Two Factors, Five Factors, or Both? External Validation Studies of Negative Symptom Dimensions in Schizophrenia

Abstract

Objectives

Negative symptom studies frequently use single composite scores as indicators of symptom severity and as primary endpoints in clinical trials. Factor analytic and external validation studies do not support this practice but rather suggest a multidimensional construct. The current study used structural equation modeling (SEM) to compare competing dimensional models of negative symptoms to determine the number of latent dimensions that best capture variance in biological, psychological, and clinical variables known to have associations with negative symptoms.

Methods

Three independent studies (total n = 632) compared unidimensional, two-factor, five-factor, and hierarchical conceptualizations of negative symptoms in relation to cognition, psychopathology, and community functioning (Study 1); trait emotional experience and defeatist performance beliefs (Study 2); and glutamate and gamma-aminobutyric acid levels in the anterior cingulate cortex quantified using proton magnetic resonance spectroscopy (Study 3).

Results

SEM favored the five-factor and hierarchical models over the unidimensional and two-factor models regardless of the negative symptom measure or external validator. The five dimensions—anhedonia, asociality, avolition, blunted affect, and alogia—proved vital either as stand-alone domains or as first-order domains influenced by second-order dimensions—motivation and pleasure and emotional expression. The two broader dimensions sometimes masked important associations unique to the five narrower domains. Avolition, anhedonia, and blunted affect showed the most domain-specific associations with external variables across study samples.

Conclusions

Five domains and a hierarchical model reflect the optimal conceptualization of negative symptoms in relation to external variables. Clinical trials should consider using the two dimensions as primary endpoints and the five domains as secondary endpoints.

Introduction

Negative symptom studies frequently use a single aggregate score to indicate the severity of negative symptoms. In associational studies, such score is correlated with external variables; and in clinical trials, might be designated as the primary endpoint.^1,2 The use of a single severity indicator is however inconsistent with the results of studies that consistently support a multidimensional conception of negative symptoms.^3,4 Most studies have used factor analysis, a multivariate analytic approach that identifies the number of distinct dimensions in data by modeling the relationship between observed variables and latent dimensions.⁵ A two-factor structure with dimensions—Motivation and Pleasure (MAP) and Emotional Expression (EXP) have received the most support in factor analytic studies with comparisons to a unidimensional model also favoring the earlier.^3,4 As added support to the two-factor model, external validation studies show differential patterns of associations between the MAP/EXP dimensions and criterion (external) variables. For example, MAP is significantly associated with executive functioning, goal-directed action, and function; whereas EXP is associated with impairments in overall cognition.^3,6–9 This two-factor model has subsequently seen extensive adoption in negative symptom studies.

Leveraging goodness-of-fit statistics in conjunction with confirmatory factor analysis, some recent studies have favored a more complex structure and rather found the two-factor model a mediocre fit to data.^10–14 Strauss et al¹⁰ compared the two-factor model to more complex models using data obtained with the Schedule for the Assessment of Negative Symptoms (SANS), Brief Negative Symptom Scale (BNSS), and the Clinical Assessment Interview for Negative Symptoms (CAINS). They found that a five-factor model with more granular domains reified by the NIMH-MATRICS consensus conference—Anhedonia, Asociality, Avolition, Blunted Affect, and Alogia—as separable dimensions provided a much stronger fit to negative symptoms data. Subsequent studies have found the five-factor model and a hierarchical model with MAP and EXP as second-order factors that influence the five aforementioned domains (first-order factors) to be strong fits to the data regardless of the nationality/language of the sample or illness-stage.^11–14 Unlike the two-factor structure, however, the five-factor and hierarchical models have yet to be extensively studied in relation to meaningful external correlates. The only study to date found MAP and two of its corresponding subdomains—Avolition and Asociality—to have a significant inverse association with Global Assessment of Functioning ratings.⁹ Neither EXP nor its subdomains—Blunted Affect and Alogia—showed significant associations.

