Hostname: page-component-5db58dd55d-pjp64 Total loading time: 0 Render date: 2026-06-03T14:59:15.491Z Has data issue: false hasContentIssue false
  • English
  • Français

The broad structure of psychopathology in the All of Us Research Program

Published online by Cambridge University Press:  14 November 2025

Alireza Ehteshami Open the ORCID record for Alireza Ehteshami [Opens in a new window]  and
Irwin D. Waldman Open the ORCID record for Irwin D. Waldman [Opens in a new window]
Alireza Ehteshami*
Affiliation:
Department of Psychology, Emory University, Atlanta, GA, USA
Irwin D. Waldman
Affiliation:
Department of Psychology, Emory University, Atlanta, GA, USA
*
Corresponding author: Alireza Ehteshami; Email: alireza.ehteshami@emory.edu
Rights & Permissions [Opens in a new window]

Abstract

Background

Given substantial comorbidity among, and considerable heterogeneity within, psychiatric diagnoses, researchers have suggested alternative systems for classifying psychopathology. The Hierarchical Taxonomy of Psychopathology (HiTOP) is a recently proposed framework for understanding mental disorders based on how symptoms and diagnoses tend to cluster across individuals. While the model is grounded in existing research and supported by recent meta-analytic evidence, its structure has not yet been directly tested using large, representative clinical datasets. In this study, we used electronic health record (EHR) data to examine the overall organization of mental disorders as proposed by HiTOP, with the goal of informing future research on biological and environmental risk factors as well as important life outcomes.

Methods

Data were drawn from the All of Us Research Program, a landmark nationwide US biobank initiative designed to advance population-scale health research, and included participants’ psychiatric diagnoses and sociodemographic correlates as documented in their EHRs. A total of 127,963 participants and 39 primary diagnoses were identified. We analyzed patterns of co-occurrence among psychiatric diagnoses to identify broader psychopathology dimensions, assess the overall structure of mental disorders, and clarify the placement of conditions that have been inconsistently categorized in past research. Several competing dimensional models were compared based on their statistical fit and complementary assessments of factor strength, specificity, and reproducibility.

Results

A model identifying six broad and correlated dimensions – Fear, Distress, Externalizing, Substance Use, Thought Problems, and Neurodevelopmental Disorders – provided the best fit to the data. This structure was highly consistent across analyses and showed strong split-half replicability and meaningful associations with relevant clinical and demographic characteristics.

Conclusions

These findings support a 6-factor model of psychopathology that broadly resembles major dimensions in the HiTOP framework. By addressing key gaps in the literature, this study advances our understanding of the structure and correlates of mental disorders. The results offer a foundation for more nuanced investigations into the etiology, progression, and treatment of mental health conditions.

Keywords

TaxonomyClassificationNosologyTransdiagnosticFactor analysisElectronic health records (EHR)

Information

Type
Original Article
Information
Psychological Medicine , Volume 55 , 2025 , e344
Check for updates
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press

Introduction

Psychiatric disorders and related traits (collectively termed psychopathology) are highly prevalent globally (Baumeister & Härter, Reference Baumeister and Härter2007; Kessler & Wang, Reference Kessler and Wang2008), imposing substantial personal, interpersonal, and economic burdens (Insel, Reference Insel2008). It has become evident that traditional categorical nosologies have hindered progress in gaining insight into the etiology, treatment, and prevention of psychopathology (Hyman, Reference Hyman2010; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017, Reference Kotov, Krueger, Watson, Cicero, Conway, DeYoung and Wright2021). The utility of categorical diagnoses for assessment, applied research, and clinical practice is significantly limited by factors such as pervasive comorbidity (Plana-Ripoll et al., Reference Plana-Ripoll, Pedersen, Holtz, Benros, Dalsgaard, Jonge and McGrath2019), excessive heterogeneity, low symptom specificity, poor reliability, and poor diagnostic coverage (Conway, Mansolf, & Reise, Reference Conway, Mansolf and Reise2019; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017; Krueger et al., Reference Krueger, Kotov, Watson, Forbes, Eaton, Ruggero and Zimmermann2018; Wright et al., Reference Wright, Krueger, Hobbs, Markon, Eaton and Slade2013). Addressing many of these limitations, the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017, Reference Kotov, Krueger, Watson, Cicero, Conway, DeYoung and Wright2021) has emerged as an alternative framework that synthesizes extant research on the structure of psychopathology. HiTOP aims to deliver reliable and valid phenotypes that can be used to advance understanding of the genetic, neurobiological, and environmental bases of psychopathology.

HiTOP organizes psychiatric signs, symptoms, and diagnoses into a dimensional hierarchy, from broad spectra to specific symptoms and traits. These hierarchical dimensions are derived by statistically analyzing how symptoms, maladaptive traits, and diagnoses co-occur across diverse samples—including clinical, community, and epidemiological populations—revealing which features consistently appear together. This analysis reveals that traditionally distinct disorders (e.g. major depression, generalized anxiety disorder, PTSD) share substantial common variance, suggesting they reflect underlying latent dimensions. The higher-order dimensions (e.g. Internalizing, Externalizing, and Thought Disorder) represent these latent constructs, accounting for covariation among lower-order syndromes, symptoms, and traits. For example, the Internalizing spectrum emerges from the observed clustering of depressive, anxiety, and trauma-related disorders, hypothesized to reflect shared processes such as elevated negative affectivity, while Externalizing captures the co-occurrence of substance use disorders and antisocial behaviors, potentially reflecting common disinhibitory mechanisms. Within this hierarchy, clinical features group into increasingly broad levels – from specific symptoms and traits to subfactors (e.g. Fear vs. Distress) to spectra – allowing the framework to explain comorbidity through shared liabilities while preserving clinically meaningful distinctions. Thus, higher-order dimensions effectively account for and parsimoniously explain diagnostic comorbidity by capturing common risk factors, developmental trajectories, and treatment responses (Barlow et al., Reference Barlow, Farchione, Bullis, Gallagher, Murray-Latin, Sauer-Zavala and Cassiello-Robbins2017; Dalgleish, Black, Johnston, & Bevan, Reference Dalgleish, Black, Johnston and Bevan2020; Mansell, Harvey, Watkins, & Shafran, Reference Mansell, Harvey, Watkins and Shafran2009; Parkes et al., Reference Parkes, Moore, Calkins, Cook, Cieslak, Roalf and Bassett2021), facilitating research into shared and unique etiological mechanisms while supporting transdiagnostic intervention development.

The HiTOP model was initially developed through a narrative synthesis of prior factor analytic studies, an approach that integrates findings across the literature but does not provide direct estimates of the strength or consistency of associations among disorders. Although many individual studies have supported dimensional models of psychopathology, much of this work has relied on smaller or specialized samples with limited disorder coverage – particularly for certain diagnoses such as OCD and certain neurodevelopmental disorders. These constraints limit generalizability and make it difficult to fully capture the higher-order structure of psychopathology (Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017). In addition, many prior studies have modeled individual dimensions in isolation (e.g. internalizing, externalizing), rather than examining how a broader set of dimensions covary or relate to external validators such as biological correlates or treatment outcomes (Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017, Reference Kotov, Krueger, Watson, Cicero, Conway, DeYoung and Wright2021). Most studies have also tested only a limited number of competing structural models, further restricting our ability to evaluate the relative robustness or utility of alternative dimensional frameworks (Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023). A recent meta-analysis (Ringwald, Forbes, & Wright, Reference Ringwald, Forbes and Wright2023) addressed many of these limitations by quantitatively synthesizing results across dozens of studies and providing strong support for much of the HiTOP framework. Meta-analytic methods are well-suited to account for heterogeneity and increase statistical precision, but they remain limited by the available literature – particularly with respect to the range of disorders examined and the comparability of measurement models across studies. Consequently, important questions remain about the consistency and magnitude of correlations among disorders and higher-order dimensions when evaluated directly in a single, well-powered dataset. The present study addresses this need by leveraging a large, diverse sample with broad diagnostic coverage and using a comprehensive model comparison framework to test the structure and external correlates of transdiagnostic psychopathology dimensions.

To address these limitations, we utilized psychiatric diagnoses extracted from electronic health records (EHR) in the All of Us Research Program (All of Us) – a large and demographically diverse sample representative of the US population (Epstein, Reference Epstein2022; “All of Us” Research Program Investigators et al., Reference Denny, Rutter, Goldstein, Philippakis, Smoller and Dishman2019). We used Confirmatory Factor Analyses (CFAs), which test whether patterns of observed diagnostic co-occurrence fit hypothesized dimensional structures informed by HiTOP. This approach allows us to statistically evaluate the extent to which psychiatric diagnoses can be captured by interpretable higher-order dimensions, despite the fact that diagnoses themselves are heterogeneous and shaped by clinical judgment. While less granular than symptom-level data, diagnoses remain the most widely used and scalable phenotypes available in large-scale health systems, enabling investigation of broad structural patterns in real-world settings.

