A comprehensive search was conducted in PubMed, Embase, Web of Sciences, the Cochrane Database of Systematic Reviews, Scopus, and Google Scholar for studies published before 15 April 2024. These studies examined the impact of protective factors and resilience-promoting interventions on resilience development in a global context. No restrictions were placed on the study start year or participant age. Only English language articles were included. The search employed medical subject heading (MeSH) terms, keywords, and their combinations. Additionally, reference lists of retrieved articles were manually searched for potential studies. Five key concepts were identified for each search term, forming the basis of the search strategies.

Concept 1 (adversity): adverse childhood experience, life challenges, unpleasant, misfortune, hardship, distress, suffering, and sorrow. Concept 2 (resilience): resilience, resilien*, recovery, over-coming, resiliency, and adaptive function. Concept 3 (protective factors): religion, spirituality, support, self-esteem, and coping. Concept 4 (interventions): training, CBT, education, mindfulness, counseling, promotion, intervention, and meditation. Concept 5 (SRM): meta-analysis’, ‘systematic review’, and ‘review’.

The literature search was carried out by two reviewers independently, with discrepancies resolved by consensus. Articles with incomplete reported data were handled by contacting corresponding authors. We used the search terms independently and/or in combination using “OR” or “AND” [(“adversity” OR “adverse childhood experience” OR “life challenges” OR “unpleasant” OR “misfortune” OR “hardship” OR “distress” OR “suffering” OR “sorrow”) AND (“resilience” OR “resilien*” OR “recovery” OR “over-coming” OR “resiliency” OR “adaptive function”) AND (“religious” OR “spirituality” OR “support” OR “self-esteem,” OR “coping”) AND (“training” OR”CBT” OR “education” OR “mindfulness” OR “counseling” OR “promotion” OR “intervention” OR “meditate”)]. A sample of the literature search strategy, the PubMed search strategy, developed using a combination of MeSH terms and free texts is presented as a Supplementary Data Sheet 1 ( Supplementary Table 1 ). In addition to the systematic database search, an article search was carried out using the reference list of the included studies and the ‘cited by’ and ‘related articles’ functions of PubMed.

Retrieved articles were imported into Endnote v8 for duplicate removal. The screening and selection of studies were conducted in two stages. Two reviewers (BBA and MAB) independently screened titles and abstracts, followed by a full-text assessment using pre-specified inclusion criteria. To be included, SRMs had to report the effects of protective factors and resilience-promoting interventions on resilience development in a global context and be published in English. Additionally, studies were required to (a) present a defined literature search strategy, (b) appraise included studies using a relevant tool, and (c) employ a standardized approach to pooling studies and reporting summary estimates. Studies were excluded for the following reasons: (a) a lack of reporting on the measures of interest, (b) a language other than English, and (c) study type (narrative reviews, editorials, correspondence, abstracts, and methodological studies). Disagreements regarding study inclusion were resolved through consensus among reviewers.

The methodological quality of included systematic reviews was assessed independently by two reviewers (AKS and AMK) using the AMSTAR tool (21). AMSTAR evaluates the quality of systematic reviews based on 11 items assessing the rigor of methods used for pooling and summarizing studies. Given its reliability and widespread use, AMSTAR was selected for this review. Studies were categorized as high (8–11), medium (4–7), or low (≤3) quality based on their total AMSTAR score. Only reviews with a medium- or high-quality rating (score ≥4) were included in this study. A predetermined quality threshold was established and agreed upon by all reviewers before the assessment process. Although the quality of the included systematic reviews was assessed, the quality of primary studies within these reviews was not explicitly evaluated for this umbrella review.

Data from the included SRMs were extracted using a standardized Excel-based data abstraction form. For each SRM, the following information was collected: (a) study identification (first author, publication year), (b) review aim and type, (c) estimate of resilience among adversity-exposed populations, (d) effects of adversity protective factors, (e) effects of resilience-promoting interventions, (f) included primary studies (design type and sample size), (g) sample size of included studies, (h) publication bias assessment methods and results, (i) quality assessment methods and scores, (j) data synthesis method (random or fixed-effects), and (k) authors’ main conclusions ( Table 1 ).

Data extracted from the included SRMs using Microsoft Excel were imported into STATA v14.0 for analysis. A combined narrative and quantitative approach was used to summarize review estimates. When multiple estimates were available, a range of estimates and a pooled estimate were presented. Pooled estimates for the effects of adversity, protective factors, and resilience-promoting interventions on later-life resilience were calculated with 95% confidence intervals (CIs) and visualized using forest plots (22). The pooled estimates with 95% CIs were presented using forest plots.