In the absence of extensive external validation, the more complex five-factor and hierarchical models have seen only limited adoption. Moreover, no studies have compared competing dimensional models with a range of external variables (eg, cognitive, functional, symptom, biological, or attitudinal measures) often studied in relation to negative symptoms. Such endeavor would inform about whether the comparative validity of competing dimensional models differs as a function of the type of external validator. Second, distinctions among negative symptom dimensions may be meaningful for some but not all types of external variables. For example, several studies support the MAP/EXP distinction in associations with function measures, but inconsistently for measures of cognition and symptoms.^{6–9,15–18} Despite the salience of neuroanatomic and neurochemical alterations distributed in several frontal and subcortical networks to the pathophysiology of negative symptoms, few studies have leveraged neurobiological variables as external validators.^19,20 The anterior cingulate cortex (ACC) works in tandem with other frontal regions to support reward valuation and effort-cost computation needed to obtain valuated rewards.^21,22 Reduced ACC connectivity with other frontal regions likely contributes to impaired goal-directed pursuits in schizophrenia. The levels of gamma-aminobutyric acid (GABA) and glutamate, which modulate functionally connected regions,²³ have been reported to be abnormal in schizophrenia.^24,25 Reduced GABA and glutamate in the ACC may therefore show domain-specific associations with MAP and/or subdomains subsumed within MAP including Anhedonia, Asociality, or Avolition. Although yet to be evaluated as an external validator of negative symptom dimensions, defeatist performance beliefs have been linked to motivational deficits.^26,27 Heightened negative affectivity also features prominently in negative symptoms—in particular anhedonia.^28–30 Studies have confirmed associations between high negative affectivity and both physical and social anhedonia.^31,32 In contrast, positive affect demonstrates an inverse association with the severity of negative symptoms in general with the absence of any symptom-specific associations.³³ Overall, defeatist beliefs and heightened negative affectivity may show particular associations with MAP and/or its corresponding subdomains. In contrast, positive affectivity is expected to show associations with all negative symptom dimensions.

The current article presents three studies that examine the association of negative symptom dimensions with external variables. Study 1 used structural equation modeling (SEM) to examine the comparative validity of four competing models—a unidimensional (one-factor) model, the two-factor MAP/EXP model, the five-factor model, and the hierarchical model—vis-à-vis their associations with cognition, function, and symptoms. Study 2 similarly used SEM to compare the four models in associations with defeatist performance beliefs, and trait positive and negative affect. Study 3 examined the association of negative symptom dimensions with glutamate and GABA levels in the ACC. All three studies examined if the five consensus domains produced associations with external variables not captured by the MAP/EXP dimensions.

Methods

Participants

Participants were people with schizophrenia or schizoaffective disorder enrolled in studies that administered a negative symptom rating scale and measures of external variables. At the time of the studies, participants were receiving services in outpatient settings. Local Institutional Review Boards reviewed and approved each study and participants provided written informed consent at each study location.

Study 1 used data from three negative symptom samples—Samples 1–3. Sample 1 included data from 308 individuals from outpatient clinics in the Maryland Psychiatric Research Center (MPRC), Maryland, USA. In this sample, negative symptoms were rated using a 16-item version of the SANS that excludes items not currently considered negative symptoms such as attention, inappropriate affect, etc. (Supplemental table S1). To assess cognitive abilities, participants in this sample completed the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS), the Wechsler Adult Intelligence Scale (WAIS-III), and Wide Range Achievement Test (WRAT-3). The Level of Function Scale was administered to assess psychosocial function. The Brief Psychiatric Rating Scale (BPRS) served to measure the severity of positive, negative, and general symptoms, and the Calgary Depression Scale measured the severity of depressive symptoms. Sample 2 included data from 146 individuals recruited from outpatient clinics in Spain (n = 115) and the United Kingdom (n = 31). These participants were administered the Spanish versions of the BNSS, MATRICS Consensus Cognitive Battery, and the Personal and Social Performance Scale as measures of negative symptoms, cognition, and functioning respectively. Participants also completed the Positive and Negative Syndrome Scale (PANSS) to assess the severity of positive, negative, and general symptoms. Sample 3 included data from 178 individuals recruited from assertive community treatment teams in the San Diego County Mental Health System. These participants completed the CAINS and the SANS to assess negative symptoms. They also completed the Brief Assessment of Cognition in Schizophrenia as a measure of cognition, the BPRS as a measure of symptoms, and the Independent Living Skills Survey and the Specific Level of Functioning as measures of functioning. The use of multiple distinct samples from different geographical regions and the use of divergent measures ensure that conclusions about the best fitting model are consistent across many different samples.

