Kessler RC, Merikangas KR, Berglund P, Eaton WW, Koretz DS, Walters EE. Mild Disorders Should Not Be Eliminated From the DSM-V. Arch Gen Psychiatry. 2003;60(11):1117-1122. doi:10.1001/archpsyc.60.11.1117
High prevalence estimates in epidemiological surveys have led to concerns that the DSM system is overly inclusive and that mild cases should be excluded from future DSM editions.
To demonstrate that the DSM-III-R disorders in the baseline National Comorbidity Survey (NCS) can be placed on a severity gradient that has a dose-response relationship with outcomes assessed a decade later in the NCS follow-up survey (NCS-2) and that no inflection point exists at the mild severity level.
The NCS was a nationally representative household survey of DSM-III-R disorders in the 3-year time span 1990-1992. The NCS-2 is a follow-up survey of 4375 NCS respondents (76.6% conditional response rate) reinterviewed in 2000 through 2002. The NCS-2 outcomes include hospitalization for mental health or substance disorders, work disability due to these disorders, suicide attempts, and serious mental illness.
Twelve-month NCS/DSM-III-R disorders were disaggregated into 3.2% severe, 3.2% serious, 8.7% moderate, and 16.0% mild case categories. All 4 case categories were associated with statistically significantly (P<.05, 2-sided tests) elevated risk of the NCS-2 outcomes compared with baseline noncases, with odds ratios of any outcome ranging monotonically from 2.4 (95% confidence interval, 1.6-3.4) to 15.1 (95% confidence interval, 10.0-22.9) for mild to severe cases. Odds ratios comparing mild to moderate cases were generally nonsignificant.
There is a graded relationship between mental illness severity and later clinical outcomes. Retention of mild cases in the DSM is important to represent the fact that mental disorders (like physical disorders) vary in severity. Decisions about treating mild cases should be based on cost-effectiveness not current severity. Cost-effectiveness analysis should include recognition that treatment of mild cases might prevent a substantial proportion of future serious cases.
TWO MAJOR psychiatric epidemiological surveys have been carried out in the United States since 1980—the Epidemiologic Catchment Area Study1 and the National Comorbidity Survey (NCS).2 Both surveys estimated that about 30% of the respondents in the 18- to 54-year age range met criteria for 1 or more of the 12-month DSM-III (Epidemiologic Catchment Area) or DSM-III-R (NCS) psychiatric disorders assessed in the surveys.3,4 These are lower-bound estimates, as neither of the surveys included the full range of DSM disorders in their assessments and even this range of disorders is likely to be noninclusive.5,6 Nevertheless, these rates seemed high, leading to the suggestion that the lay administered diagnostic interviews in these surveys might be upwardly biased.7,8 However, clinical calibration studies showed that the prevalence estimates in these surveys were not upwardly biased,9,10 leading critics to conclude that the DSM system itself is overly inclusive.3,11,12 This conclusion was instrumental in causing an American Psychiatric Association Task Force to add a clinical significance criterion to many disorders in the DSM-IV to remind readers of the basic definition of a mental disorder in the "Introduction" of the manual. However, even when this additional requirement was applied post hoc to the Epidemiologic Catchment Area and NCS data, the 12-month prevalence of a DSM disorder, equivalent to about 37 million adults in the United States, continued to exceed substantially the number who could be helped with current treatment resources.13
In recognition of this problem, several more restrictive definitions have been proposed that can be used to narrow the number of persons qualifying for treatment. The National Institute of Mental Health National Advisory Mental Health Council,14 for example, distinguished persons with severe and persistent mental illness (SPMI) from other persons with mental illness, while the Alcohol, Drug Abuse, and Mental Health Administration Reorganization Act stipulated that state mental health block grant funds can be used only to treat persons with serious mental illness (SMI).15 Many health care plans have followed suit by restricting mental health coverage to a subset of DSM disorders that they consider to be biologically based. Darryl Regier, MD, director of the Division of Research at the American Psychiatric Association, who is expected to play a prominent role in the development of DSM-V, has supported similar restrictions using more elaborate requirements for clinically significant distress or impairment than those in DSM-IV.13,16,17
The SPMI and SMI proposals are different from the Regier proposal for DSM-V. The definitions of SPMI and SMI are practical attempts to focus treatment resources on the subset of cases with the greatest need. In comparison, the Regier proposed restriction of cases in DSM-V is an attempt to declare that mild cases do not exist. This radical approach has implications not only for the definition of current unmet need for treatment but also for current research and consideration of future treatment needs. Research shows that many syndromes currently defined as mental disorders are extremes on continua that appear not to have meaningful thresholds.18,19 These results are important in at least 2 ways. First, exploration of the full continua rather than the currently established diagnostic thresholds might yield greater power in studies of genetic and environmental risk factors.20 Second, development of early interventions to prevent progression along a given severity continuum might reduce the prevalence of serious cases.21 Removal of current mild cases from the DSM system would undercut both of these advantages as well as distort the reality that mental disorders (like physical disorders) vary widely in seriousness.22,23
The current article presents data designed to argue against using the deletion of mild cases from future editions of the DSM as an approach to restrict the number of persons who are defined as being in need of treatment. The argument is based on the expansion of an analysis of the NCS that was recently reported in the ARCHIVES by Narrow, Regier, and colleagues.13 These authors divided 12-month NCS/DSM-III-R cases into those considered either clinically significant mental illness (CSMI) or clinically nonsignificant mental illness (CNMI) based on respondent reports about interference and treatment. Comparison of these 2 subgroups showed, not surprisingly, that various indicators of illness severity (eg, days out of role [ie, the number of days when a person is unable to carry out their normal activities], history of suicide attempts) were higher in CSMI than CNMI. Narrow, Regier, and colleagues asserted, based on these results, that mild cases should be excluded from DSM-V.
