Introduction
Controversy has surrounded the diagnosis of generalized anxiety disorder (GAD) since its introduction into the Diagnostic and Statistical Manual of Mental Disorders (DSM)-III (APA, 1980). Prior to that time, GAD was conceptualized as one of the two core components of anxiety neurosis, the other being panic (APA, 1968). Recognition that GAD and panic, while often co-occurring, are sufficiently distinct to be considered independent disorders led to their separation in DSM-III, where a diagnosis of GAD required uncontrollable and diffuse anxiety or worry that was excessive or unrealistic in relation to objective life circumstances that persisted for 1 month or longer in addition to a number of psycho-physiological symptoms.
Early clinical studies in clinical samples found that DSM-III GAD seldom occurred in the absence of major depression (MD) (Breslau, Reference Breslau1985; Breslau & Davis, Reference Breslau and Davis1985a), leading some to the suggestion that GAD might best be conceptualized as a prodrome, residual, or severity marker of MD (Cloninger et al. Reference Cloninger, Martin, Guze, Clayton, Maser and Cloninger1990; Offord et al. Reference Offord, Boyle, Campbell, Cochrane, Goering, Lin, Rhodes and Wong1994; Brown et al. Reference Brown, Chorpita and Barlow1998). However, this co-morbidity weakened as duration of GAD increased (Breslau & Davis, Reference Breslau and Davis1985b), leading the GAD duration requirement to be increased to 6 months in the revised DSM-III (DSM-III-R) (APA, 1987) and DSM-IV (APA, 1994).
The issue of a strong association between GAD and MD was not resolved, though, with this change, as community epidemiological studies using DSM-III-R and DSM-IV criteria continued to find strong co-morbidity between GAD and MD (Kessler et al. Reference Kessler, Nelson, McGonagle, Liu, Swartz and Blazer1996, Reference Kessler, Chiu, Demler, Merikangas and Walters2005; Grant et al. Reference Grant, Hasin, Stinson, Dawson, June Ruan, Goldstein, Smith, Saha and Huang2005). Attempts to explore the factor structure of co-morbidity among axis I disorders concluded that GAD and MD are both strongly related to a general ‘distress’ factor (Krueger et al. Reference Krueger, Caspi, Moffitt and Silva1998; Krueger, Reference Krueger1999; Vollebergh et al. Reference Vollebergh, Iedema, Bijl, de Graaf, Smit and Ormel2001) that also includes dysthymia, post-traumatic stress disorder (PTSD), social phobia, and, in at least one analysis (Slade & Watson, Reference Slade and Watson2006), neurasthenia. However, longitudinal analysis showed meaningful divergence between GAD and MD both in risk factors (Moffitt et al. Reference Moffitt, Caspi, Harrington, Milne, Melchior, Goldberg and Poulton2007) and in illness course (Fergusson et al. Reference Fergusson, Horwood and Boden2006), arguing that more is involved than a single latent factor leading to both disorders.
The prospective evidence of somewhat different environmental risk factors for GAD and MD is consistent with the results of genetic epidemiological studies, which suggest that while the genes for GAD and MD are very similar or possibly even identical, the environmental determinants are less strongly related (Kendler et al. Reference Kendler, Neale, Kessler, Heath and Eaves1992, Reference Kendler, Gardner, Gatz and Pedersen2007; Roy et al. Reference Roy, Neale, Pedersen, Mathe and Kendler1995; Kendler, Reference Kendler1996). Taken together with the prospective evidence for differential risk factors, these data suggest that, despite strong similarities that might lead them to be placed in the same diagnostic category in the upcoming revisions of the DSM and International Classification of Diseases (ICD) systems (Watson, Reference Watson2005), GAD and MD are distinct disorders. This conclusion is consistent with the results of neurobiological studies, which find numerous differences that argue against GAD and MD being the same neurobiological disorder (Martin & Nemeroff, in press). However, the body of prospective epidemiological risk factor evidence that elaborates the implications of this conclusion is thin. The current report presents prospective data that add to this body of evidence by examining patterns and risk factors for onset and persistence of GAD and MD in a nationally representative two-wave panel survey of the household population of the USA.
