Identifying Factors That Predict Longitudinal Outcomes of Untreated Common Mental Disorders

In order to guide policy on allocation of appropriate mental health services, there has been a worldwide effort to estimate mental health services need. Meeting criteria for a common mental disorder has been used as a proxy for the need for mental health services. However, most worldwide surveys have found high rates of common mental disorders in conjunction with low mental health service use (1–4). Among people with a mental disorder, the most important barrier to service use has been the lack of perceived need for help (1,5–7).

In the context of these findings, there has been increasing concern that a mental disorder diagnosis by itself is not a clear indicator of need for services. Our group recently published the first study examining the three-year outcomes of untreated common mental disorders (8). Using a longitudinal survey, we found that approximately 60% of people with an untreated mental disorder at baseline had a remission of their disorder without any subsequent treatment (and with no development of a comorbid disorder or any suicide attempt). These findings indicate that a substantial proportion of people who meet diagnostic criteria for a common mental disorder recover without needing formal treatment. On the other hand, approximately 40% had persistence of the same disorder, developed a comorbid disorder, or attempted suicide, and therefore they may have benefited from outreach interventions.

The aim of this study was to extend the literature by examining, with a longitudinal community-based survey, the risk factors for poor outcomes among individuals with untreated common mental disorders. Although previous research has examined predictors of persistence and recurrence of mental disorders in special populations (such as older adults or individuals with major depression) (9,10), no studies have considered a broad range of untreated mental disorders in the adult population (age ≥18) in a longitudinal sample. Such work could help identify people at risk of poor outcomes and inform the development of targeted interventions. Considering previous work showing that various sociodemographic factors (female sex, higher education, and middle age), childhood maltreatment, comorbid mental disorders, and comorbid general medical conditions are associated with higher likelihood of mental health service use and persistence (or recurrence) of mental disorders (6,9–14), we hypothesized that in the absence of mental health treatment, these factors would be associated with poor longitudinal outcomes.

Methods

Survey

Data were from the National Epidemiologic Survey of Alcohol and Related Conditions (NESARC), a longitudinal, nationally representative survey of the U.S. adult population (age ≥18) (15,16). In wave 1, after obtaining informed consent from participants, trained lay interviewers conducted face-to-face interviews with 43,093 noninstitutionalized respondents between 2001 and 2002 (15,17). Wave 2 followed up with 34,653 wave 1 respondents between 2004 and 2005, with a cumulative response rate of 70.2% for the two waves (16,18,19). The U.S. Census Bureau and U.S. Office of Management and Budget reviewed and provided ethical approval for the NESARC (17).

Measures

Axis I mental disorders.

The Alcohol Use Disorder and Associated Disabilities Interview Schedule—DSM-IV Version (AUDADIS-IV) (20) derived past-year axis I mental disorders according to DSM-IV (21) criteria. Axis I disorders include any mood disorder (major depression, dysthymia, and bipolar disorder), any anxiety disorder (social phobia, panic disorder, specific phobia, agoraphobia, and generalized anxiety disorder), and any substance use disorder (alcohol use disorders and illicit drug use disorders).

Respondents with a psychotic illness or bipolar disorder were excluded for several reasons. First, psychosis is difficult to measure in epidemiologic samples (22); in this study psychotic illness was measured through self-report, rather than with a diagnostic algorithm, as was done with all other mental disorders. Second, the debate around allocation of mental health services has been limited to nonpsychotic illness; few critics would argue that psychotic individuals are not in need of treatment. Finally, bipolar disorder was excluded due to difficulty defining remission; according to DSM-IV, once a person has had an episode of mania (or a major depressive episode and a hypomanic episode) he or she has a diagnosis of bipolar disorder. As a result, our analysis examined (nonbipolar and nonpsychotic) depressive, anxiety, and substance use disorders.

Mental health service use.