We build on this analysis in 2 ways. First, we use data from the recently completed NCS-2,24 a survey that reinterviewed NCS respondents a decade after their baseline interviews, to examine the associations of baseline NCS 12-month illness severity with clinically significant outcomes assessed in NCS-2. Second, we expand the number of illness severity categories from 2 to 4 by dividing the cases that Narrow, Regier, and colleagues defined as having CSMI into severe, serious, and moderate cases. We show that differences in the risk of clinically significant outcomes in NCS-2 across these severity categories are as large as, and in some cases larger than, those between moderate and mild (ie, CNMI) cases. We also show that the elevated risk of the NCS-2 outcomes among mild cases vs noncases is consistently larger than the elevated risk among moderate cases vs mild cases. These results call into question the suggestion that the DSM-V case threshold should be set above CNMI rather than at any other arbitrary point on the severity gradient.
The baseline NCS was a nationally representative household survey of 8098 respondents in the age range of 15 to 54 years carried out in the 3-year time span of 1990-1992.4 Respondents were selected from a stratified, multistage area probability sample of the noninstitutionalized civilian population in the 48 coterminous states, with a supplemental sample of students living in campus housing. The survey response rate was 82.4%. All NCS respondents were administered a part I diagnostic interview (described in the "Baseline Diagnostic Assessment" section), while a part II risk factor interview was administered to a probability subsample of 5877 respondents who included all those in the age range of 15 to 24 years, all those who screened positive for any of the DSM-III disorders assessed in part I, and a 1 in 6 random subsample of all remaining part I respondents. The part II sample was weighted to adjust for differential probabilities of selection as well as for nonresponse bias.25
The NCS-2 attempted to trace and reinterview all 5877 part II NCS respondents a decade after the baseline NCS using an expanded version of the baseline interview that reconstructed information on illness onset, course, and severity during the years between the 2 surveys. A total of 5463 target respondents were successfully traced, of whom 166 were deceased, and 4375 were interviewed for a conditional response rate of 76.6%. The unconditional response rate, which considers the baseline NCS response rate of 82.4%, is 63.1% (0.766 × 0.824). The NCS-2 respondents differ significantly from other baseline NCS respondents in higher probabilities of being female, well educated, and residents of rural areas (Table 1, part I). A propensity score adjustment weight26 was used to correct the NCS-2 sample for these compositional biases. There was remarkably little difference, in comparison, between NCS-2 respondents and nonrespondents either in the prevalence of baseline NCS/DSM-III-R disorders or in the severity of these disorders (Table 1, part II).
The baseline NCS diagnostic interview was a modified version of the World Health Organization Composite International Diagnostic Interview (CIDI).27,28 Diagnoses were based on DSM-III-R criteria. The 12-month diagnoses available in the part II sample, which are the focus of this article, are mood disorders (major depression, mania, dysthymia), anxiety disorders (panic disorder, agoraphobia without panic, social phobia, simple phobia, generalized anxiety disorder, posttraumatic stress disorder), substance use disorders (alcohol or other drug abuse or dependence), and nonaffective psychosis (NAP) (schizophrenia, schizophreniform disorder, schizoaffective disorder, delusional disorder, or psychosis not otherwise specified). Criteria for at least one of these disorders in the 12 months before the baseline interview was met by 31.2% of the weighted part II NCS sample.