Within the context of evaluating the overarching HiTOP structure, we addressed several specific a priori questions regarding the classification of psychopathology. First, we tested whether substance use disorders and non–substance-related externalizing disorders are best represented by distinct yet correlated latent dimensions. This question was motivated by dimensional models of psychopathology that conceptualize conduct disorder (CD), antisocial personality disorder (ASPD), and substance use disorders as reflecting a shared, genetically mediated liability for externalizing behavior (Krueger et al., Reference Krueger, Hobbs, Conway, Dick, Dretsch and Eaton2021; Krueger, Markon, Patrick, Benning, & Kramer, Reference Krueger, Markon, Patrick, Benning and Kramer2007). While many latent variable studies support a unidimensional externalizing construct, others have questioned whether substance use reflects unique etiological and clinical features that warrant separate classification (Jablensky, Reference Jablensky2009; Poore et al., Reference Poore, Hatoum, Mallard, Sanchez-Roige, Waldman, Palmer and Dick2023; Verona, Javdani, & Sprague, Reference Verona, Javdani and Sprague2011; Voorhees et al., Reference Voorhees, Dennis, Elbogen, Clancy, Hertzberg, Beckham and Calhoun2014). For instance, Verona et al. (Reference Verona, Javdani and Sprague2011) found that in a sample of adolescents, a three-factor model that distinguished Substance Use from other externalizing behaviors (e.g. conduct problems, ADHD, oppositional defiant disorder) provided superior fit compared to two-factor models, suggesting meaningful differentiation even in youth samples. Similarly, in a sample of adult veterans, Voorhees et al. (Reference Voorhees, Dennis, Elbogen, Clancy, Hertzberg, Beckham and Calhoun2014) identified a three-factor structure comprising Internalizing, Externalizing, and Substance Abuse, reinforcing the possibility that substance use may emerge as a separate psychopathological dimension across age groups and clinical populations. Hence, we were particularly interested in whether the inclusion of a broader and more diverse range of psychopathology indicators might allow a stronger, more meaningful distinction between substance use and other externalizing disorders to emerge – something that might not be detectable in more constrained models.

Second, we examined whether Distress and Fear should be represented as separate (yet related) dimensions rather than as a single overarching Internalizing dimension. This distinction is supported by research indicating that depressive and anxiety-related disorders cluster into separable, though correlated, emotional syndromes, which differ in their patterns of comorbidity, external validators, and developmental trajectories (Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017; Watson, Clark, Simms, & Kotov, Reference Watson, Clark, Simms and Kotov2022).

Third, building on these specific modeling decisions, we then asked whether a model comprising six correlated higher-order factors – Fear, Distress, Externalizing, Substance Use, Thought Problems, and Neurodevelopmental Disorders – would yield superior fit relative to more parsimonious structures.

Fourth, we assessed whether bifactor models – often favored for their apparent superior statistical fit – would actually perform worse relative to correlated factors models on conventional model fit as well as alternative indices meant to assess the coherence and interpretability of factors. This reflects growing concern that bifactor models may capitalize on idiosyncrasies in the data and produce misleading or unstable factors (Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023; Watts, Greene, Bonifay, & Fried, Reference Watts, Greene, Bonifay and Fried2024).

Finally, we hypothesized that the six dimensions would be meaningfully distinct in their associations with external correlates, supporting their validity as separate constructs within a multidimensional framework of psychopathology.

In addition to testing alternative overarching structural models, we conducted targeted analyses to clarify the placement of disorders with ambiguous or debated positions within the HiTOP framework. These included examining whether ADHD better reflects Externalizing or Neurodevelopmental domains, where eating disorders and OCD fit within the hierarchy, and how personality disorders like ASPD and BPD may reflect multiple spectra. Full details of these disorder-specific analyses are provided in eMethods in Supplement 1.

We also refined our models to better capture relations between closely related disorders that share features beyond their primary dimensions (such as different ADHD subtypes or OCD with obsessive-compulsive personality disorder). Additionally, we used more data-driven approaches to identify potential improvements to the model, while carefully validating these changes to ensure they reflected genuine patterns rather than statistical noise (MacCallum, Roznowski, & Necowitz, Reference MacCallum, Roznowski and Necowitz1992). These refinement procedures and validation methods are detailed in the eMethods in Supplement 1.

We went beyond standard statistical measures to evaluate our models in several ways. We tested whether our identified dimensions remained stable when individual diagnoses were removed, verified that our findings held true when analyzing different halves of the sample, and confirmed our models performed better than models with random arrangements of diagnoses. Finally, we examined the criterion validity of the higher-order psychopathology dimensions through their associations with relevant correlates. This project was pre-registered on the Open Science Framework (https://osf.io/y5svr/).

Methods

Participants

Participants for this study were drawn from the All of Us Research Program (All of Us), a nationwide, prospective cohort study that aims to investigate the impact of lifestyle, environment, and genomics on health outcomes. Sponsored by the US National Institutes of Health (NIH), All of Us aims to recruit ≥1 million adults over the next several years (Epstein, Reference Epstein2022; Investigators et al., Reference Denny, Rutter, Goldstein, Philippakis, Smoller and Dishman2019). A central focus of the program is to equitably include populations traditionally underrepresented in biomedical research (Investigators et al., Reference Denny, Rutter, Goldstein, Philippakis, Smoller and Dishman2019; Mapes et al., Reference Mapes2020).

Recruitment primarily occurs through participating healthcare organizations and Federally Qualified Health Centers, although interested individuals can also enroll directly at community-based sites. Participants complete necessary procedures on the All of Us website (https://joinallofus.org), including providing informed consent and completing baseline health surveys. After enrollment, participants may undergo a basic physical examination and biospecimen collection at a partnered healthcare site. Follow-up is conducted actively through periodic surveys and passively through linkage with electronic health records (EHRs).

As of October 2022, over 537,000 individuals aged 16–65 (40% male, 60% female) had joined the program. Of these participants, more than 324,000 shared electronic health records (EHRs), 372,380 completed surveys on overall health, lifestyle, and healthcare access & utilization, 311,300 provided physical measurements, and 320,000 donated at least one biospecimen. This study analyzed data from May 6, 2018, to June 6, 2022 (release 6, N = 372,380) in compliance with the All of Us Code of Conduct.

Measures

We examined EHR lifetime psychiatric diagnoses in our primary analyses. Despite criticisms of their limitations, researchers have recently begun utilizing EHR psychiatric diagnoses in psychopathology research (Chen et al., Reference Chen, Castro, Minnier, Charney, Stahl and Smoller2018; Linder, Bastarache, Hughey, & Peterson, Reference Linder, Bastarache, Hughey and Peterson2021; Smoller, Reference Smoller2018; Smoller et al., Reference Smoller, Andreassen, Edenberg, Faraone, Glatt and Kendler2019), including in All of Us (Barr, Bigdeli, & Meyers, Reference Barr, Bigdeli and Meyers2022). To standardize EHR data across various input sources, All of Us employs the Observational Medical Outcomes Partnership (OMOP) Common Data Model (Klann, Joss, Embree, & Murphy, Reference Klann, Joss, Embree and Murphy2019). We selected all relevant OMOP codes for psychiatric conditions (listed in eTable 1 in Supplement 1). Additionally, we clustered codes for highly similar diagnoses (e.g. autism spectrum disorder, autism, autistic disorder, and Asperger’s syndrome; a full list of these clusters can be found in eTable 2 in Supplement 1). We used sex assigned at birth, annual household income, education level, and sleep disturbance as external correlates for the psychopathology dimensions identified in our preferred structural models (eMethods in Supplement 1 contains a full description of the external correlates).

Statistical analysis

Our approach involved conducting factor analyses of the EHR diagnoses on a continuum ranging from fully confirmatory to more exploratory, within a Confirmatory Factor Analytic (CFA) framework. We began by contrasting a priori hypothesized models to test progressive elaborations of the HiTOP structure, then tested alternative classifications of disorders with ambiguous placements, and finally used a more exploratory approach (involving modification indices) to search for cross-loadings and correlated residuals that might improve model fit. We relied primarily on correlated factor models but also contrasted these with bifactor models. In correlated factors models, indicators (i.e. diagnoses herein) load on a given factor, and the resulting factors are allowed to correlate with one another. In bifactor models, all indicators load on a general factor (i.e. general psychopathology, or p) and on a specific factor (e.g. Internalizing or Externalizing), allowing for the separate contributions of general versus domain-specific variance in disorders. In contrast, correlated factors models assume that variance in disorders is due only to partially overlapping dimensions but not to a general factor.

