Introduction
Psychosis is now widely believed to occur on a continuum, with an estimated 7.2% of the general population experiencing subthreshold psychotic experiences (PEs) at some point in their lifetimes (Linscott & van Os, Reference Linscott and van Os2013). Population-based studies of the co-occurrence of PEs and common mental health conditions have found widespread associations across a variety of diagnoses, including depressive, anxiety and substance use disorders. For example, the prevalence of PEs is reported to be greater in adults with major depressive disorder or any anxiety disorder in Europe (van Nierop et al. Reference van Nierop, van Os, Gunther, Myin-Germeys, de Graaf, ten Have, van Dorsselaer, Bak and van Winkel2012; Wigman et al. Reference Wigman, van Nierop, Vollebergh, Lieb, Beesdo-Baum, Wittchen and van Os2012) and in Australia, with or without post-traumatic stress disorder (PTSD) (Varghese et al. Reference Varghese, Scott, Welham, Bor, Najman, O'Callaghan, Williams and McGrath2011; Saha et al. Reference Saha, Scott, Varghese and McGrath2012), which is no longer classified as an anxiety disorder (APA, 2013). PTSD itself has been associated prospectively with prior PEs (Fisher et al. Reference Fisher, Caspi, Poulton, Meier, Houts, Harrington, Arseneault and Moffitt2013) and in cross-sectional data specifically with delusion-like experiences (Scott et al. Reference Scott, Chant, Andrews, Martin and McGrath2007). This finding is consistent with other studies showing elevated prevalence of trauma exposure history among individuals reporting PEs (Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; Lataster et al. Reference Lataster, van Os, Drukker, Henquet, Feron, Gunther and Myin-Germeys2006; Kelleher et al. Reference Kelleher, Harley, Lynch, Arseneault, Fitzpatrick and Cannon2008). PEs have similarly been associated with substance use disorders, including both alcohol and drug dependence, in England (Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004). Other studies have shown associations with drug use across a broad range of substances (opiate, cannabis, alcohol, tobacco; Degenhardt & Hall, Reference Degenhardt and Hall2001) and particularly cannabis use (Rössler et al. Reference Rössler, Riecher-Rössler, Angst, Murray, Gamma, Eich, van Os and Gross2007).
PEs in the context of general psychopathology
One possible explanation for widespread associations between PEs and nearly all classes of psychopathology is that risk for psychosis may increase with greater illness severity, regardless of the particular type of illness. Overall severity of psychopathology may be indicated by quantifying the number of co-occurring disorders, or by measuring symptom severity within disorders or phenotypic dimensions. Several studies have found dose–response relationships between severity of co-occurring disorders and risk for PEs. Among children and adolescents in Ireland, risk for PEs was shown to increase linearly as the child met criteria for an increasing number of diagnosable conditions across four population-based data sets using different categories of disorder and means of assessment (Kelleher et al. Reference Kelleher, Keeley, Corcoran, Lynch, Fitzpatrick, Devlin, Molloy, Roddy, Clarke, Harley, Arseneault, Wasserman, Carli, Sarchiapone, Hoven, Wasserman and Cannon2012) and among a help-seeking clinical sample (Kelleher et al. Reference Kelleher, Devlin, Wigman, Kehoe, Murtagh, Fitzpatrick and Cannon2013). Similarly, within a single disorder, risk for delusion-like experiences increases with increasing severity of major depression (Saha et al. Reference Saha, Scott, Varghese and McGrath2012). These findings, coupled with the widespread associations of PEs with specific mental health conditions, suggest that the extent of psychopathology (i.e. the number of co-occurring diagnoses) may be equally important to, or more important than, the nature of psychopathology (i.e. the presence of particular diagnoses) in determining risk for PEs.