A clinical calibration of all NCS/DSM-III-R diagnoses other than NAP was carried out with blind follow-up interviews of 356 NCS respondents using the lifetime version of the Structured Clinical Interview for DSM-III-R (SCID).29 The vast majority of CIDI prevalence estimates were lower than the SCID prevalence estimates.10 However, only 2 differences were statistically significant—an overestimated prevalence of major depression in the CIDI compared with the SCID and an underestimated prevalence of simple phobia in the CIDI compared with the SCID. A separate calibration study for NAP showed that, although the CIDI could be used as a useful first-stage screen for NAP, no calibration rule could generate valid diagnoses of NAP from the CIDI data.30 Based on this result, an attempt was made to administer a clinical reappraisal interview for NAP to all NCS respondents who screened positive for NAP in the baseline NCS. Clinician-diagnosed NAP was subsequently used instead of CIDI-diagnosed NAP in all NCS reports, including the current report.
The baseline illness severity scale was created by using 3 previously published definitions: (1) The National Mental Health Advisory Council definition of SPMI;14 (2) the Substance Abuse and Mental Health Services Administration definition of SMI15;and (3) the definition of CSMI recently made by Narrow, Reiger, and colleagues.13 The operational definitions of SPMI and SMI have been described in detail elsewhere.31,32 Briefly, SPMI was defined as either (1) meeting criteria for 12-month NAP or mania or (2) having a 12-month anxiety or mood disorder that required hospitalization or treatment with antipsychotic medications. Serious mental illness was defined as either (1) meeting criteria for SPMI or (2) having a 12-month anxiety or mood disorder that was associated with at least 1 indicator of serious role impairment (suicidality, mental health-related work disability, marital violence, or extreme social isolation). Following Narrow, Reiger, and colleagues,13 CSMI was defined as having a 12-month NCS/DSM-III-R diagnosis that either (1) was described by the respondent as interfering "a lot" with his or her life and activities or (2) was treated at some time in the 12 months before the baseline NCS interview. Clinically nonsignificant mental illness, finally, was defined as any 12-month NCS/DSM-III-R disorder that did not meet the requirements for CSMI. Using these definitions and defining categories hierarchically, 3.2%(SE, 0.3%) of part II NCS respondents met 12-month criteria at baseline for SPMI, an additional 3.2% (SE, 0.2%) met criteria for SMI but not SPMI, an additional 8.7% (SE, 0.4%) met criteria for CSMI but neither SPMI nor SMI, and an additional 16.0% (SE, 0.5%) met criteria for CNMI but not SPMI, SMI, or CSMI.
The 4 NCS-2 outcomes considered herein were defined as follows: (1) Hospitalization was defined as being hospitalized for mental health or substance problems at any time in the decade between the NCS and NCS-2 interviews. This outcome was reported by 2.5% (SE, 0.3%) of the NCS-2 respondents. (2) Work disability was defined as being out of work on disability for a mental health or substance problem for a minimum of 4 consecutive weeks at any time between the 2 interviews. This outcome was reported by 1.4% (SE, 0.2%) of the NCS-2 respondents. (3) Suicide attempt was defined as reporting an attempted suicide at any time between the 2 interviews. This outcome was reported by 1.4% (SE, 0.2%) of the NCS-2 respondents. (4) Serious mental illness was defined as meeting the criteria described earlier for SMI in the 12 months before the NCS-2 interview. This outcome was reported by 5.6% (SE, 0.5) of the NCS-2 respondents. At least 1 of the 4 outcomes was reported by 8.5% (SE, 0.6%) of the NCS-2 respondents.
Controlling for respondent age and sex, logistic regression analysis was used to predict the NCS-2 outcomes from dummy variables that distinguished NCS respondents with baseline 12-month severe, serious, moderate, and mild DSM-III-R/NCS disorders compared with respondents who did not meet criteria for any of these disorders at the time of the original NCS. Parallel logistic regression equations were estimated to compare outcomes associated with the 4 contiguous pairs of illness severity categories (severe vs serious, serious vs moderate, moderate vs mild, mild vs noncase). Logistic regression coefficients were transformed to odds ratios (ORs) and are reported with design-based 95% confidence intervals (CIs) estimated using the Taylor series linearization method,33 implemented in the SUDAAN software package.34 Wald χ2 tests correcting for the clustering and weighting of the NCS-2 data were used to evaluate overall model fit. Simulations using an SAS (SAS Inc, Cary, NC) macro estimated the population attributable risk proportion of each outcome associated with each category of the baseline clinical severity distinction. The population attributable risk proportion was operationalized in this macro as the proportional reduction in the prevalence of the outcome estimated from the logistic regression equation when the OR associated with a given category of severity was set to 1.0. This ratio can be interpreted as the proportion of respondents with the outcome who would no longer have the outcome in the absence of the elevated risk associated with the specified disorder severity category. All analyses of statistical significance were based on P = .05, 2-sided tests.
