Identifying Transdiagnostically Relevant Risk and Protective Factors for Internalizing Psychopathology: An Umbrella Review of Longitudinal Meta-Analyses

Abstract

Internalizing mental disorders are highly comorbid with one another, and evidence suggests that etiological processes contributing to these disorders often overlap. This systematic umbrella review aimed to synthesize meta-analytic evidence from observational longitudinal studies to provide a comprehensive overview of potentially modifiable risk and protective factors across the depressive, anxiety, and eating disorder psychopathology domains. Six databases were searched from inception to August 2022. Only meta-analyses of longitudinal studies that accounted for baseline psychopathology (either via exclusion of baseline cases or statistical adjustment for baseline symptoms) were included. Methodological quality of meta-analyses was evaluated using the AMSTAR 2, and quality of evidence for each analysis was rated using GRADE. Study selection, quality assessment, and data extraction were conducted in duplicate by independent reviewers. The protocol for this review was registered with PROSPERO (CRD42020185575). Sixty-one meta-analyses were included, corresponding to 137 meta-analytic estimates for unique risk/protective factor-psychopathology relationships. Most potential risk/protective factors, however, were examined only in relation to depressive psychopathology. Concern over mistakes and self-esteem were the only risk and protective factors, respectively, identified as statistically significant across depressive, anxiety, and eating disorder psychopathology domains. Eight risk factors and four protective factors also emerged as having transdiagnostic relevance across depressive and anxiety domains. Results suggest intervention targets that may be valuable for preventing/treating the spectrum of internalizing psychopathology and reducing comorbidity. However, few factors were identified as transdiagnostically relevant across all three internalizing domains, highlighting the need for more research investigating similar sets of potential risk/protective factors across internalizing domains.

Introduction

Depressive and anxiety disorders have long been recognized as involving a tendency to direct negative emotions inwardly, or to “internalize” them (e.g., Achenbach, 1966; Krueger, 1999). More recently, with the development of the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al., 2017)—a dimensional classification of psychopathology based on structural evidence—it has become apparent that along with depressive and anxiety disorders, eating disorders also fall within the internalizing spectrum (Forbush et al., 2010, 2017; Kotov et al., 2017, 2021). Internalizing mental disorders affect roughly one in three individuals at some point in their lives (Hudson et al., 2007; Kessler et al., 2005; Nuyen et al., 2021) and account for a substantial burden of disease that has increased over recent decades (Murray et al., 2012; Rehm & Shield, 2019). Disorders across the internalizing spectrum are highly comorbid with one another (Hudson et al., 2007; Kessler et al., 2005), and evidence suggests that etiological processes contributing to these disorders often overlap (Conway et al., 2019). Therefore, adopting a transdiagnostic approach to prevention and treatment by targeting shared modifiable risk and protective factors may translate to improved efficiency and dissemination of interventions. Efficient treatments with relevance beyond a single disorder would be particularly valuable in contexts in which traditional treatments remain unaffordable or unavailable (Farchione & Bullis, 2014) and may also lead to improved treatment outcomes for individuals with psychiatric comorbidities across the internalizing spectrum.

Numerous meta-analyses have examined a wide range of risk and protective factors for psychopathology in the depressive, anxiety, and eating disorder domains separately. For example, prior work has found low socioeconomic status to be a risk factor for depression (Su et al., 2021), physical activity to be a protective factor in relation to anxiety (McDowell et al., 2019), and dieting to be a risk factor for eating disorders (Stice, 2002). However, synthesis of these literatures with consistent methodological evaluation across domains is needed to understand (a) which potentially modifiable risk and protective factors are shared across depressive, anxiety, and eating psychopathology domains, (b) which are domain-specific, and (c) which have been identified in one or two domains but have not yet been studied across all three. Advancing understanding in these three areas is crucial for delineating risk/protective factors to target in transdiagnostic interventions, providing insights regarding domain-specific targets that might be uniquely effective for mental disorders within the relevant domain, and identifying knowledge gaps pertaining to transdiagnostic processes, respectively.

Umbrella reviews, which aggregate findings from published systematic reviews/meta-analyses using a uniform approach, allow for comparison across results from previous research syntheses (Aromataris et al., 2015; Fusar-Poli & Radua, 2018; Ioannidis, 2009). As such, umbrella reviews represent a valuable approach for synthesizing a broad scope of evidence. Umbrella reviews have previously been conducted to examine factors associated with depression (Köhler et al., 2018), anxiety (Fullana et al., 2020), and eating disorder (Solmi et al., 2021) domains separately, offering insights into possible risk processes within each domain. Arango and colleagues (2021) have also summarized the results of these umbrella reviews. However, factors identified in prior umbrella reviews do not necessarily represent true risk or protective factors according to the definition of risk/protective factors introduced by Kraemer and colleagues (1997) and detailed further by Stice (2002). Per this definition, longitudinal evidence must establish that risk/protective factors predict either (a) subsequent onset of the outcome among individuals determined not to have the outcome as baseline (for dichotomous outcomes) or (b) changes in symptoms over time (for continuous outcomes; Kraemer et al., 1997; Stice, 2002), yet prior umbrella reviews did not require the factors examined to clearly precede the outcome. For example, even when restricting to meta-analyses deemed by the authors as prospective, factors such as physical abuse in childhood were reported retrospectively (e.g., Mandelli et al., 2015), and baseline levels of psychopathology were often not accounted for in prospective studies (e.g., Clauss & Blackford, 2012). Additionally, methodological and reporting differences (e.g., variation in eligibility criteria, use of different quality assessment tools) across these umbrella reviews make it difficult to meaningfully compare the evidence across the three internalizing psychopathology domains.

To overcome the aforementioned limitations of prior reviews, the present study used an umbrella review approach to provide a comprehensive overview of potentially modifiable risk and protective factors that have been demonstrated to temporally precede psychopathology across the depressive, anxiety, and eating disorder domains. Specifically, this umbrella review aimed to synthesize meta-analyses—which represent the highest level of evidence synthesis below umbrella reviews (Fusar-Poli & Radua, 2018)—of observational, longitudinal studies that accounted for baseline psychopathology. The focus on risk and protective factors that are potentially modifiable was driven by the ultimate goal of this review, which is to inform intervention efforts.

Methods

The systematic literature search was conducted according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines (Page et al., 2021). The protocol for this umbrella review was registered in PROSPERO (CRD42020185575). Changes in the methods from the registered protocol were as follows: (1) we used the AMSTAR 2 rather than the original AMSTAR to assess methodological quality of the included meta-analyses, (2) we evaluated the quality of evidence for each analysis using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach (Balshem et al., 2011) rather than classifying the level of evidence according to the criteria outlined by Fusar-Poli and Radua (2018), and (3) we revised our approach to dealing with overlapping meta-analyses (i.e., when more than one meta-analysis examined the association between the same risk/protective factor and the same psychopathology domain) for meta-analyses identified after the original literature search. These changes are described in further detail in Supplemental Appendix A online.

Literature Search

We systematically searched MEDLINE, PsycINFO, Embase, Scopus, JBI Database of Systematic Reviews and Implementation Reports, and Cochrane Database of Systematic Reviews from inception to December 31, 2019 in an initial search, and we updated the search through August 3, 2022 using the same search strategy. The search was restricted to peer-reviewed meta-analyses of human studies published in English, as it was not feasible to translate articles published in other languages. The search algorithm used in MEDLINE is presented in Table 1. This search strategy was adapted for each additional database (details provided in Supplemental Appendix B online). In addition, reference lists of any relevant umbrella reviews or the equivalent (e.g., overviews of reviews) identified during the screening process were screened for additional articles.

Table 1.

Search Algorithm Used for the MEDLINE Database

Fields	Terms
Title	(depress* OR dysthym* OR anxiety OR panic OR phobia OR agoraphobia OR “eating disorder” OR “eating disorders” OR “disordered eating” OR “eating pathology” OR bulimi* OR anorexi* OR “binge eating” OR “binge-eating” OR internalizing OR internalising OR psychopathology OR “mental health” OR “mental disorders” OR “psychiatric disorders” OR “affective disorders” OR “mood disorders”)
	AND
Title/abstract	(“systematic review” OR “systematic literature review” OR “meta-analysis” OR “meta-analytic”)
	AND
All fields	((longitudinal* OR prospective*) OR
Title	(risk OR protective))

Note. The search was limited to human studies published in English.

Eligibility Criteria

Peer-reviewed systematic reviews of observational longitudinal studies with meta-analytic summary estimates derived from at least two primary studies meeting all criteria detailed below were eligible for inclusion. Systematic reviews without meta-analysis, meta-analyses of data not identified via systematic review, and umbrella reviews or the equivalent (e.g., overview of reviews) were excluded.