Fit indices evaluate how well a factor model captures the correlational patterns in the data by quantifying the discrepancy between the covariances implied by the model and those observed in the data. Following standard practices in the field, we used multiple fit indices (CFI, TLI, RMSEA, and SRMR) to evaluate how well our dimensional models captured the structure of psychiatric disorders (see eMethods in Supplement 1 for more details). However, accumulating evidence suggests that conventional fit indices may at times favor unnecessarily complex models or fail to detect meaningful distinctions among competing models (Bonifay & Cai, Reference Bonifay and Cai2017; Bonifay, Lane, & Reise, Reference Bonifay, Lane and Reise2017; Forbes et al., Reference Forbes, Greene, Levin-Aspenson, Watts, Hallquist, Lahey and Krueger2021; Greene et al., Reference Greene, Eaton, Li, Forbes, Krueger, Markon and Kotov2019; Morgan, Hodge, Wells, & Watkins, Reference Morgan, Hodge, Wells and Watkins2015; Murray & Johnson, Reference Murray and Johnson2013; Watts, Poore, & Waldman, Reference Watts, Poore and Waldman2019). To address these limitations, we supplemented conventional indices with alternative indices that evaluate three critical properties of the relationships between diagnoses and their assigned dimensions: magnitude (the strength of association between each disorder and its respective dimension), precision (the degree of statistical certainty with which these associations are estimated), and consistency (the extent to which the magnitudes of these associations are comparable across disorders within the same dimension) (Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023). We expected these alternative indices to provide incremental utility in discriminating between competing models beyond conventional fit indices.

We examined the external validity of factors in our preferred model using simple linear regressions, estimated within structural equation models (SEMs), to characterize similarities and differences among the factors in their associations with the sociodemographic and sleep variables.

We also conducted several sensitivity analyses to assess the replicability and robustness of our preferred models. These included replicating the fit of competing models in randomly split halves of the All of Us sample, assessing the sensitivity of factor loadings to each factor’s constituent diagnoses, and contrasting the fit indices and alternative indices of the preferred model with models in which diagnoses were randomly assigned to the factors.

Results

Our analytic sample comprised 127,963 individuals with at least one psychiatric diagnosis recorded in their EHRs, ensuring that all participants had been evaluated for psychopathology. The prevalence of all 39 primary diagnoses is illustrated in eFigure 1 in Supplement 1, and the sociodemographic information for the analytic sample is shown in eTable 3 in Supplement 1.

Correlations among psychiatric diagnoses

The tetrachoric correlations among the 39 psychiatric diagnoses are illustrated in the heatmap in Figure 1. While appreciable correlations among disorders were observed across psychopathology domains, correlations were highest among diagnoses within each domain. This pattern of correlations suggests that several correlated higher-order dimensions might best account for the comorbidity among the 39 psychiatric diagnoses.

Figure 1.

Heatmap of the correlations among 39 lifetime psychiatric diagnoses. Note. The heatmap is partitioned into six domains representing our most complex hypothesized model of psychopathology comprising: Fear, Distress, Thought Problems, Substance Use, Externalizing, and Neurodevelopmental Disorders.

Tests of a priori hypothesized models

We next employed CFAs to test alternative hypothesized models. The main models we contrasted are displayed in Table 1a, and the complete set in eTable 4 in Supplement 1 (alternative indices and factor correlations for each model can be found in Supplement 2).

Table 1.

Comparison of alternative models of psychopathology in All of Us

Note: ADHD-C, ADHD combined subtype; ADHD-HI, ADHD hyperactive–impulsive subtype; ADHD-IN, ADHD inattentive subtype; AGOR, agoraphobia; CD, conduct disorder; CFI, comparative fit index; FL, median factor loading; ODD, oppositional defiant disorder; PANIC, panic disorder; RMSEA, root mean square error of approximation; SCZ, schizophrenia; SCZAFF, schizoaffective; SD, standard deviation of loadings; SE, mean standard error; SRMR, standardized root mean squared residual; TLI, tucker-lewis index.

Model in bold (9.5) is our preferred model.

As shown in Table 1, all correlated factors models (Models 2–9) demonstrated better fit than the single General factor model (Model 1). The most comprehensive model, incorporating six distinct factors (Model 9), showed the best fit. Specifically, separating Substance Use from Externalizing emerged as a critical distinction (Model 9 vs. Model 8), as reflected by both superior fit indices and alternative indices. This separation also was justified by differential associations of the external correlates with distinct Substance Use and Externalizing dimensions (discussed below), rather than an overarching combined Externalizing dimension (see eFigure 2a in Supplement 1). Models including a broad Internalizing dimension versus distinct Fear and Distress dimensions (Model 2–5 vs. 6–9) showed equivalent overall fit but differed in their alternative indices (shown in eTable 5 in Supplement 1), as discussed below.

Comparing correlated factors models to their bifactor equivalents revealed mixed results. While traditional fit indices generally favored bifactor models with some exceptions (eTable 6 in Supplement 1), alternative indices strongly supported correlated factors models, which showed more stable factor loadings with less variability and lower standard errors (for more details, see eResults in Supplement 1). The 6 correlated factors model (shown in Figure 2) performed best overall, including on traditional fit indices. These findings demonstrate the importance of using both traditional and alternative indices when evaluating structural models of psychopathology.

Figure 2.

Path diagrams for the preferred correlated factors model (Model 9.5). Note: D, distress; E, externalizing; F, fear; I, internalizing; N, neurodevelopmental disorders; S, substance use; T, thought problems.

Comparisons among factors in their alternative indices

We next compared the alternative indices of the six factors from our preferred model (Model 9.5) to examine the extent to which the factors were well-represented by their indicators. In the preferred model, average factor loadings for diagnoses ranged from .6 to .7, with lower loadings on the Fear and Distress factors, ranging from .43 to .52. Factor loadings were estimated very precisely across all factors, with standard errors (SEs) ranging from .006 to .028. Diagnoses’ factor loadings were fairly homogeneous within factors, with standard deviations ranging from .014 for Distress to .17 for Thought Problems, except for Fear, which had a standard deviation of .30. Due to the limited number of diagnoses per factor (5–8), we were unable to statistically compare these alternative indices. Detailed statistics for each factor are presented in eTable 7 in Supplement 1.

The distinction between Substance Use and Externalizing was supported by their alternative indices, as median Factor Loadings were higher and the variability and Standard Errors were lower in Model 9 vs. Model 8. Although models including a broad Internalizing dimension versus distinct Fear and Distress dimensions (Model 2–5 vs. 6-9) showed virtually equivalent fit, they differed in their alternative indices. As shown in eTable 5 in Supplement 1, for Distress diagnoses, factor loadings were comparable across the models. For Distress diagnoses, median factor loadings and standard deviations of factor loadings were similar between models. For Fear diagnoses, Model 9 showed higher median loadings with slightly increased variability. While Fear and Distress were highly correlated (r = .898), their distinct patterns of correlations with other dimensions (eFigure 2b in Supplement 1) support treating them as separate factors (for more details, see eResults in Supplement 1). The six-factor model distinguishing Substance Use from Externalizing and Fear from Distress was therefore selected for analyzing the placement of ambiguously classified disorders. Factor loadings for the diagnoses on each of the 6 factors are shown in Table 2.

Table 2.

Standardized factor loadings for the preferred six-factor model of psychopathology (Model 9.5).

Note: D, distress; E, externalizing; F, fear; I, internalizing; N, neurodevelopmental disorders; S, substance use; T , thought problems.