Study 2 used two datasets—Samples 3 and 4—to examine the association of negative symptom dimensions with defeatist beliefs and trait emotional experience. Sample 4 obtained data from 234 individuals recruited from MPRC, Binghamton University, and the University of Georgia. These individuals completed the BNSS to assess negative symptoms and the Positive and Negative Affect Scale (PANAS)—a self-report measure of trait positive and negative emotional experience. Associations between negative symptom dimensions and defeatist beliefs were also examined in Sample 3 (also used in Study 1). This study administered the Defeatist Performance Attitude Scale to assess respondents’ endorsement of dysfunctional attitudes about their ability to successfully perform goal-directed action.

Study 3 included 39 people from outpatient clinics in the MPRC and local mental health clinics. The BNSS served to rate negative symptoms in this sample. Participants also completed a 7T proton Magnetic Resonance Spectroscopy procedure to measure glutamate and GABA levels in the ACC. Methods are fully described in Rowland et al³⁴

Participants in each sample were over the age of 18. Participants’ DSM-IV or DSM-5 diagnosis was ascertained using the Structured Clinical Interview for DSM-IV (SCID) or SCID-5. Most participants in each sample were receiving antipsychotic medications and were clinically stable with no changes in medication type or dose in the weeks prior to clinical evaluation in their respective studies.

Procedures

Each sample was drawn from larger studies and measures administered in each study were part of larger protocols to illuminate the phenomenology and treatment of schizophrenia (table 1). Research staff with training in negative symptom assessment completed ratings in each study. All raters had completed at least a bachelor’s degree. Rater reliability was ascertained through reviews of instrument manuals, interview training, rating of training videos, discussions of gold standard ratings, and ongoing supervision.

Data Analysis

We estimated and compared structural equation models that included the negative symptom dimensions and domains of the external variables in studies 1 and 2. SEM is a confirmatory approach to multivariate data analysis that combines factor analytic models of the association of observed measures such as clinical ratings with latent factors, and structural regression paths that depict association among two or more latent variables. For each external variable, we estimated four structural models with a single negative symptom factor (unidimensional model); the two-factor MAP/EXP dimensions; the five NIMH-MATRICS consensus domains as separate dimensions; and the hierarchical model with MAP and EXP as second-order factors that influence the five consensus domains. Supplemental tables S1–S5 list negative symptom items used in the study, the average rating of each item, and their factor locations in the competing factor models.

In Study 1, we designated each external variable as a single-factor latent variable measured by domain scores from measures of the variable available in the dataset (see Supplemental Materials for additional details). Negative symptom items were excluded from the BPRS and PANSS in structural associations that designated “Symptoms” as the single-factor external variable. In Study 2, we designated external variables as observed variables in the structural models. Supplemental table S6 includes descriptive statistics of the study external variables.