Population

Meta-analyses focused exclusively on populations in which the etiology of internalizing psychopathology is expected to be unique were excluded, with such populations including: (a) pregnant or postpartum women, (b) the elderly, (c) specific subgroups of the population not directly relevant to the objective of the present review (e.g., caregivers, stroke patients), and (d) populations in specific contexts not directly relevant to the objective of the present review (e.g., pre- or post-surgery). Meta-analyses of human studies with participants of any age not expected to have unique etiology, any sex/gender, any race/ethnicity, any nationality, and any weight status were eligible for inclusion, including meta-analyses focused on specific subgroups of the population defined by such characteristics (e.g., adolescents, women, individuals of higher weight status).

Risk/Protective Factors

Meta-analyses of observational longitudinal studies that explored at least one potentially modifiable risk and/or protective factor were eligible for inclusion. In accordance with the definition of risk and protective factors introduced by Kraemer and colleagues (1997) in which risk/protective factors must longitudinally predict subsequent psychopathological outcomes, meta-analyses including only cross-sectional studies, retrospective studies, or studies examining subsequent psychopathology without accounting for initial levels of psychopathology were excluded. Meta-analyses exclusively focused on experimental studies, non-modifiable factors (e.g., age, sex, race/ethnicity, genetics), health conditions generally not considered to be preventable (e.g., type 1 diabetes), factors unique to a specific population (e.g., gender-affirming hormones for transgender populations), subthreshold levels of an outcome of interest predicting threshold levels of the same outcome, or one domain of internalizing psychopathology predicting another (e.g., anxiety predicting depression) were also excluded.

Outcomes

Meta-analyses were eligible for inclusion if outcome(s) included symptoms or diagnoses corresponding to unipolar depressive disorders, anxiety disorders, or eating disorders in the 5th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) that fall underneath the internalizing spectrum of the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al., 2017). Therefore, meta-analyses in which outcome(s) included symptoms or diagnoses corresponding to major depressive disorder, persistent depressive disorder (dysthymia), generalized anxiety disorder, social anxiety disorder, separation anxiety disorder, panic disorder, agoraphobia, specific phobia, anorexia nervosa, bulimia nervosa, or binge-eating disorder were eligible for inclusion. Reviews focused exclusively on suicidality, psychopathology during or after pregnancy (e.g., postpartum depression), or mental disorders not specified in inclusion criteria (e.g., bipolar disorder) were excluded. Meta-analyses focused exclusively on illness course (e.g., symptom/disorder recurrence or persistence) were also excluded, as were meta-analyses in which internalizing symptoms could not be disaggregated into distinct psychopathology domains.

Study Selection

In duplicate, independent reviewers (VMH, KES, LMS, TBM, LMA, DRD) screened titles and abstracts and then assessed full texts of selected studies in detail against the eligibility criteria. All screening was conducted in Covidence systematic review software, in which any reviewer could screen any studies in the list of studies to be screened. Any disagreements that arose between the reviewers at either stage of the selection process were resolved through discussion or by a third reviewer. If insufficient information was provided in a publication to determine eligibility (e.g., if it was not clear whether initial levels of psychopathology had been controlled for), the corresponding author of the meta-analysis was contacted to request the required information.

Methodological Quality Assessment

Eligible meta-analyses were critically appraised in duplicate by independent reviewers (VMH, KES, LMS, TBM, LMA, DRD) using the 16-item AMSTAR (A MeaSurement Tool to Assess systematic Reviews) 2, a validated instrument designed to assess the methodological quality of systematic reviews/meta-analyses including non-randomized studies (Pieper et al., 2019; Shea et al., 2017). Interrater reliability for rating of AMSTAR 2 items was adequate (Fleiss’ κ = .73). Any discrepancies were resolved through discussion or with a third reviewer before rating the overall confidence in review results as high, moderate, low, or critically low per the guidelines proposed by the developers of the AMSTAR 2 (Shea et al., 2017). Seven of the 16 domains assessed by the AMSTAR 2 are considered critical domains; these critical domains encompass (1) establishment of review methods prior to conducting the review, (2) comprehensiveness of literature search strategy, (3) justification for excluded studies, (4) risk of bias assessment, (5) appropriateness of meta-analytical methods, (6) consideration of risk of bias when interpreting results, and (7) assessment and consideration of publication bias. Missing one of these critical domains contributes to an overall rating of low methodological quality, and missing more than one of these critical domains contributes to an overall rating of critically low methodological quality (Shea et al., 2017).

Overlapping Meta-Analyses

Overlapping evidence represents a unique challenge in umbrella review methodology, as using data from primary studies more than once without accounting for such overlap will overrepresent such results, thereby potentially leading to misleading conclusions (Ballard & Montgomery, 2017; Cooper & Koenka, 2012). A proposed approach for dealing with overlapping evidence when conducting umbrella reviews is to develop a priori criteria for selecting a single research synthesis for inclusion when multiple syntheses on the same topic are available (Ballard & Montgomery, 2017; Cooper & Koenka, 2012). Following this approach, we included only the meta-analysis with the highest methodological quality according to AMSTAR 2 ratings in instances of overlapping meta-analyses in the present study. In the event that there were overlapping meta-analyses with identical AMSTAR 2 ratings, we included only the most recent meta-analysis, with some exceptions in cases of overlap across meta-analyses from the initial and updated literature searches. In such instances, we retained data that had already been extracted from meta-analyses identified in a prior search if they (a) had been published within the last five years as of the time of the updated search and were of equal or higher methodological quality than the overlapping meta-analysis identified via the updated search and/or (b) considered a broader segment of the general population than the overlapping meta-analysis identified via the updated search (e.g., included studies of both youth and adults rather than only youth or only adults).

Data Extraction

A coding form in Covidence systematic review software was used to extract relevant data for each eligible analysis from each of the included meta-analyses, including the risk/protective factor examined (using the term designated by the authors of the meta-analysis), the psychopathology domain examined, the number of independent samples and total sample size included in each analysis, the type of pooled effect estimate reported (e.g., odds ratio, β coefficient), the pooled effect estimate and its corresponding 95% confidence interval and p-value from a covariate-adjusted, random effects model when possible (or from an unadjusted and/or fixed effects model when a covariate-adjusted, random effects model was not conducted, appropriately downgrading the methodological quality of the meta-analysis), and heterogeneity statistics. To rate the quality of evidence for each analysis (i.e., the confidence that an effect has been correctly estimated)—versus the overall methodological quality of the systematic review/meta-analysis as assessed with the AMSTAR 2—data were also extracted on risk of bias, inconsistency, indirectness, imprecision, publication bias, effect size, and evidence of a dose-response relationship. Based upon these data, the quality of evidence for each analysis was rated as high, moderate, low, or very low using the GRADE approach (Balshem et al., 2011). Data were extracted in duplicate by independent reviewers (VMH, KES, LMS, TBM, LMA, DRD), and any discrepancies were resolved through discussion or with a third reviewer. If insufficient information was provided in a publication, the corresponding author of the meta-analysis was contacted to request additional information. The coding form and all extracted data are available upon request from the corresponding author.

Statistical Analysis

Consistent with published methodological guidance for the conduct of an umbrella review (Aromataris et al., 2015), we summarized data that had already been meta-analyzed rather than re-synthesizing data from primary studies. Results were considered statistically significant at p < .05. To facilitate comparison across different types of pooled effect estimates reported in different meta-analyses as recommended by Fusar-Poli and Radua (2018), we converted pooled effect estimates to equivalent odds ratios (eORs) using the effectsize package in R (Ben-Shachar et al., 2020). We report both the original pooled effect estimates and the eORs.

Results

In total, we screened the titles and abstracts of 2,365 unique records, assessed 296 full-text articles for eligibility, and ultimately included 61 meta-analyses in this umbrella review (Figure 1). The most common reason for exclusion was not having established temporal precedence (69 articles). The 61 included meta-analyses provided 137 meta-analytic estimates for unique risk/protective factor-psychopathology relationships, of which 94 fell under the depressive psychopathology domain, 28 fell under the anxiety psychopathology domain, and 15 fell under the eating disorder psychopathology domain. Based on AMSTAR 2 evaluation of methodological quality, seven meta-analyses (11.5%) were rated as high quality, five (8.2%) were rated as moderate quality, 13 (21.3%) were rated as low quality, and 36 (59.0%) were rated as critically low quality (Table 2). The items that most commonly contributed to low and critically low ratings were item 2 (55.7% of meta-analyses did not include an explicit statement in the article that the review methods were established prior to the conduct of the review) and item 13 (60.7% of meta-analyses did not account for risk of bias when interpreting/discussing results of the review).

