Although significant associations of childhood adversities (CAs) with adult mental disorders have been documented consistently in epidemiological surveys, these studies generally have examined only 1 CA per study. Because CAs are highly clustered, this approach results in overestimating the importance of individual CAs. Multivariate CA studies have been based on insufficiently complex models.
To examine the joint associations of 12 retrospectively reported CAs with the first onset of DSM-IV disorders in the National Comorbidity Survey Replication using substantively complex multivariate models.
Cross-sectional community survey with retrospective reports of CAs and lifetime DSM-IV disorders.
Household population in the United States.
Nationally representative sample of 9282 adults.
Main Outcome Measures
Lifetime prevalences of 20 DSM-IV anxiety, mood, disruptive behavior, and substance use disorders assessed using the Composite International Diagnostic Interview.
The CAs studied were highly prevalent and intercorrelated. The CAs in a maladaptive family functioning (MFF) cluster (parental mental illness, substance abuse disorder, and criminality; family violence; physical abuse; sexual abuse; and neglect) were the strongest correlates of disorder onset. The best-fitting model included terms for each type of CA, number of MFF CAs, and number of other CAs. Multiple MFF CAs had significant subadditive associations with disorder onset. Little specificity was found for particular CAs with particular disorders. Associations declined in magnitude with life course stage and number of previous lifetime disorders but increased with length of recall. Simulations suggest that CAs are associated with 44.6% of all childhood-onset disorders and with 25.9% to 32.0% of later-onset disorders.
The fact that associations increased with length of recall raises the possibility of recall bias inflating estimates. Even considering this, the results suggest that CAs have powerful and often subadditive associations with the onset of many types of largely primary mental disorders throughout the life course.
Significant associations between retrospectively reported childhood adversities (CAs) and adult illness have been documented in numerous studies.1,2 The first such studies focused on only a single CA, such as parental death or neglect,3,4 and 1 mental disorder, most often depression.5,6 Subsequent studies showed that retrospectively reported CAs are often highly clustered,7,8 requiring examination of multiple CAs to avoid overestimating associations involving particular CAs.2,9,10 These studies also found that CAs are often nonspecific in their associations with many different mental disorders,10-12 making it useful to examine multiple outcomes to avoid overly narrow interpretations.
Subsequent studies13-15 created summary CA scales and documented dose-response relationships with adult outcomes. However, such indices implicitly assumed that each CA has the same effect and that joint effects are additive. These assumptions are almost certainly incorrect.16 Indeed, a preliminary examination of these assumptions in the National Comorbidity Survey (NCS)17 showed that some CAs have stronger associations with adult outcomes than do others and that joint associations are nonadditive.10 That study also found that these associations sometimes attenuate with age, a specification generally, but not always,12,18 ignored in subsequent studies.
The present study builds on these earlier NCS findings by analyzing the CAs assessed in the NCS Replication (NCS-R).19 Associations between retrospectively reported CAs and mental disorders can be upwardly biased owing to recall failure, nevertheless, it is useful to examine associations based on such retrospective data because they provide upper estimates that avoid the problem of downward bias due to systematic sample attrition in estimates based on long-term prospective data. We examine associations of CAs with the first onset of diverse DSM-IV disorders based on several competing multivariate models. A companion study20 examines associations of CAs with lifetime persistence of the same disorders.
The NCS-R is a face-to-face survey of English-speaking adults performed between February 5, 2001, and April 7, 2003, in a multistage clustered area probability sample of the US household population.19 The response rate was 70.9%. Recruitment began with a letter and a study fact brochure followed by in-person interviewer visits to explain study aims and procedures before obtaining informed consent. Respondents were paid $50 for participation. Recruitment and consent procedures were approved by the human subjects committees of Harvard Medical School, Boston, Massachusetts, and the University of Michigan, Ann Arbor.
