A, F1 indicates global functioning; N1; expressive negative symptoms; N2, experiential negative symptoms; P1, hallucinations; P2, delusions; P3, bizarre behavior; P4, thought disorder. B, F1 indicates global functioning; N3, flat affect; N4, avolition; N5 anhedonia; N6, alogia; P1, hallucinations; P2, delusions; P3, bizarre behavior; P4, thought disorder. C, F2 indicates work functioning; F3, independent living; F4, family relationships; F5, social network; N3, flat affect; N4, avolition; N5 anhedonia; N6, alogia; P1, hallucinations; P2, delusions; P3, bizarre behavior; P4, thought disorder.
F2 indicates work functioning; F3, independent living; F4, family relationships; F5, social network; N3, flat affect; N4, avolition; N5 anhedonia; N6, alogia; P1, hallucinations; P2, delusions; P3, bizarre behavior; P4, thought disorder.
eAppendix 1. Missing Data
eAppendix 2. Analysis of Network Robustness and Stability
eAppendix 3. Analysis of an additional community structure
eAppendix 4. Statistical Code Availability
eTable 1. Demographic Characteristic and Differences Between Participants with Complete and Missing Data
eFigure 1. Bootstrapped 95% Confidence Intervals of Edge Weights for All Network Analyses
eFigure 2. Results of Case-Dropping Subset Bootstrap Procedure to Assess Stability of Expected Influence Across All Network Analyses
eFigure 3. Bootstrapped exploratory graph analysis (EGA) for community detection of role functioning and negative symptoms
eTable 2. Item replicability corresponding to the EGA community structure
eTable 3. Network loadings corresponding to the EGA community structure
Members of the Consortium on the Genetics of Schizophrenia-2
Customize your JAMA Network experience by selecting one or more topics from the list below.
Abplanalp SJ, Braff DL, Light GA, Nuechterlein KH, Green MF, Consortium on the Genetics of Schizophrenia-2. Understanding Connections and Boundaries Between Positive Symptoms, Negative Symptoms, and Role Functioning Among Individuals With Schizophrenia: A Network Psychometric Approach. JAMA Psychiatry. 2022;79(10):1014–1022. doi:10.1001/jamapsychiatry.2022.2386
What do network psychometrics tell us about the connections and structural overlap between positive symptoms, negative symptoms, and role functioning in schizophrenia?
This cross-sectional study using network psychometric approach and including 979 outpatients with schizophrenia indicated that anhedonia was the central variable. Positive symptoms were the least central and showed unstable structural overlap, and negative symptoms and role functioning showed highly stable structural overlap, as negative symptoms and work functioning formed one network cluster.
The findings in this study indicate that anhedonia could be a rational treatment target for enhancing functional domains and avolition could be a reasonable target to enhance work functioning.
Improved understanding of the boundaries and connections between positive symptoms, negative symptoms, and role functioning in schizophrenia is critical, given limited empirical support for clear distinctions among these clinical areas. This study’s use of network psychometrics to investigate differential associations and structural overlap between positive symptoms, negative symptoms, and functional domains in schizophrenia may contribute to such understanding.
To apply network analysis and community detection methods to examine the interplay and structure of positive symptoms, negative symptoms, and functional domains in individuals with schizophrenia.
Design, Setting, and Participants
Cross-sectional study in 5 geographically distributed research centers in the US as part of the Consortium on the Genetics of Schizophrenia-2 from July 1, 2010, through January 31, 2014. Data were analyzed from November 2021 to June 2022. Clinically stable outpatients with schizophrenia or schizoaffective disorder were included. Participants were excluded if they had evidence of neurologic or additional Axis I psychiatric disorders. Other exclusion criteria included head injury, stroke, and substance abuse. Of 1415 patients approached, 979 were included in the final analysis.
Main Outcomes and Measures
Measures included the Scale for the Assessment of Positive Symptoms, the Scale for the Assessment of Negative Symptoms, and the Role Functioning Scale. Main outcomes were expected influence, which assesses the relative importance of items to the network and is defined as the association of an item with all others, and community detection and stability, defined as the presence of statistical clusters and their replicability.