The placement of ambiguously classified disorders

We conducted targeted CFAs to determine the optimal placement of disorders with ambiguous classification (Table 1b; detailed procedures in eResults in Supplement 1 and full results in eTable 4). For several disorders, placement decisions were informed by small differences in model fit and factor loadings. ADHD subtypes were best represented with ADHD-Hyperactive–Impulsive loading on Externalizing and both ADHD-Combined and ADHD-Inattentive loading on Neurodevelopmental Disorders, with correlated residuals between the subtypes to reflect their additional covariation not captured by the Externalizing - Neurodevelopmental Disorders factor correlation. Eating disorders showed equivalent support for loading on Distress alone or on both Distress and Thought Problems. OCD and OCPD demonstrated better fit loading on Fear or Distress rather than on Thought Problems, while Antisocial Personality Disorder better reflected Substance Use or both Substance Use and Externalizing rather than Externalizing alone. Model fit improved with Borderline Personality Disorder loading on both Distress and Thought Problems (or Externalizing) versus Thought Problems alone. Thus, these analyses were more useful in ruling out alternative models for the placement of disorders with ambiguous classification than for deciding on their optimal placement. In the final model, the addition of three theoretically justified correlated residuals (between schizophrenia and schizoaffective disorder, OCD and OCPD, and MDD and dysthymia) to reflect their additional covariation not captured by their factor loadings improved model fit while maintaining favorable alternative fit indices.

Replication and robustness across random sample halves and randomly assigned diagnoses

We validated our preferred six correlated factors model through split-half replication analyses and comparison with randomly generated alternative models. Split-half analyses demonstrated strong replication of the six-factor model results, with high intra-class correlations (.90–1.0) for most fit indices and minimal differences between halves (all < .01; eTable 8 in Supplement 1). Bifactor models showed much less consistency across split halves, with lower intra-class correlations for RMSEA (.79), SRMR (.85), and particularly for the mean (.56) and standard deviation (.52) of factor loadings (eTables 9–10 in Supplement 1). To test the specificity of diagnosis-to-factor assignments, we compared our preferred six-factor model to 1,000 random variations of the six-factor model in which diagnoses were randomly assigned to factors. Our model demonstrated superior fit across alternative indices compared to the random models (eFigure 3 in Supplement 1).

The sensitivity of factors to their constituent indicators

We examined the sensitivity of the factors to their constituent diagnoses by re-running the CFAs excluding one diagnosis at a time and examining changes in the remaining diagnoses’ factor loadings. As shown in Figure 3, factor loadings for diagnoses changed only slightly after removing each diagnosis in turn [M = −.0007, Median = .0002, SD = .013, minimum = −.006, maximum = .002]. Taken together, these leave-one-out CFAs indicate that the latent factors are robust: no single diagnosis exerted undue influence on the remaining loadings.

Figure 3.

Impact of successive removal of each diagnosis on factor loadings. Note: Entries above the diagonal indicate the magnitude of changes in factor loadings for each diagnosis shown on the Y-axis as a function of removing each other diagnosis shown on the X-axis in turn, whereas entries below the diagonal indicate the magnitude of changes in the factor loadings for all other diagnoses shown on the Y-axis as a function of removing the diagnosis shown on the X-axis.

External correlates of the higher-order psychopathology dimensions

All associations with external correlates (i.e. sex at birth, sleep disturbance, education level, and income) significantly differed across dimensions (see Figure 4; all p < 5*10−10). Males were particularly high in Substance Use and Externalizing, whereas females were higher in Fear and Distress. Sleep disturbance showed the strongest positive association with Distress, while Substance Use showed negative associations with both education and income. Consistent with our expectations, the six psychopathology dimensions showed distinct patterns of external correlates, validating their distinction in the model. Specifically, Substance Use was more strongly associated than Externalizing with male sex and lower education and income, and the two factors showed opposite associations with sleep disturbance (positive for Externalizing, negative for Substance Use). Similarly, Distress showed stronger associations than Fear with female sex and sleep disturbance, supporting their distinction. Detailed results are presented in eTable 11 in Supplement 1.

Figure 4.

Differences in external correlates among the six higher-order psychopathology dimensions (Model 9.5). Note: D, distress; E, externalizing; F, fear; I, internalizing; N, neurodevelopmental disorders; S, substance use; T , thought problems.

Discussion

In this study, we employed both confirmatory and exploratory analyses within a CFA framework to test alternative models for the structure of psychopathology in All of Us. Our findings support a six-dimensional structure comprising Fear, Distress, Externalizing, Substance Use, Thought Problems, Neurodevelopmental Disorders. Correlated factors models outperformed bifactor models across multiple criteria, including replicability and fit indices. The model’s robustness was demonstrated through sensitivity analyses and replication across random sample halves. External validity analyses revealed distinct patterns of association between the six factors and demographic variables, though we did not replicate previous findings regarding educational attainment in Neurodevelopmental Disorders (Kuriyan et al., Reference Kuriyan, Pelham, Molina, Waschbusch, Gnagy, Sibley and Kent2013; Schmengler et al., Reference Schmengler, Peeters, Stevens, Hartman, Oldehinkel and Vollebergh2023; Toft et al., Reference Toft, Liu, Menon, Schendel, Loss and Ehrenstein2021) or sleep disturbance in Substance Use Disorders (Ara, Jacobs, Bhat, & McCall, Reference Ara, Jacobs, Bhat and McCall2016; Conroy & Arnedt, Reference Conroy and Arnedt2014; Roehrs & Roth, Reference Roehrs and Roth2015). Targeted CFAs suggested some refinements in the placement of OCD, OCPD, Tics, Tourette’s Syndrome, and ASPD, while ADHD subtypes showed associations with both Externalizing and Neurodevelopmental Disorders. Nonetheless, these analyses were more useful in ruling out alternative models for the placement of disorders with ambiguous classification than for deciding on their optimal placement.

The present study advances our understanding of the structure of psychopathology by addressing key gaps in the literature. Our study included a more extensive and diverse array of psychopathology indicators (i.e. diagnoses) and tested a broader range of alternative models, utilizing a sample size that surpasses nearly all extant studies in this domain. Our approach supports a hierarchical dimensional representation of psychopathology, as represented in the HiTOP model (Forbes et al., Reference Forbes, Sunderland, Rapee, Batterham, Calear, Carragher and Krueger2020; Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017, Reference Kotov, Krueger, Watson, Cicero, Conway, DeYoung and Wright2021; Ringwald et al., Reference Ringwald, Forbes and Wright2023), addressing comorbidity and heterogeneity in categorical diagnoses using higher-order dimensions, while also representing unique variance specific to individual diagnoses. This method allows for a more nuanced characterization of psychopathology, representing both broad, overarching dimensions and specific diagnostic features, thus providing a more accurate reflection of the complex nature of mental health presentations in research and clinical practice (Conway, Forbes, et al., Reference Conway, Forbes, Forbush, Fried, Hallquist, Kotov and Eaton2019; Conway et al., Reference Conway, Krueger, Board, Cicero, DeYoung, Eaton and Wright2021; DeYoung et al., Reference DeYoung, Kotov, Krueger, Cicero, Conway, Eaton and Wright2021).

The inclusion of a broad set of diagnoses and a large sample size enabled us to test alternative placements for disorders whose classification has traditionally been ambiguous, as well as permitting more rigorous tests of bifactor models. The results of alternative indices and of split-half and random placement analyses argued against a general psychopathology factor (Fried, Greene, & Eaton, Reference Fried, Greene and Eaton2021; Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023). Our analyses also underscored the limitations of relying solely on fit indices for model evaluation (Forbes et al., Reference Forbes, Greene, Levin-Aspenson, Watts, Hallquist, Lahey and Krueger2021; Greene et al., Reference Greene, Eaton, Li, Forbes, Krueger, Markon and Kotov2019; Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023), as placing diagnoses on different factors often made very little difference to traditional fit indices, highlighting the importance of supplementing these with alternative indices. Moreover, factor correlations and external validity results supported distinguishing certain factors (e.g. Fear from Distress) even when fit indices suggested combining them (e.g. within a broader Internalizing factor). These findings demonstrate the value of a multifaceted approach to model adjudication for validly capturing the nuanced structures of psychopathology.

Despite its strengths, this study has several limitations that should be considered when interpreting the findings. First, while we analyzed a broad array of psychiatric diagnoses in a large, diverse sample, our reliance on diagnostic categories – rather than symptom-level data – limits the granularity of our modeling. Diagnoses are inherently heterogeneous, overlapping, and shaped by clinical conventions, which may obscure finer distinctions among syndromes. Although recent studies using symptom-level data have generally supported higher-order structures similar to HiTOP (Forbes et al., Reference Forbes, Sunderland, Rapee, Batterham, Calear, Carragher and Krueger2020; Levin-Aspenson, Reference Levin-Aspenson2023; Ringwald et al., Reference Ringwald, Forbes and Wright2023; Waldman & Poore, Reference Waldman and Poore2023), we expect that future large-scale research incorporating symptom indicators will better resolve the placement of diagnostically ambiguous conditions than we were able to herein.