The survey was administered in 2 parts. Part I included a core diagnostic assessment and was administered to all the respondents (n = 9282). Part II, which was generally administered on the same occasion as part I, included questions about correlates and additional disorders administered to all part I respondents who met lifetime criteria for any part I disorder plus a probability subsample of other part I respondents (n = 5692). The part I sample was weighted to adjust for differential probabilities of selection and intensity of recruitment effort in hard-to-recruit cases. The part II sample, the focus of the present study, was additionally weighted for the lower selection probabilities of part I respondents without a mental disorder. A final weight adjusted the sample to match the 2000 census population on the cross-classification of numerous geographic and sociodemographic variables. All the analyses used these weights. As a result, the sociodemographic characteristics of the weighted part II sample closely match those of the population (eg, 42% female, 71% non-Hispanic white, 24% aged 18-29 years, and 21% ≥60 years old). More detailed information on NCS-R sampling, design, weighting, and sociodemographic distribution is reported elsewhere.21
The NCS-R lifetime diagnoses are based on the World Health Organization Composite International Diagnostic Interview (CIDI),22 a fully structured, lay-administered interview that generates diagnoses according to the definitions and criteria of the Internaaational Classification of Diseases, 10th Revision and the DSM-IV. The DSM-IV criteria are used herein. The lifetime diagnoses include 4 broad classes of 20 specific disorders: mood disorders (major depressive disorder, dysthymic disorder, bipolar I disorder, bipolar II disorder, and subthreshold bipolar disorder), anxiety disorders (panic disorder, agoraphobia without a history of panic disorder, generalized anxiety disorder, specific phobia, social phobia, posttraumatic stress disorder, and separation anxiety disorder), disruptive behavior disorders (intermittent explosive disorder, attention-deficit/hyperactivity disorder, oppositional-defiant disorder, and conduct disorder), and substance use disorders (alcohol abuse, alcohol dependence with abuse, drug abuse, and drug dependence with abuse). Diagnostic hierarchy rules and organic exclusion rules were used in making diagnoses. The DSM-IV/CIDI disorder prevalence estimates in sociodemographic subsamples are reported elsewhere (http://www.hcp.med.harvard.edu/ncs). An NCS clinical reappraisal study23 found generally good concordance between diagnoses based on the CIDI and those based on blinded clinical reinterviews using the Structured Clinical Interview for DSM-IV.24
The CIDI assessed age at onset of the disorder retrospectively. Based on evidence that retrospective age-at-onset reports are often erroneous,25 a special question sequence was used to improve the accuracy of reporting. This began with questions designed to emphasize the importance of accurate responses: “Can you remember your exact age the very first time [emphasis in original] when you had [the symptom/the syndrome]?” Respondents who answered “no” were then probed for a bound of uncertainty by asking the earliest age at which they could clearly remember having the disorder. Onset was set at the upper end of the bound of uncertainty. Experimental research26 has shown that this approach yields more plausible age-at-onset distributions than do standard age-at-onset questions.
Twelve dichotomous CAs occurring before age 18 years were assessed in the NCS-R. The selection of CAs was based on a review of the literature. These CAs include 3 types of interpersonal loss (parental death, parental divorce, and other separation from parents or caregivers), 4 types of parental maladjustment (mental illness, substance abuse, criminality, and violence), 3 types of maltreatment (physical abuse, sexual abuse, and neglect), and 2 other CAs (life-threatening childhood physical illness in the respondent and extreme childhood family economic adversity). The measures of parental death, divorce, and other separation (eg, respondent placed in foster care) focus only on the biological parents, not on stepparents or other caregivers. Respondents who were born to a single mother and never experienced any further disruption of this parenting arrangement were coded as not experiencing any parental separation. We did not include information about the number of caregiver disruptions (eg, multiple divorces) or separations (eg, multiple foster care placements) but rather coded respondents dichotomously as having any vs no such disruptions because the rarity of multiple disruptions made estimates of dose-response relationships unstable.
Parental criminality, family economic adversity, and sexual abuse were assessed using a short question series developed for the baseline NCS.10 Parental criminality was assessed using questions about whether a parent either engaged in criminal activities, such as burglary or selling stolen property, or was ever arrested for criminal activity. Economic adversity was assessed using questions about whether the family received welfare or other government assistance and whether the family often lacked enough money to pay for the basic necessities of living. Sexual abuse was assessed using questions about repeated fondling, attempted rape, and rape. Parental mental illness (major depression, generalized anxiety disorder, and prior to panic disorder) and substance abuse were assessed using the Family History Research Diagnostic Criteria interview27 and its extensions.28 Family violence and physical abuse of the respondent by parents were assessed using a modified version of the Conflict Tactics Scales.29 Neglect was assessed using questions used in studies of child welfare about the frequency of not having adequate food, clothing, or medical care; having inadequate supervision; and having to do age-inappropriate chores.30 Life-threatening physical illness was assessed using a standard chronic conditions checklist.31
Tetrachoric factor analysis (promax rotation) was used to examine intercorrelations among CAs. Associations of CAs with lifetime disorders were estimated using discrete-time survival analysis, with person-years as the unit of analysis,32 controlling for respondent age at interview, sex, race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, and other), and other DSM-IV/CIDI disorders with onset before the age at onset of the disorder under investigation and before age 18 years. The controls for early-onset disorders were included to adjust for the associations of CAs with temporally secondary disorders through earlier-onset disorders that affected secondary disorders. Person-years began at age 4 years, the youngest age evaluated for possible disorder onset. Person-years were coded “0” on the dependent variables until the age at onset and “1” at the year of onset and were censored after the year of onset. Several multivariate models were estimated, with each including dummy predictor variables for CAs plus controls. The first model was additive; that is, it included a separate predictor variable for each of the 12 CAs without interaction terms. The second multivariate model included predictor variables for number of CAs without variables for types of CAs experienced. A third model included 12 predictors for type of CA and additional predictors for number of CAs, with the latter starting at exactly 2 rather than 1 because the variable for exactly 1 CA was perfectly predicted by the 12 dummy variables for the individual CAs. A variant of this third model distinguished between 2 types of CAs as described in the “Associations of CAs With the First Onset of DSM-IV/CIDI Disorders” subsection. Another variant included interactions between types of CAs and number of CAs. Finally, we considered more complex, inherently nonlinear models, but these did not improve on the fit of the simpler models and are consequently not discussed herein.