Participants with complete data included 979 outpatients (mean [SD] age, 46  years; 663 male [67.7%]; 390 participants [40%] self-identified as African American, 30 [3%] as Asian, 7 [0.7%] as Native American, 8 [0.8%] as Pacific Islander, 412 [42.1%] as White, 125 [12.8%] as more than 1 race, and 5 [0.5%] did not identify). Anhedonia had the highest expected influence in the most comprehensive network analysis, showing connections with negative and positive symptoms and functional domains. Positive symptoms had the lowest expected influence. Community detection analyses indicated the presence of 3 clusters corresponding to positive symptoms; negative symptoms and work functioning; functional domains, including independent living, family relationships, and social network; and avolition, anhedonia, and work functioning. Hallucinations and delusions replicated in 1000 bootstrapped samples (100%), while bizarre behavior and thought disorder replicated in 390 (39%) and 570 (57%), respectively. In contrast, negative symptoms and work functioning replicated between 730 (73%) and 770 (77%) samples, respectively, and the remaining functional domains in 940 samples (94%).
Conclusions and Relevance
The high centrality of anhedonia and its connections with multiple functional domains suggest that it could be a treatment target for global functioning. Interventions for work functioning may benefit from a specialized approach that focuses primarily on avolition.
Schizophrenia is a leading cause of disability and is characterized by impairments in multiple domains of functioning.1,2 While interventions designed to improve functioning may lessen impairment over time, deficits persist and are enduring characteristics of the disorder.1 Associations with clinical symptoms have been explored extensively, yet uncertainty remains regarding the nature and strength of the associations. For example, while the association of positive symptoms with functioning is equivocal,3,4 connections between negative symptoms and functioning are consistently reported.4-6
Links between negative symptoms and functional outcomes were traditionally measured using global scores.6,7 However, this approach has limitations, given the current hierarchical conceptualization of negative symptoms. Factor analytic studies conceptualize alogia and flat affect as first-order factors that create the second-order expressive factor. Meanwhile, anhedonia, avolition, and asociality are the first-order factors that create the second-order experiential factor.8,9 This hierarchical structure has led to further insight into the association between negative symptoms and functional outcomes. For instance, flat affect, but not the second-order expressive factor, is associated with worse global functioning. Further distinction is observed between experiential domains.10 The second-order experiential factor, anhedonia and avolition, but not asociality, is associated with worse global functioning.10 Together, findings illustrate negative symptoms—particularly experiential domains—and functioning show differential associations, which may partially be due to conceptual overlap.
A network psychometric approach can elucidate the boundaries between negative symptoms and functional domains. One primary method in this approach is network analysis. Networks in psychological research often display variables (eg, symptoms of a disorder) as nodes, which are then connected in the network by edges representing partial correlations.11 Network analysis allows us to understand how variables of the same construct are connected and how different constructs mutually interact. A critical aspect of network analysis is centrality, which supplies information about the relative influence nodes have on the network. Given differential associations between negative symptoms and functional domains, network analysis is ideal for examining the interplay among their various levels and domains.
Another method of network psychometrics is community detection. Communities in networks refer to the clustering of nodes so that the presence of highly connected nodes is greater within a given cluster and lower between other clusters.12,13 Community detection can be helpful if there is a high statistical or conceptual overlap between items, in which they can be clustered together.14 As such, communities are summary statistics analogous to composites found in principal component analysis, with the benefit of including links between and within communities.15 Therefore, community detection can statistically illustrate where overlap occurs between negative symptoms and functioning.
The current study aimed to examine the interplay and structure of positive symptoms, negative symptoms, and functional domains using network analysis and community detection methods. While these are data-driven analyses, we had general hypotheses surrounding the outcomes of interest. We hypothesized that (1) using network analysis, experiential negative symptoms, including avolition and anhedonia, would have high network centrality, and positive symptoms would have low network centrality; (2) using network analysis, experiential and expressive negative symptoms would show differential associations with functional domains; and (3) using community detection methods, experiential negative symptoms and functional domains would be a part of the same community.
Participants were recruited for the Consortium on the Genetics of Schizophrenia-2 (COGS-2).16 A total of 1415 outpatients diagnosed with schizophrenia or schizoaffective disorder, depressed type, were recruited from 1 of 5 testing sites: University of California, San Diego; University of California, Los Angeles; University of Washington, Seattle; University of Pennsylvania, Philadelphia; and Mount Sinai School of Medicine, New York, New York. Among these, 1020 participants (72.1%) were prescribed atypical antipsychotics, 108 (7.6%) typical antipsychotics, and 135 (9.5%) both. Diagnoses were verified using the patient edition of the Structured Clinical Interview for DSM-IV.17 Participants were excluded if they had evidence of neurologic or additional Axis I psychiatric disorders. Other exclusion criteria included head injury, stroke, and substance abuse. Written consent was obtained from all participants, and the study was approved by the human research protection committees affiliated with each testing site. Race data were collected by free response to provide additional demographic information that may aid in future research on the topic and to provide further description of the study sample.