We estimated all structural equation models in Mplus5.0 using the Weighted Least-Squares and Maximum Likelihood with robust standard errors which are both robust to non-normally distributed variables.³⁵ Goodness-of-fit statistics including the model Loglikelihood chi-square (χ²), the Comparative Fit Index (CFI), Tucker Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and the Weighted Root Mean Squared Residual (WRMR) served to evaluate the absolute fit of each model to the data. We sought consensus across fit indices to adjudge the absolute fit of estimated models including CFI and TLI ≥ 0.95, RMSEA ≤ 0.08, and WRMR ≤ 1.00.^36,37 We used the Akaike’s Information Criterion (AIC), Bayesian Information Criteria (BIC), and the sample-size adjusted BIC (SSA-BIC) to compare the relative fits of alternate structural equation models to the same data with models with lower values preferred.³⁸ The CFI, RMSEA, and AIC are presented in the main text and all goodness-of-fit estimates are provided in the Supplemental Materials. We first adjudged the absolute fit of each structural model to the data using CFI, TLI, RMSEA, and WRMR. Next, we adjudged the comparative validity of the competing models by comparing information criteria indices. Models with lower information criteria estimates are preferred and adjudged as having better comparative validity.

For each structural model, we examined path coefficients of the association of negative symptom dimensions with external variables in Studies 1 and 2. We examined patterns of associations with external variables to determine whether more granular domains show domain-specific associations missed by broader dimensions.

In Study 3, we examined correlations between MAP/EXP two-factor subscales and ACC glutamate and GABA signaling, and correlations between the five-factor subscales and ACC glutamate and GABA signaling.

Results

Sample Characteristics

Table 1 summarizes the demographic and clinical characteristics of participants in each study. The majority in each sample were men diagnosed with schizophrenia.

Study 1. Comparative Validity of Negative Symptom Dimensions and Clinical Variables

Model Fit.

The results of fitting the alternate structural equation models with measures of cognition, function, and symptoms as external variables were largely consistent across each of the study samples (table 2 and Supplemental tables S7–S9). Overall, models that included the single (one-factor) dimension or the MAP/EXP (two-factor) dimensions proved to be mediocre fits to the data. The majority of fit statistics—CFI/TLI, RMSEA, and WRMR—failed to consistently meet the required thresholds for acceptable fit in all three samples. There were some exceptions with the two-factor model. The two-factor model produced acceptable CFI, TLI, and WRMR values in its association with the Cognition latent variable in Sample 3 and Symptoms in Sample 2.

In contrast, structural models that included the five factors or the hierarchical dimensions were strong fits in all three samples. The CFI/TLI estimates for these structural models most often exceeded 0.95; their RMSEAs were less than 0.08 and WRMR values less than or equal to 1.00 in associations with Cognition, Function, and Symptom latent variables. The only exceptions were the low CFI and slightly high RMSEA values obtained with both the CAINS and the SANS in Sample 3 in association with Function.

In all three samples, the hierarchical model produced the lowest information criteria estimates in associations with cognition, function, and symptom measures. This suggests that the hierarchical model had the strongest comparative validity in associations with clinical measures.

Association of Negative Symptom Dimensions With Clinical Variables

Cognition.

Both the second-order MAP and EXP factors showed a significant inverse association with Cognition in Samples 1 and 3 (table 3). In Sample 2, only MAP was significantly associated with Cognition. With the exception of the SANS in Sample 3, the five first-order dimensions showed domain-specific associations. Avolition achieved a statistically significant inverse association with cognition in all three samples. Blunted Affect achieved significance in Samples 2 and 3. Anhedonia and Alogia were significant in Sample 1 and in Sample 3 with the SANS as the negative symptoms scale. In Sample 2, although EXP showed no significant association with Cognition, Blunted Affect achieved statistical significance.

Function.

MAP had a significant inverse association with Function in all three samples. EXP associations with Function did not achieve significance in any of the samples except in association with the SANS in Sample 3. Of the first-order factors, Anhedonia and Avolition produced significant inverse associations with Function in all three samples. Remarkably, Blunted Affect and Alogia were significantly associated with Function in Sample 2 despite EXP not achieving significance. Only Blunted Affect was in the expected inverse direction of association, however.

Symptoms.

MAP and its subdomains—Anhedonia, Asociality, and Avolition—showed significant direct associations with the Symptoms latent variable in all three samples. EXP and its subdomains—Blunted Affect and Alogia—achieved significance in Samples 1 and 2. In Sample 3, although EXP did not achieve a significant association, Alogia was significantly associated with Symptoms.