The Akaike information criterion33 was used to select the best multivariate model for the overall data array (ie, the consolidated data file that stacked the 20 separate disorder-specific person-year files and included 19 dummy predictor variables to distinguish among these files, thereby forcing the estimated slopes of disorders on CAs to be constant across disorders). This best-fitting model was then estimated again in subsamples defined by disorder, class of disorder (mood, anxiety, disruptive behavior, and substance use disorders), life course stage, and the conjunction of life course stage with class of disorder. Survival coefficients and their standard errors were exponentiated and are reported as odds ratios (ORs) and 95% confidence intervals, respectively.
The population-attributable risk proportion (PARP) of the outcomes was computed for the best-fitting model. The PARP is the proportion of observed outcomes that would have been prevented in the absence of CAs if the ORs were due to causal effects of CAs.34 In the more realistic case in which the associations of CAs with outcomes are partly due to common causes, the PARP reflects overall associations. All of the PARPs were calculated using simulation methods to generate individual-level predicted probabilities of the outcome disorders from the coefficients in the best-fitting model with and without coefficients for CAs. The PARP is 1 minus the ratio of the predicted prevalence estimates in the 2 specifications. The PARP for a pooled data set is the average PARP across all disorders included in the calculation based on a constant model across disorders.
All statistical significance tests were evaluated using 2-sided tests (P < .05). Because the NCS-R data are clustered and weighted, the design-based Taylor series method35 implemented in the SUDAAN software system36 was used to estimate standard errors of ORs.
PREVALENCE AND CO-OCCURRENCE OF CAs
Approximately 53.4% of NCS-R respondents reported having at least 1 CA (Table 1). The most common CAs were parental divorce (17.5%), family violence (14.0%), family economic adversity (10.6%), and parental mental illness (10.3%). Multiple CAs were the norm in respondents with each CA, from 51.2% in those with death of a parent to 95.1% in those with parental neglect; there were a mean of 3.2 CAs in respondents with more than 1 CA.
Most tetrachoric correlations between pairs of CAs (94%) are positive. (Detailed results are available on request from the corresponding author.) Negative values are small (range, −0.09 to −0.01). Positive values have a median of 0.11 and an interquartile range (25th-75th percentiles) of 0.04 to 0.19. Factor analysis found 3 meaningful factors (Table 1). Most CAs have significant loadings on the first factor of maladaptive family functioning (MFF) (eg, parental substance abuse, criminality, domestic violence, and abuse and neglect), with factor loadings of 0.32 to 0.67. The second factor represents parental death and other loss with associated economic adversity (factor loadings, 0.50-0.67). The third factor represents parental divorce with associated economic adversity (factor loadings, 0.48-0.83). The CAs in factor 1 are referred to herein as MFF CAs and the remaining CAs as other CAs.
ASSOCIATIONS OF CAs WITH THE FIRST ONSET OF DSM-IV/CIDI DISORDERS
In the bivariate models (ie, only 1 CA considered at a time) of the pooled associations of CAs with the first onset of the 20 DSM-IV/CIDI disorders, all but 1 CA (parental death) was significant, with ORs of 1.5 to 1.9 for MFF CAs and 1.0 to 1.5 for other CAs (Table 2). The ORs are generally smaller in the additive multivariate model, with 8 CAs significant and ORs of 1.0 to 1.4 for MFF CAs and 1.0 to 1.3 for other CAs. The χ2 test for associations of all CAs is significant (χ212 = 884.5, P < .001), although the ORs are substantively modest. We can reject the hypothesis that the ORs are the same for all CAs (χ211 = 286.6, P < .001).
The multivariate model that considers only number rather than type of CAs shows generally increasing ORs with number of CAs, from 1.3 for 1 CA (compared with respondents who had no CAs) to highs of 3.4 for 6 CAs and 3.2 for 7 or more CAs. The χ2 test for the joint associations is significant (χ27 = 822.0, P < .001). The model that includes measures of types of CAs and number of CAs fits the data better than the previous models in terms of Akaike information criterion, as indicated by the types-of-CA measures being significant after controlling for number of CAs (χ212 = 86.9, P < .001) and the number of CAs measures being significant after controlling for types (χ26 = 63.7, P < .001). (Detailed results of model fitting are available on request from the author.) The hypothesis that the ORs are the same for all types of CAs can be rejected (χ211 = 60.0, P < .001). The MFF CAs consistently have higher ORs than do other CAs. The ORs associated with types are mostly higher than in the additive model, indicating that the additivity assumption led to a downward bias in the estimated associations of individual CAs with the outcome. The reason for this is that the ORs associated with number of CAs in the more complex model are for the most part less than 1.0 and become increasingly smaller as number of CAs increases. This means that although the odds of disorder onset increase with increasing number of CAs, they increase at a significantly decreasing rate.
