Failure and Delay in Initial Treatment Contact After First Onset of Mental Disorders in the National Comorbidity Survey Replication FREE

Research consistently shows that a high proportion of people with prevalent mental disorders in the United States are untreated despite their disorders causing substantial distress and impairment, and despite effective treatments being available.^{1 - 5} It is of considerable public health importance to uncover modifiable reasons for this lack of treatment. Successful treatment requires people suffering from the disorder to take a sequence of steps in the help-seeking process.^{6 - 7} A critical early step is establishing initial contact with a health care provider following first onset of the disorder. Relatively little is known about patterns and correlates of this early phase of the help-seeking process because most mental health services research focuses on recent treatment of current episodes rather than initial treatment of incident cases.^{8 - 14} The few studies that have examined the latter consistently document 2 important facts^{15 - 21} : that despite low current treatment of prevalent cases, the vast majority of lifetime cases eventually make treatment contact; and that it typically takes quite a number of years after first onset of the disorder for initial treatment contact to occur. The baseline National Comorbidity Survey (NCS), for example, estimated that approximately 80% of all people in the United States with a mental disorder eventually seek treatment, but that the median delay between first onset of the disorder and first treatment contact is nearly a decade.^{18 - 21} Studies of delay in initial treatment contact in other developed countries have found similar results.^{19 - 20}

In the decade since the baseline NCS was carried out, important changes have occurred in the following categories: public attitudes and awareness of mental disorders, clinical diagnostic procedures, the availability of psychotropic medications and evidence-based psychotherapies, and the organization and financing of mental health care.^{22 - 41} Unfortunately, the current extent and timing of initial treatment seeking is unknown. The current report addresses this issue by analyzing data from the recently completed National Comorbidity Survey Replication (NCS-R).⁴² We began by constructing cumulative lifetime probability of treatment contact curves to estimate the current lifetime probability of treatment contact for mental disorders and the typical duration of delay. We then examined socio-demographic predictors of failure and delay in making initial treatment contact. Because recent changes in awareness and management have not occurred uniformly across disorders, with mood disorders in particular receiving considerable attention,²⁵ we examined the extent and predictors of failure and delay separately for each of the core disorders assessed in the NCS-R.

As described in more detail elsewhere,^{43 - 44} the NCS-R is a nationally representative, multistage clustered area probability sample of English-speaking respondents aged 18 years and older in the noninstitutionalized civilian population of the 48 coterminous states. Fieldwork was carried out by the professional survey interview field staff of the Institute for Social Research at the University of Michigan, Ann Arbor, between February 2001 and April 2003. A total of 9282 face-to-face interviews were completed. All respondents were administered a part I diagnostic interview of core diagnoses. A subsample of 5692 part I respondents, consisting of all those who met lifetime criteria for a core disorder plus a probability subsample of other respondents, were also administered a part II interview that assessed correlates and disorders of secondary focus. The response rate was 70.9%. Interviewers explained the study and obtained verbal informed consent prior to beginning all interviews. The NCS-R recruitment, consent, and field procedures were approved by the Human Subjects Committees of both Harvard Medical School, Boston, and the University of Michigan.

Diagnoses of DSM-IV disorders were made using the World Health Organization’s World Mental Health Survey Initiative version of the Composite International Diagnostic Interview (WMH-CIDI),^{44 - 45} a fully structured lay-administered diagnostic interview that generates diagnoses according to the definitions and criteria of both the ICD-10⁴⁶ and DSM-IV⁴⁷ diagnostic systems. Criteria for DSM-IV disorders are used in the current report. The disorders considered in this report include (1) mood disorders, including major depressive episode (MDE), dysthymia (DYS), and bipolar disorder (BPD) I and II studied together for increased statistical power; (2) anxiety disorders, including panic disorder (PD), agoraphobia without panic (AG), specific phobia (SP), social phobia (SoP), generalized anxiety disorder (GAD), posttraumatic stress disorder (PTSD), and separation anxiety disorder (SAD); (3) substance disorders, including alcohol abuse (AA), alcohol dependence (AD), drug abuse (DA), and drug dependence (DD); and (4) impulse control disorders, including intermittent explosive disorder (IED), oppositional defiant disorder (ODD), and attention-deficit/ hyperactivity disorder (ADHD). Lifetime prevalence and age of onset were assessed separately for each disorder.⁴⁴ It is noteworthy that obsessive-compulsive disorder (OCD) and conduct disorder (CD) were also assessed in the NCS-R and are included in separate reports in substantive analyses, but are not included in the current report because the sample of cases was too small for powerful analysis of OCD. This was because OCD was a secondary disorder assessed in only a fraction of the part II subsample and because age of first treatment seeking was not assessed for CD. All diagnoses are considered with organic exclusions and without diagnostic hierarchy rules. As described by Kessler et al,⁴⁴ a blind clinical reappraisal study using the Structured Clinical Interview for DSM-IV (SCID)⁴⁸ showed generally good concordance between DSM-IV diagnoses based on the WMH-CIDI and the SCID for anxiety, mood, and substance disorders. The WMH-CIDI diagnoses of impulse control disorders have not been validated.