Study 2: Negative Symptom Dimensions and Psychological Variables

Model Fit.

The Positive and Negative Affect subscales of the PANAS were included as external variables in separate structural models in Sample 4. In Sample 3, the defeatist performance beliefs (DPB) total score served as external variable. In each sample, models with the one-factor and two-factor MAP/EXP dimensions were less than acceptable fits due to high RMSEA and WRMR values (table 4 and Supplemental table S10). The CFI/TLI values did meet acceptable thresholds in associations with positive and negative affect, however.

Models with either the five-factor or hierarchical dimensions proved to be excellent fits with each external variable. The information criteria favored the hierarchical model with Positive Affect or DPB score as the external variable; whereas they favored the five factors with Negative Affect as external variable.

Association of Negative Symptom Dimensions With Psychological Variables

MAP but not EXP showed a significant direct relationship with Negative Affect (table 5). Anhedonia and Avolition also produced significant direct associations with Negative Affect. Remarkably, although EXP showed no such association, Alogia had a significant inverse association with Negative Affect. This association was in the unexpected direction and its magnitude is rather low. Unlike the three multidimensional models, the unidimensional model failed to produce a significant path coefficient with Negative Affect.

Both MAP and EXP second-order dimensions showed a significant inverse relationship with Positive Affect and the DPB score (table 5). Similarly, all five first-order dimensions showed significant inverse associations with Positive Affect and DPB score; except Avolition which was non-significant with the DPB score.

Study 3: Negative Symptom Dimensions and ACC Glutamate and GABA Levels

In Sample 5, BNSS dimensional scores were created by multiplying clinical ratings by standardized factor loadings obtained from a larger study¹¹ and aggregating the product (see formulas in Supplemental Materials). The unidimensional score showed significant inverse correlation with GABA (r = −0.34, P = .036) but did not achieve significance with glutamate levels (r = −0.291, P = .073). The severity of MAP showed significant inverse associations with both GABA (r = −0.38, P = .018) and glutamate (r = −0.37, P = .02) whereas associations with EXP did not achieve significance. The Avolition dimension showed a significant inverse association with both GABA (r = −0.43, P = .007) and glutamate expression (r = −0.54, P < .001). None of the other first-order dimensions achieved significant associations with GABA or glutamate (Supplemental table S11).

Discussion

The current study examined the relative validity of four competing dimensional models of negative symptoms vis-à-vis their concurrent associations with clinical, psychological, and biological external validators. In Studies 1 and 2, structural models that included the five-factor or the hierarchical dimensions almost always proved to be excellent fits to data regardless of the negative symptom scale, or the external validator. In contrast, models that included either a single-factor or the two-factor MAP/EXP dimensions failed to consistently meet acceptable fit thresholds. Notably, indices of relative fit usually favored structural models with the hierarchy over the consensus five factors. The only exception was in concurrent associations with Negative Affect in Sample 4 in which a model with the five factors was preferred. Fit indices consistently favored the hierarchical and the five-factor models, a state of affairs unlikely to be due to model overfitting due to consistency across samples and measures. Avolition, anhedonia, and blunted affect showed the most preponderance of domain-specific associations with external variables across study samples.

The preference for the hierarchical model is not support for the two-factor MAP/EXP model that has gained traction in the literature over the last decade. In the hierarchical model, MAP/EXP directly influence the five domains but in the latter, MAP/EXP directly influence negative symptom ratings. When we compared significant associations for the broad MAP/EXP dimensions and the narrower five domains, evidence highlighted the importance of the five consensus domains either as stand-alone domains or as first-order domains influenced by second-order MAP/EXP dimensions (ie, the five consensus domains also showed associations not captured by MAP/EXP). For example, Blunted Affect showed significant association with Cognition and Alogia showed significant association with Function in Sample 2, even though EXP showed no corresponding associations. Similarly, Alogia was significantly associated with Negative Affect, whereas EXP was not. In these cases, the broader MAP/EXP dimensions masked associations driven by the five consensus domains. Thus, while the current study replicated some prior findings regarding external correlates of the MAP and EXP dimensions,^{6–9,25,26,39,40} it also extended earlier work in an important way by showing unique associations between external correlates and the individual domains that make up the broader MAP and EXP dimensions. The result of the domain correlations with ACC GABA and Glutamate showed that their significant association with MAP and total negative symptoms is driven almost exclusively by Avolition. This finding further speaks to the centrality of avolition in negative symptoms⁴¹ and the possibility that GABA and Glutamate ACC effects on negative symptoms may be driven exclusively by their impact on this domain.