We also evaluated more complex models but found that they generally did not fit as well as the model with types and number of CAs. One refinement did improve fit, although by distinguishing number of MFF CAs from number of other CAs. The significant subadditive interactions associated with number of CAs are found for MFF CAs (χ26 = 61.7, P < .001) but not for other CAs (χ23 = 5.2, P = .16). (The test had only 3 df because no respondents had all 5 MFF CAs.) This was the model used in subsequent disaggregated analyses.
DIFFERENTIAL ASSOCIATIONS BY CLASS OF DSM-IV/CIDI DISORDER
Disaggregation shows that CAs are significantly associated with the first onset of each class of disorders (mood, anxiety, disruptive behavior, and substance use). The ORs associated with types of CAs are always associated with increased odds (χ212 = 44.5-193.7, P < .001). Those for MFF CAs are more consistently significant (χ27 = 38.2-115.2, P < .001) than are those for other CAs (χ25 = 7.4-57.5, P = .19 to <.001) (Table 3). The ORs associated with number of CAs are always associated with decreased odds, although they are largely confined to MFF CAs (χ26 = 19.4-50.9, P = <.001 to .004).
Close inspection finds what seems to be meaningful variation in the ORs associated with some MFF CAs, such as parental criminality consistently having its lowest OR and parental substance abuse its highest OR predicting respondent substance use disorders. The more striking pattern, though, is that each MFF CA is significantly associated with each disorder class with rather consistent ORs. The ORs of other CAs are less consistent, with only 25% significant at P < .05. Again, there seems to be some meaningful variation, most notably family economic adversity and respondent physical illness associated with anxiety but not mood disorders, but these differences are not statistically significant.
Differential associations by life course stage and number of previous disorders
Disaggregation by life course stage (childhood: aged 4-12 years, adolescence: aged 13-19 years, early adulthood: aged 20-29 years, and middle-later adulthood: aged ≥30 years) shows that the significant ORs of some, but not all, CAs persist throughout the life course (Table 4). The ORs associated with other CAs decline with age, but these declines are generally not statistically significant. The exceptions are significant declines with age in ORs for parental death (χ23 = 8.1, P = .04), physical abuse (χ23 = 22.9, P < .001), sexual abuse (χ23 = 40.3, P < .001), and physical illness (χ23 = 13.7, P = .003). The persistence of the OR for other parental loss throughout the life course is striking compared with the OR for parental death being significant only in childhood. More highly disaggregated analyses showed that age-related declines involving sexual abuse were consistent across all disorder classes (although significant only for mood disorders), whereas declines associated with physical abuse, parental death, and physical illness varied by class of disorder. (Detailed results are available on request.)
We also examined differential associations of CAs with the first onset of DSM-IV/CIDI disorders as a function of the number of previous lifetime disorders. (Detailed results are available on request from the author.) We found that the ORs associated with most CAs become smaller as the number of previous disorders becomes larger. This means that CAs are more strongly associated with the onset of temporally primary vs secondary disorders. The sign pattern of the associations between types of CAs and onset of disorders remains largely positive (ie, ORs >1.0) when number of previous disorders is 0 (11 of 12 ORs >1.0, 9 of 12 significant at P < .05), 1 (7 of 12 ORs >1.0, 0 of 12 significant at P < .05), or 2 or more (7 of 12 ORs >1.0, 6 of 12 significant at P < .05), but the magnitude of ORs is considerably stronger when number of previous disorders is 0, with median (interquartile range) values of the ORs being higher when number of previous disorders is 0 (1.6 [1.2-1.7]) rather than either 1 (1.2 [1.1-1.2]) or 2 or more (1.2 [1.1-1.3]).
POPULATION-LEVEL ASSOCIATIONS OF CAs WITH DISORDER ONSET
We calculated the PARPs associated with CAs based on the best-fitting model. Results show that CAs explain (in a predictive sense) 32.4% of all disorders, 41.2% of disruptive behavior disorders, 32.4% of anxiety disorders, 26.2% of mood disorders, and 21.0% of substance use disorders (Table 5). The CAs explain a higher proportion of childhood-onset disorders (44.6%) than adolescent-onset disorders (32.0%) and adult-onset disorders (28.6% and 25.9%). This decline is largely explained by the PARPs for mood disorders decreasing with age from a high of 57.1% for childhood-onset cases to a low of 20.5% for onsets in the age range of 30 years or older. The PARPs also decrease with age for anxiety disorders, but less dramatically than for mood disorders (from 39.5% of childhood-onset cases to 29.8% of onsets in the age range of ≥30 years). The PARPs do not decrease with age, in comparison, for substance use disorders. The number of disruptive behavior disorders that occur for the first time in adulthood is so small that we could not calculate the PARPs for these disorders beyond adolescence.