Method
Sample
Data come from the 5001 respondents who participated in the 1990–1992 National Comorbidity Survey (NCS) and 2001–2003 NCS follow-up survey (NCS-2). The NCS (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994) was a nationally representative US survey of 8098 respondents aged 15–54 years carried out between 1990 and 1992. The response rate was 82.4%. Interviews were conducted by professional survey interviewers and administered in two parts. Part I, which included the core diagnostic interview, was administered to all respondents. Part II, which included additional disorders and risk factors, was administered to a probability subsample of 5877 respondents including all respondents aged 15–24 years, all others with any lifetime DSM-III-R disorder assessed in Part I, and a random subsample of remaining Part I respondents. The Part II sample was weighted to adjust for differential probabilities of selection and for non-response bias. Importantly for the purposes of this paper, the non-response adjustment weight was based on the results of a brief screening survey carried out in a representative subsample of initial survey non-respondents. Diagnostic stem questions for both MD and GAD were included in that weight, which means that adjustments were made for non-response bias based on MD and GAD. Further details about the NCS design, the non-response survey and weighting are reported elsewhere (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994).
The NCS-2 sought to trace and re-interview the Part II NCS respondents a decade after the NCS. Of those, 5463 were successfully traced, of whom 166 were deceased and 5001 re-interviewed, for a conditional response rate of 87.6%. The unconditional response rate is 72.2% (0.876×0.824). NCS-2 respondents were assessed using an expanded version of the baseline interview that assessed onset and course of disorders between the two surveys. Relative to other baseline NCS respondents, NCS-2 respondents were significantly more likely to be female, well educated and residents of rural areas. A propensity score adjustment weight (Rosenbaum & Rubin, Reference Rosenbaum and Rubin1983) corrected for these discrepancies. No difference existed between NCS-2 respondents and non-respondents in baseline history of either major depressive episodes (MDE) [χ2(1)=0.8, p=0.39] or GAD [χ2(1)=2.1, p=0.16], although we have no way to know if subsequent onset-persistence of MDE or GAD differed for follow-up respondents versus non-respondents.
Diagnostic assessment
Lifetime DSM-III-R disorders were assessed in the baseline NCS with the World Health Organization Composite International Diagnostic Interview (CIDI) version 1.1 (Robins et al. Reference Robins, Wing, Wittchen, Helzer, Babor, Burke, Farmer, Jablenski, Pickens, Regier, Farmer, Jablenski, Pickens, Regier, Sartorious and Towle1988), a fully structured, lay-administered diagnostic interview. DSM-IV disorders that had first onsets in the decade between the two interviews were assessed in the NCS-2 using CIDI version 3.0 (Kessler & Ustun, Reference Kessler and Ustun2004). DSM organic exclusion rules were used in making diagnoses in both surveys. The NCS-2 assessment also considered first onsets of DSM-IV disorders prior to the time of the baseline interview that were not reported at baseline. This inconsistency in reporting was resolved by coding such disorders as having occurred prior to baseline despite not being reported at baseline in order to make lower-bound estimates on age of onset. This inconsistency in reporting was uncommon (fewer than 5% of all lifetime cases of MDE and GAD) and equivalent for the two disorders.
Persistence of baseline disorders in the decade between interviews was assessed in NCS-2 with a life history calendar (Belli, Reference Belli1998) that charted prevalence for each year in the decade between the interviews. The life history calendar method has been shown experimentally to produce significantly more accurate retrospective recall than more conventional survey methods (Belli et al. Reference Belli, Shay and Stafford2001). Blinded clinical re-interviews assessed concordance of CIDI diagnoses with clinical diagnoses using the Structured Clinical Interview for DSM-III-R (Spitzer et al. Reference Spitzer, Williams, Gibbon and First1992) in the NCS and the Structured Clinical Interview for DSM-IV (SCID; First et al. Reference First, Spitzer, Gibbon and Williams2002) in the NCS-2. Good CIDI–SCID concordance was found for MDE and for GAD in both surveys, with areas under the curve of 0.78 for MDE and 0.71 for GAD in the NCS (Kessler et al. Reference Kessler, Wittchen, Abelson, McGonagle, Schwarz, Kendler, Knäuper and Zhao1998) and 0.75 for MDE and 0.83 for GAD in the NCS replication (Haro et al. Reference Haro, Arbabzadeh-Bouchez, Brugha, di Girolamo, Guyer, Jin, Lepine, Mazzi, Reneses, Vilagut Saiz, Sampson and Kessler2006).