The study results have several implications. First, the hierarchical and the five-factor models provide the best option for studying the clinical, psychological, and biological correlates of negative symptoms. In both models, five factors directly influence negative symptom ratings. Their strong goodness-of-fit suggest that the more granular five consensus domains should serve as the basis for negative symptoms studies. Effects found with the broader MAP/EXP may be specific to one or more of the five domains. Moreover, the presence of distinct associations at the level of the five domains improves the interpretability of concurrent associations. The broader MAP/EXP dimensions may also occasionally mask significant associations with one of the narrower five domains. The disadvantage of broader aggregate scores is even more apparent with the unidimensional model which masked domain-level associations with Negative Affect and glutamate levels.

Second, the presence of domain-specific associations between first-order dimensions and external variables demonstrate that the five domains are distinct and future scale development should represent symptoms within these domains. Rating scales that incorporate supplemental global severity ratings of MAP/EXP that are informed by ratings on each of the five domains would be consistent with a hierarchical model of assessment. Third, distinct psychological and pathophysiological processes may underlie the five first-order domains. A hierarchical model, however, suggests that if MAP/EXP dimensions subsume the five consensus domains, the latter may share overlapping or interlinked causal chains. Investigations informed by the hierarchical model should focus on the five domains and processes known to support multiple domains that share the same second-order factor, such as hedonic capacity and motivation for subcomponents of MAP.

Finally, findings have implications for how to operationalize primary treatment endpoints in clinical trials and associational studies. Due to multiple comparisons and recruitment constraints, clinical trials have tended to utilize single total scores as primary outcome measures. However, factor analytic and external validation results suggest that this practice is not ideal. Broader aggregate scores may mask significant effects when improvements are not ubiquitous but rather limited to a subset of symptoms. However, should analyses that are more granular focus on the two broad dimensions (MAP and EXP) or the narrower five domains—Anhedonia, Avolition, Asociality, Alogia, Blunted Affect? We suggest that informed by specific hypotheses about the intervention’s target(s), primary analyses should focus on one or both MAP/EXP dimensions whereas secondary analyses should focus on the narrower five domains. This approach should maximize the potential for finding significant effects of a drug or psychosocial intervention on their most relevant treatment targets. Indeed, a recent study showing the efficacy of the Phase 2B Roluperidone (MIN-101) trial supports this approach—a significant change on the Avolition domain drove this effect.⁴¹ Notably, adopting the proposed primary (two-dimensional) and secondary (five-domain) approach would necessitate careful scale selection. Given their limited scope, first-generation scales (eg, PANSS, negative symptom assessment) commonly used as outcome measures in clinical trials do not adequately capture the broad MAP/EXP dimensions or the five narrower domains. Second-generation scales (BNSS, CAINS) are more appropriate, informed by modern conceptualizations and able to capture the hierarchical and five-factor structures.^{10–14,42–45} The brevity of the five-factor subscales in the BNSS and the CAINS might limit their reliability estimates compared to the MAP/EXP subscales. Examining the more granular five subdomains only if significant effects are observed with the broader scales may attenuate some of the reliability issues with the shorter five-factor subscales. Studies that use the SANS should endeavor to exclude items not considered negative symptoms and limit measurement to items included in Supplemental table S1.