Effects of time to recall
The use of retrospective data introduces the possibility of recall bias. We investigated this possibility by examining age differences in the reported prevalence of CAs and in the ORs of CAs with disorder onset. (Detailed results are available on request from the author.) Reported death of a parent when the respondent was a child was positively related to age, whereas parental divorce when the respondent was a child was inversely related to age. These patterns are consistent with historical trends. Respondent age was unrelated, in comparison, to reports of other parental loss, neglect, or life-threatening childhood physical illness. Respondent age of 65 years and older was significantly related to low reports of parental mental illness, substance abuse, criminality, family violence, physical abuse, and sexual abuse, whereas age was generally unrelated to these CAs in the age range 18 to 64 years. These patterns could be due to a genuinely low prevalence of some CAs in older respondents, underrepresentation of elderly people with these CAs in the sample (due to early death or differential participation), or underreporting of these CAs in elderly respondents (due to differential recall or differential willingness to report). Although we have no way to know which of these processes are at work, any bias in prevalence estimates is likely conservative in the total sample because of lower reporting in the elderly respondents.
Analysis of age differences in associations at given life course stages found generally good consistency between ORs estimated in the youngest cohorts only (aged 18-29 years at interview) and in the entire sample. Of the 48 coefficients for individual CAs (12 CAs associated with disorder onsets in the person-year ranges of 4-12, 13-19, 20-29, and ≥30), 36 were positive and 21 were significant in the youngest cohorts, compared with 41 positive and 31 significant in the total sample. Median (interquartile range) ORs were also similar in the youngest cohorts (1.3 [1.1-1.5]) and the total sample (1.4 [1.2-1.7]). In 8 of 48 cases, the ORs differed significantly for the youngest vs the older cohorts. The OR was significant but was lower in magnitude in the younger (1.2-1.8) vs the older (1.4-1.4) cohorts in 3 of these cases. The OR changed from greater than 1.0 (1.1-1.1) in the older cohorts to less than 1.0 (0.7-0.9) in the younger cohorts in 2 other cases but was nonsignificant in both. The OR was nonsignificant in the youngest cohorts (0.8-1.1) but was significant in older cohorts (1.4-1.7) in the other 3 cases, which involved associations of childhood sexual abuse with disorder onsets in the age ranges of 20 to 29 years and 30 years and older and of parental substance abuse with disorder onsets in the age range of 30 years and older. These findings are not definitive because recall failure could exist even for respondents with the shortest recall intervals, but they nonetheless show that the results are stable across a range of recall times.
Despite these results, this study is limited by the retrospective nature of the data. Methodological research suggests that recall bias can lead to underreporting of CAs,38 which would be expected to make the estimates of PARPs conservative. However, bias could be anticonservative in estimating ORs if the same respondents who did not report CAs also underreported disorders. A long-term prospective study is needed to resolve these uncertainties. Several such studies39-42 exist that could be used to evaluate these results, but these studies generally have nontrivial attrition. If this attrition is systematic (ie, respondents with the highest risk of disorders also have the highest attrition), estimates of CA effects could be biased downward. The best way to guard against this possibility is to think of retrospective and prospective studies as bounding the true values of associations (ie, retrospective studies giving upper bound estimates and prospective studies lower bound estimates).
A second study limitation is that the list of CAs, although larger than that in most previous studies, is not exhaustive. We also did not consider the timing, sequencing, persistence, recurrence, or severity of individual CAs. In some cases, such as parental mental illness, there could be complex associations remaining to be discovered that involve the number of ill parents, the number of illnesses, and the persistence and severity of these illnesses. A related limitation is that the analysis of joint CA effects did not include fine-grained evaluation of interactions but focused only on broad interaction patterns. This broad-gauged approach is probably desirable as a first approximation but inevitably misses important subtleties. For example, some research43-45 suggests that parental divorce is associated with a reduced risk of subsequent psychiatric disorders if it facilitates escape from exposure to maladaptive parenting. Future analyses need to examine such specifications against the backdrop of the broader preliminary patterns found in the present study.
In the context of these limitations, the present results are consistent with those of previous studies in suggesting that most US children are exposed to childhood family adversities that are often clustered.7,9 Neglect, in particular, almost always appears with other CAs. Even the CAs most likely to be independent co-occur with at least one other CA in most cases. Because of this high co-occurrence, it is critical for future research not to focus on one CA without considering others, because bivariate analyses artificially inflate estimates of individual CA effects.13 There are implications as well for more subtle analyses. For example, some previous research11 suggested that childhood neglect exacerbates the predictive effects of other CAs, but the present results raise the possibility that this finding is due to neglect being associated with an especially large number of other uncontrolled CAs rather than itself creating a high risk of psychiatric disorders.
The present finding that the multivariate structure of the associations between CAs and disorder onset is broadly subadditive has, to our knowledge, never before been examined. This subadditive pattern has important implications for intervention because it means that prevention or amelioration of only a single CA in youths exposed to many CAs is unlikely to have important preventive effects. The finding that this nonadditivity is confined to MFF CAs is reminiscent of the finding in the child maltreatment literature that the most severe CAs tend to be chronic intrafamilial adversities involving the use of physical force.46 This finding also reinforces the importance of considering CA persistence and severity in future research because the finding that people exposed to many co-occurring MFF CAs have a very high risk of lifetime disorders might be due at least partly to the effects of unmeasured CA persistence and severity.