Figure 1.

Study Selection Flowchart

Note. IV = independent variable; DV = dependent variable.

Table 2.

AMSTAR 2 Quality Appraisal of Included Meta-Analyses, Ranked in Order of Highest to Lowest Methodological Quality

Reference	AMSTAR 2 Item Number																Overall Quality
	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
Matison et al., 2021	Yes	Yes	Yes	Yes*	Yes	Yes	Yes	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Mac Giollabhui et al., 2021	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Haynes et al., 2019	Yes	Yes	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Rugulies et al., 2017	Yes	Yes	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Phelan et al., 2018	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Baranyi et al., 2021	Yes	Yes	Yes	Yes	Yes	Yes	Yes*	Yes*	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Pagliai et al., 2021	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★★
Braithwaite et al., 2019	Yes	Yes	Yes	Yes*	Yes	No	Yes	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★★☆
Anglin et al., 2013	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes	Yes	Yes*	No	Yes	No	Yes	Yes	Yes	Yes	★★★☆
Yosaee et al., 2022	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	Yes	No	Yes	Yes	Yes	Yes	★★★☆
Dishman et al., 2021	Yes	Yes*	Yes	Yes*	No	Yes	Yes*	Yes	Yes*	No	Yes	No	Yes	Yes	Yes	Yes	★★★☆
Scott et al., 2021	Yes	Yes	Yes	Yes	No	No	Yes*	Yes	Yes*	No	Yes	No	Yes	Yes	Yes	Yes	★★★☆
Richardson et al., 2015	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	No	Yes	Yes	Yes	★★☆☆
Lassale et al., 2019	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes	Yes*	No	Yes	Yes	No	Yes	Yes	Yes	★★☆☆
Atlantis et al., 2013	Yes	No	Yes	Yes*	No	Yes	Yes	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★☆☆
Schuch et al., 2016	Yes	Yes*	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	No	Yes	Yes	No	Yes	★★☆☆
Su et al., 2021	Yes	Yes	Yes	Yes	Yes	Yes	Yes*	Yes*	Yes*	No	Yes	No	No	Yes	Yes	Yes	★★☆☆
Twomey, 2017	Yes	No	Yes	Yes	No	No	Yes*	Yes	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★☆☆
Rönnblad et al., 2019	Yes	Yes*	Yes	Yes*	Yes	No	No	Yes	Yes	No	Yes	Yes	Yes	No	Yes	Yes	★★☆☆
Chaplin et al., 2021	Yes	Yes	Yes	Yes*	No	Yes	Yes*	Yes	Yes*	No	Yes	Yes	No	Yes	Yes	Yes	★★☆☆
McDowell et al., 2019	Yes	Yes*	Yes	Yes*	No	Yes	Yes*	Yes	Yes*	No	Yes	No	Yes	Yes	No	Yes	★★☆☆
Luppino et al., 2010	Yes	No	Yes	Yes*	No	No	Yes*	Yes*	Yes*	No	Yes	Yes	Yes	Yes	Yes	Yes	★★☆☆
Gong et al., 2022	Yes	Yes	Yes	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	No	No	Yes	No	Yes	Yes	★★☆☆
Leung et al., 2022	Yes	Yes*	Yes	Yes*	Yes	No	Yes*	Yes	Yes*	No	Yes	No	No	No	Yes	Yes	★★☆☆
Molendijk et al., 2018	Yes	Yes	Yes	Yes*	No	No	Yes*	Yes	Yes*	No	Yes	No	No	Yes	Yes	No	★★☆☆
Hu et al., 2019	Yes	No	Yes	Yes*	Yes	Yes	Yes	Yes	Yes*	No	No	Yes	Yes	Yes	Yes	Yes	★☆☆☆
Madsen et al., 2017	Yes	Yes*	Yes	Yes	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	No	Yes	No	Yes	★☆☆☆
Kim et al., 2020	Yes	No	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	No	Yes	Yes	Yes	★☆☆☆
Virtanen et al., 2018	Yes	No	Yes	Yes	Yes	No	Yes	Yes	Yes*	No	Yes	Yes	No	Yes	Yes	Yes	★☆☆☆
Pan et al., 2012	Yes	No	Yes	Yes	Yes	Yes	Yes*	Yes	Yes*	No	Yes	Yes	No	No	Yes	Yes	★☆☆☆
Pedersen et al., 2022	Yes	Yes	Yes	Yes*	Yes	Yes	Yes*	Yes	Yes*	No	Yes	No	No	Yes	No	Yes	★☆☆☆
Huang et al., 2020	Yes	No	Yes	Yes*	No	No	Yes*	Yes	Yes*	No	No	Yes	Yes	Yes	Yes	Yes	★☆☆☆
Jadambaa et al., 2019	Yes	Yes*	Yes	Yes*	No	No	Yes*	Yes	Yes*	No	No	No	Yes	No	No	Yes	★☆☆☆
Zhang et al., 2022	Yes	No	Yes	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	No	Yes	No	Yes	Yes	Yes	★☆☆☆
Gobbi et al., 2019	Yes	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No	Yes	No	No	Yes	No	Yes	★☆☆☆
Angerer et al., 2017	Yes	No	Yes	Yes*	Yes	Yes	No	Yes	Yes	No	Yes	No	No	Yes	Yes	Yes	★☆☆☆
Ding & Zhang, 2022	Yes	No	Yes	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Kim & von dem Knesebeck, 2016	Yes	No	Yes	Yes*	Yes	No	Yes	Yes	No	No	Yes	No	No	Yes	Yes	Yes	★☆☆☆
Bean et al., 2022	Yes	No	Yes	Yes	No	Yes	Yes*	Yes	No	No	Yes	No	No	Yes	Yes	Yes	★☆☆☆
Dong et al., 2015	Yes	No	Yes	Yes*	No	Yes	Yes*	Yes*	Yes*	No	No	Yes	No	Yes	Yes	Yes	★☆☆☆
Pinquart, 2017	Yes	No	Yes	Yes*	No	Yes	Yes*	Yes*	No	No	No	No	Yes	Yes	Yes	No	★☆☆☆
Zhang et al., 2017	Yes	No	No	Yes*	Yes	Yes	Yes*	Yes*	Yes*	No	No	No	No	No	Yes	Yes	★☆☆☆
Theorell et al., 2015	Yes	No	Yes	No	Yes	No	No	Yes	Yes*	No	No	Yes	Yes	Yes	Yes	Yes	★☆☆☆
Wang et al., 2016	Yes	No	Yes	Yes*	Yes	Yes	No	Yes*	No	No	Yes	No	No	Yes	Yes	Yes	★☆☆☆
Colmsee et al., 2021	Yes	No	Yes	Yes*	Yes	Yes	Yes*	Yes	No	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Kehayes et al., 2019	Yes	No	Yes	Yes	Yes	Yes	Yes	Yes	No	No	No	No	No	No	Yes	Yes	★☆☆☆
Smith et al., 2018	Yes	No	Yes	Yes*	Yes	No	Yes	Yes	No	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Khazanov & Ruscio, 2016	Yes	No	Yes	Yes*	No	Yes	No	Yes*	No	No	Yes	No	No	Yes	Yes	Yes	★☆☆☆
Nielson et al., 2021	Yes	Yes	Yes	Yes*	Yes	Yes	Yes*	Yes	No	No	No	No	No	No	No	Yes	★☆☆☆
Fitzallen et al., 2021	Yes	No	Yes	Yes*	Yes	Yes	Yes*	No	Yes*	No	No	No	No	Yes	No	Yes	★☆☆☆
Maes et al., 2019	Yes	No	Yes	Yes*	No	No	No	Yes	No	No	Yes	No	No	No	Yes	No	★☆☆☆
Prieto-Fidalgo et al., 2022	Yes	Yes*	No	Yes*	No	No	No	Yes	No	No	No	No	No	No	Yes	Yes	★☆☆☆
Sowislo & Orth, 2013	Yes	No	Yes	Yes	Yes	Yes	No	Yes	No	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Esmaeelzadeh et al., 2018	Yes	No	Yes	No	Yes	No	No	Yes*	Yes*	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Huang, 2015a	Yes	No	Yes	Yes	No	No	No	Yes	No	No	No	No	No	Yes	Yes	Yes	★☆☆☆
Liao et al., 2022	Yes	No	Yes	Yes*	No	No	No	Yes*	No	No	No	No	No	Yes	Yes	No	★☆☆☆
Huang, 2015b	Yes	No	Yes	Yes*	No	No	No	No	No	No	No	No	No	Yes	Yes	No	★☆☆☆
Rood et al., 2009	Yes	No	Yes	Yes	No	No	No	Yes*	No	No	No	No	No	No	No	Yes	★☆☆☆
Jeronimus et al., 2016	Yes	No	Yes	No	No	No	Yes*	Yes*	No	No	No	No	No	No	No	No	★☆☆☆
Smith et al., 2016	Yes	No	Yes	No	Yes	Yes	No	Yes	No	No	No	No	No	Yes	No	No	★☆☆☆
Stice, 2002	Yes	No	Yes	No	No	No	No	Yes*	No	No	No	No	No	Yes	No	Yes	★☆☆☆
% of meta-analyses scored as “No”	0	56	3	8	36	38	25	3	31	98	41	61	61	20	20	12

Note. Shaded columns indicate critical domains.