Racial/ethnic differences in co-occurrence of common disorders with psychosis
All of the above-mentioned studies of co-occurrence of PEs and general mental health conditions were conducted with samples entirely or almost entirely made up of white respondents. Threshold schizophrenia is known to occur in every country worldwide (Tandon et al. Reference Tandon, Keshavan and Nasrallah2008), and the World Health Organization (WHO) World Health Survey (Nuevo et al. Reference Nuevo, Chatterji, Verdes, Naidoo, Arango and Ayuso-Mateos2012) identified subthreshold PEs in all 52 countries studied. As such, the psychosis continuum is likely to be a universal phenomenon, conceivably with a similar clinical profile across disparate cultural and ethnic groups. However, studies have found PEs to be more prevalent among racial/ethnic minorities both in the UK (Johns et al. Reference Johns, Nazroo, Bebbington and Kuipers2002, Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; King et al. Reference King, Nazroo, Weich, McKenzie, Bhui, Karlsen, Stansfeld, Tyrer, Blanchard, Lloyd, McManus, Sproston and Erens2005; Morgan et al. Reference Morgan, Fisher, Hutchinson, Kirkbride, Craig, Morgan, Dazzan, Boydell, Doody, Jones, Murray, Leff and Fearon2009) and in the USA (Cohen & Marino, Reference Cohen and Marino2013), whereas risk for common mental health conditions does not seem to be greater among racial/ethnic minority groups (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994; Breslau et al. Reference Breslau, Aguilar-Gaxiola, Kendler, Su, Williams and Kessler2006). Given this discrepancy, co-occurrence of psychosis and common mental disorders may be less common among minority groups. Generalizability of findings of co-occurrence within and across racial/ethnic groups remains an important and, as yet, unanswered issue of considerable interest.
Aims and hypotheses
The current study used four population-level data sets, including one nationally representative survey and three surveys of distinct racial/ethnic groups, to examine the co-occurrence of mental health conditions and PEs in the general adult population of the USA. The primary hypothesis was that PEs would be highly prevalent across disorders and across surveys, and that the prevalence of these experiences would increase with higher levels of morbidity in a dose–response fashion. A strong dose–response relationship with multi-morbidity (i.e. the number of subtypes of a disorder) in the context of weak relationships with individual disorders would be consistent with the proposition that psychosis may derive from more extensive and severe psychopathology rather than from specific types of disorder (i.e. quantity rather than quality). Although the absolute rate of co-occurrence between PEs and common mental disorders was expected to be lower among racial/ethnic minority groups relative to the nationally representative sample, there was no clear a priori evidence or reason to expect that the nature of the relationship between psychosis and common mental disorders would vary across racial/ethnic groups. Despite expected racial/ethnic variation in prevalence rates of PEs, common mental disorders and their co-occurrence, we thus hypothesized a consistent overall pattern of findings across all surveys, in terms of whether PEs were associated primarily with particular diagnostic categories or the overall extent of multi-morbidity.
Method
Participants
The analyses presented here were conducted separately, with four samples drawn from three population-level surveys in the USA. The surveys used similar methodology and sampling design, and collectively comprise the Collaborative Psychiatric Epidemiology Surveys (CPES; Alegría et al. Reference Alegría, Jackson, Kessler and Takeuchi2007a ). The four samples are drawn from: (1) the National Comorbidity Survey Replication (NCS-R; Kessler et al. Reference Kessler, Berglund, Chiu, Demler, Heeringa, Hiripi, Jin, Pennell, Walters, Zaslavsky and Zheng2004; Alegría et al. Reference Alegría, Jackson, Kessler and Takeuchi2007b ), (2) the National Latino and Asian American Study (NLAAS; Alegría et al. Reference Alegría, Takeuchi, Canino, Duan, Shrout, Meng, Vega, Zane, Vila, Woo, Vera, Guarnaccia, Aguilar-Gaxiola, Sue, Excobar, Lin and Gong2004; Alegría & Takeuchi, Reference Alegría, Takeuchi, Alegría, Jackson, Kessler and Takeuchi2007) and (3) the National Survey of American Life (NSAL; Jackson et