Despite considerable early theorizing to suggest unique effects of particular CAs on particular mental disorders, such as of childhood parental death on adult depression,47 we found remarkably little specificity of this sort in the NCS-R data. Most CAs we studied, especially MFF CAs, were associated with all the disorder classes we considered. This pattern was found even in the models that controlled for number of CAs, in which ORs associated with specific CA types can be interpreted under the model as the associations of pure CAs (ie, having a particular 1 and only 1 CA vs none) with disorder onset, thereby removing the confounding effects of CA co-occurrences. We also controlled for comorbid child-adolescent disorders to increase the ability to detect specificities of this sort. Previous studies48,49 found some evidence of specificity in predicting prevalent cases, but inspection of coefficients in the best-fitting models at the level of disorder class and the level of individual disorders (the latter results are available on request) yielded little evidence of specificity. The obvious implication is that the causal pathways that link CAs to the onset of psychiatric disorders are quite general.
In considering the theme of causal pathways, note that these results do not confirm that CAs have causal effects. An alternative possibility is that unmeasured third variables caused CAs and subsequent mental disorders. Genetic factors are possible confounding variables of this type. This is most obviously true for parental mental illness, which can predict respondent mental illness through genetic pathways unrelated to CAs, but the same might be true for other CAs to the extent that they are indicators of genetic liability. Gene × environment interactions could also be involved to the extent that the people exposed to CAs have an elevated genetic risk of psychiatric disorders and are exposed to stressful experiences related to their CAs that potentiate this genetic liability. Genetically informative designs (eg, twin-family and adoption studies) are needed to evaluate these possibilities rigorously.
Another class of potentially important third variables is respondent behaviors and behavioral predispositions that elicit some CAs, such as abuse, and cause the subsequent onset of respondent mental disorders. Prospective studies that measure these proposed constructs repeatedly would be in the best position to evaluate this possibility. In the ideal case, such studies would have multiple informants to assess reporting bias.
A final noteworthy finding is that the associations of many, but not all, CAs with first onset of DSM-IV disorders persist into adulthood. Future research needs to investigate the causal pathways responsible for this specification. Although previous research15,50 has documented long-term associations of some CAs with adult disorders, these studies almost entirely focused on prevalent cases rather than on first onsets. It is much more striking to document, as we did herein, that CAs continue to be related to first onsets of DSM-IV disorders beyond early adulthood. Indeed, the PARPs calculated herein suggest that CAs are associated with more than one-fourth of all new disorders in adulthood. Although several hypotheses could be advanced to explain this finding,50-52 nothing in these results sheds light on them. The indirect retrospective documentation of long-term multivariate associations is nonetheless important in providing an empirical justification for conducting further analyses to explore such hypotheses to investigate mediators, developmental sequences, and dynamic relationships between CAs and adult-onset disorders.
Future research also needs to distinguish between associations of CAs with disorder onset and disorder persistence. As reported in a companion article,20 we found a rather different association of CAs with disorder persistence than reported herein with disorder onset. In addition, future research should integrate the kind of broad-based analyses of joint effects presented herein with more focused investigations of specific adversities53,54 and important adversity clusters.2,55 Future studies should also examine the moderating effects of early disorders on the associations of CAs with later disorders,56 a line of study that could be important in focusing clinical attention on preventing the onset of secondary disorders. Finally, future studies should try to identify risk and protective factors in adulthood (eg, personality, social support, and adult stressors) that mediate or modify the relationships of CAs with adult disorders.
Correspondence: Ronald C. Kessler, PhD, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02115 (email@example.com).
Submitted for Publication: February 27, 2009; final revision received August 6, 2009; accepted August 9, 2009.
Financial Disclosure: Dr Kessler has been a consultant for GlaxoSmithKline Inc, Kaiser Permanente, Pfizer Inc, Sanofi-Aventis, Shire Pharmaceuticals, and Wyeth-Ayerst; has served on advisory boards for Eli Lilly & Co and Wyeth-Ayerst; and has had research support for his epidemiological studies from Bristol-Myers Squibb, Eli Lilly & Co, GlaxoSmithKline, Johnson & Johnson Pharmaceuticals, Ortho-McNeil Pharmaceutical Inc, Pfizer Inc, and Sanofi-Aventis.
Funding/Support: The NCS-R is supported by grant U01-MH60220 from the National Institute of Mental Health (NIMH) with supplemental support from the National Institute on Drug Abuse (NIDA), the Substance Abuse and Mental Health Services Administration, grant 044780 from the Robert Wood Johnson Foundation, and the John W. Alden Trust. These activities were supported by grant R01 MH070884 from the NIMH; the John D. and Catherine T. MacArthur Foundation; the Pfizer Foundation; grants R13-MH066849, R01-MH069864, and R01 DA016558 from the US Public Health Service; Fogarty International Research Collaboration Award R03-TW006481 from the Fogarty International Center; the Pan American Health Organization; Eli Lilly & Co; Ortho-McNeil Pharmaceutical Inc; GlaxoSmithKline; and Bristol-Myers Squibb.
Role of the Sponsors: The funding organizations had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.