Negative symptoms were assessed using the Scale for the Assessment of Negative Symptoms (SANS).18 We used first-order (avolition, anhedonia or apathy, flat affect, and alogia) and second-order (experiential and expressive) factors in analyses. First-order factors used global scores corresponding to each domain, and second-order factors were calculated using the mean of global scores. SANS items were scored on a 0 to 5 scale, with higher scores indicating greater symptom severity. Attention items were not included. Positive symptoms were assessed using the Scale for the Assessment of Positive Symptoms (SAPS).19 The SAPS measures 4 domains on the same 0 to 5 scale: hallucinations, delusions, bizarre behavior, and thought disorder.
The Role Functioning Scale (RFS)20 assessed 4 interviewer-rated role functions: work functioning, independent living and self-care, family relationships, and immediate social network relationships. All RFS items were rated initially on a 1 to 7 scale, with higher score indicating better functioning. However, to aid in interpretation and consistency with symptom scales, all RFS ratings were reverse scored so that higher scores indicated worse functioning.
Statistical analyses were performed using R version 4.0.2 (R Foundation); relevant code is linked in eAppendix 4 in Supplement 1. Network analysis is a graphical representation of nodes. Lines between nodes are referred to as edges and represent partial correlations after accounting for all other correlations. Correlation magnitude is based on edge thickness, such that wider and thicker edges equate to stronger correlations. Blue edges represent positive correlations, and red edges represent negative correlations. Polychoric correlations were used to calculate correlations between nodes, and networks were estimated using the graphical Gaussian model (GGM).21
Due to the possibility of false-positive edges, we regularized the networks using the adaptive least absolute shrinkage and selection operator (LASSO).22 LASSO shrinks all edges in the network and sets small edges to zero. We estimated GGMs using the R package qgraph,23 applying extended bayesian information criterion (EBIC).24 EBIC estimates 100 different network models with varying degrees of sparsity and selects the model with the lowest EBIC. Node placement was decided by the Fruchterman-Reingold algorithm, which arranges nodes with stronger correlations near the center of the graph.25
To quantify centrality, expected influence was computed.26 Expected influence measures the association of an item with others in the network model and is the sum of all partial correlations. Items with high expected influence values are interconnected and may activate large network portions.26 Other standard centrality metrics, such as closeness and degree centrality, were not reported due to their tendency to be unstable.27,28 We used a bootstrapping approach from bootnet27 with 95% CIs to examine expected influence stability (eAppendix 2 and eFigures 1 and 2 in Supplement 1).27,29
A bootstrapped exploratory graph analysis (EGA)13,30 using 1000 bootstrapped samples was used to explore communities using the same variables as in the network analyses and was applied using EGAnet.31 EGA calculates polychoric correlations and uses the walktrap algorithm to find the number of communities of the partial correlation matrix. The walktrap algorithm measures similarities between vertices based on random walks. We also examined item stability, defined as the proportion of times specific items clustered with their community across the replicated bootstrapped samples.32,33
Network loadings for the communities was calculated using EGAnet.31,34 These loadings represent a node’s expected influence on each community. Nodes are allowed to load on all communities, with high loadings suggesting that a node greatly influences the emergence of a community. Effect sizes for network loadings are 0.15 for small, 0.25 for moderate, and 0.35 for large.34 Of note, while visualizing networks can be important for showing connections and communities, our analyses and results mainly focused on quantitative measures, including expected influence and network loadings, as relying on network visualization may lead to inaccurate inferences.35
The study sample in this cross-sectional study included 979 outpatients diagnosed with schizophrenia or schizoaffective disorder with complete data. There were no differences in self-reported age, education, or sex between participants with complete data and those with missing data (eAppendix 1 and eTable 1 in Supplement 1). The mean (SD) age of the sample was 46 (11) years, and 663 participants (67.7%) identified as male. A total of 390 participants identified (40%) as African American, 30 (3%) as Asian, 7 (0.7%) as Native American, 8 (0.8%) as Pacific Islander, 412 (42.1%) as White, 125 (12.8%) as more than one race, and 5 (0.5%) did not identify. The mean (SD) age at illness onset was 22.47 (7.05) years. All network analysis variables were normally distributed (Table 1).