Yes* indicates partial yes.

★★★★ = high methodological quality, ★★★☆ = moderate methodological quality, ★★☆☆ = low methodological quality, ★☆☆☆ = critically low methodological quality.

Results by Psychopathology Domain

Risk and Protective Factors for Depressive Psychopathology

In total, 56 statistically significant risk factors and 15 statistically significant protective factors were identified for depressive psychopathology (Table 3). Bullying in the workplace was identified as the strongest risk factor (eOR = 2.82), and self-esteem was identified as the strongest protective factor (eOR = 0.56). Vitamin D deficiency was identified as the strongest risk factor (eOR = 2.21) with the highest quality of evidence as rated by the GRADE approach (moderate), and caffeine consumption was identified as the strongest protective factor (eOR = 0.84) with the highest quality of evidence as rated by the GRADE approach (moderate).

Table 3.

Risk/Protective Factors Examined in Relation to Depressive Psychopathology, Ranked in Order of Strongest Risk Factor to Strongest Protective Factor

Reference	Risk/Protective Factor	k	N	Pooled Effect Size					I2	AMSTAR 2 Rating	GRADE Rating
				Metric	Estimate	95% CI	p < .05	eOR
Potential risk factors (eOR > 1.00)
Theorell et al., 2015	Bullying in the workplace	4	12,173	OR	2.82FE	(2.21, 3.59)	Yes	2.82	NR	★☆☆☆	⊕ ⊕ ◯ ◯
Smith et al., 2016	Perfectionistic attitudes	3	250	r	0.24RE	(0.11, 0.35)	Yes	2.45	0%	★☆☆☆	⊕ ◯ ◯ ◯
Pedersen et al., 2022	Parental depression	3	1,700	OR	2.30RE	(0.73, 7.24)	No	2.30	66%	★☆☆☆	⊕ ◯ ◯ ◯
Anglin et al., 2013	Vitamin D deficiency	3	8,815	HR	2.21RE	(1.40, 3.49)	Yes	2.21	21%	★★★☆	⊕ ⊕ ⊕ ◯
Zhang et al., 2022	Prenatal alcohol exposure	2	5,731	OR	2.20FE	(1.19, 4.05)	Yes	2.20	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Nielson et al., 2021	Reduced reward-related positivity	5	1,112	r	0.18RE	(0.04, 0.30)	Yes	1.94	74%	★☆☆☆	⊕ ◯ ◯ ◯
Gong et al., 2022	Maternal obesity	2	1,108,865	RR	1.92FE	(1.72, 2.11)	Yes	1.92	0%	★★☆☆	⊕ ◯ ◯ ◯
Jeronimus et al., 2016	Neuroticism	16	49,585	d	0.33NR	(0.18, 0.52)	Yes	1.82	NR	★☆☆☆	⊕ ◯ ◯ ◯
Madsen et al., 2017	Job strain	7	27,461	OR	1.77RE	(1.47, 2.13)	Yes	1.77	24%	★☆☆☆	⊕ ⊕ ◯ ◯
Schuch et al., 2016	Low cardiorespiratory fitness	3	1,131,330	HR	1.76RE	(1.61, 1.91)	Yes	1.76	12%	★★☆☆	⊕ ◯ ◯ ◯
Jadambaa et al., 2019	Bullying victimization	23	NR	RR	1.73FE	(1.46, 2.05)	Yes	1.73	71%	★☆☆☆	⊕ ⊕ ◯ ◯
Atlantis et al., 2013	Chronic obstructive pulmonary disease	6	7,439,159	RR	1.69RE	(1.45, 1.96)	Yes	1.69	71%	★★☆☆	⊕ ◯ ◯ ◯
Scott et al., 2021	Any sleep disturbances	9	11,524	OR	1.62RE	(1.43, 1.82)	Yes	1.62	64%	★★★☆	⊕ ◯ ◯ ◯
Smith et al., 2016	Doubts about actions	6	914	r	0.13RE	(0.07, 0.19)	Yes	1.61	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Smith et al., 2016	Socially prescribed perfectionism	9	1,402	r	0.13RE	(0.07, 0.18)	Yes	1.61	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Rönnblad et al., 2019	Job insecurity	6	23,648	OR	1.61RE	(1.29, 2.00)	Yes	1.61	44%	★★☆☆	⊕ ⊕ ◯ ◯
Pedersen et al., 2022	Negative family environment	8	6,192	OR	1.60RE	(0.82, 3.10)	No	1.60	89%	★☆☆☆	⊕ ◯ ◯ ◯
Luppino et al., 2010	Obesity	6	7,866	OR	1.56RE	(1.02, 2.40)	Yes	1.56	38%	★★☆☆	⊕ ◯ ◯ ◯
Smith et al., 2016	Self-criticism	5	861	r	0.12RE	(0.07, 0.20)	Yes	1.55	0%	★☆☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Maternal postnatal depression	3	10,379	OR	1.52RE	(0.92, 2.50)	No	1.52	55%	★★☆☆	⊕ ◯ ◯ ◯
Rugulies et al., 2017	Effort-reward imbalance at work	8	84,963	OR	1.49RE	(1.23, 1.80)	Yes	1.49	59%	★★★★	⊕ ◯ ◯ ◯
Gong et al., 2022	Maternal hypertensive disorders	2	12,709	RR	1.49FE	(1.11, 1.86)	Yes	1.49	3%	★★☆☆	⊕ ◯ ◯ ◯
Pan et al., 2012	Metabolic syndrome	10	26,936	OR	1.49RE	(1.19, 1.87)	Yes	1.49	57%	★☆☆☆	⊕ ⊕ ◯ ◯
Su et al., 2021	Born to parents < 20 years old	9	93,850	OR	1.46RE	(1.22, 1.74)	Yes	1.46	32%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Low birth weight	10	27,574	OR	1.44RE	(1.17, 1.76)	Yes	1.44	62%	★★☆☆	⊕ ◯ ◯ ◯
Smith et al., 2016	Concern over mistakes	9	1,402	r	0.10RE	(0.05 ,0.15)	Yes	1.44	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Smith et al., 2016	Personal standards	6	809	r	0.10RE	(0.04, 0.17)	Yes	1.44	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Huang, 2015a	Negative attributional style	47	9,482	β	0.10RE	NR	NR	1.44	NR	★☆☆☆	⊕ ◯ ◯ ◯
Nielson et al., 2021	Reduced striatal reward processing	9	1,692	r	0.10RE	(0.03, 0.18)	Yes	1.44	5%	★☆☆☆	⊕ ◯ ◯ ◯
Angerer et al., 2017	Nighttime shift work	7	26,294	RR	1.42RE	(0.92, 2.19)	No	1.42	74%	★☆☆☆	⊕ ◯ ◯ ◯
Esmaeelzadeh et al., 2018	Alcohol use	2	14,628	OR	1.39RE	(0.95, 2.03)	No	1.39	89%	★☆☆☆	⊕ ◯ ◯ ◯
Bean et al., 2022	Dampening of positive affect	12	3,760	β	0.09RE	(0.05, 0.13)	Yes	1.39	22%	★☆☆☆	⊕ ◯ ◯ ◯
Liao et al., 2022	Peer victimization	64	NR	β	0.09RE	(0.08, 0.11)	Yes	1.39	NR	★☆☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Preterm birth	7	128,001	OR	1.38RE	(1.00, 1.90)	Yes	1.38	62%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	< 9 years of maternal education	9	41,768	OR	1.37RE	(1.19, 1.57)	Yes	1.37	40%	★★☆☆	⊕ ⊕ ◯ ◯
Gobbi et al., 2019	Cannabis use during adolescence	7	11,606	OR	1.37RE	(1.16, 1.62)	Yes	1.37	0%	★☆☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Maternal prenatal depression	8	36,985	OR	1.36RE	(1.07, 1.72)	Yes	1.36	76%	★★☆☆	⊕ ◯ ◯ ◯
Virtanen et al., 2018	Long working hours	13	43,887	OR	1.35RE	(1.07, 1.71)	Yes	1.35	NR	★☆☆☆	⊕ ⊕ ◯ ◯
Haynes et al., 2019	Perceived overweight	4	11,689	OR	1.35RE	(1.16, 1.58)	Yes	1.35	25%	★★★★	⊕ ⊕ ◯ ◯
Smith et al., 2016	Self-oriented perfectionism	6	809	r	0.08RE	(0.01, 0.15)	Yes	1.34	0%	★☆☆☆	⊕ ◯ ◯ ◯
Matison et al., 2021	High Dietary Inflammatory Index	3	NR	OR	1.33RE	(1.04, 1.70)	Yes	1.33	43%	★★★★	⊕ ◯ ◯ ◯