Prospective risk factors
Three risk factor sets were examined: childhood adversities, parental history of common mental-substance disorders, and respondent personality. All three have been found to predict both MDE and GAD in previous studies (Kendler et al. Reference Kendler, Davis and Kessler1997; Hettema et al. Reference Hettema, Neale, Myers, Prescott and Kendler2006; Moffitt et al. Reference Moffitt, Caspi, Harrington, Milne, Melchior, Goldberg and Poulton2007). The childhood adversities included three dimensions of maltreatment (neglect, physical abuse, sexual abuse) and three dimensions of loss (parental death, parental divorce, other long-term separation). The parent disorders included MDE, GAD, panic disorder, antisocial personality disorder, and alcohol-drug dependence. The personality dimensions included neuroticism, extroversion, and openness to experience. Detailed discussions of measurement have been presented elsewhere (Kendler et al. Reference Kendler, Davis and Kessler1997; Kessler et al. Reference Kessler, Davis and Kendler1997; Mickelson et al. Reference Mickelson, Kessler and Shaver1997) and will not be repeated here other than to note that childhood adversities were assessed with questions developed specifically for the NCS, parental psychopathology with the Family-History Research Diagnostic Criteria interview (Endicott et al. Reference Endicott, Andreasen and Spitzer1978), and personality with a modified version of the Goldberg (Reference Goldberg1992) personality scales. Childhood adversities and parental history were defined dichotomously. Personality was defined continuously using standardized scales with a mean of zero and variance of one.
Statistical analyses
Cross-tabulations were used to estimate lifetime prevalence and persistence. Discrete-time survival analysis with person-year as the unit of analysis (Efron, Reference Efron1988) was used to study prospective predictors of first onset and persistence. A logistic link function was used in the survival models. Survival analysis was used instead of logistic regression to consider information in the life history calendar about speed of onset and the number of years of persistence in the decade between interviews. Prior lifetime history of MDE was studied as a time-varying predictor on GAD and vice versa. MDE was ‘time-varying’ in the sense that we took into consideration retrospectively reported information about age-of-onset (AOO) of MDE in predicting later GAD and vice versa. Standard errors (s.e.s) and confidence intervals (CIs) were estimated using the Taylor series method (Wolter, Reference Wolter1985) implemented in the sudaan software system (Research Triangle Institute, 2002) to adjust for the weighting and clustering of the data. Multivariate significance was evaluated using Wald χ2 tests based on design-corrected coefficient variance–covariance matrices. Statistical significance was evaluated consistently using two-tailed 0.05-level tests.
Results
Prevalence and lifetime co-morbidity between MDE and GAD
As of the time of the baseline NCS, 21.2% (s.e.=0.8%) of NCS-2 respondents met lifetime criteria for DSM-III-R MDE. Of respondents without baseline lifetime MDE, 9.6% (s.e.=0.8%) had an onset by NCS-2, resulting in a 28.8% (s.e.=0.9%) lifetime prevalence at NCS-2. GAD was much less prevalent, with 8.6% (s.e.=0.5%) lifetime prevalence at baseline, 3.6% (s.e.=0.4%) onset between waves and 11.9% (s.e.=0.6%) lifetime prevalence at NCS-2. Lifetime MDE and GAD were strongly co-morbid at baseline, with an odds ratio (OR) of 7.5 (95% CI 5.5–10.3). The OR at NCS-2 was 6.6 (95% CI 3.3–13.3).