al. Reference Jackson, Torres, Caldwell, Neighbors, Nesse, Taylor, Trierweiler and Williams2004, Reference Jackson, Caldwell, Chatters, Neighbors, Nesse, Taylor, Trierweiler, Williams, Alegría, Jackson, Kessler and Takeuchi2007). All surveys used multi-stage sampling designs, drawing participants from households in the 48 contiguous states. The NCS-R is a nationally representative survey of 9282 individuals (predominantly Caucasian, reflecting the general population of the USA), of which a random subsample (n = 2322) completed the psychosis screen. The NLAAS is a survey of Latino (n = 2554) and Asian Americans (n = 2095), which was analyzed as two separate samples (divided by ethnicity) for this study. Finally, the NSAL is a nationally representative sample of African-American households (n = 3570), with Afro-Caribbean (n = 1621) and Caucasian (n = 891) respondents drawn from the same source neighborhoods, although Caucasians were not assessed with the psychosis screen. Response rates for the NCS-R, NLAAS Latino sample, NLAAS Asian sample, and NSAL were 70.9, 75.5, 65.6 and 72.3% respectively, which are consistent with other large-scale psychiatric epidemiology studies. Because the focus of this study was subthreshold psychosis, participants were excluded if they had a lifetime psychotic disorder, conservatively defined as a self-reported lifetime diagnosis of schizophrenia or meeting WHO Composite International Diagnostic Interview (WHO-CIDI) criteria for lifetime bipolar disorder (regardless of the presence of psychosis). Participants with missing data for any of the variables of interest, including sociodemographic confounders, were also excluded. Respondents excluded due to missing demographic data (n = 28) did not vary significantly from those included on age or prevalence of PEs in any data set, but were more likely to be female in the Latino sample [χ 2 (df = 1) = 4.33, p = 0.037]. Following exclusion, the sample sizes and percentages included for each survey were: NCS-R: n = 2216 (95.4%); NLAAS Latino: n = 2539 (99.4%); NLAAS Asian: n = 2077 (99.1%); NSAL: n = 4884 (94.1%).
Measures
Demographic variables (age, sex, race/ethnicity, foreign birth, education, employment status, and marital status) were self-reported by respondents. Lifetime PEs were assessed using the WHO-CIDI 3.0 psychosis screen, which included items assessing: (1) visual hallucinations, (2) auditory hallucinations, (3) thought insertion, (4) thought control, (5) telepathy and (6) delusions of persecution. PEs were excluded if the experience took place solely in the context of falling asleep, dreaming, or substance use.
Respondents were assessed for lifetime DSM-IV Axis I disorders using the WHO-CIDI, a widely used and reliable structured interview assessment for diagnosis (Wittchen, Reference Wittchen1994; Haro et al. Reference Haro, Arbabzadeh-Bouchez, Brugha, De Girolamo, Guyer, Jin, Lepine, Mazzi, Reneses, Vilagut, Sampson and Kessler2006). Participants were coded positive for affective disorders if they met criteria for major depressive disorder or dysthymia; for anxiety disorders if they met criteria for generalized anxiety disorder, agoraphobia with and without panic disorder, panic disorder or social phobia; for substance use disorders if they met criteria for abuse or dependence of alcohol or other substances; and individually for PTSD.
Analyses
All analyses were conducted using the complex sample features of SPSS version 21 (SPSS Inc., USA), and confirmed through independent analysis using Stata SE version 13 (Stata Corporation, USA). All statistical estimates were weighted using sampling weights for the corresponding survey to account for individual-level sampling factors including non-response and unequal probabilities of selection. Design-based analyses were used to estimate standard errors that accounted for the complex multistage clustered design of each survey, using the Taylor series linearization method. The analytic plan was conducted independently but following identical procedures for each of the four surveys (NCS-R, NLAAS Latino and Asian samples, NSAL). Post-hoc analyses tested for differences between the Afro-Caribbean and African-American subsamples of the NSAL in terms of associations of mental health conditions and PEs; all main results were in the same direction and of similar effect size magnitude. NSAL results are therefore presented as a single set of analyses.