The strengths of the study include multiple independent samples that obtained ratings with different negative symptom scales. The use of different measures suggests that findings are unlikely measure-specific. Another is an extensive investigation of the concurrent validity of negative symptom dimensions on a range of external validators—cognition, function, symptoms, trait affect, defeatist beliefs, and glutamate and GABA. Some limitations are noteworthy. The first is the exclusive use of outpatient samples in each of the three studies, which may have limited the severity range of negative symptoms in these analyses. Second, the current study did not investigate if the associations demonstrated are equivalent across all forms of negative symptoms. This is important given that non-arbitrary subgroups, such as individuals with primary vs secondary negative symptoms, differ on some of the external variables including cognition, function, and the severity of symptoms.^46–49 Third, despite the contention that domain-specific associations support the distinction among the five domains, there was remarkable variability across samples in the statistical significance of path coefficients. Despite considerable convergence, the same domains did not always show domain-specific associations with the designated external variable across samples. Moreover, domain-specific associations were almost absent in Sample 3 with the SANS as the negative symptom measure. Of note, the observed variability was similarly present in associations with MAP/EXP dimensions. It could be that across-sample differences in measures of external variables explain some of the observed variability in path coefficients. What is compelling and supportive of the hierarchical and five-factor models is the prevalence of these domain-specific associations at the level of the five consensus domains.

Acknowledgments

The study was supported in part by National Institute of Health grants R01MH091057 (Granholm, PI); R01MH096263 (Barker/Rowland, PIs).

Disclosures

Dr Ahmed has received consulting fees from for Minerva Neurosciences.

Dr Kirkpatrick receives licensing royalties from ProPhase LLC for use of the Brief Negative Symptom Scale (BNSS) by for-profit groups; these fees are donated to the Brain and Behavior Research Foundation. He has also received honoraria and travel support from ProPhase LLC for training pharmaceutical company raters on the BNSS; consulting fees and/or travel support from Lundbeck, Acadia, ProPhase LLC, Otsuka, and Minerva Neurosciences; and fees from anonymized investors through Guideposts and Decision Resources Group. He is part owner of Quantic Innovations, which provides services related to digital phenotyping.

Dr Granholm receives consulting fees from Click Therapeutics Inc., receives fees from Granholm Consulting Inc. for training workshops on cognitive-behavioral social skills training, and is co-founder/part owner of BioSignal Solutions LLC, which develops biosignal-based digital therapeutics.

Dr Rowland received consulting fees from Otsuka America Pharmaceutical, Inc for educational purposes only for the platform PsychU.

Dr Gold receives royalty payments from Vera Sci and has consulted for Acadia.

Dr Buchanan serves as a data safety and monitoring board member for Newron, Roche; he serves on the advisory boards for Acadia, Avanir, Boehringer Ingelheim GMBH, GW Pharmaceutical, plc.; Minerva, Roche; and serves as a consultant for Boehringer Ingelheim GMBH.

Dr Bernardo has been a consultant for, received grant/research support and honoraria from, and has been on the speakers/advisory board of ABBiotics, Adamed, Angelini, Casen Recordati, Janssen-Cilag, Menarini, Rovi, and Takeda.

Dr Garcia-Portilla has been a consultant to and/or has received honoraria/grants from Angelini, Alianza Otsuka-Lundbeck, Instituto de Salud Carlos III, Janssen-Cilag, Lundbeck, Otsuka, Pfizer, and SAGE Therapeutics.

Dr Mané has received honoraria and travel support from Otsuka, Angellini, and Janssen; and grants from the Spanish Ministry of Health, Instituto de Salud Carlos III.

Dr Strauss is one of the original developers of the Brief Negative Symptom Scale (BNSS) and receives royalties and consultation fees from ProPhase LLC in connection with commercial use of the BNSS and other professional activities; these fees are donated to the Brain and Behavior Research Foundation. He has received honoraria and travel support from ProPhase LLC for training pharmaceutical company raters on the BNSS. In the past 2 years, he has consulted for and/or been on the speaker bureau for Minerva Neurosciences, Acadia, and Lundbeck pharmaceutical companies.

All other authors have no conflicts of interest to report.

References