Hu et al., 2019	Sugar-sweetened beverage consumption	4	536,895	RR	1.30FE	(1.19, 1.41)	Yes	1.30	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Pinquart, 2017	Parental harsh control	12	NR	r	0.07RE	(0.02, 0.13)	Yes	1.29	0%	★☆☆☆	⊕ ◯ ◯ ◯
Rood et al., 2009	Emotion-focused rumination	9	2,387	r	0.07NR	(0.03, 0.11)	Yes	1.29	NR	★☆☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Low socioeconomic status	6	28,710	OR	1.29RE	(1.10, 1.52)	Yes	1.29	60%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Paternal smoking	4	26,237	OR	1.28RE	(1.11, 1.48)	Yes	1.28	33%	★★☆☆	⊕ ⊕ ◯ ◯
Mac Giollabhui et al., 2021	Tumor necrosis factor α	3	1,063	r	0.06RE	(0.00, 0.12)	Yes	1.25	0%	★★★★	⊕ ◯ ◯ ◯
Chaplin et al., 2021	Smoking	4	9,134	OR	1.23RE	(1.02, 1.49)	Yes	1.23	11%	★★☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental psychological control	29	NR	r	0.05RE	(0.01, 0.09)	Yes	1.20	0%	★☆☆☆	⊕ ◯ ◯ ◯
Pagliai et al., 2021	High consumption of ultra-processed foods	2	41,637	RR	1.20RE	(1.03, 1.40)	Yes	1.20	42%	★★★★	⊕ ◯ ◯ ◯
Mac Giollabhui et al., 2021	Interleukin-6	7	7,807	r	0.05RE	(0.02, 0.07)	Yes	1.19	10%	★★★★	⊕ ⊕ ◯ ◯
Huang et al., 2020	Sedentary behavior	8	80,565	RR	1.15RE	(1.05, 1.27)	Yes	1.15	50%	★☆☆☆	⊕ ◯ ◯ ◯
Matison et al., 2021	High adherence to a Western diet	7	79,917	OR	1.15RE	(1.04, 1.26)	Yes	1.15	46%	★★★★	⊕ ◯ ◯ ◯
Luppino et al., 2010	Overweight	5	6,118	OR	1.14RE	(0.83, 1.55)	No	1.14	14%	★★☆☆	⊕ ◯ ◯ ◯
Richardson et al., 2015	Poor neighborhood socioeconomic conditions	10	6,835,959	OR	1.14RE	(1.01, 1.28)	Yes	1.14	82%	★★☆☆	⊕ ◯ ◯ ◯
Zhang et al., 2017	Meat consumption	3	20,072	RR	1.13FE	(1.03, 1.24)	Yes	1.13	19%	★☆☆☆	⊕ ⊕ ◯ ◯
Su et al., 2021	Born to parents > 35 years old	6	73,208	OR	1.13RE	(1.07, 1.20)	Yes	1.13	0%	★★☆☆	⊕ ◯ ◯ ◯
Baranyi et al., 2021	Neighborhood crime	16	175,121	r	0.03RE	(0.01, 0.05)	Yes	1.12	NR	★★★★	⊕ ◯ ◯ ◯
Kim & von dem Knesebeck, 2016	Unemployment	11	60,258	OR	1.12RE	(1.08, 1.17)	Yes	1.12	40%	★☆☆☆	⊕ ◯ ◯ ◯
Braithwaite et al., 2019	Long-term PM2.5 exposure	2	69,114	OR	1.11RE	(1.00, 1.23)	Yes	1.11	20%	★★★☆	⊕ ◯ ◯ ◯
Kim et al., 2020	Aspirin use	7	1,944,481	OR	1.11RE	(1.08, 1.14)	Yes	1.11	0%	★☆☆☆	⊕ ⊕ ⊕ ◯
Leung et al., 2022	Prescription opioid use	8	113,467	OR	1.10RE	(0.89, 1.36)	No	1.10	NR	★★☆☆	⊕ ◯ ◯ ◯
Molendijk et al., 2018	High adherence to unhealthy food groups	7	97,632	OR	1.09RE	(1.00, 1.19)	No	1.09	26%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Maternal anxiety	2	13,127	OR	1.09RE	(1.02, 1.17)	Yes	1.09	0%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Maternal drinking	2	13,976	OR	1.09RE	(0.96, 1.24)	No	1.09	0%	★★☆☆	⊕ ◯ ◯ ◯
Su et al., 2021	Second or later in birth order	5	21,246	OR	1.08RE	(0.91, 1.29)	No	1.08	30%	★★☆☆	⊕ ◯ ◯ ◯
Mac Giollabhui et al., 2021	C-reactive protein	15	40,627	r	0.02RE	(0.00, 0.03)	Yes	1.07	40%	★★★★	⊕ ◯ ◯ ◯
Molendijk et al., 2018	High adherence to unhealthy dietary patterns	10	84,870	OR	1.05RE	(0.99, 1.12)	No	1.05	45%	★★☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Observing facet of dispositional mindfulness	15	1,161	β	0.01RE	NR	No	1.04	NR	★☆☆☆	⊕ ◯ ◯ ◯
Potential protective factors (eOR < 1.00)
Matison et al., 2021	High adherence to a “healthy” diet	8	75,965	OR	0.97RE	(0.95, 1.00)	No	0.97	30%	★★★★	⊕ ⊕ ◯ ◯
Prieto-Fidalgo et al., 2022	Describing facet of dispositional mindfulness	16	1,508	β	−0.02RE	NR	No	0.93	NR	★☆☆☆	⊕ ◯ ◯ ◯
Matison et al., 2021	High adherence to a Mediterranean diet	3	10,343	OR	0.93RE	(0.84, 1.04)	No	0.93	81%	★★★★	⊕ ◯ ◯ ◯
Matison et al., 2021	High vegetable intake	4	176,659	OR	0.91RE	(0.87, 0.96)	Yes	0.91	0%	★★★★	⊕ ⊕ ◯ ◯
Molendijk et al., 2018	High adherence to neutral food groups	7	98,084	OR	0.91RE	(0.84, 1.00)	No	0.91	43%	★★☆☆	⊕ ◯ ◯ ◯
Rood et al., 2009	Distraction response style	4	1,063	r	−0.03NR	(−0.09, 0.03)	No	0.90	NR	★☆☆☆	⊕ ◯ ◯ ◯
Molendijk et al., 2018	High adherence to healthy food groups	18	147,011	OR	0.89RE	(0.83, 0.95)	Yes	0.89	71%	★★☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental behavioral control	30	NR	r	−0.04RE	(−0.09, 0.01)	No	0.86	0%	★☆☆☆	⊕ ◯ ◯ ◯
Matison et al., 2021	High fruit intake	4	176,659	OR	0.85RE	(0.81, 0.90)	Yes	0.85	0%	★★★★	⊕ ⊕ ◯ ◯
Dong et al., 2015	Higher tea consumption	5	9,000	RR	0.85RE	(0.72, 0.99)	Yes	0.85	32%	★☆☆☆	⊕ ⊕ ⊕ ◯
Wang et al., 2016	Caffeine consumption	4	29,033	RR	0.84RE	(0.75, 0.93)	Yes	0.84	0%	★☆☆☆	⊕ ⊕ ⊕ ◯
Huang, 2015b	Academic achievement	27	12,982	β	−0.06RE	(−0.10, −0.03)	Yes	0.80	0%	★☆☆☆	⊕ ◯ ◯ ◯
Dishman et al., 2021	Physical activity	179	NR	OR	0.79RE	(0.75, 0.82)	Yes	0.79	88%	★★★☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental warmth	74	NR	r	−0.07RE	(−0.10, −0.04)	Yes	0.78	0%	★☆☆☆	⊕ ◯ ◯ ◯
Lassale et al., 2019	High Healthy Eating Index score	4	45,533	OR	0.76RE	(0.57, 1.02)	No	0.76	81%	★★☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Acting with awareness facet of dispositional mindfulness	8	8,747	β	−0.08FE	NR	Yes	0.75	NR	★☆☆☆	⊕ ◯ ◯ ◯
Khazanov & Ruscio, 2016	Positive emotionality	58	24,229	β	−0.08RE	(−0.09, −0.06)	Yes	0.75	28%	★☆☆☆	⊕ ◯ ◯ ◯
Ding & Zhang, 2022	High dietary vitamin E intake	2	14,777	RR	0.74FE	(0.41, 1.32)	No	0.74	46%	★☆☆☆	⊕ ◯ ◯ ◯
Theorell et al., 2015	Decision latitude at work	18	61,867	OR	0.73FE	(0.68, 0.77)	Yes	0.73	NR	★☆☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Non-judging facet of dispositional mindfulness	16	1,508	β	−0.09RE	NR	No	0.72	NR	★☆☆☆	⊕ ◯ ◯ ◯
Phelan et al., 2018	Adherence to a diet high in magnesium	2	18,156	OR	0.71RE	(0.40, 1.02)	No	0.71	4%	★★★★	⊕ ◯ ◯ ◯
Yosaee et al., 2022	High zinc intake	4	15,852	RR	0.66RE	(0.50, 0.82)	Yes	0.66	14%	★★★☆	⊕ ⊕ ◯ ◯
Prieto-Fidalgo et al., 2022	Non-reacting facet of dispositional mindfulness	16	1,788	β	−0.12RE	NR	Yes	0.65	NR	★☆☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental autonomy granting	10	NR	r	−0.12RE	(−0.20, −0.03)	Yes	0.65	0%	★☆☆☆	⊕ ◯ ◯ ◯
Sowislo & Orth, 2013	Self-esteem	77	35,501	β	−0.16RE	(−0.18, −0.14)	Yes	0.56	66%	★☆☆☆	⊕ ◯ ◯ ◯