Descriptive data for demographic variables and prevalence of PEs, by diagnosis, are reported across all four samples. Associations between PEs (dependent variable) and each diagnostic group (depressive, anxiety, substance use and PTSD) were tested using logistic regression, first without adjustment as separate bivariate analyses, and then together in a single model, with adjustment for potential demographic confounds (age, sex, foreign birth, marital status, employment status and education). Diagnostic categories were then summed to create a measure of multi-morbidity to test for a dose–response relationship between the number of disorder subtypes and risk for PEs. This was also tested using logistic regression (excluding variables indicating specific diagnoses), with adjustment for potential demographic confounds.
Statistical significance was assessed for all tests using Wald χ 2, with two-tailed α = 0.05. Effect sizes were reported as unadjusted (bivariate analyses) or adjusted (multiple logistic regression analyses) odds ratios (ORs) with 95% confidence intervals (CIs).
Results
Descriptive data for all demographic and clinical variables in each data set, which differed primarily in terms of racial/ethnic make-up, are presented in Table 1. PEs and DSM-IV diagnoses are both notably uncommon among the Asian-American sample compared to the others, possibly limiting statistical power in terms of significance testing in further analyses with this group. Prevalence of PEs was numerically greater in the context of every individual disorder and diagnostic subtype, across all data sets, with the exception of agoraphobia (without panic disorder) among the Asian sample (see online Supplementary Table S1). All remaining analyses used diagnostic subtypes (depressive, anxiety, PTSD, substance) rather than individual disorders. In unadjusted bivariate analyses, lifetime prevalence of PEs was significantly greater among respondents with each subtype of mental health condition (depressive, anxiety and substance use disorders, and PTSD diagnosis) compared to the remainder of the sample, across all four data sets (Table 2).
Table 1. Descriptive data for all variables, divided by data set
NCS-R, National Comorbidity Survey Replication; NLAAS, National Latino and Asian American Study; NSAL, National Survey of American Life; PTSD, post-traumatic stress disorder.
Values for categorical variables given as weighted percentage (standard error).
a Diagnoses are not mutually exclusive.
Table 2. Unadjusted odds for psychotic experiences within each particular diagnostic category
NCS-R, National Comorbidity Survey Replication; NLAAS, National Latino and Asian American Study; NSAL, National Survey of American Life; PTSD, post-traumatic stress disorder; OR, odds ratio; CI, confidence interval; df, degrees of freedom.
Each statistical test separately compares respondents with the particular diagnosis to the remainder of the sample, unadjusted for the presence of other clinical diagnoses.
Relationships were then tested between each diagnostic category and PEs using a fully adjusted model, which controlled for demographic variables and also each remaining subtype of mental disorder. For instance, associations between PEs and depressive disorders controlled for the remaining disorders of anxiety, PTSD and substance use disorders, and so on for each disorder category. In the fully adjusted model, odds for PEs were still greater within each data set and each mental health condition, except for depressive disorders in the NLAAS Asian data set (Table 3). However, the effect sizes were reduced relative to those from the unadjusted analyses and many were no longer statistically significant.
Table 3. Final adjusted model of the association between each diagnostic category and psychotic experiences, within each data set
NCS-R, National Comorbidity Survey Replication; NLAAS, National Latino and Asian American Study; NSAL, National Survey of American Life; PTSD, post-traumatic stress disorder; OR, odds ratio; CI, confidence interval; Ref., reference.
Values presented as OR (95% CI), with each OR adjusted for the remaining clinical conditions and for sociodemographic factors. Significant associations are presented in bold.
Specifically, adjusted associations were only significant with depressive disorders in the NCS-R, NLAAS Latino and NSAL samples, with anxiety disorders in the NCS-R and NSAL samples, with PTSD in the NSAL sample, and with substance use disorders in the NCS-R, NLAAS Latino and NSAL samples. Sociodemographic associations with PEs also varied across the data sets, with respondents reporting PEs more likely to be female in the NCS-R only, more likely to be never married in the NLAAS Latino only, previously married in the NSAL, and unemployed in the NLAAS Asian and NSAL, and more likely to be out of the labor force in the NSAL but less likely in the NLAAS Asian sample. Mental disorders were common among respondents with PEs, although with notably lower rates of co-occurrence among the ethnic minority samples across disorder categories (except PTSD in the NSAL and substance use among NLAAS Latinos), particularly among Asian-Americans (Table 4). There was a significant linear relationship between the number of mental health conditions and psychosis risk across all four data sets (Table 5).