Time-lagged associations involving first onset of the MDE and GAD
We examined time-lagged associations of first onset using retrospective AOO reports. The OR of temporally primary MDE predicting subsequent first onset of GAD is 2.7 (Table 1, top). The OR of temporally primary GAD predicting subsequent first onset of MDE is 3.2 (Table 1, bottom). Both associations are somewhat stronger among respondents who were adolescents (15–24 years of age) than adults (25–54 years of age) at baseline (3.7 v. 1.7–2.4 for MDE predicting GAD; 3.6 v. 3.0–3.1 for GAD predicting GAD).
Table 1. Cross-sectional and time-lagged associations of one disorder, either MDE or GAD, predicting the first onset of the other disorder as a function of time since onset of the primary disorder and cohort in the NCS-2 panel sample (n=5001)Footnote a
a Based on a discrete-time survival model with person-year as the unit of analysis, controlling for cohort (age at interview), sex, race-ethnicity and person-year.
* Significant at the 0.05 level (two-sided test).
The time-lagged ORs decrease dramatically with time since onset of the temporally primary disorder, although the elevated ORs remain statistically significant more than a decade after onset. All these ORs are somewhat more pronounced among respondents in the younger than older cohorts. Strikingly, the ORs associated with GAD and MDE starting in the very same year are much higher than any of the time-lagged ORs (54.1–54.4 v. 2.0–8.9). We cannot distinguish temporal priority between MDE and GAD within a year, though, so we have no way to know if same-year onsets involve MDE starting before, after, or simultaneous with GAD.
Time-lagged associations involving persistence of MDE and GAD
Some 46.2% (s.e.=1.8%) of respondents with lifetime MDE at baseline had recurrences of MDE in the intervening decade. On average, baseline cases reported having recurrent MDE in 19.8% (s.e.=1.0%) of the years between the two interviews, again with an inverse relationship with age and a relatively narrow range from the youngest to oldest age groups (21.9–16.4%). Inter-temporal persistence of GAD was slightly higher, with 49.7% (s.e.=2.4%) of baseline lifetime cases having episodes in at least one inter-current year and a mean of 27.4% (s.e.=1.7%) of years in episode.
We examined time-lagged association between baseline MDE and persistence of GAD and vice versa. An important asymmetry was found: that prior GAD significantly predicted persistence of MDE (1.8), while prior MDE did not significantly predict persistence of GAD (1.2) (Table 2). These prediction equations controlled for AOO of the outcome disorder, which was consistently associated with significantly decreased risk of recurrence. The association between co-morbid GAD and persistence of MDE was unrelated to temporal priority in first onset of MDE and GAD.
Table 2. Time-lagged associations of baseline MDE predicting the persistence of GAD and of baseline GAD predicting the persistence of MDE by cohort in the NCS-2 panel sampleFootnote a
a Based on a discrete-time survival model with person-year as the unit of analysis, controlling for cohort (age at interview), sex, race-ethnicity and person-year.
b Unweighted sample sizes.
* Significant at the 0.05 level (two-sided test).
Prospective risk factors for first onset of MDE and GAD
When considered one at a time, most of the childhood adversities considered here were associated with significantly elevated risk of subsequent GAD, with ORs in the range 1.4–1.8. (Table 3). More detailed analysis (results available on request) showed that these effects are largely confined to childhood–adolescent onsets. Associations of childhood adversity with subsequent onset of MDE were more complex. As with GAD, most childhood adversities were associated with significantly elevated risk of MDE in the total sample (1.4–2.2). However, unlike GAD, where significant ORs were largely confined to childhood–adolescent onsets, more detailed analysis (results available on request) showed that childhood adversities also predict first onset of MDE into early adulthood (ages 25–39 years). Furthermore, parental death significantly predicts adolescent onsets of MDE (1.6) but not GAD (0.8).