We examined 3 networks moving from less to greater differentiation of variables. By taking this approach, we aimed to examine whether broad global scores potentially mask associations between domains of positive symptoms, negative symptoms, and role functioning. The first network included the second-order negative symptom factors, the 4 positive symptom domains, and a global role functioning score (Figure 1A). The experiential factor, but not the expressive factor, was associated with global role functioning, and hallucinations and bizarre behavior were associated with global role functioning. The experiential factor showed the highest expected influence value (1.20) but was not significantly different from delusions (0.85).
The second network included the first-order negative symptom factors, the 4 positive symptom domains, and a global role functioning score (Figure 1B). These variables were chosen to examine differential associations between second- and first-order negative symptoms and global role functioning. Positive symptoms were relatively isolated; anhedonia was the only negative symptom node associated with positive symptoms. Global functioning was associated with all negative symptom nodes, with the strongest edge being anhedonia. Additionally, global functioning was inversely associated with flat affect. Anhedonia and avolition had the highest expected influences values (1.20 and 0.89, respectively).
The final network included the first-order negative symptom factors, the 4 positive symptom domains, and the 4 role functioning domains (Figure 1C). These variables were chosen to identify whether specific domains of positive and negative symptoms show differential associations with functional domains. Positive symptoms were again isolated. Flat affect was negatively associated with independent living, and alogia was not associated with any role functional domain. Avolition and work functioning showed the strongest edge in the network; avolition was not related to any other functional domain. Regarding expected influence, anhedonia was slightly higher than avolition. Although similar in value, the expected influence values of anhedonia (1.15) and avolition (1.09) were distinct. The expected influence value of anhedonia stemmed from edges with nodes representing positive symptoms, negative symptoms, and role functioning, while the expected influence value of avolition primarily stemmed from its edge with work functioning.
A bootstrapped EGA to detect communities was conducted (Figure 2). Results showed a 3-community structure was present in 610 bootstrapped samples (61%). The 3-community structure was composed of a community of positive symptom domains, a community of first-order negative symptom factors and work functioning, and a community of the remaining role functioning domains. Overall, the positive symptom community showed low stability (Table 2). Although hallucinations and delusions were replicated in 1000 bootstrapped samples (100%), bizarre behavior and thought disorder were only replicated in 390 (39%) and 570 (57%), respectively. This low replicability was due to the weak connections between bizarre behavior and thought disorder with delusions and hallucinations and their cross-connections to negative symptoms and functional domains. Negative symptoms and work functioning replicated between 730 (73%) and 770 (77%) of samples, and the remaining role functional domains in 940 samples (94%). Due to the low stability of positive symptoms, we conducted another bootstrapped EGA excluding the positive symptom domains (eAppendix 3, eFigure 3, and eTables 2 and 3 in Supplement 1). A 3-community structure represented the data in 970 samples (97%).
Network loadings for the EGA community structure are in Table 3. Several items had moderate to high loadings on multiple communities. Most notably, anhedonia loaded highly on the negative symptoms and work functioning community (0.39) and the role functioning community (0.29). Additionally, work functioning loaded more strongly with negative symptoms (0.26) than other role functioning domains (0.17). Overlap between communities suggests that negative symptoms, particularly avolition and anhedonia, and role functioning domains may not have adequate separation to be treated as independent.
Network psychometrics were used to explore the interplay and community structure of positive symptoms, negative symptoms, and role functioning in a large sample of patients with schizophrenia. Negative symptoms showed higher expected influence than positive symptom domains. Additionally, experiential and expressive symptoms showed differential associations with role functioning. Negative symptoms formed a community with work functioning but no other functional domain. These findings suggest that negative symptoms, particularly experiential symptoms, and role functioning have blurred boundaries measured by clinical interviews.
Our initial network included positive symptom domains, second-order negative symptom factors, and a global role functioning score. One takeaway is that the expressive factor was isolated, while the experiential factor was connected to all other nodes in the network. However, results clearly show that global scores provide an incomplete picture of the association between negative symptoms and role functioning domains. For example, in the second network, alogia and flat affect were associated with global functioning; surprisingly, lower levels of flat affect were associated with poorer global functioning.
The third network examined more detailed associations between negative symptoms and role functioning by including specific functional domains. We observed that positive symptoms were isolated, as they showed limited associations with functional domains. Interestingly, anhedonia showed a strong negative association with independent living, such that greater severity of anhedonia was associated with better independent living. While surprising on its surface, item content may explain this finding. As anhedonia is rated alongside asociality on the SANS, multiple items deal with a lack of pleasure in social situations. On the other hand, independent living on the RFS primarily assesses self-sufficiency. As a result, more severe anhedonia may involve less tendency to receive assistance with daily tasks.