A consistent monotonic relationship exists between the baseline illness severity categories and the 4 NCS-2 outcomes (Table 2), with only 1 inversion in monotonicity (the OR for mild cases was larger than the OR for moderate cases in predicting work disability). The largest ORs associated with severe cases are in the range of 5.60 to 29.7, while the smallest ORs associated with mild cases are in the range of 1.3 to 2.7. Three of the 3 ORs associated with mild cases are statistically significant at the .05 level.
A comparison of pairwise differences in outcomes across contiguous categories of the baseline illness severity gradient (Table 3) finds 10 statistically significant differences (P<.05, 2-sided tests) of 20 comparisons. Importantly, the differences between moderate cases vs mild cases are consistently smaller than either the differences between severe cases vs serious cases or serious cases vs moderate cases. The moderate vs mild distinction is statistically significant in only 1 comparison (SMI in the year before the NCS-2 interview). The mild vs none distinction, in comparison, is significant in 3 comparisons and consistently larger than the moderate vs mild distinction.
The population attributable risk proportions associated with each category of baseline illness severity (Table 4) vary less than the ORs. This is true because the higher baseline prevalences of the less severe categories counterbalance their lower effects on the outcomes. As a result, although severe cases are consistently associated with the highest population attributable risk proportions, there is no consistent rank order of population attributable risk proportions among the other categories.
These results show that a continuous graded relationship exists between the baseline NCS illness severity categories and the NCS-2 outcomes. No consistent inflection point in the gradient of outcome risk can be found at the point on the distribution proposed by Narrow, Regier, and colleagues13 as distinguishing between CSMI and CNMI. The results also show that the elevated risk of NCS-2 outcomes among mild cases vs noncases is consistently larger than the elevated risk among moderate cases vs mild cases. These results call into question the suggestion that the case threshold should exclude mild cases.
This is not to say that more principled considerations, based either on future epidemiological, biological, or taxometric studies, might not lead to the conclusion that diagnostic thresholds for certain DSM disorders should be changed. Nor is it to say that the problem that motivated Narrow, Regier, and colleagues13 to propose narrowing the DSM criteria (the number of persons who meet current criteria is much larger than the number of persons who can be treated with available treatment resources) is unimportant. However, the solution to this problem proposed by Narrow, Regier, and colleagues to define the problem out of existence is ill conceived. The definition of a case should not be considered synonymous with need for treatment any more than with clinically significant distress or impairment.35
The problem of unmet need for treatment should be addressed by developing comprehensive triage rules that allocate available resources based on evidence-based assessments of the cost-effectiveness of available treatments across the severity threshold of the disorder. Severity gradients are widely used in this way in other branches of medicine.36 In the absence of such rules, which do not exist, ad hoc decision making is inevitable.37 In developing these rules for mental disorders, consideration should be given not only to current distress and impairment but also to the risk of progression from a mild to a more severe disorder. It is unclear whether these rules, once they are developed, would define treatment of mild cases as cost-effective. Even if they did not, mild cases should be retained in the definition of disorders both to acknowledge that mental disorders (like physical disorders) vary in severity and to remind us that the development of cost-effective treatments for mild disorders might prevent a substantial proportion of future serious disorders.
Corresponding author and reprints: Ronald C. Kessler, PhD, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02115 (e-mail: email@example.com).
Submitted for publication October 11, 2002; final revision received April 30, 2003; accepted May 1, 2003.
The baseline NCS was supported by grants MH46376, MH49098, and MH52861 from the National Institute of Mental Health (NIMH), with supplemental support from grant 90135190 from the William T. Grant Foundation, New York, NY. Dr Kessler was the principal investigator of the baseline NCS. His participation in the survey was supported by NIMH Research Scientist Award MH00507. The NCS-2 is funded by grant DA12058 from the National Institute of Drug Abuse, with supplemental support from the NIMH.
Drs Kessler and Merikangas are coprincipal investigators. Collaborators include Jim Anthony, PhD, Johns Hopkins University; Jane Costello, PhD, Duke University, Durham, NC; William Eaton, PhD, Johns Hopkins University; Doreen Koretz, PhD, NIMH; Dan Offord, MD, McMaster University, Hamilton, Ontario; and Elaine Wethington, PhD, Cornell University, Ithaca, NY.
A complete list of NCS publications along with information about NCS-2 can be obtained from the NCS Web site home page at: http://www.hcp.med.harvard.edu/ncs.
The helpful comments of Bernie Arons, MD, David Mechanic, PhD, and the journal reviewers are gratefully acknowledged.
The views and opinions expressed in this article are those of the authors and should not be construed to represent the views of any of the sponsoring organizations, agencies, or US government.