For each depressive and anxiety disorder at wave 1, four questions examined whether a respondent ever sought treatment from a counselor, therapist, physician, psychologist, or similar mental health worker; went to an emergency room; stayed in a hospital overnight; or was prescribed medication for a mood disorder or for an anxiety disorder. In addition, respondents were asked at wave 1 whether they had ever sought treatment from a physician, counselor, Alcoholics or Narcotics Anonymous, or any other community agency or professional for a substance use problem. At wave 2 similar questions were posed to respondents, except the phrasing of each question regarding timing of service use was changed from “ever” to “since your last interview.”

Suicide attempts.

At wave 2 respondents were asked whether they had ever attempted suicide and if so, when the most recent attempt occurred.

Axis II personality disorders.

The AUDADIS-IV assessed respondents for the lifetime presence of all ten DSM-IV personality disorders: schizoid, paranoid, antisocial, dependent, histrionic, obsessive-compulsive, avoidant, borderline, narcissistic, and schizotypal.

Sociodemographic factors.

We included several sociodemographic factors, as assessed at wave 1, in our analyses: age, household income (<$20,000, $20,000–$34,999, $35,000–$59,999, and ≥$60,000), sex, marital status (married or common-law status; never married; or widowed, divorced, or separated), education (less than high school, high school, or some college or more), and race-ethnicity (white, black, Hispanic, or other).

Childhood maltreatment.

Questions regarding childhood maltreatment were adapted from the Conflict Tactics Scale (23,24) and the Childhood Trauma Questionnaire (25). Participants were classified as having experienced childhood maltreatment if they experienced childhood physical, sexual, or emotional abuse; physical neglect; or emotional neglect. Our definitions of these types of abuse and neglect were consistent with those used in the Adverse Childhood Experiences Study (26,27) and identical to those used in previous research with the NESARC (28,29).

General medical conditions.

Respondents answered questions about the presence of the following general medical conditions diagnosed by a physician: arteriosclerosis, hypertension, liver disease, chest pain or angina pectoris, tachycardia, heart attack, other heart disease, stomach ulcer, gastritis, and arthritis. A count of the number of past-year general medical conditions at wave 1 was created on the basis of these responses.

Health-related quality of life.

The valid and reliable 12-Item Medical Outcomes Study–Short Form (SF-12) (30) measured respondents’ past-month health-related quality of life. The SF-12 produces mental component summary (MCS) and physical component summary (PCS) scores ranging from 0 to 100. These scales are standardized to a mean of 50, with higher scores indicating better quality of life.

Main Outcomes

We determined the outcomes of individuals who had past-year depressive, anxiety, and substance use disorders at baseline who had not received any lifetime mental health services. Individuals who endorsed any service use during the three-year follow-up were excluded from the analysis. We then created two mutually exclusive outcomes concerning depressive, anxiety, and substance use disorder: nonremission—defined as mental disorder persistence in the past year at wave 2, presence of a comorbid disorder in the past year at wave 2, or any suicide attempt during the three-year follow-up period—or spontaneous remission—defined as neither meeting criteria for any axis I mental disorder in the past year at wave 2 nor making a suicide attempt during the follow-up. We created four separate outcome variables to represent each class of disorder (that is, any depressive disorder, any anxiety disorder, any substance use disorder, and any disorder). It is important to note that we used a hierarchy rule such that respondents experiencing remission of their baseline disorder could be categorized in the remission group only if they did not have any of the other outcomes (suicide attempt during the three-year follow-up or meeting criteria for at least one axis I comorbid disorder during the past year at wave 2).

Analytic Strategy

Weights supplied by the NESARC were used to ensure that the analyses were representative of the U.S. population according to the 2000 Decennial Census (18). To account for the complex sampling frame of the survey, Taylor-series linearization was adopted when estimating regression coefficients and standard errors with SUDAAN 10.0.1 statistical software (31,32). We examined the associations among all predictors before conducting regression analyses to check for multicollinearity.

We used cross-tabulations to describe the data. Logistic regression analyses tested the associations between the sociodemographic factors; comorbid axis I, II, or III disorders; and childhood maltreatment history and nonremission (as defined above). We first examined the unadjusted association between each predictor of nonremission and then entered all predictors into the regression model simultaneously. Unadjusted and adjusted odds ratios and 95% confidence intervals are reported. Log-likelihood ratio statistics and Cox and Snell R² statistics assessed the goodness of fit for the model.