Near the end of each WMH-CIDI diagnostic section, respondents were asked whether they ever in their life talked to a medical doctor or other professional about the disorder under investigation. In asking this question, the interviewer clarified that the term “other professional” was meant to apply broadly to include psychologists, counselors, spiritual advisors, herbalists, acupuncturists, and any other healing professionals. Respondents who reported ever talking to any of these professionals about the disorder in question were then asked how old they were the first time they did so. The response to this question was used to define age of first treatment contact.

Predictor variables include age at onset of the focal disorder (coded into the categories 0-12, 13-19, 20-29, and 30 or more years of age), cohort (defined by age at interview in the categories 18-29, 30-44, 45-59, 60 or more years), sex, race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, and other), education (categorized as either current students or nonstudents with 0-11, 12, 13-15, or 16 or more years of education), and marital status (categorized as either currently married/cohabitating, previously married, or never married). The last 2 of these predictors vary within a given individual over time. Information was obtained in the NCS-R on timing of marital histories (ie, ages at marriage and marital dissolution), allowing marital status to be coded for each year of each respondent’s life. Information on years of education was also coded as a time-varying predictor by assuming an orderly educational history for each respondent in which 8 years of education corresponds to being a student up to age 14 years and other lengths of education are associated with ages consistent with this benchmark (eg, 12 years of education is assumed to correspond to being a student up to age 18 years).

The data were weighted to adjust for differential probabilities of selection of respondents within households and differential nonresponse as well as to adjust for residual differences between the sample and the United States population on the cross-classification of sociodemographic variables. An additional weight zwas used in the part II sample to adjust for differences in probability of selection into that sample. These procedures are described in more detail by Kessler et al.⁴³ As most disorders considered here were assessed in part I, cumulative probability of lifetime treatment contact curves were estimated using part I data and weights whenever possible. Part II data and weights were used for disorders that were assessed in part II (PTSD, ODD, CD, and ADHD). Survival analysis was used to make estimated projections of cumulative lifetime probability of treatment contact from year of onset. The actuarial method,⁴⁹ implemented in the Statistical Analysis System version 8.2 (SAS Institute, Cary, NC),⁵⁰ was used rather than the more familiar Kaplan-Meier method⁵¹ of generating survival curves because the former has an advantage over the latter in estimating onsets within a year. Separate curves were generated for each disorder. The typical duration of delay in initial treatment contact was defined as the median number of years from disorder onset to first treatment contact among cases that eventually made treatment contact based on these curves. Time to initial treatment contact in the subset of respondents with BPD was defined as the duration of time from the initial onset of a manic or hypomanic episode to first treatment of a manic or hypomanic episode. Treatment for MDE in the course of BPD was not considered because the current analyses only focused on treatment seeking for specific syndromes and not for comorbid conditions. Discrete-time survival analysis⁵² with person-year as the unit of analysis was used to examine correlates of treatment contact separately for each disorder. Predictors included both time-invariant predictors (age at onset of the disorder, cohort, sex, race/ethnicity) as well as several time-varying predictors (number of years since first onset of the disorder, education, marital status). Each model was estimated twice for a given disorder: once among all respondents with a history of the disorder to study the predictors of ever making a treatment contact; and then a second time among the subsample who eventually made a treatment contact to study the predictors of delay in initial contact. The Taylor series linearization method⁵³ implemented in the SUDAAN software system⁵⁴ was used to adjust for the effects of the weighting and clustering of the NCS-R data on significance tests. Multivariate significance tests in the discrete-time survival analyses were made with χ² tests using Taylor series design-based coefficient variance-covariance matrices. Statistical significance was consistently evaluated using 0.05 level, 2-sided tests.