Heterogeneity among the included SRMs was assessed using Galbraith plots, inverse variance (I² statistics), and Cochran’s Q statistic. I² values, ranging from 0 to 100%, quantify the proportion of total variance attributable to heterogeneity (23, 24). I² of 25%, 50%, and 75% thresholds for low, moderate, and high heterogeneity, respectively, with a threshold of I² < 50%, is often considered indicative of low heterogeneity (25, 26). Additionally, τ² was calculated to estimate between-study variance, as I² can be influenced by study size (27). To further explore heterogeneity, 95% prediction intervals (PIs) for the standardized mean difference were calculated, providing a range for potential effect sizes in future studies (27). I², τ², and 95% PIs were computed for each SRM (28).

For heterogeneous data, a DerSimonian–Laird random-effects model was applied. Subgroup analyses were conducted based on protective factors and resilience-promoting intervention types. Sensitivity analysis was performed to evaluate the impact of individual studies on overall estimates. Publication bias was assessed using funnel plots and Egger’s regression test (29).

The number of studies included in each SRM ranged from 2 (30) to 49 (30). The minimum and maximum sample sizes were 1,557 and 69,619, respectively. Out of 44 included SRMs, 27 investigated adversity protective factors for resilience development (Wan X et al. (31), Yule K et al. (30), Lee JH et al. (32), Lavoie J et al. (33) and Schwalm FD et al. (34)). Fourteen SRMs examined the effects of resilience-promoting interventions on resilience development (Wan X et al. (31), Ang WHD et al. (35), Han S-J, Yeun Y-R et al. (36), Llistosella M et al. (37), Liu JJ et al. (38), Joyce S et al. (39), Joyce S et al. (39), Ma L et al. (40), Wu Y et al. (41), Pinto TM et al. (42), Liu X et al. (43), Hodder RK et al. (44)). Five SRMs reported the association between adversity and resilience (Wan X et al. (31), Ang WHD et al. (35), Castillo‐González A et al. (45), Lavoie J et al. (33), Morgan CA et al. (46)) ( Table 1 ).

The effects of protective factors were determined from 27 SRMs (Wan X et al. (31), Yule K et al. (30), Lee JH et al. (32), Lavoie J et al. (33), and Schwalm FD et al. (34)). The effects of protective factors ranged from 0.01 (95% CI: 0.14, 4.53) (30) to 0.45 (95% CI: 0.37, 0.519) (31). Random-effects model analysis of these studies revealed a pooled effect of protective factors on resilience development following adversity of 0.31 (95% CI: 0.22, 0.40) (I² = 78.46%; p < 0.001) ( Figure 2 )

Heterogeneity among the 27 included SRMs was assessed using Galbraith plots and I² statistics. Random-effects meta-analysis revealed a pooled effect of adversity protective factors on resilience development of 0.31 (95% CI: 0.22, 0.40) with substantial heterogeneity (I² = 78.46%, p < 0.001), as visualized in the Galbraith plot ( Supplementary Figure 1 ).

Subgroup analysis was conducted based on the protective factor type. Pooled effects for resilience development were 0.26 (95% CI: −0.04, 0.55) for availability of support, 0.09 (95% CI: −0.01, 0.19) for cognitive ability, 0.05 (95% CI: −0.71, 0.80) for community cohesion, 0.34 (95% CI: 0.14, 0.54) for positive self-perception, 0.23 (95% CI: −0.10, 0.57) for religious involvement, and 0.43 (95% CI: 0.37, 0.48) for self-regulation ( Supplementary Figure 2 ).

A funnel plot exhibited a symmetrical distribution ( Supplementary Figure 3 ). The Egger’s regression test yielded a value of 0.278, indicating no evidence of publication bias ( Supplementary Figure 4 ). Consequently, a trim and fill analysis was not conducted ( Supplementary Figure 5 ).

A leave-one-out sensitivity analysis was conducted to assess the influence of individual studies on the pooled estimate of protective factors for resilience development. Results indicated that the pooled findings were robust to the exclusion of any single study. The pooled estimate ranged from 0.297 (0.20–0.38) to 0.34 (0.28–0.41) after removing a single study ( Supplementary Figure 6 ).

Fourteen of the 44 included SRMs reported the effects of resilience-promoting interventions on resilience development following adversity (Wan X et al. (31), Ang WHD et al. (35), Han S-J, Yeun Y-R et al. (36), Llistosella M etal. (37), Liu JJ et al. (38), Joyce S et al. (39), Joyce S et al. (39), Ma L et al. (40), Wu Y et al. (41), Pinto TM et al. (42), Liu X et al. (43), Hodder RK et al. (44)). Random-effects model analysis of these studies revealed a pooled effect of resilience-promoting interventions on resilience development of 0.42 (95% CI: 0.30, 0.55), with substantial heterogeneity (I² = 78.46%; p < 0.001) ( Figure 3 ).

Heterogeneity among the 14 included SRMs was assessed using Galbraith plots and I² statistics ( Supplementary Figure 7 ). Random-effects meta-analysis revealed a pooled effect of resilience-promoting interventions on resilience development of 0.42 (95% CI: 0.30, 0.55), with substantial heterogeneity (I² = 78.46%, p < 0.001).