Note. CI = confidence interval; eOR = equivalent odds ratio; OR = odds ratio; RR = risk ratio; RE = random effects model; FE = fixed effects model; NR = not reported.

k represents the number of samples; N represents the number of participants; I² represents heterogeneity. AMSTAR 2 rating indicates overall methodological quality of the review, where ★☆☆☆ = critically low, ★★☆☆ = low, ★★★☆ = moderate, and ★★★★ = high. GRADE rating indicates quality of evidence for the specific risk/protective factor, where ⊕ ◯ ◯ ◯ = very low, ⊕ ⊕ ◯ ◯ = low, ⊕ ⊕ ⊕ ◯ = moderate, and ⊕ ⊕ ⊕ ⊕ = high.

Risk and Protective Factors for Anxiety Psychopathology

In total, 11 statistically significant risk factors and 5 statistically significant protective factors were identified for anxiety psychopathology (Table 4). Job insecurity was identified as the strongest risk factor (eOR = 2.63), and physical activity was identified as the strongest protective factor (eOR = 0.66). Maternal severe obesity was identified as the strongest risk factor (eOR = 1.56) with the highest quality of evidence as rated by the GRADE approach (low), and physical activity, which was the strongest protective factor overall, was also the protective factor with the highest quality of evidence as rated by the GRADE approach (low).

Table 4.

Risk/Protective Factors Examined in Relation to Anxiety Psychopathology, Ranked in Order of Strongest Risk Factor to Strongest Protective Factor

Reference	Risk/Protective Factor	k	N	Pooled Effect Size					I2	AMSTAR 2 Rating	GRADE Rating
				Metric	Estimate	95% CI	p < .05	eOR
Potential risk factors (eOR > 1.00)
Fitzallen et al., 2021	Preterm birth	4	1,197	OR	2.63FE	(0.87, 7.95)	No	2.63	0%	★☆☆☆	⊕ ⊕ ◯ ◯
Rönnblad et al., 2019	Job insecurity	2	7,910	OR	1.77RE	(1.18, 2.65)	Yes	1.77	52%	★★☆☆	⊕ ◯ ◯ ◯
Smith et al., 2018	Doubts about actions	5	1,191	r	0.13RE	(0.05, 0.20)	Yes	1.61	33%	★☆☆☆	⊕ ◯ ◯ ◯
Gong et al., 2022	Maternal severe obesity	2	1,107,912	RR	1.56FE	(1.44, 1.68)	Yes	1.56	0%	★★☆☆	⊕ ⊕ ◯ ◯
Maes et al., 2019	Loneliness	10	3,995	β	0.12RE	(0.04, 0.21)	Yes	1.55	72%	★☆☆☆	⊕ ◯ ◯ ◯
Jadambaa et al., 2019	Bullying victimization	20	NR	RR	1.52FE	(1.35, 1.72)	Yes	1.52	34%	★☆☆☆	⊕ ⊕ ◯ ◯
Smith et al., 2018	Concern over mistakes	5	1,264	r	0.11RE	(0.06, 0.17)	Yes	1.49	0%	★☆☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental psychological control	18	NR	r	0.10RE	(0.05, 0.15)	Yes	1.44	0%	★☆☆☆	⊕ ◯ ◯ ◯
Jeronimus et al., 2016	Neuroticism	9	14,646	d	0.18NR	(0.06, 0.27)	Yes	1.39	NR	★☆☆☆	⊕ ◯ ◯ ◯
Gong et al., 2022	Maternal moderate obesity	2	1,107,912	RR	1.36FE	(1.28, 1.44)	Yes	1.36	0%	★★☆☆	⊕ ⊕ ◯ ◯
Pinquart, 2017	Parental harsh control	2	NR	r	0.08RE	(−0.06, 0.21)	No	1.34	0%	★☆☆☆	⊕ ◯ ◯ ◯
Smith et al., 2018	Personal standards	4	962	r	0.08RE	(0.02, 0.15)	Yes	1.34	0%	★☆☆☆	⊕ ◯ ◯ ◯
Gong et al., 2022	Maternal hypertensive disorders	3	30,517	RR	1.32RE	(0.84, 1.79)	No	1.32	57%	★★☆☆	⊕ ◯ ◯ ◯
Liao et al., 2022	Peer victimization	22	NR	β	0.07RE	(0.05, 0.09)	Yes	1.29	NR	★☆☆☆	⊕ ◯ ◯ ◯
Smith et al., 2018	Socially prescribed perfectionism	9	2,665	r	0.04RE	(0.00, 0.09)	No	1.16	19%	★☆☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental behavioral control	18	NR	r	0.03RE	(−0.03, 0.09)	No	1.12	0%	★☆☆☆	⊕ ◯ ◯ ◯
Smith et al., 2018	Self-oriented perfectionism	8	2,363	r	0.03RE	(−0.01, 0.07)	No	1.12	0%	★☆☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Observing facet of dispositional mindfulness	13	495	β	0.03RE	NR	No	1.12	NR	★☆☆☆	⊕ ◯ ◯ ◯
Twomey, 2017	Cannabis use	7	53,719	OR	1.09RE	(0.97, 1.24)	No	1.09	18%	★★☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Non-judging facet of dispositional mindfulness	14	842	β	0.01RE	NR	No	1.04	NR	★☆☆☆	⊕ ◯ ◯ ◯
Potential protective factors (eOR < 1.00)
Prieto-Fidalgo et al., 2022	Describing facet of dispositional mindfulness	14	842	β	−0.02RE	NR	No	0.93	NR	★☆☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental autonomy granting	14	NR	r	−0.04RE	(−0.10, 0.02)	No	0.86	0%	★☆☆☆	⊕ ◯ ◯ ◯
Pinquart, 2017	Parental warmth	20	NR	r	−0.05RE	(−0.11, 0.01)	No	0.83	0%	★☆☆☆	⊕ ◯ ◯ ◯
Khazanov & Ruscio, 2016	Positive emotionality	26	11,494	β	−0.06RE	(−0.09, −0.04)	Yes	0.80	30%	★☆☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Non-reacting facet of dispositional mindfulness	13	495	β	−0.09RE	NR	Yes	0.72	NR	★☆☆☆	⊕ ◯ ◯ ◯
Sowislo & Orth, 2013	Self-esteem	18	3,597	β	−0.10RE	(−0.14, −0.06)	Yes	0.69	18%	★☆☆☆	⊕ ◯ ◯ ◯
Prieto-Fidalgo et al., 2022	Acting with awareness facet of dispositional mindfulness	17	1,175	β	−0.11RE	NR	Yes	0.67	NR	★☆☆☆	⊕ ◯ ◯ ◯
McDowell et al., 2019	Physical activity	3	17,725	OR	0.66RE	(0.53, 0.82)	Yes	0.66	62%	★★☆☆	⊕ ⊕ ◯ ◯

Note. CI = confidence interval; eOR = equivalent odds ratio; OR = odds ratio; RR = risk ratio; RE = random effects model; FE = fixed effects model; NR = not reported.

k represents the number of samples; N represents the number of participants; I² represents heterogeneity. AMSTAR 2 rating indicates overall methodological quality of the review, where ★☆☆☆ = critically low, ★★☆☆ = low, ★★★☆ = moderate, and ★★★★ = high. GRADE rating indicates quality of evidence for the specific risk/protective factor, where ⊕ ◯ ◯ ◯ = very low, ⊕ ⊕ ◯ ◯ = low, ⊕ ⊕ ⊕ ◯ = moderate, and ⊕ ⊕ ⊕ ⊕ = high.