Results

Table 1 shows the sociodemographic distribution of the nonremission group and the spontaneous remission group. Across all disorder categories, adults in the nonremission group were younger than those in the spontaneous remission group. There was a greater proportion of women with depressive disorders in the nonremission group compared with the spontaneous remission group, whereas for substance use disorders, there was a greater proportion of men in the nonremission group.

Table 2 shows the results of the unadjusted logistic regressions examining the association between risk factors and persistence, comorbidity, and suicide attempt in untreated baseline depressive, anxiety, and substance use disorders. Meeting criteria for any personality disorder was associated with an increased likelihood of nonremission for all classes of disorders. Two factors—having two or more axis I mental disorders at baseline and childhood maltreatment—were also associated with an increased likelihood of nonremission for all disorders. Age was inversely associated with nonremission for anxiety, substance use disorders, and any disorder. Compared with men, women with a substance use disorder or any disorder were significantly less likely to have nonremission. Compared with individuals who were married or cohabiting, those who were never married had increased odds of nonremission of an anxiety disorder and for the composite any-disorder category. Number of general medical conditions was inversely associated with nonremission for the composite any-disorder category. Lower SF-12 MCS scores were associated with increased odds of nonremission for anxiety disorders, whereas higher SF-12 PCS scores were associated with increased odds of nonremission for the composite any-disorder category.

Table 3 provides the results of multivariable analyses for estimating risk factors associated with nonremission compared with spontaneous remission of baseline disorders. Two risk factors were significant across all depressive, anxiety, and substance use disorders. Participants with a baseline history of multiple axis I disorders had increased odds of nonremission for all depressive, anxiety, and substance use disorders. Childhood maltreatment also was associated with an increased likelihood of nonremission for depressive, anxiety, and substance use disorders, as well as for the any-disorder category. Axis II personality disorders were significantly associated with nonremission among those with an anxiety or substance use disorder (and also with depressive disorders, although not significantly so) and among individuals in the any-disorder category.

Other factors were associated with disorder persistence, comorbidity, and suicide attempt for specific depressive, anxiety, and substance use disorders but not for others. For example, there were sex differences in outcomes across different disorder types. Among persons with a baseline depressive disorder, women had significantly greater odds of nonremission. Being male was significantly associated with nonremission for substance use disorders and for the any-disorder category. Compared with whites, Hispanics had significantly lower odds of nonremission for substance use disorders, although this association may not be unique to substance use disorders, given that the trend was similar across other disorder categories. Among people with an anxiety disorder or a substance use disorder and those in the composite any-disorder category, older age was associated with decreased likelihood of nonremission. After adjustment for other covariates, analyses showed that number of general medical conditions at baseline and SF-12 scores were no longer associated with outcomes.

Discussion

This study is the first to examine predictors of poor outcome among individuals age 18 and older with a broad range of common mental disorders but without a history of mental health treatment, and the findings extend the literature in several important ways. First, these results suggest that comorbidity with other axis I and II disorders should be considered as an additional measure for estimating need for mental health services in a population. In this study, comorbid disorders were robustly associated with a lower likelihood of remission without treatment across depressive, anxiety, and substance use disorders. It is likely that the presence of multiple psychiatric disorders indicates more severe or more complex cases that are most in need of intervention. Mackenzie and colleagues (9) also found that disorder severity and complexity were robust correlates of persistent mood and anxiety disorders among older adults, although their study included individuals with a history of mental health treatment whereas ours focused exclusively on those without a history of treatment. Similarly, greater symptom severity and comorbid depression or anxiety have both been found to be associated with persistence of depression and anxiety disorders in longitudinal studies of individuals with and without a history of mental health treatment (33,34).