Table 3. Time-lagged associations of childhood adversities, parental history of mental disorders, and respondent personality with the subsequent first onset of GAD and MDE in the NCS-2 panel sample (n=5001)Footnote a
a Based on a discrete-time survival model with person-year as the unit of analysis, controlling for cohort (age at interview), sex, race-ethnicity and person-year.
b Estimated only in person-years subsequent to the baseline interview and among those without an onset of the outcome disorder prior to the baseline interview due to the fact that personality was only assessed as of the time of the baseline interview (n=4470).
* Significant at the 0.05 level (two-sided test).
Parent history of MDE, GAD, substance disorder and panic disorder all predicted GAD (1.5–1.6). More detailed analyses (results available on request) showed little evidence of variation based on number of parents with the disorder, sex of the parent with the disorder in cases where only one had it, or match between sex of disordered parent and respondent. As with childhood adversities, more detailed analyses (results available on request) showed effects confined to childhood–adolescent onsets. The effects of parental disorders on MDE were somewhat more consistent and stronger. As with GAD, more detailed analyses (results available on request) showed little evidence of variation in effects based on number of parents with the disorder, sex of parent with the disorder in cases where only one had it, or match between sex of disordered parent and respondent. However, unlike GAD, effects of parent disorders on MDE also were significant in adulthood.
As personality was assessed only in the baseline NCS, the associations of personality with onset of GAD were examined only in the person-years subsequent to the baseline interview. Neuroticism is the only dimension of personality significantly associated with elevated odds of GAD. Unlike childhood adversity, more detailed analyses (results available on request) found a significantly elevated OR of neuroticism with subsequent GAD in early adulthood (25–39 years of age; OR 1.4). In addition, extroversion is associated with significantly reduced odds of GAD onset in early adulthood (OR 0.7), but not in the total sample or at other parts of the life course examined here. The ORs of neuroticism predicting subsequent onset of MDE are similar in magnitude to those predicting onset of GAD, although not significant. While the association between extroversion and MDE is significant in two of the cohorts it is not consistently correlated (positive during childhood or adolescence, negative in middle age). Openness to experience, finally, is more consistently correlated (positively) with MDE than GAD.
When controls were introduced for all the above risk factors, the significant time-lagged associations (ORs) of temporally primary MDE with subsequent first onset of GAD attenuated somewhat, but remained statistically significant in the total sample (1.8), increased among respondents who were adolescents at baseline (5.0), and became insignificant in older cohorts (1.4–1.8). The introduction of controls led to more modest attenuation of the OR between GAD and subsequent MDE, which remained statistically significant in the total sample (2.8) as well as in the early adulthood and middle-age subsamples (2.7–3.3).
Prospective risk factors for persistence of MDE and GAD
Persistence of GAD in the total sample was significantly predicted by several childhood adversities (1.4–1.8) and by neuroticism (1.1), but not by any of the measures of parental history of mental or substance disorders. (results not presented, but available on request.) We were unable to study the stability of these associations across cohorts due to sparse data. Persistence of MDE in the total sample was significantly predicted by a somewhat different set of childhood adversities (1.3–1.5), by parental panic disorder (1.3) and by two personality dimensions: neuroticism (1.1) and openness to experience (1.1). As with GAD, subsample associations in cohorts could not be examined because of sparse data. When controls were introduced for childhood adversities, parental history of mental and substance disorders, and personality, the significant time-lagged associations (ORs) of baseline history of MDE with the subsequent persistence of GAD remained insignificant (1.1). In the case of GAD predicting persistence of MDE, the introduction of controls led to modest attenuation of the association, but the association remained statistically significant (1.7).
Discussion
The results reported here are limited in four ways. First, the assessments of MDE and GAD were based on fully structured diagnostic interviews that are likely to be less accurate than clinician-administered diagnostic interviews. Second, AOO and persistence were assessed retrospectively. Third, although the risk factor data were gathered prospectively, they were based on retrospective reports. Fourth, although analyses were carried out in parallel to predict MDE and GAD, formal modeling procedures were not used to adjust for the influence of a latent ‘internalizing’ variable so as to determine the extent to which the effects of predictors on MDE and GAD are mediated through such a construct.