Second, because our data were limited to DSM-based diagnoses rather than individual symptoms, we were only able to examine structure at a single level of the HiTOP hierarchy – that of broad diagnostic spectra. This prevented us from modeling the full hierarchical organization of psychopathology, including finer-grained symptom clusters or intermediate subfactors. This limitation follows directly from the nature of EHR diagnostic data and thus constrains our ability to evaluate how the structure of psychopathology may differ across levels of analysis. As future large-scale datasets begin to include rich symptom-level phenotypes, more comprehensive tests of the full HiTOP hierarchy will become possible.

Third, our modeling approach imposed certain constraints that warrant consideration. For example, in line with traditional confirmatory factor analysis, we initially prohibited cross-loadings of diagnoses on multiple factors. Although a limited number of theory-driven cross-loadings were introduced during model refinement, this constraint may have artificially increased the number of latent dimensions needed to account for comorbidity among diagnoses. More flexible methods – such as exploratory structural equation modeling (ESEM) – may offer a different representation of the underlying structure, particularly for diagnostically complex conditions that span multiple domains.

Fourth, while our preferred model demonstrated strong overall fit and robustness, several alternative models yielded similarly acceptable fit using conventional indices. Consequently, decisions regarding the placement of certain disorders – particularly those with ambiguous nosological status – were partially guided by theoretical plausibility amid small empirical differences. This underscores the fact that fit indices may be particularly useful for winnowing out less viable models than for selecting a ‘best’ model (Waldman et al., Reference Waldman, King, Poore, Luningham, Zinbarg, Krueger and Zald2023) and highlights the interpretive nature of structural model adjudication and reinforces the need for replication and triangulation across methods.

Fifth, despite its importance, we did not formally evaluate whether the factor structure varied across demographic subgroups using measurement and structural invariance analyses. Although we examined associations between latent dimensions and variables such as sex, income, and education, we did not assess how such associations may vary across demographic groups. Given the demographic diversity of the All of Us cohort, future research should test whether the structure of psychopathology generalizes across sex, socioeconomic strata, and racial or ethnic groups.

Sixth, our use of diagnoses extracted from EHRs also presents several unresolved challenges, including concerns about reliability, validity, and the optimal algorithms for diagnostic classification (Chen et al., Reference Chen, Castro, Minnier, Charney, Stahl and Smoller2018; Davis, Sudlow, & Hotopf, Reference Davis, Sudlow and Hotopf2016; Denny et al., Reference Denny, Bastarache, Ritchie, Carroll, Zink, Mosley and Roden2013; Hripcsak & Albers, Reference Hripcsak and Albers2012; Linder et al., Reference Linder, Bastarache, Hughey and Peterson2021; Smoller, Reference Smoller2018; Wei et al., Reference Wei, Bastarache, Carroll, Marlo, Osterman, Gamazon and Denny2017, Reference Wei, Teixeira, Mo, Cronin, Warner and Denny2016; Wu et al., Reference Wu, Gifford, Meng, Li, Campbell, Varley and Wei2019; Zheutlin et al., Reference Zheutlin, Dennis, Linnér, Moscati, Restrepo, Straub and Smoller2019). Diagnostic practices vary across providers, health systems, and time, and are influenced by differences in access to care, comorbidity documentation, and clinician judgment. These sources of variability may affect the completeness and consistency of diagnosis-based phenotyping. Despite these limitations, EHR-based diagnoses serve a valuable practical purpose by enabling scalable access to clinically meaningful psychiatric data in very large and diverse samples. They offer a foundation for constructing more representative cohorts and open the door to powerful downstream opportunities – such as integrating hierarchical dimensional models like HiTOP with the extensive genetic, biological, behavioral, and environmental data available in large-scale biobanks such as All of Us. As methodological innovations – including natural language processing, machine learning, and probabilistic phenotyping – continue to improve the validity of EHR data (An et al., Reference An, Pazokitoroudi, Alvarez, Huang, Bacanu, Schork and Sankararaman2023; Dahl et al., Reference Dahl, Thompson, An, Krebs, Appadurai, Border and Cai2023; Papini et al., Reference Papini, Iturralde, Lu, Greene, Barreda, Sterling and Liu2023; Weng et al., Reference Weng, Baur, Daswani, Chen, Harrell, Kakarmath and Ardila2024; Yun et al., Reference Yun, Cosentino, Behsaz, McCaw, Hill, Luben and Hormozdiari2024), their utility in psychiatric research is likely to be greatly enhanced.

Finally, the structure of psychopathology represented in our preferred model may be influenced in part by our reliance on lifetime diagnoses. While this approach captures a broad history of disorder, it fails to account for the time-specific manifestations and fluctuations of symptoms across the lifespan. Psychopathology is known to exhibit age-related shifts in presentation, severity, and comorbidity patterns, particularly during key developmental transitions such as adolescence and late adulthood (Rutter, Reference Rutter2013; Skodol & Bender, Reference Skodol and Bender2016). As such, lifetime diagnoses may obscure important temporal dynamics, including periods of remission, recurrence, or the emergence of new symptoms/syndromes. Future research should aim to incorporate longitudinal diagnostic data to model within-person changes over time, ideally using large, developmentally diverse samples. Leveraging repeated measures, time-varying covariates, and growth modeling techniques could help clarify how the structure and correlates of psychopathology evolve, and the extent to which transdiagnostic dimensions remain stable or shift across developmental stages. Such approaches are important not only for refining nosological models like HiTOP but also for informing prevention and intervention strategies tailored to individuals at different developmental stages.

Hierarchical dimensional models, such as HiTOP, offer a framework that parsimoniously represents patterns of comorbidity at higher levels of the taxonomy while preserving critical distinctions at lower levels. Unlike traditional diagnostic systems like the DSM, which collapse individual differences into broad, heterogenous, highly comorbid categories, this hierarchical dimensional approach explicitly maintains individual differences at multiple levels, enabling analyses of psychopathology at different levels of generality and specificity. The comprehensive model we propose herein leverages these advantages to provide a foundation for more nuanced investigations into the causes and outcomes of both broad, higher-order psychopathology dimensions and more specific, lower-order factors. This multi-tiered approach is particularly valuable for examining similarities and differences in psychopathology across diverse ethnicities (Cicero & Ruggero, Reference Cicero and Ruggero2021; Eaton et al., Reference Eaton, Keyes, Krueger, Noordhof, Skodol, Markon and Hasin2013; Forbes et al., Reference Forbes, Greene, Levin-Aspenson, Watts, Hallquist, Lahey and Krueger2021; He & Li, Reference He and Li2021; He, Rodriguez-Seijas, Waldman, & Li, Reference He, Rodriguez-Seijas, Waldman and Li2023; Moriarity, Joyner, Slavich, & Alloy, Reference Moriarity, Joyner, Slavich and Alloy2022; Ringwald, Forbes, & Wright, Reference Ringwald, Forbes and Wright2022; Rodriguez-Seijas et al., Reference Rodriguez-Seijas, Li, Balling, Brandes, Bernat, Boness and Eaton2023), as well as exploring how genetic and environmental risk factors may influence psychopathology at various levels of the hierarchy across different ancestral backgrounds.

Supplementary material

The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725102407.

Acknowledgements

We gratefully acknowledge All of Us participants for their contributions, without whom this research would not have been possible. We also thank the National Institutes of Health’s All of Us Research Program for making available the participant data used in this study.

Author contribution

All authors contributed to and have approved the final manuscript.

Funding statement

This research received no specific grant from any funding agency, commercial or not-for-profit sectors.

Competing interests

The authors declare none.