Survival curves were used to make projections of the proportion of cases who will eventually make treatment contact for each disorder assessed in the NCS-R. These proportions are estimated to be 88.1% for MDE, 90.2% for BPD, and 94.2% for DYS (Figure 1), with no statistically significant differences in the survival curves of the individual disorders in this class (χ²₂ = 0.7; P = .72). Projected proportions are much more variable for anxiety disorders (Figure 2) and are 27.3% for SAD, 50.1% for SoP, 50.1% for SP, 65.3% for PTSD, 66.5% for AG, 86.1% for GAD, and 95.3% for PD, with statistically significant differences in the survival curves of the individual disorders in this class (χ²₅ = 242.4; P<.001). Projected proportions for impulse control disorders (Figure 3) are 33.9% for ODD, 50.4% for IED, and 51.8% for ADHD, with significant differences in the survival curves of the individual disorders in this class (χ²₂ = 6.0; P = .05). Projected proportions for substance disorders (Figure 4) are 52.7% for AA, 57.0% for DA, 69.8% for AD, and 76.9% for DD, with significant differences in the survival curves of the individual disorders in this class (χ²₃ = 44.8; P<.001).

The survival curves were also used to estimate the proportion of cases that made treatment contact in the year of first onset of the disorder and the median delay among people who eventually made treatment contact after the year of first onset (Table 1). The proportion of cases that made treatment contact in the year of disorder onset ranges from highs of 37.4% to 41.6% for the mood disorders to lows of 1.0% to 3.4% for SP, SoP, and SAD. Median years of delay also differ greatly across disorders, from lows of 6 to 8 years for mood disorders to highs of 20 to 23 years for SP and SAD.

Predictors of failure to ever make a treatment contact among respondents with a given lifetime disorder (Tables 2-4) are very similar to predictors of duration of delay among respondents who eventually made treatment contact. The most consistent predictors of failure to make a treatment contact are cohort and age at onset. Cohort is significantly related to lifetime treatment contact in 14 of 17 comparisons (0.05 level, 2-sided tests), with the dominant pattern being for treatment contact to increase in more recent cohorts. This pattern is even stronger (17 of 17 comparisons) in predicting delay in treatment contact among eventual patients. Age of onset is significantly related to treatment contact in 15 of 17 comparisons, the exceptions being 2 childhood-onset disorders (SAD, ADHD), with a consistent pattern of increasing treatment contact with increasing age at onset. A very similar pattern (17 of 17 significant comparisons) is found in predicting delay in treatment among eventual patients.

Sociodemographic predictors are less consistent. Women have significantly higher odds of treatment contact than men for 4 of the 17 outcomes (MDE, DYS, BPD, SoP), while men never had higher odds than women for any outcome. Sex differences are even weaker (1 of 17 comparisons significant) in predicting delays. Race/ethnic differences are significant in 8 of 17 comparisons, with non-Hispanic whites always having higher odds of treatment than 1 or more minority groups. This pattern is weaker in predicting delays (5 of 17 comparisons significant), even though there are some cases in which minorities have significantly shorter delays than non-Hispanic whites. Education is significant in 8 of 17 comparisons, but the pattern of the association varies considerably across outcomes. The most consistent element in the pattern is that students generally have higher odds of treatment than people who have completed their education. This is quite different from the pattern with respect to delays among eventual cases (10 of 17 comparisons significant), where college graduates generally have the shortest delays. Finally, marital status is significantin 5 of 17 comparisons, with never married respondents consistently more likely and previously married people sometimes more likely than married people to make treatment contact. A similar pattern (7 of 17 comparisons significant) exists in predicting delays among people who eventually make treatment contacts.