Risk and Protective Factors for Eating Disorder Psychopathology

In total, 12 statistically significant risk factors and 1 statistically significant protective factor were identified for eating disorder psychopathology (Table 5). Evaluative concerns (a component of perfectionism) and modeling of body image/eating disturbance by family members and peers were identified as the strongest risk factors (both eORs = 1.80), and self-esteem was identified as the only protective factor (eOR = 0.73). All risk and protective factors identified for eating disorder psychopathology had very low quality of evidence as rated by the GRADE approach.

Table 5.

Risk/Protective Factors Examined in Relation to Eating Disorder Psychopathology, Ranked in Order of Strongest Risk Factor to Strongest Protective Factor

Reference	Risk/Protective Factor	k	N	Pooled Effect Size					I2	AMSTAR 2 Rating	GRADE Rating
				Metric	Estimate	95% CI	p < .05	eOR
Potential risk factors (eOR > 1.00)
Kehayes et al., 2019	Evaluative concerns perfectionism	4	1,586	r	0.16RE	(0.10, 0.22)	Yes	1.80	32%	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Family/peer modeling of body image and eating disturbances	2	449	r	0.16FE	NR	Yes	1.80	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Dieting	9	10,190	r	0.15FE	NR	Yes	1.73	NR	★☆☆☆	⊕ ◯ ◯ ◯
Kehayes et al., 2019	Concern over mistakes	3	571	r	0.14RE	(0.06, 0.22)	Yes	1.67	0%	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Body dissatisfaction	12	18,366	r	0.13FE	NR	Yes	1.61	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Perceived pressure to be thin	5	7,895	r	0.12FE	NR	Yes	1.55	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Negative affect	12	18,445	r	0.09FE	NR	Yes	1.39	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Thin-ideal internalization	4	15,128	r	0.08FE	NR	Yes	1.34	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Impulsivity	4	1,669	r	0.07FE	NR	Yes	1.29	NR	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Substance use	5	2,972	r	0.07FE	NR	Yes	1.29	NR	★☆☆☆	⊕ ◯ ◯ ◯
Kehayes et al., 2019	Socially prescribed perfectionism	2	409	r	0.06RE	(−0.04, 0.16)	No	1.24	0%	★☆☆☆	⊕ ◯ ◯ ◯
Kehayes et al., 2019	Personal/parental standards	5	1,418	r	0.06RE	(0.01, 0.11)	Yes	1.24	0%	★☆☆☆	⊕ ◯ ◯ ◯
Kehayes et al., 2019	Personal standards	4	1,394	r	0.04RE	(−0.06, 0.13)	No	1.16	66%	★☆☆☆	⊕ ◯ ◯ ◯
Stice, 2002	Body mass	11	11,063	r	0.04FE	NR	Yes	1.16	NR	★☆☆☆	⊕ ◯ ◯ ◯
Potential protective factors (eOR < 1.00)
Colmsee et al., 2021	Self-esteem	9	3,711	r	−0.09RE	(−0.14, −0.03)	Yes	0.73	61%	★☆☆☆	⊕ ◯ ◯ ◯

Note. CI = confidence interval; eOR = equivalent odds ratio; OR = odds ratio; RR = risk ratio; RE = random effects model; FE = fixed effects model; NR = not reported.

k represents the number of samples; N represents the number of participants; I² represents heterogeneity. AMSTAR 2 rating indicates overall methodological quality of the review, where ★☆☆☆ = critically low, ★★☆☆ = low, ★★★☆ = moderate, and ★★★★ = high. GRADE rating indicates quality of evidence for the specific risk/protective factor, where ⊕ ◯ ◯ ◯ = very low, ⊕ ⊕ ◯ ◯ = low, ⊕ ⊕ ⊕ ◯ = moderate, and ⊕ ⊕ ⊕ ⊕ = high.

Results Across Psychopathology Domains

Risk and Protective Factors With Transdiagnostic Relevance

Concern over mistakes (i.e., a central dimension of perfectionism characterized by a tendency to overgeneralize mistakes and failures; Frost et al., 1990) and self-esteem were the only risk and protective factors, respectively, identified as statistically significant across depressive, anxiety, and eating disorder psychopathology domains (Figure 2). Neuroticism, bullying victimization, peer victimization, doubts about actions, job insecurity, personal standards (i.e., setting unreasonably high goals for oneself; Frost et al., 1990), maternal obesity, and parental psychological control were identified as statistically significant risk factors shared across depressive and anxiety psychopathology domains, and physical activity, positive emotionality, and two facets of dispositional mindfulness (non-reacting and acting with awareness) were identified as statistically significant protective factors across depressive and anxiety psychopathology domains. Higher weight was identified as a statistically significant risk factor across depressive and eating disorder psychopathology domains, though it was operationalized as obesity (body mass index ≥ 30 kg/m²; dichotomous) when examined in relation to depressive psychopathology and as body mass (continuous) when examined in relation to eating disorder psychopathology. No other risk or protective factors with transdiagnostic relevance were identified. Figure 3 summarizes the overlap in risk and protective factors across internalizing psychopathology domains.

Figure 2.

Risk/Protective Factors Identified as Transdiagnostically Relevant Across Depressive, Anxiety, and Eating Disorder Psychopathology Domains

Note. eOR = equivalent odds ratio; CI = confidence interval. eORs > 1.00 indicate risk factors and eORs < 1.00 indicate protective factors.

Figure 3.

Summary of Risk/Protective Factors Identified Across Depressive, Anxiety, and Eating Disorder Psychopathology Domains

Note. COPD = chronic obstructive pulmonary disease. This Venn diagram summarizes risk/protective factors identified as statistically significant either across domains or for a specific domain. Many factors identified for specific domains were not examined across other domains and thus are not necessarily domain-specific. Due to the large number of risk/protective factors identified for depressive psychopathology, some are grouped into overarching categories in this figure (e.g., dietary patterns); each specific risk/protective factor for depressive psychopathology can be found in Table 3. Refer to Tables 3–5 for information regarding quality of the evidence and effect sizes, including risk versus protective status of each factor.

Domain-Specific Risk and Protective Factors

Aside from concern over mistakes and self-esteem, which were significant across all three internalizing domains, personal standards and socially prescribed perfectionism were the only other potential risk/protective factors examined across all three internalizing psychopathology domains. Personal standards was a statistically significant risk factor only for depressive and anxiety psychopathology, and socially prescribed perfectionism was a statistically significant risk factor only for depressive psychopathology. Pre-term birth, self-oriented perfectionism, cannabis use, maternal hypertensive disorders, and parental harsh control were examined as potential risk factors across depressive and anxiety psychopathology domains, and parental autonomy granting and parental warmth were examined as potential protective factors across depressive and anxiety psychopathology domains, but these potential risk/protective factors were identified as statistically significant only for depressive psychopathology. Their roles in relation to eating disorder psychopathology were not examined.

Risk and Protective Factors Only Studied in Relation to One Psychopathology Domain

Except for the domain-specific risk/protective factors mentioned above, all other statistically significant risk/protective factors displayed in a specific psychopathology domain in Figure 3 (i.e., not displaying overlap across domains) were only examined in relation to that single psychopathology domain.

