition-sensitive implementation. Future trials should standardize reporting of intervention content, treatment fidelity, and therapist competence. There is a need for large, multicenter, direct head-to-head comparisons across formats and doses, along with research on hybrid CBT and mindfulness models in diverse cultural and resource settings in low- and middle-income countries. Implementation studies should evaluate equity, access, digital adaptations, and real-world effectiveness using routine outcome measurement, with particular attention to task-sharing models and cost-effectiveness in health systems across Africa, Asia, and Latin America. Furthermore, future research should incorporate more granular country-level analyses and implementation-focused data to better understand how contextual factors influence the delivery of interventions and their outcomes in LMIC settings. There is also a clear need for large, multicenter, head-to-head trials that directly compare CBT and MBIs across conditions, formats, and dosages. In conclusion, cognitive behavioral therapy and mindfulness-based interventions both improve depression, anxiety, and stress in adults with diabetes or cancer. CBT demonstrates greater clinical benefit, particularly for depression and among individuals with diabetes, with the strongest improvements observed when delivered in groups and at higher doses. CBT was associated with larger effect sizes, particularly for depression and among individuals with diabetes; however, these findings are based on indirect comparisons and substantial heterogeneity and should be interpreted with caution. MBIs may be beneficial in briefer, less structured programs and offer scalable options in oncology settings, although these observations are similarly based on indirect evidence and require cautious interpretation. These findings highlight the potential for condition-sensitive approaches to psychosocial care that consider modality, format, and dose; however, further direct comparative research is needed to inform clinical and service-level decision-making. Supplementary information Supplementary information Supplementary information