Table 4. Prevalence of disorders among respondents with psychotic experiences
NCS-R, National Comorbidity Survey Replication; NLAAS, National Latino and Asian American Study; NSAL, National Survey of American Life; PTSD, post-traumatic stress disorder.
Values given as percentage (standard error).
Table 5. Logistic regression tests for dose–response relationships between increasing numbers of diagnoses and odds of psychotic experiences (PEs)
NCS-R, National Comorbidity Survey Replication; NLAAS, National Latino and Asian American Study; NSAL, National Survey of American Life; s.e., standard error; df, degrees of freedom.
All analyses were fully adjusted for all potential sociodemographic confounders.
Discussion
Main findings
There is increasing evidence that subthreshold PEs occur frequently in the general population, particularly among individuals with one or more diagnosable disorders (Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; Scott et al. Reference Scott, Chant, Andrews, Martin and McGrath2007; Varghese et al. Reference Varghese, Scott, Welham, Bor, Najman, O'Callaghan, Williams and McGrath2011; Kelleher et al. Reference Kelleher, Keeley, Corcoran, Lynch, Fitzpatrick, Devlin, Molloy, Roddy, Clarke, Harley, Arseneault, Wasserman, Carli, Sarchiapone, Hoven, Wasserman and Cannon2012; Saha et al. Reference Saha, Scott, Varghese and McGrath2012; van Nierop et al. Reference van Nierop, van Os, Gunther, Myin-Germeys, de Graaf, ten Have, van Dorsselaer, Bak and van Winkel2012; Wigman et al. Reference Wigman, van Nierop, Vollebergh, Lieb, Beesdo-Baum, Wittchen and van Os2012). In these analyses, there were widespread moderate associations with PEs across all types of psychopathology, replicated in four data sets including a nationally representative sample and three samples within specific racial/ethnic groups. These associations were consistently weakened and only variably significant, although still in the predicted direction, when controlling for sociodemographic confounders and, perhaps more importantly, the remaining subtypes of disorder. However, the relationship between the extent of multi-morbidity (i.e. the number of subtypes of diagnosable disorders) was robust to controls for sociodemographic factors, exhibiting a strong linear relationship across all four data sets, suggesting that PEs may specifically be related to the quantity (number of co-occurring disorders) rather than the quality (specific diagnoses) of co-occurring mental health conditions. Although the prevalence of common mental disorders among individuals with PEs was generally lower in the racial/ethnic minority samples, the overall pattern of a linear association with multi-morbidity was consistent across groups, thus suggesting universality in the relationship of prevalence of common mental disorders with PEs within the USA.
Is psychosis risk related to the quality or the degree of psychopathology?
The prevalence of PEs was elevated in the context of every individual diagnosis and diagnostic category examined, across all samples, with a single exception (agoraphobia without panic disorder among Asian-Americans). Several studies have now tested for associations between PEs and specific psychiatric diagnoses, generally finding significantly greater prevalence of PEs among those with the disorder in question compared to those without the disorder, regardless of the disorder (i.e. depressive, anxiety and substance use disorders, and PTSD; Johns et al. Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; Scott et al. Reference Scott, Chant, Andrews, Martin and McGrath2007; Varghese et al. Reference Varghese, Scott, Welham, Bor, Najman, O'Callaghan, Williams and McGrath2011; Saha et al. Reference Saha, Scott, Varghese and McGrath2012; van Nierop et al. Reference van Nierop, van Os, Gunther, Myin-Germeys, de Graaf, ten Have, van Dorsselaer, Bak and van Winkel2012; Wigman et al. Reference Wigman, van Nierop, Vollebergh, Lieb, Beesdo-Baum, Wittchen and van Os2012). In a recent study by Fisher et al. (Reference Fisher, Caspi, Poulton, Meier, Houts, Harrington, Arseneault and Moffitt2013), childhood PEs were prospectively associated with schizophrenia and PTSD in adulthood, but not significantly associated with persistent anxiety or depressive disorders or substance dependence, although nearly all individuals with childhood PEs later met criteria for at least one of these diagnoses.