Most previous studies have not considered personality disorders in estimates of treatment need, for several reasons. First, most large cross-national studies have not assessed personality disorders (with the exception of antisocial personality disorder). Second, for most personality disorders, with the exception of borderline personality disorder, there is little evidence for effective interventions. Two large longitudinal studies have found personality disorders to be associated with persistence of major depression (35,36). When these results are taken as a whole and considered in the context of accumulating evidence of the deleterious outcomes associated with several personality disorders (37–41), evidence suggests that estimates of treatment need should also consider the presence of personality disorders.

Another important finding of this study is that childhood maltreatment was consistently associated with poor outcomes for depressive, anxiety, and substance use disorders. Our results are consistent with previous work demonstrating that childhood maltreatment is associated with the onset and persistence of mental disorders (42,43). They are also in line with a study that found that harsh physical punishment was associated with persistent depressive symptoms in a nonclinical sample (44). Childhood maltreatment has also been found to be associated with help seeking and perceived need for care, even in the absence of a lifetime disorder (12). Thus in addition to considering criteria that indicate presence of a mental disorder, we suggest that consideration also be given to persons with a history of childhood maltreatment when estimating the need for mental health services.

Moreover, our findings suggest that certain groups with untreated mental disorders (younger individuals, women with depressive disorders, and men with substance use disorders) should be specifically targeted for intervention because they may have a lower likelihood of remitting without treatment. These findings are consistent with previous work that has examined correlates of help seeking (6,11–13), as well as with studies that have examined persistence of depression and alcohol use disorders in population-based samples of individuals who have and who have not sought treatment (45–47).

Surprisingly, household income and number of general medical conditions were not associated with remission without treatment in fully adjusted models. These results are in contrast to previous work that has consistently shown that low socioeconomic status is associated with poor mental and general medical outcomes (48). General health problems are commonly comorbid with mental disorders, and we had suspected that disease burden would be associated with outcomes. One possible explanation for these null findings is that the general medical conditions contained in the NESARC are of varying severity, and specific conditions may be associated with more distress than others. Unfortunately, it was beyond the scope of this study to examine specific general medical conditions and their association with remission. However, future studies should consider this issue.

The study had several limitations. First, mental disorder diagnoses were generated by a single data source, the AUDADIS-IV diagnostic algorithms, and these results may not match those based on clinical interviews (18). Fortunately, research has indicated that there are no significant differences in diagnoses based on trained lay interviews and professional clinical interviews (49). Second, given the retrospective nature of the interviews, recall bias may have affected the recollection of when specific symptoms occurred. However, focusing on past-year disorders at both time points would likely attenuate the impact of recall bias.

Third, not all axis I disorders were assessed at both waves of the NESARC (for example, obsessive-compulsive disorder and posttraumatic stress disorder), which prevented us from examining predictors of poor outcome of all depressive, anxiety, and substance use disorders. It is unclear whether the results would also generalize to these disorders. Fourth, a recent reliability study found self-reported suicide attempts in the NESARC to be only modestly reliable (50).

Fifth, the study was based on DSM-IV diagnostic criteria. As such, the results may not be applicable to DSM-5 mental disorders. Sixth, to reduce the number of comparisons, we chose to limit our analyses to categories of mental disorder rather than investigating individual disorders. It is possible that individual depressive, anxiety, and substance use disorders have unique correlates of persistence, comorbidity, and suicide attempt.

Seventh, we were unable to examine symptom severity as a predictor of persistence, comorbidity, and suicide attempt because there are not adequate measures of mental disorder severity in the NESARC. Finally, it is unclear whether the results of this study generalize to other countries where access to services and attitudes toward mental health and treatment providers differ. Future studies should investigate whether the associations found in this study hold true in other areas of the world.

Conclusions

This study suggests that beyond looking at the presence of a depressive, anxiety, or substance use disorder, policy makers should consider other important factors, such as childhood maltreatment and comorbid axis I and axis II disorders, when estimating need for mental health services. These results warrant the development of targeted prevention and intervention strategies aimed to help individuals with multiple disorders, childhood maltreatment histories, or both, especially when we consider these results in the context of low treatment seeking among these individuals (1,3,4). Future studies should test public health outreach interventions for persons with untreated depressive, anxiety, and substance use disorders to determine whether such interventions improve outcomes and are cost-effective.