Additionally, anhedonia had the highest expected influence in the network, contrasting with some prior work suggesting that avolition is the most influential negative symptom.36-38 The use of different centrality metrics may partially explain conflicting findings. Prior studies that reported high centrality for avolition used other centrality metrics (eg, closeness and degree centrality) that may not be appropriate for psychological networks.28,39 Closeness assumes that all nodes are connected—an assumption rarely satisfied in psychological networks.28 Moreover, degree centrality calculates absolute edge weights and thus does not account for negative edges. Therefore, expected influence could be a more accurate network centrality metric.26,40
Our hypothesis that functioning and experiential negative symptoms would identify as part of the same community was partially supported. Avolition and anhedonia (along with alogia and flat affect) formed a community with work functioning. Blurred boundaries between negative symptoms and functional domains have been reported41,42; however, to our knowledge, this is the first study to find that avolition, anhedonia, and work functioning were part of the same cluster. The presence of this community may be partly due to measurement overlap. It is also possible that work functioning is qualitatively different from other functional domains. For instance, success at work requires integration of multiple factors and skills sets, including the ability to engage in work opportunities, the ability to perform duties at work, ability to get along with coworkers, and local limitations on opportunities for work. Given these complexities, experiential negative symptoms may have more opportunities to affect work functioning than other RFS domains.
As network loadings were designed to circumvent the limitation of latent confounding that is inherent in network centrality metrics,34 we can combine the information obtained from them and expected influence. The expected influence of anhedonia in the network analyses stemmed from its multidimensionality and influence on all communities. Even though avolition had a similar value, it stemmed from its influence on the emergence of the negative symptoms and work functioning community. So, even though anhedonia and avolition had similar expected influence values, network loadings showed that their values were qualitatively different.
Our results offer suggestions for interventions. The high expected influence of anhedonia and connections across communities make it a rational treatment target for interventions focused on multiple functional domains. Additionally, network loadings suggest that avolition may be an appropriate treatment target for work functioning or other employment-based outcomes. These findings contrasted with structural equation modeling studies of specific negative symptom domains, which suggested that anhedonia is directly associated with work functioning,43 and that other negative symptoms, including avolition and blunted affect, are more strongly associated with global functioning.10
However, we note that centrality and network loadings do not automatically translate to clinical relevance.44 For example, anhedonia or avolition could be end points themselves (ie, caused by other variables in the model),45 so interventions that target them would not induce changes in the network. Also, nodes with low centrality still could be clinically relevant. In our networks, positive symptoms were isolated yet could still play a role in network activation. For example, if positive symptoms and anhedonia are both associated with dopaminergic dysregulation,46,47 hallucinations could lead to delusions, leading to anhedonia causing downstream effects on functioning. Longitudinal network studies are needed to fully understand the causal pathways relevant to interventions.48 Even so, our results highlight the value of network psychometrics for generating hypotheses surrounding the associations among symptoms and functioning.
The main strength of this study was our network psychometric approach, which allowed us to examine the interplay of symptoms and functioning. More specifically, we examined differential associations between first- and second-order negative symptoms and functional domains without the need for a priori models. Most prior work mainly applied theory-driven structural equation models that limited opportunities to observe potentially meaningful associations between symptoms and functioning. By using a data-driven approach, we were able to examine a wide range of associations simultaneously.
This study has limitations. First, the cross-sectional design limits our conclusions, and our results should be extended using longitudinal designs that could better test for causality. Second, the SANS is an older measure of negative symptoms and may not fully represent current conceptualizations.49 For example, the SANS contains inappropriate affect and poverty of content of speech, which are not current considered to be negative symptoms. However, our use of SANS global scores may have minimized the influence of these 2 items, as global scores reflect more salient items and general clinical impressions. A further limitation of the SANS is its reliance on measuring external behavior over internal experience for the experiential scores. In contrast, second-generation negative symptom scales (ie, BNSS50 and CAINS51) measure both behavior and internal experience of motivation and pleasure. However, recent evidence suggests that the BNSS, CAINS, and SANS do not differ in their associations with functioning.10 Moreover, we did not include potential confounders that may impact negative symptoms, including depression. Negative symptoms and depression are often moderately correlated; however, evidence is mixed on their association in network analyses. For instance, one network analysis52 showed that depression was associated with avolition, while another study53 showed that depression was not associated with anhedonia but was associated with expressive negative symptoms.