Overall, therefore, the literature points to widespread associations between PEs and psychopathology in general, with few examples of null associations. Given that the data presented here also demonstrate consistent associations with every disorder subtype (in bivariate analysis), it would be parsimonious to conclude that the odds of PEs are increased in the context of psychopathology in general, rather than in any particular subtype of mental health condition. Such non-specific co-occurrence has similarly been found among people with schizophrenia, a population with an elevated prevalence of co-morbidity across the entire spectrum of mental health conditions (Buckley et al. Reference Buckley, Miller, Lehrer and Castle2009), and among individuals with a family history of schizophrenia, who are similarly at elevated risk for a broad range of psychiatric disorders (Dean et al. Reference Dean, Stevens, Mortensen, Murray, Walsh and Pedersen2010; DeVylder & Lukens, Reference DeVylder and Lukens2013).
Although psychosis may not be related to the type of co-occurring psychopathology, it may be related to the severity of co-occurring psychopathology, as indicated by the number of co-occurring disorders. Such an explanation is consistent with the linear dose–response relationship identified here across all four data sets, in accordance with past studies showing that odds for PEs increased with more extensive multi-morbidity in adolescents (Kelleher et al. Reference Kelleher, Keeley, Corcoran, Lynch, Fitzpatrick, Devlin, Molloy, Roddy, Clarke, Harley, Arseneault, Wasserman, Carli, Sarchiapone, Hoven, Wasserman and Cannon2012, Reference Kelleher, Devlin, Wigman, Kehoe, Murtagh, Fitzpatrick and Cannon2013) and with greater severity within a single diagnosis (Saha et al. Reference Saha, Scott, Varghese and McGrath2012). This finding supports a construct of subthreshold psychosis as an independent syndrome or dimension of psychopathology, the risk for which is increased with greater co-morbidity but which is not bound to a single disorder. Notably, this association is robust across racial/ethnic groups.
Mechanisms of multi-morbidity
There are numerous potential explanations for the co-occurrence of psychosis with common mental health disorders, none of which could be distinguished based on the presented data but all of which should be explored in future studies. Recent evidence has shown that diagnostically disparate mental health conditions (i.e. schizophrenia, autism, attention-deficit hyperactivity disorder, bipolar disorder and major depressive disorder) share genetic risk loci (Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013), confirming through genetic analysis previous conclusions from research with twins (Kendler et al. Reference Kendler, Prescott, Myers and Neale2003) and in family history studies with population-level samples (Dean et al. Reference Dean, Stevens, Mortensen, Murray, Walsh and Pedersen2010; DeVylder & Lukens, Reference DeVylder and Lukens2013). As such, inherited genes may exert pleiotropic effects such that a common set of ‘risk’ genes may lead to different and seemingly disparate phenotypic expressions (Weiser et al. Reference Weiser, van Os and Davidson2005), which may then be more or less consistent with particular diagnostic criteria.