The results reported here should be interpreted in light of 5 potential limitations. Perhaps the most concerning is the possibility of recall failure of lifetime events. If respondents who did not seek treatment were more likely either to forget or to normalize symptoms than respondents who received treatment, leading to a false-negative report of lifetime disorder, the prevalence of disorder would be underestimated and the probability of eventual treatment contact would be underestimated. We have no way of evaluating this possibility, although it is noteworthy that in such a case the estimate of the raw number of people with a disorder who had significant delays in initial treatment contact would be conservative.

A second possibility is that despite accurate reports of disorders and treatment contacts occurring, dating was inaccurate. As discussed in more detail by Kessler and Ustun,⁴⁵ special efforts were made in the NCS-R to help respondents recall age at onset and age at initial treatment contact by asking questions that focused memory search and bounded recall uncertainty.⁵⁵ Although it is unclear how remaining dating errors would affect results, telescoping (recalling past experiences as having occurred more recently than they actually did occur) is the most likely type of error, which would lead to downward bias in estimates of delays.⁵⁶

Third, the NCS-R questions about initial treatment contact provide no information about whether treatment was actually obtained or, if so, about the nature, intensity, or adequacy of these treatments. We also did not distinguish between contacts made in health care vs non–health care sectors, as we have done in previous analyses of the original NCS.²¹ As a result, the findings presented here should be interpreted as upper bounds on the proportions of these cases that received treatment and lower bounds on the typical durations of delays until treatment was received. Furthermore, some meaningful proportion of treated cases received treatment that did not meet minimal standards for treatment adequacy.⁵ Even though we have no way of pinpointing the specific cases that received inadequate treatment at the point of initial treatment contact, we know that the proportions of cases projected to make eventual treatment contact are greater than the proportions that obtain adequate treatment.

Fourth, predictors of failure and delay in initial contact were necessarily limited to ones that could be retrospectively dated. In the socioeconomic domain, for example, we considered the effects of educational attainment but not of income, as we had no information about each respondent’s income in each year of his or her life. In addition, we limited predictors to variables for which a priori hypotheses have been raised.^{18 - 21} Although categories of some predictors (eg, ages at interview and disorder onset, education) were kept broad so as to contain sufficient cases, we still may have lacked the statistical power to identify important differences (eg, between racial and ethnic minority groups). Furthermore, it is important to recognize that even though our outcomes were failures and delays in initial treatment contacts, the estimated effects of the predictors represent amalgamated effects on a number of steps in the help-seeking process that lead up to making an initial treatment contact, such as becoming aware of the disorder, perceiving need for treatment, and accessing care.

Finally, we cannot conclude with certainty how patterns and predictors of failure and delay have changed over the past decade because the current report on the NCS-R differs from earlier analyses of the original NCS^{18 - 19 ,21} in terms of disorders, covariates, and outcomes examined. Even subtle differences between analyses (eg, use of diagnostic hierarchy rules) could have important effects. Exploring temporal trends between the NCS and NCS-R remains an important area for future research.

With these limitations in mind, our results document 2 types of unmet need for mental health treatment in the United States. First, a large number of people never make treatment contacts for the DSM-IV/WMH-CIDI disorders evaluated in the NCS-R. This is especially true for substance and impulse control disorders, where nearly half of all lifetime cases failed to make any treatment contact. Consistent with the finding regarding substance disorders, previous research has shown that people with substance disorders often do not perceive a need for treatment, actively resist treatment, and do not seek treatment until their disorders have become highly debilitating.^{9 ,57 - 58} The low proportion of cases that ever seek treatment for impulse control disorders could reflect the perceptions, both on the part of the people with the disorders and of society at large, that their problems are less relevant to the mental health care system than to other systems (eg, social services, education, criminal justice). Although impulse control disorders can be highly distressing to others, they are often associated with more externalization and less internal dysphoria for cases, resulting in less motivation for afflicted individuals to seek treatment. The facts that these disorders have only recently become recognized as DSM disorders and that treatments are only beginning to emerge are other likely contributing factors to the low rate of treatment of impulse control disorders.^{47 ,59 - 60}

The second source of unmet need for mental health care documented in the NCS-R concerns pervasive delays in initial treatment contact. The typical delays documented here persist for years or even decades for some disorders. This has not been a focus of previous research, as mental health services research has traditionally focused on treatment of current episodes for established cases.^{8 - 14} This focus led naturally to a concern with lack of treatment rather than with treatment delay. The NCS-R findings suggest that this focus needs to be expanded, as eventual treatment is the typical state of affairs for most of the disorders considered here, while delay in initial treatment seeking is pervasive.