References

An, U., Pazokitoroudi, A., Alvarez, M., Huang, L., Bacanu, S., Schork, A. J., & Sankararaman, S. (2023). Deep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries. Nature Genetics, 55(12), 2269–2276. https://doi.org/10.1038/s41588-023-01558-w.CrossRefGoogle ScholarPubMed
Ara, A., Jacobs, W., Bhat, I. A., & McCall, W. V. (2016). Sleep disturbances and substance use disorders: A bi-directional relationship. Psychiatric Annals, 46(7), 408–412. https://doi.org/10.3928/00485713-20160512-01.CrossRefGoogle Scholar
Barlow, D. H., Farchione, T. J., Bullis, J. R., Gallagher, M. W., Murray-Latin, H., Sauer-Zavala, S., & Cassiello-Robbins, C. (2017). The unified protocol for Transdiagnostic treatment of emotional disorders compared with diagnosis-specific protocols for anxiety disorders: A randomized clinical trial. JAMA Psychiatry, 74(9), 875. https://doi.org/10.1001/jamapsychiatry.2017.2164.CrossRefGoogle ScholarPubMed
Barr, P. B., Bigdeli, T. B., & Meyers, J. L. (2022). Prevalence, comorbidity, and sociodemographic correlates of psychiatric disorders reported in the all of us research Program. JAMA Psychiatry, 79(6), 622–628. https://doi.org/10.1001/jamapsychiatry.2022.0685.CrossRefGoogle ScholarPubMed
Baumeister, H., & Härter, M. (2007). Prevalence of mental disorders based on general population surveys. Social Psychiatry and Psychiatric Epidemiology, 42(7), 537–546. https://doi.org/10.1007/s00127-007-0204-1.CrossRefGoogle ScholarPubMed
Bonifay, W., & Cai, L. (2017). On the complexity of item response theory models. Multivariate Behavioral Research, 52(4), 465–484. https://doi.org/10.1080/00273171.2017.1309262.CrossRefGoogle ScholarPubMed
Bonifay, W., Lane, S. P., & Reise, S. P. (2017). Three concerns with applying a bifactor model as a structure of psychopathology. Clinical Psychological Science, 5(1), 184–186. https://doi.org/10.1177/2167702616657069.CrossRefGoogle Scholar
Chen, C.-Y., Castro, P. H., Minnier, J., Charney, A. W., Stahl, E. A., & Smoller, J. W. (2018). Genetic validation of bipolar disorder identified by automated phenotyping using electronic health records. Translational Psychiatry, 8(1), 86. https://doi.org/10.1038/s41398-018-0133-7.CrossRefGoogle ScholarPubMed
Cicero, D., & Ruggero, C. (2021). Commentary– Opening a can of worms: The importance of testing the measurement invariance of hierarchical models of psychopathology – A commentary on He and Li (2020). Journal of Child Psychology and Psychiatry, 62(3), 299–302. https://doi.org/10.1111/jcpp.13353.CrossRefGoogle Scholar
Conroy, D. A., & Arnedt, J. T. (2014). Sleep and substance use disorders: An update. Current Psychiatry Reports, 16(10), 487. https://doi.org/10.1007/s11920-014-0487-3.CrossRefGoogle ScholarPubMed
Conway, C. C., Forbes, M. K., Forbush, K. T., Fried, E. I., Hallquist, M. N., Kotov, R., & Eaton, N. R. (2019). A hierarchical taxonomy of psychopathology can transform mental health research. Perspectives on Psychological Science, 14(3), 419–436. https://doi.org/10.1177/1745691618810696.CrossRefGoogle ScholarPubMed
Conway, C. C., Krueger, R. F., Board, H. C. E., Cicero, D. C., DeYoung, C. G., Eaton, N. R., & Wright, A. G. C. (2021). Rethinking the diagnosis of mental disorders: Data-driven psychological dimensions, not categories, as a framework for mental-Health Research, treatment, and training. Current Directions in Psychological Science, 30(2), 151–158. https://doi.org/10.1177/0963721421990353.CrossRefGoogle Scholar
Conway, C. C., Mansolf, M., & Reise, S. P. (2019). Ecological validity of a quantitative classification system for mental illness in treatment-seeking adults. Psychological Assessment, 31(6), 730–740. https://doi.org/10.1037/pas0000695.CrossRefGoogle ScholarPubMed
Dahl, A., Thompson, M., An, U., Krebs, M., Appadurai, V., Border, R., & Cai, N. (2023). Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder. Nature Genetics, 55(12), 2082–2093. https://doi.org/10.1038/s41588-023-01559-9.CrossRefGoogle ScholarPubMed
Dalgleish, T., Black, M., Johnston, D., & Bevan, A. (2020). Transdiagnostic approaches to mental health problems: Current status and future directions. Journal of Consulting and Clinical Psychology, 88(3), 179–195. https://doi.org/10.1037/ccp0000482.CrossRefGoogle ScholarPubMed
Davis, K. A. S., Sudlow, C. L. M., & Hotopf, M. (2016). Can mental health diagnoses in administrative data be used for research? A systematic review of the accuracy of routinely collected diagnoses. BMC Psychiatry, 16(1), 263. https://doi.org/10.1186/s12888-016-0963-x.CrossRefGoogle ScholarPubMed
Denny, J. C., Bastarache, L., Ritchie, M. D., Carroll, R. J., Zink, R., Mosley, J. D., & Roden, D. M. (2013). Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nature Biotechnology, 31(12), 1102–1111. https://doi.org/10.1038/nbt.2749.CrossRefGoogle ScholarPubMed
DeYoung, C. G., Kotov, R., Krueger, R. F., Cicero, D. C., Conway, C. C., Eaton, N. R., … Wright, A. G. C. (2021). Answering questions about the hierarchical taxonomy of psychopathology (HiTOP): Analogies to whales and sharks miss the boat. Clinical Psychological Science, 216770262110493. https://doi.org/10.1177/21677026211049390Google ScholarPubMed
Eaton, N. R., Keyes, K. M., Krueger, R. F., Noordhof, A., Skodol, A. E., Markon, K. E., & Hasin, D. S. (2013). Ethnicity and psychiatric comorbidity in a national sample: Evidence for latent comorbidity factor invariance and connections with disorder prevalence. Social Psychiatry and Psychiatric Epidemiology, 48(5), 701–710. https://doi.org/10.1007/s00127-012-0595-5.CrossRefGoogle Scholar
Epstein, H.-A. B. (2022). All of us: A national effort to improve health. Journal of Consumer Health On The Internet, 26(4), 452–459. https://doi.org/10.1080/15398285.2022.2138058.CrossRefGoogle Scholar
Forbes, M. K., Greene, A. L., Levin-Aspenson, H. F., Watts, A. L., Hallquist, M., Lahey, B. B., & Krueger, R. F. (2021). Three recommendations based on a comparison of the reliability and validity of the predominant models used in research on the empirical structure of psychopathology. Journal of Abnormal Psychology, 130(3), 297–317. https://doi.org/10.1037/abn0000533.CrossRefGoogle ScholarPubMed
Forbes, M. K., Sunderland, M., Rapee, R. M., Batterham, P. J., Calear, A. L., Carragher, N., & Krueger, R. F. (2020). A detailed hierarchical model of psychopathology: From individual symptoms up to the general factor of psychopathology. Clinical Psychological Science, 9(2), 139–168. https://doi.org/10.1177/2167702620954799.CrossRefGoogle Scholar
Fried, E. I., Greene, A. L., & Eaton, N. R. (2021). The p factor is the sum of its parts, for now. World Psychiatry, 20(1), 69–70. https://doi.org/10.1002/wps.20814.CrossRefGoogle Scholar
Greene, A. L., Eaton, N. R., Li, K., Forbes, M. K., Krueger, R. F., Markon, K. E., & Kotov, R. (2019). Are fit indices used to test psychopathology structure biased? A simulation study. Journal of Abnormal Psychology, 128(7), 740–764. https://doi.org/10.1037/abn0000434.CrossRefGoogle ScholarPubMed
He, Q., & Li, J. J. (2021). Factorial invariance in hierarchical factor models of mental disorders in African American and European American youths. Journal of Child Psychology and Psychiatry, 62(3), 289–298. https://doi.org/10.1111/jcpp.13243.CrossRefGoogle ScholarPubMed
He, Q., Rodriguez-Seijas, C. A., Waldman, I., & Li, J. J. (2023). Racial and ethnic invariance of HiTOP dimensions in the adolescent brain and cognitive development (ABCD) study. https://doi.org/10.31234/osf.io/aq7ru.CrossRefGoogle Scholar
Hripcsak, G., & Albers, D. J. (2012). Next-generation phenotyping of electronic health records. Journal of the American Medical Informatics Association, 20(1), 117–121. https://doi.org/10.1136/amiajnl-2012-001145.CrossRefGoogle ScholarPubMed
Hyman, S. E. (2010). The diagnosis of mental disorders: The problem of reification. Annual Review of Clinical Psychology, 6(1), 155–179. https://doi.org/10.1146/annurev.clinpsy.3.022806.091532.CrossRefGoogle ScholarPubMed
Insel, T. R. (2008). Assessing the economic costs of serious mental illness. American Journal of Psychiatry, 165(6), 663–665. https://doi.org/10.1176/appi.ajp.2008.08030366.CrossRefGoogle ScholarPubMed
Investigators, A. of U. R. P, Denny, J. C., Rutter, J. L., Goldstein, D. B., Philippakis, A., Smoller, J. W., & Dishman, E. (2019). The “All of Us” research program. New England Journal of Medicine, 381(7), 668–676. https://doi.org/10.1056/nejmsr1809937.Google Scholar
Jablensky, A. (2009). A meta-commentary on the proposal for a meta-structure for DSM-V and ICD-11. Psychological Medicine, 39(12), 2099–2103. https://doi.org/10.1017/s0033291709991292.CrossRefGoogle ScholarPubMed
Kessler, R. C., & Wang, P. S. (2008). The descriptive epidemiology of commonly occurring mental disorders in the United States*. Annual Review of Public Health, 29(1), 115–129. https://doi.org/10.1146/annurev.publhealth.29.020907.090847.CrossRefGoogle ScholarPubMed
Klann, J. G., Joss, M. A. H., Embree, K., & Murphy, S. N. (2019). Data model harmonization for the all of us research Program: Transforming i2b2 data into the OMOP common data model. PLoS One, 14(2), e0212463. https://doi.org/10.1371/journal.pone.0212463.CrossRefGoogle ScholarPubMed
Kotov, R., Krueger, R. F., Watson, D., Achenbach, T. M., Althoff, R. R., Bagby, R. M., & Zimmerman, M. (2017). The hierarchical taxonomy of psychopathology (HiTOP): A dimensional alternative to traditional Nosologies. Journal of Abnormal Psychology, 126(4), 454–477. https://doi.org/10.1037/abn0000258.CrossRefGoogle ScholarPubMed
Kotov, R., Krueger, R. F., Watson, D., Cicero, D. C., Conway, C. C., DeYoung, C. G., & Wright, A. G. C. (2021). The hierarchical taxonomy of psychopathology (HiTOP): A quantitative nosology based on consensus of evidence. Annual Review of Clinical Psychology, 17(1), 1–26. https://doi.org/10.1146/annurev-clinpsy-081219-093304.CrossRefGoogle ScholarPubMed
Krueger, R. F., Hobbs, K. A., Conway, C. C., Dick, D. M., Dretsch, M. N., Eaton, N. R., & Workgroup, H. U. (2021). Validity and utility of hierarchical taxonomy of psychopathology (HiTOP): II. Externalizing superspectrum. World Psychiatry, 20(2), 171–193. https://doi.org/10.1002/wps.20844.CrossRefGoogle ScholarPubMed
Krueger, R. F., Kotov, R., Watson, D., Forbes, M. K., Eaton, N. R., Ruggero, C. J., & Zimmermann, J. (2018). Progress in achieving quantitative classification of psychopathology. World Psychiatry, 17(3), 282–293. https://doi.org/10.1002/wps.20566.CrossRefGoogle ScholarPubMed
Krueger, R. F., Markon, K. E., Patrick, C. J., Benning, S. D., & Kramer, M. D. (2007). Linking antisocial behavior, substance use, and personality: An integrative quantitative model of the adult externalizing Spectrum. Journal of Abnormal Psychology, 116(4), 645–666. https://doi.org/10.1037/0021-843x.116.4.645.CrossRefGoogle ScholarPubMed