Psychosis and common mental health conditions also share environmental risk factors. For example, childhood trauma may lead to PTSD (e.g. Cloitre et al. Reference Cloitre, Stolbach, Herman, Kolk, Pynoos, Wang and Petkova2009), but is also more prevalent among people with depressive and anxiety disorders (Douglas et al. Reference Douglas, Chan, Gelernter, Arias, Anton, Weiss, Brady, Poling, Farrer and Kranzler2010; Hovens et al. Reference Hovens, Wiersma, Giltay, Van Oppen, Spinhoven, Penninx and Zitman2010) and substance abuse disorders (Wilsnack et al. Reference Wilsnack, Vogeltanz, Klassen and Harris1997; Medrano et al. 1999; Ducci et al. Reference Ducci, Roy, Shen, Yuan, Yuan, Hodgkinson, Goldman and Goldman2009; Douglas et al. Reference Douglas, Chan, Gelernter, Arias, Anton, Weiss, Brady, Poling, Farrer and Kranzler2010), and notably among people with PEs (Galletly et al. Reference Galletly, Van Hooff and McFarlane2011; Mackie et al. Reference Mackie, Castellanos-Ryan and Conrod2011; Soosay et al. Reference Soosay, Silove, Bateman-Steel, Steel, Bebbington, Jones, Chey, Ivancic and Marnane2012). Similar non-specific associations are found with social support, employment status, socio-economic status and other known risk factors for PEs (Linscott & van Os, Reference Linscott and van Os2013). Psychosis and common mental disorders also share putative neural substrates, including dysfunction in the hippocampus, hypothalamic–pituitary–adrenal axis, prefrontal cortex or dopaminergic system, which may mediate the relationship between genetic and environmental risk factors and symptom presentation.
Although it is parsimonious to assume that multi-morbidity is caused by shared risk factors, it is possible that it is caused by shared risk factors in addition to interactive effects between diagnoses (or dimensions of symptoms). A pilot analysis of the temporal order of diagnoses and PEs in this sample, based on retrospective estimates of age of onset, did not seem to be reliable or consistent across data sets (data not shown), leaving this an open question that would be better answered using prospective cohort data.
Psychosis and common disorders across racial/ethnic groups
The four data sets examined in this study were drawn from the same source country using similar methods, with race/ethnicity as the primary factor distinguishing the samples. No single prior study has attempted to replicate findings with PEs across racial/ethnic groups. In this analysis, prevalence rates of PEs were higher among the racial/ethnic minority samples (except for Asian-Americans) but lower for common mental disorders, in concurrence with prior research (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994; Johns et al. Reference Johns, Nazroo, Bebbington and Kuipers2002, Reference Johns, Cannon, Singleton, Murray, Farrell, Brugha, Bebbington, Jenkins and Meltzer2004; King et al. Reference King, Nazroo, Weich, McKenzie, Bhui, Karlsen, Stansfeld, Tyrer, Blanchard, Lloyd, McManus, Sproston and Erens2005; Breslau et al. Reference Breslau, Aguilar-Gaxiola, Kendler, Su, Williams and Kessler2006; Morgan et al. Reference Morgan, Fisher, Hutchinson, Kirkbride, Craig, Morgan, Dazzan, Boydell, Doody, Jones, Murray, Leff and Fearon2009; Cohen & Marino, Reference Cohen and Marino2013). However, findings regarding the relationship between common disorders and psychosis were consistent across data sets, particularly in that PEs were associated with (1) all classes of psychopathology in bivariate analyses and (2) multi-morbidity in a linear dose–response fashion, across all four samples. Our findings thus argue for the universality of the relationship between multi-morbidity of common mental disorders and the prevalence of PEs, at least across racial/ethnic groups in the USA.