Subgroup analysis was conducted based on the types of resilience-promoting interventions. Pooled effects for resilience development were 0.30 (95% CI: 0.05, 0.55) for CBT interventions, 0.21 (−0.24, 0.66) for mindfulness-based interventions, 0.51 (95% CI: 0.12, 0.90) for mixed interventions, and 0.52 (95% CI: 0.42, 0.62) for resilience-promoting interventions ( Supplementary Figure 8 ).

A funnel plot showed a symmetrical distribution ( Supplementary Figure 9 ). The Egger’s regression test value was 0.647, which indicated the absence of publication bias ( Supplementary Figure 10 ). Owing to the absence of publication bias, we did not employ a trim and fill analysis ( Supplementary Figure 11 ).

Five SRMs reported the association between previous adversity and later-life resilience (Wan X et al. (31), Ang WHD et al. (35), Castillo‐González A et al. (45), Lavoie J et al. (33), Morgan CA et al. (46)). The random-effects model analysis from those studies revealed that the pooled effect of adversity for the development of resilience after previous adversity was 0.25 (95% CI: 0.05, 0.46) (I² = 85.82%; p < 0.001) ( Figure 4 ).

The Galbraith plot and I² were used to assess heterogeneity among the five included SRMs ( Supplementary Figure 12 ). Random-effects meta-analysis revealed a pooled effect of adversity on resilience of 0.25 (95% CI: 0.05, 0.46), with substantial heterogeneity (I² = 85.82%; p < 0.001). The Galbraith plot also indicated heterogeneity.

A funnel plot exhibited a symmetrical distribution ( Supplementary Figure 13 ). The Egger’s regression test yielded a value of 0.39, indicating no evidence of publication bias ( Supplementary Figure 14 ). Consequently, a trim and fill analysis was not conducted.

A leave-one-out sensitivity analysis was conducted to assess the influence of individual studies on the pooled estimate of adversity on resilience development. Results indicated that the pooled findings were robust to the exclusion of any single study. The pooled estimate ranged from 0.18 (0.04–0.42) to 0.30 (0.10–0.50) after removing individual studies ( Supplementary Figure 15 ).

Protective factors, such as caregiver support, can increase resilience without necessarily directly impacting psychological symptoms, although their relationship to maladjustment warrants further investigation (63). To accurately measure resilience in children exposed to violence, studies should consider both indicators of healthy development and protective factors (63).

Research emphasizes the importance of strong relationships and understanding broader cultural contexts in fostering children’s resilience, calling for improved prevention strategies (64, 65). Effective prevention and health promotion across diverse cultural groups requires understanding how sociocultural factors influence protective factors within families, schools, and peer groups (66). The Resilience Portfolio Model offers a structured framework for examining factors influencing children’s adjustment processes (67, 68).

This study emphasizes the importance of universal prevention, sensitive caregiving, supportive family relationships, and parental intervention programs in fostering children’s resilience (69). Caregiver programs promoting self-regulation, including emotion socialization strategies, have demonstrated improvements in socio-emotional competencies and reduced behavioral issues among preschoolers (70).

Schools are transitioning from punitive discipline toward promoting healthy development through social emotional learning (SEL) principles, cultivating supportive relationships and individual strengths among students and teachers (71–73). SEL programs, encompassing self-regulation, teacher-student relationships, positive self-perception, and symptom reduction, can be integrated into school-based initiatives (74). Mindfulness-based interventions (75), such as compassion and attention in schools, which teach attention focus and mental/physical self-control (76, 77), have shown benefits for self-perception, well-being, self-regulation, coping, and mental health in children and adolescents (78, 79).

Trauma-sensitive schools adopt a comprehensive approach, including SEL curricula, support services, and targeted interventions, to address the impact of trauma on children’s behavior (80–82). Research affirms the potential of trauma-sensitive schools to enhance children’s functioning, regardless of trauma exposure (82). Although this meta-analysis highlights the significance of protective factors in parenting and school-based programs, their effectiveness in enhancing children’s functioning is often underestimated (83). Program evaluations investigating the impact of self-regulation and supportive teacher-student relationships on children’s health and well-being contribute to resilience research by examining their additive and buffering effects.

The findings of this study should be interpreted with caution due to several limitations. First, the included studies exhibited substantial heterogeneity, indicating variability in the methodologies, populations, and outcomes reported. Although a random-effects model and subgroup analysis were employed to account for this heterogeneity, it is important to acknowledge that the pooled estimate may not accurately represent the true effect size in all contexts. Second, the geographical distribution of the included studies was uneven, with a potential overrepresentation of certain regions. This limitation may restrict the generalizability of the findings to populations from underrepresented geographical areas. Finally, the inclusion of only English language studies might have led to the exclusion of relevant research conducted in other languages. This language bias could potentially limit the comprehensiveness of the evidence base.