Kuriyan, A. B., Pelham, W. E., Molina, B. S. G., Waschbusch, D. A., Gnagy, E. M., Sibley, M. H., & Kent, K. M. (2013). Young adult educational and vocational outcomes of children diagnosed with ADHD. Journal of Abnormal Child Psychology, 41(1), 27–41. https://doi.org/10.1007/s10802-012-9658-z.CrossRefGoogle ScholarPubMed
Levin-Aspenson, H. F. (2023). To fully leverage fine-grained clinical phenomena, we have to think beyond DSM-based concepts and the presumption of diagnostic kinds. Journal of Psychopathology and Clinical Science, 132(7), 881–887. https://doi.org/10.1037/abn0000876.CrossRefGoogle ScholarPubMed
Linder, J. E., Bastarache, L., Hughey, J. J., & Peterson, J. F. (2021). The role of electronic health records in advancing genomic medicine. Annual Review of Genomics and Human Genetics, 22(1), 1–20. https://doi.org/10.1146/annurev-genom-121120-125204.CrossRefGoogle ScholarPubMed
MacCallum, R. C., Roznowski, M., & Necowitz, L. B. (1992). Model modifications in covariance structure analysis: The problem of capitalization on chance. Psychological Bulletin, 111(3), 490–504. https://doi.org/10.1037/0033-2909.111.3.490.CrossRefGoogle ScholarPubMed
Mansell, W., Harvey, A., Watkins, E., & Shafran, R. (2009). Conceptual foundations of the Transdiagnostic approach to CBT. Journal of Cognitive Psychotherapy, 23(1), 6–19. https://doi.org/10.1891/0889-8391.23.1.6.CrossRefGoogle Scholar
Mapes, B. M., et al. (2020). Diversity and inclusion for the all of us research program: A scoping review. PLoS One, 15(7), e0234962. https://doi.org/10.1371/journal.pone.0234962.CrossRefGoogle ScholarPubMed
Morgan, G., Hodge, K., Wells, K., & Watkins, M. (2015). Are fit indices biased in favor of bi-factor models in cognitive ability research?: A comparison of fit in correlated factors, higher-order, and bi-factor models via Monte Carlo simulations. Journal of Intelligence, 3(1), 2–20. https://doi.org/10.3390/jintelligence3010002.CrossRefGoogle Scholar
Moriarity, D. P., Joyner, K. J., Slavich, G. M., & Alloy, L. B. (2022). Unconsidered issues of measurement noninvariance in biological psychiatry: A focus on biological phenotypes of psychopathology. Molecular Psychiatry, 1–5. https://doi.org/10.1038/s41380-021-01414-5.Google ScholarPubMed
Murray, A. L., & Johnson, W. (2013). The limitations of model fit in comparing the bi-factor versus higher-order models of human cognitive ability structure. Intelligence, 41(5), 407–422. https://doi.org/10.1016/j.intell.2013.06.004.CrossRefGoogle Scholar
Papini, S., Iturralde, E., Lu, Y., Greene, J. D., Barreda, F., Sterling, S. A., & Liu, V. X. (2023). Development and validation of a machine learning model using electronic health records to predict trauma- and stressor-related psychiatric disorders after hospitalization with sepsis. Translational Psychiatry, 13(1), 400. https://doi.org/10.1038/s41398-023-02699-6.CrossRefGoogle ScholarPubMed
Parkes, L., Moore, T. M., Calkins, M. E., Cook, P. A., Cieslak, M., Roalf, D. R., & Bassett, D. S. (2021). Transdiagnostic dimensions of psychopathology explain individuals’ unique deviations from normative neurodevelopment in brain structure. Translational Psychiatry, 11(1), 232. https://doi.org/10.1038/s41398-021-01342-6.CrossRefGoogle ScholarPubMed
Plana-Ripoll, O., Pedersen, C. B., Holtz, Y., Benros, M. E., Dalsgaard, S., Jonge, P. d., & McGrath, J. J. (2019). Exploring comorbidity within mental disorders among a Danish National population. JAMA Psychiatry, 76(3), 259–270. https://doi.org/10.1001/jamapsychiatry.2018.3658.CrossRefGoogle ScholarPubMed
Poore, H. E., Hatoum, A., Mallard, T. T., Sanchez-Roige, S., Waldman, I. D., Palmer, A. A., & Dick, D. M. (2023). A multivariate approach to understanding the genetic overlap between externalizing phenotypes and substance use disorders. Addiction Biology, 28(9), e13319–e13319. https://doi.org/10.1111/adb.13319.CrossRefGoogle ScholarPubMed
Ringwald, W. R., Forbes, M. K., & Wright, A. G. C. (2022). Meta-analytic tests of measurement invariance of internalizing and externalizing psychopathology across common methodological characteristics. Journal of Psychopathology and Clinical Science, 131(8), 847–856. https://doi.org/10.1037/abn0000785.CrossRefGoogle ScholarPubMed
Ringwald, W. R., Forbes, M. K., & Wright, A. G. C. (2023). Meta-analysis of structural evidence for the hierarchical taxonomy of psychopathology (HiTOP) model. Psychological Medicine, 53(2), 533–546. https://doi.org/10.1017/s0033291721001902.Google ScholarPubMed
Rodriguez-Seijas, C., Li, J. J., Balling, C., Brandes, C., Bernat, E., Boness, C. L., & Eaton, N. R. (2023). Diversity and the hierarchical taxonomy of psychopathology (HiTOP). Nature Reviews Psychology, 2(8), 483–495. https://doi.org/10.1038/s44159-023-00200-0.CrossRefGoogle Scholar
Roehrs, T. A., & Roth, T. (2015). Sleep disturbance in substance use disorders. Psychiatric Clinics of North America, 38(4), 793–803. https://doi.org/10.1016/j.psc.2015.07.008.CrossRefGoogle ScholarPubMed
Rutter, M. (2013). Developmental psychopathology: A paradigm shift or just a relabeling? Development and Psychopathology, 25(4pt2), 1201–1213. https://doi.org/10.1017/s0954579413000564.CrossRefGoogle ScholarPubMed
Schmengler, H., Peeters, M., Stevens, G. W. J. M., Hartman, C. A., Oldehinkel, A. J., & Vollebergh, W. A. M. (2023). ADHD symptoms and educational level in adolescents: The role of the family, teachers, and peers. Research on Child and Adolescent Psychopathology, 51(7), 1051–1066. https://doi.org/10.1007/s10802-023-01047-y.CrossRefGoogle ScholarPubMed
Skodol, A. E., & Bender, D. S. (2016). Psychiatry, 487–524. https://doi.org/10.1002/9780470515167.ch29.Google Scholar
Smoller, J. W. (2018). The use of electronic health records for psychiatric phenotyping and genomics. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 177(7), 601–612. https://doi.org/10.1002/ajmg.b.32548.CrossRefGoogle ScholarPubMed
Smoller, J. W., Andreassen, O. A., Edenberg, H. J., Faraone, S. V., Glatt, S. J., & Kendler, K. S. (2019). Psychiatric genetics and the structure of psychopathology. Molecular Psychiatry, 24(3), 409–420. https://doi.org/10.1038/s41380-017-0010-4.CrossRefGoogle ScholarPubMed
Toft, G., Liu, C., Menon, J., Schendel, D., Loss, G., & Ehrenstein, V. (2021). Assessment of educational attainment and employment among individuals with autism Spectrum disorder in Denmark. JAMA Pediatrics, 175(6), 601–608. https://doi.org/10.1001/jamapediatrics.2021.0124.CrossRefGoogle ScholarPubMed
Verona, E., Javdani, S., & Sprague, J. (2011). Comparing factor structures of adolescent psychopathology. Psychological Assessment, 23(2), 545–551. https://doi.org/10.1037/a0022055.CrossRefGoogle ScholarPubMed
Voorhees, E. E. V., Dennis, P. A., Elbogen, E. B., Clancy, C. P., Hertzberg, M. A., Beckham, J. C., & Calhoun, P. S. (2014). Personality assessment inventory internalizing and externalizing structure in veterans with posttraumatic stress disorder: Associations with aggression. Aggressive Behavior, 40(6), 582–592. https://doi.org/10.1002/ab.21554.CrossRefGoogle ScholarPubMed
Waldman, I. D., King, C. D., Poore, H. E., Luningham, J. M., Zinbarg, R. M., Krueger, R. F., & Zald, D. (2023). Recommendations for adjudicating among alternative structural models of psychopathology. Clinical Psychological Science, 11(4), 616–640. https://doi.org/10.1177/21677026221144256.CrossRefGoogle Scholar
Waldman, I. D., & Poore, H. E. (2023). Evaluating alternative models of youth externalizing using quantitative genetic analyses. Journal of Psychopathology and Clinical Science, 132(7), 833–846. https://doi.org/10.1037/abn0000874.CrossRefGoogle ScholarPubMed
Watson, D., Clark, L. A., Simms, L. J., & Kotov, R. (2022). Classification and assessment of fear and anxiety in personality and psychopathology. Neuroscience & Biobehavioral Reviews, 142, 104878. https://doi.org/10.1016/j.neubiorev.2022.104878.CrossRefGoogle ScholarPubMed
Watts, A. L., Greene, A. L., Bonifay, W., & Fried, E. I. (2024). A critical evaluation of the p-factor literature. Nature Reviews Psychology, 3(2), 108–122. https://doi.org/10.1038/s44159-023-00260-2.CrossRefGoogle Scholar
Watts, A. L., Poore, H. E., & Waldman, I. D. (2019). Riskier tests of the validity of the bifactor model of psychopathology. Clinical Psychological Science, 7(6), 1285–1303. https://doi.org/10.1177/2167702619855035.CrossRefGoogle Scholar
Wei, W.-Q., Bastarache, L. A., Carroll, R. J., Marlo, J. E., Osterman, T. J., Gamazon, E. R., & Denny, J. C. (2017). Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record. PLoS One, 12(7), e0175508. https://doi.org/10.1371/journal.pone.0175508.CrossRefGoogle ScholarPubMed
Wei, W.-Q., Teixeira, P. L., Mo, H., Cronin, R. M., Warner, J. L., & Denny, J. C. (2016). Combining billing codes, clinical notes, and medications from electronic health records provides superior phenotyping performance. Journal of the American Medical Informatics Association, 23(e1), e20–e27. https://doi.org/10.1093/jamia/ocv130.CrossRefGoogle ScholarPubMed
Weng, W.-H., Baur, S., Daswani, M., Chen, C., Harrell, L., Kakarmath, S., & Ardila, D. (2024). Predicting cardiovascular disease risk using photoplethysmography and deep learning. PLOS Global Public Health, 4(6), e0003204. https://doi.org/10.1371/journal.pgph.0003204.CrossRefGoogle ScholarPubMed
Wright, A. G. C., Krueger, R. F., Hobbs, M. J., Markon, K. E., Eaton, N. R., & Slade, T. (2013). The structure of psychopathology: Toward an expanded quantitative empirical model. Journal of Abnormal Psychology, 122(1), 281–294. https://doi.org/10.1037/a0030133.CrossRefGoogle ScholarPubMed
Wu, P., Gifford, A., Meng, X., Li, X., Campbell, H., Varley, T., & Wei, W.-Q. (2019). Mapping ICD-10 and ICD-10-CM codes to Phecodes: Workflow development and initial evaluation. JMIR Medical Informatics, 7(4), e14325. https://doi.org/10.2196/14325.CrossRefGoogle ScholarPubMed
Yun, T., Cosentino, J., Behsaz, B., McCaw, Z. R., Hill, D., Luben, R., & Hormozdiari, F. (2024). Unsupervised representation learning on high-dimensional clinical data improves genomic discovery and prediction. Nature Genetics, 56(8), 1604–1613. https://doi.org/10.1038/s41588-024-01831-6.CrossRefGoogle Scholar
Zheutlin, A. B., Dennis, J., Linnér, R. K., Moscati, A., Restrepo, N., Straub, P., & Smoller, J. W. (2019). Penetrance and pleiotropy of polygenic risk scores for schizophrenia in 106,160 patients across four health care systems. American Journal of Psychiatry, 176(10), 846–855. https://doi.org/10.1176/appi.ajp.2019.18091085.CrossRefGoogle Scholar
Figure 0