Strengths, limitations and conclusions
The primary strength of this study is the use of four large general population data sets with varying racial/ethnic composition, allowing multiple replications and cross-cultural validation within a single study. Furthermore, diagnoses and the presence of PEs were determined using the WHO-CIDI, widely considered a valid measure for use in general population survey samples. However, despite validation with clinical interviews (Haro et al. Reference Haro, Arbabzadeh-Bouchez, Brugha, De Girolamo, Guyer, Jin, Lepine, Mazzi, Reneses, Vilagut, Sampson and Kessler2006), which themselves may be fallible, the assessment of psychiatric diagnoses and psychosis using lay-interviews may be subject to misclassification. Epidemiologists have argued intermittently for the use of psychometric or dimensional measures of psychopathology rather than categorical diagnoses (e.g. Dohrenwend, Reference Dohrenwend1990), which may produce more valid measures when administered by lay-interviewers, and indeed would have been consistent with the assumption, inherent in studying subthreshold psychosis, that psychopathology is continuously distributed in the population. Controversy over the validity of single diagnoses in lay-administered surveys is offset by the use of subgroups of disorders rather than individual diagnoses. However, lifetime diagnoses may be nonetheless underestimated because of extensive recall periods (Takayanagi et al. Reference Takayanagi, Spira, Roth, Gallo, Eaton and Mojtabai2014), but were chosen over 12-month diagnoses to increase statistical power for looking at rare constructs (i.e. PEs and multi-morbidity of several diagnoses), consistent with prior studies of multi-morbidity and psychosis (Kelleher et al. Reference Kelleher, Keeley, Corcoran, Lynch, Fitzpatrick, Devlin, Molloy, Roddy, Clarke, Harley, Arseneault, Wasserman, Carli, Sarchiapone, Hoven, Wasserman and Cannon2012, Reference Kelleher, Devlin, Wigman, Kehoe, Murtagh, Fitzpatrick and Cannon2013) and studies of multi-morbidity in these data (e.g. Ruscio et al. Reference Ruscio, Stein, Chiu and Kessler2010). Conversely, lifetime measures are limiting in that they allow examination of co-occurrence within individuals but not necessarily within overlapping time periods for those individuals. The use of clinical samples with greater prevalence of both PEs and disorders (relative to the general population) may provide sufficient statistical power to examine co-occurrence of psychosis and multi-morbidity within a specified time-frame (e.g. 12-month prevalence). Future studies should therefore replicate general population findings in clinical samples to improve temporal resolution and understand the significance of co-occurrence among help-seeking individuals. A further potential limitation is that PTSD and anxiety disorders are separated based on DSM-5 guidelines, although the diagnoses in the surveys used DSM-IV criteria. However, separating PTSD and anxiety disorders seemed necessary because of prior evidence that PTSD may be particularly closely related to PEs (Fisher et al. Reference Fisher, Caspi, Poulton, Meier, Houts, Harrington, Arseneault and Moffitt2013). Overall, concerns over measurement are partially allayed by the study design, which allowed up to four replications of any significant findings; measurement limitations, which would generally introduce random error, are unlikely to sufficiently explain several replications of a statistical association.
Survey response rates were in the typical range for national surveys, notably lowest in the NLAAS Asian data set (65.6%) and highest in the NLAAS Latino data set (75.5%). Higher rates of psychopathology have been found among survey non-responders (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994), which would potentially lead to underestimation of prevalence rates of psychiatric diagnoses and PEs proportionate to the response rate, although response rates may have less of an effect on prevalence rates than previously assumed and typically do not notably influence the relative associations between clinical variables (Galea & Tracy, Reference Galea and Tracy2007).
Finally, PE prevalence rates could not be compared statistically across racial/ethnic groups because of the use of separate data sets. Specifically, because the psychosis screen was given to a random subsample of the NCS-R, to the entire sample of the NLAAS and systematically only to non-white respondents in the NSAL, comparisons between data sets and between racial/ethnic groups using the combined data set (CPES) would be improperly weighted and therefore of questionable validity (DeVylder, Reference DeVylder2014). Despite this limitation, our prevalence rates are consistent with previous reports using the combined CPES data set showing significant racial/ethnic prevalence differences (Cohen & Marino, Reference Cohen and Marino2013, Reference Cohen and Marino2014).
Given the replication of findings across four large data sets, this study provides strong evidence that PEs are associated with multi-morbidity of common mental health conditions, both in the general population of the USA and within specific racial/ethnic categories. Further prospective research would be useful in disentangling the causal relationship between PEs and co-occurring disorders.
Supplementary material
For supplementary material accompanying this paper, please visit http://dx.doi.org/10.1017/S0033291714000944.
Acknowledgments
We thank F. Chen, B. Link and E. Lukens for helpful comments on an earlier version of this manuscript, and H. Oh for independent verification of statistical analyses. We also thank the investigators responsible for the CPES for making their data publicly available to the research community.
Declaration of Interest
None.