Substantial inter-disorder variation was found both in the probability of eventually making treatment contacts and in typical durations of delay. The pattern of this variation is quite consistent with the pattern found in previous studies of initial treatment contact.^{18 - 21} The finding of initial treatment contact being higher and delays shorter for mood disorders than other classes of disorders may be because of the fact that mood disorders have been targeted by educational campaigns, primary care quality improvement programs, and treatment advances.^{25 - 26 ,28 ,31 ,61} The prominent and dysphoric somatic symptoms often found in conjunction with panic disorder may account for the finding that treatment contact is more common and occurs more quickly for panic disorder than other types of anxiety disorder.^{62 - 63} Greater failure and delays for some anxiety disorders such as phobias and SAD could be because of their generally early age at onset, fewer associated impairments, and even fear of providers or treatments involving social interactions (eg, talking therapies, group settings, waiting rooms).^{9 ,12 ,64}

Shorter delays and higher proportions of cases with eventual treatment contact in more recent cohorts provide grounds for optimism, as they provide some evidence that patterns of help-seeking have improved in the recent past. This secular change could be, in part, because of recent programs that destigmatize and increase awareness of mental illness, screening and outreach initiatives, the introduction and direct-to-consumer promotion of new treatments, and expansion of some insurance programs,^{22 - 41} although a more fine-grained mapping of time series information would be needed to document effects of specific initiatives. Drug dependence is a notable exception to the general pattern of increased treatment in recent cohorts. There are both longer delays and fewer initial contacts in younger cohorts than older ones. Whether this is the result of methodological (eg, failure to capture contacts with self-help groups which now play increasingly important roles in the treatment of substance disorders) or substantive processes is unclear, although one might expect methodological factors to have effects that emerged more consistently across the full range of conditions. To the extent that the pattern is substantive, it might be because of changes in funding of drug treatment programs, changes in public attitudes toward drug dependence, or a combination of those factors.^{65 - 67} Given the importance of timely drug treatment, more detailed investigation is warranted.

We found that early-onset disorders are consistently associated with longer delays and a lower overall probability of initial treatment contact. The same pattern has consistently been found in previous studies of delays in initial treatment contact.^{18 - 21} Minors may be less likely to receive timely treatment because they need the help of parents or other adults and recognition is often low among these adults unless symptoms are extreme.^{68 - 69} In addition, child- and adolescent-onset mental disorders might be associated with normalization of symptoms or the development of coping strategies (eg, social withdrawal in social phobias) that interfere with help-seeking during adulthood. The paucity of available or accessible child mental health services may also be an important factor.

Men have been shown in earlier research to be slower than women at translating nonspecific feelings of distress into conscious recognition that they have emotional problems, perhaps explaining the NCS-R finding that males sometimes have longer delays and lower rates of treatment contact than women.^{70 - 71} The longer delays and lower odds of ever making treatment contacts found among minorities compared with non-Hispanic whites are broadly consistent with findings in earlier studies that minorities often receive suboptimal mental health care,^{2 - 3 ,12} but we are not aware of any previous research that explicitly examined race/ethnic differences in initial treatment contact. Negative attitudes toward treatment on the part of minorities could play an important part in accounting for these results.⁷² However, the fact that the race/ethnic differences in delays and eventual treatment vary substantially across outcomes suggests that more than broad attitudinal factors are likely to be involved. Greater knowledge and financial resources to pay for treatments could help explain the positive associations between education and initial help-seeking for many mental disorders; on the other hand, greater help-seeking for substance disorders by those less educated could reflect their greater acceptance and lower financial barriers to addiction services, many of which are self-help groups.⁷³ Difficulties forming or maintaining relationships and lack of social support from a partner may be strong impetuses for those not married to rapidly seek mental health treatments.^{6 ,12}