Discussion

This umbrella review provides a systematic and comprehensive evaluation of existing meta-analytic evidence regarding potentially modifiable risk and protective factors for depressive, anxiety, and eating disorder psychopathology. In doing so, this umbrella review offers insights into risk and protective factors that demonstrate transdiagnostic relevance across these internalizing psychopathology domains, identifies some domain-specific risk and protective factors, and highlights areas in which more research is needed. From the literature to date, concern over mistakes emerged as the only risk factor with evidence for transdiagnostic relevance across the depressive, anxiety, and eating disorder psychopathology domains, and self-esteem emerged as the only protective factor with evidence for transdiagnostic relevance across all three domains. Notably, only four potential risk/protective factors were examined across all three domains in total, with three of these four being different dimensions of perfectionism. Thus, although the potential value of identifying shared etiological processes contributing to disorders across the internalizing spectrum has been emphasized in the literature (e.g., Conway et al., 2019), the relevant evidence base is limited. Specifically, longitudinal evidence at the meta-analytic level—one of the strongest levels of scientific evidence (Fusar-Poli & Radua, 2018)—is limited. A broader range of potential risk and protective factors was examined across both the depressive and anxiety domains, with eight risk factors and four protective factors emerging as transdiagnostically relevant across these two domains. However, the overwhelming majority of potential risk/protective factors were examined only in relation to depressive psychopathology.

Consistent with previous literature suggesting that perfectionism and self-esteem may serve transdiagnostic roles in relation to depressive, anxiety, and eating disorder psychopathology (Egan et al., 2011; Mann et al., 2004), the longitudinal, meta-analytic evidence synthesized in this umbrella review demonstrates transdiagnostic relevance of concern over mistakes and self-esteem across all three internalizing psychopathology domains. The emergence of concern over mistakes, but not other dimensions of perfectionism, as transdiagnostically relevant across all three domains may be related to the centrality of concern over mistakes in perfectionism, as this factor has been found to account for the largest amount of variance in overall perfectionism as measured by the Frost Multidimensional Perfectionism Scale (Frost et al., 1990). According to the multidimensional conceptualization of perfectionism, wherein perfectionism is subdivided into adaptive and maladaptive components, concern over mistakes is considered a maladaptive dimension and relates to a range of psychopathology (see Egan et al., 2011 for a review). However, the present results also coincide with unidimensional conceptualizations of clinical perfectionism. Clinical perfectionism is posited to encompass any aspect of perfectionism—even those previously thought of as adaptive, such as organization and striving—if self-worth is overly dependent on achievement of rigidly held standards (Shafran et al., 2002). The items most commonly used to assess concern over mistakes (i.e., the Concern Over Mistakes subscale of the Frost Multidimensional Perfectionism Scale; Frost et al., 1990) also capture such overdependence. For example, in one item, respondents rate the extent to which they agree or disagree with the statement, “If I do not do as well as other people, it means I am an inferior human being” (Frost et al., 1990). Moreover, evidence suggests that individuals high versus low in perfectionistic concern over mistakes do not differ in number or quality of mistakes, but only in their personal reactions to mistakes, such that those with higher concern over mistakes indicate stronger and more negative reactions to mistakes, more cognitive rumination, and perceptions of their mistakes as more serious (Frost et al., 1997). Therefore, this overdependence of self-worth on perfectionistic strivings and achievement may be key for explaining why concern over mistakes was the sole dimension of perfectionism identified as a transdiagnostically relevant risk factor across all three internalizing domains. Relatedly, it is possible that concern over mistakes and self-esteem may contribute to transdiagnostic risk for internalizing psychopathology in an interdependent fashion, in which individuals with lower self-esteem may be more likely to have concern over mistakes, and concern over mistakes may also lead to lower self-esteem. Indeed, the cognitive behavioral model of clinical perfectionism posits that these factors interact in a negative, self-reinforcing cycle (see Lo & Abbott, 2013 for a review). Therefore, it is likely that concern over mistakes and self-esteem may be important transdiagnostic targets to intervene upon simultaneously.

Strengths and Limitations

This umbrella review has several strengths. We used systematic search methods, both the study selection and data extraction were conducted by independent reviewers, we used a validated instrument to assess the methodological quality of the included meta-analyses (Shea et al., 2017), and we converted effect estimates to a common measure to facilitate comparison across risk/protective factors and psychopathology domains. Moreover, by restricting the scope of our umbrella review to meta-analyses in which temporal precedence of the potential risk/protective factors was established, factors identified in this review represent true risk/protective factors as defined by Kraemer and colleagues (1997) and detailed further by Stice (2002). However, important limitations must also be noted. While this review aimed to identify transdiagnostically relevant risk and protective factors for internalizing psychopathology, evidence regarding which mental disorders fit within the internalizing spectrum is still evolving (Kotov et al., 2021), and the scope of internalizing domains may be broader than those examined in this review. We focused on psychopathology in the unipolar depressive, anxiety, and eating disorder domains because substantial evidence indicates that several disorders within each of these three DSM-5 categories fit within the internalizing spectrum (Kotov et al., 2021). However, there is also evidence suggesting that specific disorders in other DSM-5 categories (i.e., obsessive-compulsive disorder, posttraumatic stress disorder, borderline personality disorder) and symptoms of sexual dysfunction may fit within the internalizing spectrum (Kotov et al., 2021), and we did not examine risk/protective factors for these outcomes in the present review. Further, although the results of this review suggest risk/protective factors with transdiagnostic potential across the internalizing domains examined, we could not account for comorbidity across these domains and thus cannot conclude with certainty that these factors are in fact transdiagnostic. Another limitation is that by restricting the scope of our umbrella review to meta-analyses published in English, we may have missed important studies outside of this type of publication. Additionally, definitions and measurement of each construct may have differed to some extent across meta-analyses. Finally, a key limitation is that the majority of meta-analyses—and all meta-analyses in the eating disorder psychopathology domain—included in this umbrella review had critically low methodological quality according to the AMSTAR 2. However, the most common reasons contributing to these critically low ratings, which were (a) not including an explicit statement about having established the review methods prior to the conduct of the review and (b) not commenting on risk of bias when interpreting review results, may in part reflect reduced thoroughness in reporting rather than reduced methodological rigor in review conduct. Relatedly, the AMSTAR 2 has been recognized as having a tendency to yield lower methodological quality scores than other methodological quality assessment tools (De Santis & Kaplan, 2020). For example, systematic reviews rated as medium to high quality by the original AMSTAR have been rated as low to critically low quality by the AMSTAR 2 (De Santis & Kaplan, 2020). Thus, while there is certainly room to improve the methodological quality of future meta-analyses, the methodological quality of the meta-analyses included in this umbrella review seems to be on par with the methodological quality of meta-analyses in other research areas as assessed by the AMSTAR 2 (e.g., De Santis & Kaplan, 2020; Lorenz et al., 2019).

Remaining Gaps in the Literature

A number of remaining gaps in the literature were identified via this umbrella review; these gaps may need to be filled at the level of primary studies and/or at the level of meta-analysis. For one, the large number of meta-analyses excluded due to temporal precedence of the potential risk/protective factor not having been established—even when including search terms intended to target longitudinal studies—highlights the need for more longitudinal research. In addition, considering that far more potential risk factors were examined in the included meta-analyses than were potential protective factors, more research on potential protective factors is needed to balance out the existing focus on risk factors. Finally, in contrast with the extensive range of potential risk and protective factors examined in relation to depressive psychopathology, more research is needed to identify risk and protective factors in the anxiety and eating disorder psychopathology domains.

Conclusions

Concern over mistakes and self-esteem emerged as transdiagnostically relevant risk and protective factors, respectively, across depressive, anxiety, and eating disorder psychopathology in this umbrella review, and a number of risk and protective factors were also identified as transdiagnostically relevant across depressive and anxiety psychopathology domains. These findings suggest intervention targets that may be valuable for preventing and treating the spectrum of internalizing psychopathology and for reducing comorbidity. In particular, it may prove fruitful to target concern over mistakes and self-esteem—which have both been demonstrated to be modifiable (Haney & Durlak, 1998; Lloyd et al., 2015)—in transdiagnostic intervention efforts. Encouragingly, cognitive behavioral therapy (CBT) for perfectionism, which has been shown to reduce concern over mistakes (Galloway et al., 2022), and CBT for low self-esteem have in fact each demonstrated efficacy in reducing symptoms of depression, anxiety, and eating disorders in randomized controlled trials (Biney et al., 2022; Galloway et al., 2022; Waite et al., 2012). Further research to investigate the transdiagnostic potential of CBT for perfectionism—specifically, concern over mistakes perfectionism—and low self-esteem is therefore warranted. Overall, however, few transdiagnostically relevant risk/protective factors were identified, highlighting the need for more research investigating similar sets of potential risk/protective factors across internalizing domains.

Highlights.

• We synthesized evidence across depressive, anxiety, and eating disorder domains

• Concern over mistakes and self-esteem emerged as relevant across each domain

• Additional factors emerged as relevant across depressive and anxiety domains

• Few factors were studied across all three internalizing domains