Collaborating NCS-R Investigators: Ronald C. Kessler, PhD (principal investigator, Harvard Medical School); Kathleen Merikangas, PhD (co-principal investigator, NIMH); James Anthony, PhD (Michigan State University); William Eaton, PhD (The Johns Hopkins University); Meyer Glantz, PhD (NIDA); Doreen Koretz, MD (Harvard University); Jane McLeod, PhD (Indiana University); Mark Olfson, MD, MPH (New York State Psychiatric Institute, College of Physicians and Surgeons of Columbia University); Harold Pincus, MD (University of Pittsburgh); Greg Simon, MD, and Michael Von Korff, ScD (Group Health Cooperative); Philip S. Wang, MD, DrPH (NIMH); Kenneth Wells, MD, MPH (UCLA); Elaine Wethington, PhD (Cornell University); and Hans-Ulrich Wittchen, PhD (Max Planck Institute of Psychiatry, Technical University of Dresden).
Disclaimer: The views and opinions expressed herein are those of the authors and should not be construed to represent the views of any of the sponsoring organizations, agencies, or the US government. A complete list of NCS publications and the full text of all NCS-R instruments can be found at http://www.hcp.med.harvard.edu/ncs. Send correspondence to firstname.lastname@example.org.
Additional Information: A complete list of World Mental Health (WMH) publications can be found at http://www.hcp.med.harvard.edu/wmh/.
Additional Contributions: The NCS-R is conducted in conjunction with the World Health Organization WMH Survey Initiative. We thank the staff of the World Mental Health (WMH) Data Collection and Data Analysis Coordination Centres for assistance with instrumentation, fieldwork, and consultation on data analysis.
MP Multiple forms of child abuse and neglect: adult retrospective reports. Aggress Violent Behav
547- 578Google ScholarCrossref
GP The long-term impact of the physical, emotional, and sexual abuse of children: a community study. Child Abuse Negl
7- 21PubMedGoogle ScholarCrossref
EB Psychosocial functioning in children after the death of a parent. Am J Psychiatry
511- 513PubMedGoogle Scholar
A Emotional neglect and family structure: impact on student functioning. Child Abuse Negl
1033- 1043PubMedGoogle ScholarCrossref
TO Life events, vulnerability and onset of depression: some refinements. Br J Psychiatry
1987;15030- 42PubMedGoogle ScholarCrossref
DJ Childhood adversity, recent life events and depression in late life. J Affect Disord
243- 246PubMedGoogle ScholarCrossref
WH The interrelatedness of multiple forms of childhood abuse, neglect, and household dysfunction. Child Abuse Negl
771- 784PubMedGoogle ScholarCrossref
HA Poly-victimization: a neglected component in child victimization. Child Abuse Negl
7- 26PubMedGoogle ScholarCrossref
KS Childhood adversity and adult psychiatric disorder in the US National Comorbidity Survey. Psychol Med
1101- 1119PubMedGoogle ScholarCrossref
N Differential correlates of multi-type maltreatment among urban youth. Child Abuse Negl
393- 415PubMedGoogle ScholarCrossref
B Resilience to adult psychopathology following childhood maltreatment: evidence from a community sample. Child Abuse Negl
211- 229PubMedGoogle ScholarCrossref
WH Childhood residential mobility and multiple health risks during adolescence and adulthood: the hidden role of adverse childhood experiences. Arch Pediatr Adolesc Med
1104- 1110PubMedGoogle ScholarCrossref
WH Childhood abuse, household dysfunction, and the risk of attempted suicide throughout the life span: findings from the Adverse Childhood Experiences Study. JAMA
3089- 3096PubMedGoogle ScholarCrossref
RF Relationship between multiple forms of childhood maltreatment and adult mental health in community respondents: results from the adverse childhood experiences study. Am J Psychiatry
1453- 1460PubMedGoogle ScholarCrossref
S The impact of cumulative childhood adversity on young adult mental health: measures, models, and interpretations. Soc Sci Med
1140- 1151PubMedGoogle ScholarCrossref
KS Lifetime and 12-month prevalence of DSM-III-R
psychiatric disorders in the United States: results from the National Comorbidity Survey. Arch Gen Psychiatry
8- 19PubMedGoogle ScholarCrossref
KR The National Comorbidity Survey Replication (NCS-R): background and aims. Int J Methods Psychiatr Res
60- 68PubMedGoogle ScholarCrossref
RC Childhood adversities and adult psychiatric disorders in the National Comorbidity Survey Replication II: associations with persistence of DSM-IV
disorders. Arch Gen Psychiatry
124- 132Google ScholarCrossref
H The US National Comorbidity Survey Replication (NCS-R): design and field procedures. Int J Methods Psychiatr Res
69- 92PubMedGoogle ScholarCrossref
TB The World Mental Health (WMH) Survey Initiative Version of the World Health Organization (WHO) Composite International Diagnostic Interview (CIDI). Int J Methods Psychiatr Res
93- 121PubMedGoogle ScholarCrossref
H Clinical calibration of DSM-IV
diagnoses in the World Mental Health (WMH) version of the World Health Organization (WHO) Composite International Diagnostic Interview (WMHCIDI). Int J Methods Psychiatr Res