We found 3 symptom-functional communities: positive symptoms, negative symptoms and work functioning, and the remaining role functional domains. While the positive symptom community had low stability, the other communities were stable. Anhedonia showed connections with multiple functional domains, demonstrated high centrality, and loaded highly on all communities, making anhedonia a viable target for interventions focused on multiple functional domains. Avolition is a rational target for interventions focused on work functioning, evidenced by their strong network connection and overlapping community structure. Future work should test these connections using longitudinal designs to estimate causality.
Accepted for Publication: June 28, 2022.
Published Online: August 17, 2022. doi:10.1001/jamapsychiatry.2022.2386
Corresponding Author: Samuel J. Abplanalp, PhD, Veterans Affairs Greater Los Angeles Healthcare System, MIRECC 210A, Bldg 210, 11301 Wilshire Blvd, Los Angeles, CA 90073 (firstname.lastname@example.org).
Author Contributions: Dr Abplanalp had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: All authors.
Acquisition, analysis, or interpretation of data: Abplanalp, Braff, Light, Green.
Drafting of the manuscript: Abplanalp, Braff, Green.
Critical revision of the manuscript for important intellectual content: Abplanalp, Braff, Light, Nuechterlein.
Statistical analysis: Abplanalp, Braff, Green.
Obtained funding: Braff, Green.
Administrative, technical, or material support: Braff, Light, Nuechterlein.
Supervision: Braff, Nuechterlein.
Conflict of Interest Disclosures: Dr Light reported having been a consultant to Astellas, Boehringer Ingelheim, Heptares, Merck, and NeuroSig. Dr Nuechterlein reported receiving unrelated research support from Ortho-McNeil Janssen Scientific Affairs; grants from the National Institute of Mental Health, Janssen, and Alkermes; personal fees from ReCognify; and consulting for Wyeth and Pfizer. Dr Green reported working as a consultant to AbbVie, ACADIA, DSP, FORUM, Lundbeck, and Takeda; serving on the scientific board of Luc; receiving research support from Amgen and Forum; and grants from the University of California, Los Angeles. No other disclosures were reported.
Funding/Support: This study was supported by the University of California San Diego: Brain & Behavior Research Foundation (27428, R25 MH101072, K23-MH102420, R01-MH065571, R33-MH125114, and R33-MH123603), Veterans Affairs San Diego Medical Center (IK2 RX003395), Mount Sinai School of Medicine (R01MH065554), University of California, Los Angeles (R01-MH65707), University of Pennsylvania (R01-MH65578), University of Washington (R01-MH65558), and State University of New York Downstate: Brain & Behavior Research Foundation. Funding for PsychChip genotyping was supported by a National Institute of Mental Health grant to Professor Pamela Sklar (U01-MH096296); additional genotyping support was provided by the Stanley Center for Psychiatric Research, under the auspices of the African Ancestry Genomic Psychiatry Cohort (R01 MH104964).
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Group Information: Members of the Consortium on the Genetics of Schizophrenia-2 are listed in Supplement 2.
Additional Contributions: We thank all of the participants and nonauthor support staff who made this study possible, including the following key personnel: University of California San Diego: Barbara Haugeland, PhD, Lauren Belleville, BA, Stacy Langton, BA, Daniel Mathias, BA, Natalie McCarthy, MA, Marlena Pela, MS, MA, Erich Riesen, BA, Maria Bongiovanni; Mount Sinai School of Medicine: Rui Ferreira, MA, Carolyn Khanian, PhD, Denise Poche-Jetter, MA, Rebecca West, MA; University of California, Los Angeles: William Horan, PhD, Amanda Bender, MS, Heidi Kuppinger, PhD, Mark McGee, BS, Ana Ceci Myers, MS, Felice Reddy, PhD, Amber Tidwell, MS, Christen Chapman, MSW; University of Pennsylvania: Amy Cassidy, MS, Erich Dress, BS, Colin Gallagher, MS, Mary March, MS, Alison Port, BA, Kosha Ruparel, MSE, Chandni Singh; University of Washington: Kate B. Alvey, Andrew C. Shutes-David, BA, Sean P. Meichle, BS, Denise O. Pritzl, ACSW, Sandra Perry, MSW, Annelise Sullivan, MS, Jane Whetstone, BA, Jake Wolf-Saxon, BA. None of these people were financially compensated for their work outside their regular salaries.