The impact of the first of these limitations is reduced somewhat by evidence of good concordance between diagnoses based on the CIDI and those based on blinded clinical reappraisal interviews. The second and third limitations are inherent in the design, but are to some extent counterbalanced by the fact that we worked with prospective data. The fourth limitation needs to be addressed in future investigations, but should ideally be studied using a larger set of internalizing disorders (e.g. social phobia, panic disorder, PTSD). It is noteworthy, though, that the possibility of a latent factor completely explaining inter-temporal associations between two variables is inconsistent both with the magnitude of the cross-lagged associations between those two variables (i.e. the association between MDE at time 1 and GAD at time 2 and the association between GAD at time 1 and MDE at time 2) differing significantly from each other and with there being significant interactions involving differential associations of either time 1 variable with onset versus persistence of the other variable (Lazarsfeld, Reference Lazarsfeld, Hammer, Salzinger, Sutton and Zubin1973; Kessler, Reference Kessler1977). Both these patterns are observed in our data. Specifically, the cross-lagged associations involving persistence are asymmetric, with GAD predicting MDE more strongly than MDE predicting GAD, and the time-lagged associations are stronger in predicting onset than persistence of both MDE and GAD. Both of these patterns argue against a latent variable explaining the associations, although formal evaluation of this possibility awaits future investigation.
These limitations notwithstanding, the study provides useful new information about MDE–GAD co-morbidity and prospective predictors of MDE and GAD onset and persistence. The strong MDE–GAD co-morbidity found here is consistent with much previous research (Noyes, Reference Noyes2001; Kessler, Reference Kessler2000; Belzer & Schneier, Reference Belzer and Schneier2004; Gorwood, Reference Gorwood2004), although our prospective data allowed us to decompose this cross-sectional association to document significant reciprocal cross-lagged relationships that vary with amount of time since the onset of the temporally primary disorder in predicting first onset. The most striking aspect of this time decay is the enormous same-year ORs. These same-year associations are so strong that more than one-third of all lifetime co-occurring MDE–GAD above that expected by chance occurs among cases in which both disorders started in the same year. We have no way to know what the strong association between time since onset and risk of co-morbidity might mean, but it clearly implies that the temporally primary disorder is more than a mere marker of stable constitutional risk, as the latter would not be expected to produce an association between time since first onset of the temporally primary disorder and risk of the secondary disorder. The time decay also has implications for the argument in the literature that GAD might be nothing more than a prodrome or severity marker of MDE rather than an independent disorder (Cloninger et al. Reference Cloninger, Martin, Guze, Clayton, Maser and Cloninger1990; Offord et al. Reference Offord, Boyle, Campbell, Cochrane, Goering, Lin, Rhodes and Wong1994; Brown et al. Reference Brown, Chorpita and Barlow1998). If this argument were correct, the window of risk for secondary depression would be fairly short, which we find that it is not.
An alternative view of the causal processes linking temporally primary GAD with secondary MDE argues that secondary depression is an exhaustion response to unremitting anxiety (Akiskal, Reference Akiskal, Tuma and Maser1985). Primary anxiety, in this view, can be conceptualized as a stressor that promotes secondary depression (Akiskal, Reference Akiskal, Maser and Cloninger1990; Durham et al. Reference Durham, Allan and Hackett1997). If this were the case, though, we would expect that the OR for GAD predicting secondary MDE would become higher with the passage of time since onset of GAD. The fact that the opposite is the case argues against this interpretation. A more likely scenario is that the cross-lagged associations are due either to unmeasured common causes, to the effects of one disorder on the other, or to some combination of these two effects. We know that stable effects of the former sort exist. Indeed, as noted in the Introduction, a number of population twin studies suggest that genetic influences are important common causes of MDE and GAD (Gorwood, Reference Gorwood2004; Kendler et al. Reference Kendler, Gardner, Gatz and Pedersen2007). However, we can think of no biologically plausible mechanism whereby a genetic common cause would lead to a time decay of the sort seen in the NCS-2 data in the ORs linking onset of temporally primary disorders with subsequent onset of secondary disorders. Nor could common genetic causes account for all the observed co-morbidity between MDE and GAD, as the latter is higher than would be predicted based solely on the heritability of MDE and GAD and the strength of the association between the genes for the MDE and GAD. This means that there must also be common environmental causes of MDE and GAD. The estimates in population twin studies suggest that the latter explain between one-third and two-thirds of the significant co-morbidity between MDE and GAD, assuming that the joint effects of genetic and environmental causes are additive.