It is important to consider whether the delays and failures to make initial treatment contacts truly pose a public health problem. An alternative view is that they may only characterize less severe, short-lived, or nondebilitating mental disorders.⁷⁴ However, research in the baseline NCS data showed that many people with even severe and impairing disorders reported substantial delays in initial treatment contact.²¹ Long periods of untreated illness may also be harmful to those with less severe disorders. Preclinical studies suggest that neural “kindling” can cause untreated psychiatric disorders to become more frequent, severe, spontaneous, and treatment refractory.⁷⁵ In addition, epidemiological studies suggest that school failure, teenage child-bearing, unstable employment, early marriage, marital violence, and marital instability are associated with early-onset untreated mental disorders.^{76 - 79} Recent randomized clinical trials have shown that treatment can prevent suicidality.⁸⁰ Furthermore, most people with 1 disorder progress to develop comorbid disorders and such comorbidity is associated with an even more persistent and severe clinical course.^{81 - 82} However, it should be kept in mind that not all studies have found a relationship between the duration of untreated mental illness and long-term outcomes.⁸³ Definitively answering whether reducing treatment delays would prevent such negative outcomes requires long-term trials of aggressive outreach and treatment of new cases. Such trials are just beginning to be carried out.^{84 - 86}

Despite the absence of definitive data from such long-term outreach treatment trials, the findings reported here suggest that more effort is needed to increase prompt initial treatment contacts among people with incident episodes of mental disorders. Additional large-scale public education programs (eg, the NIMH Depression, Awareness, Recognition, and Treatment program) and expanded use of National Screening Days continue to hold great promise for hastening detection and treatment.^{23 ,31 ,68} School-based screening programs using brief self-report and/or informant scales may be needed to detect early-onset mental disorders.^{87 - 88} Demand management and other outreach strategies could also help reduce critical delays and failures in initial help-seeking once mental disorders are identified.^{89 - 90} Training non–health care professionals to recognize individuals with mental disorders and make timely referrals for health care should also be explored.^{30 ,91 - 92} A range of interventions may ultimately be needed to alleviate the burdens and hazards from untreated mental disorders.

Correspondence: Philip S. Wang, MD, DrPH, 180 Longwood Ave, Boston, MA 02115 (wang@hcp.med.harvard.edu).

Submitted for Publication: June 9, 2004; final revision received October 15, 2004; accepted November 9, 2004.

Funding/Support: The National Comorbidity Survey Replication is supported by the National Institute of Mental Health (U01-MH60220), Rockville, Md; with supplemental support from the National Institute of Drug Abuse, the Substance Abuse and Mental Health Services Administration, Rockville, Md; the Robert Wood Johnson Foundation (grant 044708), Princeton, NJ; and the John W. Alden Trust, Boston, Mass.

Disclaimer: The views and opinions expressed in this report 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.

Additional Information: Collaborating investigators include Dr Kessler, principal investigator; Kathleen Merikangas, co-principal investigator, National Institute of Mental Health, Bethesda, Md; Dr Olfson; Dr Pincus; Dr Wang; Doreen Koretz, Harvard University, Cambridge, Mass; James Anthony, Michigan State University, East Lansing; William Eaton, Johns Hopkins University, Baltimore, Md; Meyer Glantz, National Institute on Drug Abuse, Baltimore, Md; Jane McLeod, Indiana University, Bloomington; Greg Simon and Michael Von Korff, Group Health Cooperative, Seattle, Wash; Kenneth Wells, University of California, Los Angeles; Elaine Wethington, Cornell University, Ithaca, NY; and Hans-Ulrich Wittchen, Institute of Psychology, Technical University, Dresden and Max Planck Institute of Psychiatry, Munich, Germany.

Acknowledgment: We thank Jerry Garcia, Sara Belopavlovich, Eric Bourke and Todd Strauss for assistance with manuscript preparation. A complete list of NCS publications and the full text of all NCS-R instruments can be found at www.hcp.med.harvard.edu/ncs. Send correspondence to ncs@hcp.med.harvard.edu. The NCS-R is carried out in conjunction with the WMH Survey Initiative. We thank the staff of the WMH Data Collection and Data Analysis Coordination Centres for assistance with instrumentation, fieldwork, and consultation on data analysis.