Figure 1. Heatmap of the correlations among 39 lifetime psychiatric diagnoses. Note. The heatmap is partitioned into six domains representing our most complex hypothesized model of psychopathology comprising: Fear, Distress, Thought Problems, Substance Use, Externalizing, and Neurodevelopmental Disorders.

Figure 1

Table 1. Comparison of alternative models of psychopathology in All of Us

Figure 2

Figure 2. Path diagrams for the preferred correlated factors model (Model 9.5). Note: D, distress; E, externalizing; F, fear; I, internalizing; N, neurodevelopmental disorders; S, substance use; T, thought problems.

Figure 3

Table 2. Standardized factor loadings for the preferred six-factor model of psychopathology (Model 9.5).

Figure 4

Figure 3. Impact of successive removal of each diagnosis on factor loadings. Note: Entries above the diagonal indicate the magnitude of changes in factor loadings for each diagnosis shown on the Y-axis as a function of removing each other diagnosis shown on the X-axis in turn, whereas entries below the diagonal indicate the magnitude of changes in the factor loadings for all other diagnoses shown on the Y-axis as a function of removing the diagnosis shown on the X-axis.

Figure 5

Figure 4. Differences in external correlates among the six higher-order psychopathology dimensions (Model 9.5). Note: D, distress; E, externalizing; F, fear; I, internalizing; N, neurodevelopmental disorders; S, substance use; T , thought problems.

Supplementary material: File

Ehteshami and Waldman supplementary material 1

Ehteshami and Waldman supplementary material
Download Ehteshami and Waldman supplementary material 1(File)
File 1.7 MB
Supplementary material: File

Ehteshami and Waldman supplementary material 2

Ehteshami and Waldman supplementary material
Download Ehteshami and Waldman supplementary material 2(File)
File 76.2 KB