We investigated a number of potentially important prospective risk factors for MDE and GAD that might be considered common causes. Most of these were found to predict the first onset of both MDE and GAD, but fewer were significant prospective predictors of persistence. Statistical control for these common predictors only explained a small proportion of the observed cross-lagged associations between MDE and GAD involving either onset of persistence. This could mean that we simply failed to measure the common environmental causes that are important for explaining the cross-lagged associations between MDE and GAD. Another possibility is that the occurrence of one of these two disorders in itself might increase risk of the subsequent onset of the second disorder, possibly through some type of sensitization or kindling phenomenon (Post & Weiss, Reference Post and Weiss1998). An important implication of the possibility that temporally primary MDE or GAD might itself increase risk of the subsequent first onset of the other disorder is that successful treatment of the temporally primary disorder might be expected in such a case to reduce risk of onset of co-morbidity. We are aware of no empirical research on this possibility. Given the high ORs associated with one of the two temporally primary disorders predicting the subsequent first onset of the other in the first few years after first onset, though, the sample size required to detect a preventive effect of this sort in an effectiveness trial framework with long-term (5-year) follow-up would not be prohibitive. It might be that this kind of experimental investigation is the only way to resolve the uncertainly regarding whether temporally primary MDE and GAD are causal risk factors for each other or only risk markers.
Even though the prospective risk factors considered here did not explain the time-lagged associations between MDE and GAD, they are important in documenting similarities in the environmental determinants of the two disorders and some differences in the associations between personality and the subsequent onset of the two disorders. Most notable in the latter regard are associations of neuroticism with GAD but not MDE and of openness to experience with MDE but not GAD. Both of these differences, though, are small. We also found that MDE was not a significant predictor of persistence of GAD, while GAD was a significant predictor of persistence of MDE. This difference is consistent with an earlier finding based on the analysis of retrospective reports in the baseline NCS. This earlier finding was associated with an investigation of whether co-morbidity is related more strongly to the persistence and severity of GAD than other anxiety or mood disorders (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994). The rationale of this analysis was that if GAD was a prodrome, residual, or severity marker of MDE, as early commentators suggested (Cloninger et al. Reference Cloninger, Martin, Guze, Clayton, Maser and Cloninger1990; Offord et al. Reference Offord, Boyle, Campbell, Cochrane, Goering, Lin, Rhodes and Wong1994; Brown et al. Reference Brown, Chorpita and Barlow1998), the persistence-severity of GAD would be more strongly affected by co-morbidity than would persistence-severity of depression. Yonkers et al. (Reference Yonkers, Warshaw, Massion and Keller1996) reported a similar result in the prospective Harvard/Brown Anxiety Research Program (HARP) study. These results provide concrete substantiation that MDE and GAD have partially distinct environmental determinants and add to the evidence from previous prospective studies of differences in the risk factors for MDE and GAD (Moffitt et al. Reference Moffitt, Caspi, Harrington, Milne, Melchior, Goldberg and Poulton2007) and differences in inter-temporal stability of these disorders that cannot be explained by a common underlying internalizing factor (Fergusson et al. Reference Fergusson, Horwood and Boden2006). Future research needs to search for evidence of additional differences of this sort in an effort to expand our understanding of the distinction between common and distinct risk factors for MDE, GAD, and other internalizing disorders.
Acknowledgments
Funding for the NCS-2 was provided by the National Institute on Drug Abuse at the National Institutes of Health (1R01DA012058).
Declaration of Interest
None.