122- 139PubMedGoogle ScholarCrossref
JBW Structured Clinical Interview for DSM-IV Axis I Disorders, Research Version, Non-patient Edition (SCID-I/NP). New York: Biometrics Research, New York State Psychiatric Institute; 2002
M Recall of psychiatric history in cross-sectional surveys: implications for epidemiologic research. Epidemiol Rev
221- 227PubMedGoogle Scholar
RC Improving accuracy of major depression age of onset reports in the US National Comorbidity Survey. Int J Methods Psychiatr Res
39- 48Google ScholarCrossref
RL Family History Research Diagnostic Criteria. New York: Biometrics Research, New York State Psychiatric Institute; 1978
LJ The family history method: whose psychiatric history is measured? Am J Psychiatry
1501- 1504PubMedGoogle Scholar
MA Measuring intrafamily conflict and violence: the Conflict Tactics (CT) Scales. J Marriage Fam
75- 88Google ScholarCrossref
A Foster Youth Transitions to Adulthood: A Longitudinal View of Youth Leaving Care. Madison, WI: Institute for Research on Poverty; 1998
RC The impact of comorbidity of mental and physical conditions on role disability in the US adult household population. Arch Gen Psychiatry
1180- 1188PubMedGoogle ScholarCrossref
JD Investigating onset, cessation, relapse, and recovery: why you should, and how you can, use discrete-time survival analysis to examine event occurrence. J Consult Clin Psychol
952- 965PubMedGoogle ScholarCrossref
DR Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. 2nd New York, NY: Springer-Verlag; 2002
S Modern Epidemiology. 2nd Philadelphia, PA: Lippincott Williams & Wilkins; 1998
KM Introduction to Variance Estimation. New York, NY: Springer-Verlag; 1985
SUDAAN Professional Software for Survey Data Analysis [computer program]. Version 8.0.1. Research Triangle Park, NC: Research Triangle Institute; 2002
JJ Case-Control Studies: Design, Conduct, and Analysis. New York, NY: Oxford University Press; 1980
M Validity of adult retrospective reports of adverse childhood experiences: review of the evidence. J Child Psychol Psychiatry
260- 273PubMedGoogle ScholarCrossref
E Child abuse and neglect and the development of mental disorders in the general population. Dev Psychopathol
981- 999PubMedGoogle Scholar
LJ The Christchurch Health and Development Study: review of findings on child and adolescent mental health. Aust N Z J Psychiatry
287- 296PubMedGoogle ScholarCrossref
GJ Adverse events in childhood and chronic widespread pain in adult life: results from the 1958 British Birth Cohort Study. Pain
92- 96PubMedGoogle ScholarCrossref
A Why do children from socioeconomically disadvantaged families suffer from poor health when they reach adulthood? a life-course study. Am J Epidemiol
966- 974PubMedGoogle ScholarCrossref
PR Children's adjustment to divorce: theories, hypotheses, and empirical support. J Marriage Fam
23- 38Google ScholarCrossref
M The adjustment of children with divorced parents: a risk and resiliency perspective. J Child Psychol Psychiatry
129- 140PubMedGoogle ScholarCrossref
A Life with (or without) father: the benefits of living with two biological parents depend on the father's antisocial behavior. Child Dev
109- 126PubMedGoogle ScholarCrossref
TL Unique and combined contributions of multiple child abuse types and abuse severity to adult trauma symptomatology. Child Maltreat
172- 181PubMedGoogle ScholarCrossref
J Parental death in childhood and risk of adult depressive disorders: a review. Psychol Med
289- 299PubMedGoogle ScholarCrossref
S Stress and psychopathology in children and adolescents: is there evidence of specificity? J Child Psychol Psychiatry
107- 133PubMedGoogle ScholarCrossref
A Specificity of putative psychosocial risk factors for psychiatric disorders in children and adolescents. J Child Psychol Psychiatry
34- 42PubMedGoogle ScholarCrossref
HR The impact of childhood abuse and neglect on adult mental health: a prospective study. J Health Soc Behav
184- 201PubMedGoogle ScholarCrossref
J Early childhood adversity and adolescent depression: the mediating role of continued stress. Psychol Med
581- 589PubMedGoogle ScholarCrossref
MJ Direct and indirect effects of childhood adversity on depressive symptoms in young adults. J Youth Adolesc
89- 103Google ScholarCrossref
MK Traumatic events and suicide-related outcomes among Mexico City adolescents. J Child Psychol Psychiatry
654- 666PubMedGoogle ScholarCrossref
RC Child sexual abuse and subsequent psychopathology: results from the National Comorbidity Survey. Am J Public Health
753- 760PubMedGoogle ScholarCrossref
R The effect of lifetime victimization on the mental health of children and adolescents. Soc Sci Med
13- 27PubMedGoogle ScholarCrossref
W Stress sensitization and adolescent depressive severity as a function of childhood adversity: a link to anxiety disorders. J Abnorm Child Psychol
287- 299PubMedGoogle ScholarCrossref