American Journal of Epidemiology Copyright C 1992 by The Johns Hopkins University School of Hygiene and Pubfc Health Ai rights reserved
VoL 135, No. 12 PnntBd in U.S.A.
Reevaluation of Secular Trends in Depression Rates
Gregory E. Simon and Michael VonKorff
Results of numerous community surveys of psychiatric illness suggest a striking change in the occurrence of depression, with younger generations experiencing higher lifetime risk and earlier age of onset. Data from the National Institute of Mental Health Epidemiologic Catchment Area Survey (a cross-sectional survey of psychiatric morbidity in five US communities conducted between 1980 and 1984) were reexamined for evidence of methods effects which might contribute to these unexpected findings. A pattern of higher lifetime risk and earlier age of onset among recent birth cohorts was observed for every psychiatric disorder examined, with schizophrenia, major depression, and panic disorder showing equally strong trends. For respondents of all ages, reported first onset of major depression clustered in the 10-year period prior to the study interview, in contrast to the expectation that older respondents would report onset in early adulthood. Examination of individual psychiatric symptoms revealed a nearly universal pattern of decreasing lifetime prevalence among older respondents, a reversal of the expected accumulation of lifetime symptoms with age. These findings suggest that effects of study methods may contribute to the apparent temporal trends in prevalence of depression and that cross-sectional surveys may underestimate lifetime psychiatric morbidity among older respondents. Generational changes in the lifetime risk of depression or other psychiatric disorders may not be reliably assessed by crosssectional survey data. Am J Epidemiol 1992;135:1411-22. affective disorders; depression; epidemiologic methods; life tables
Several recent reports have described a dramatic secular trend in the lifetime prevalence of major depression. In 1985, Klerman et al. (1) reported data on the lifetime psychiatric histories of family members studied in National Institute of Mental Health Collaborative Program on the Psychobiology of Depression. Within this group at high risk for affective disorder, the reported lifetime
prevalence of major depression showed a large and unexpected decline among older subjects. Examination of lifetime risk for depression according to year of birth appeared to demonstrate that more recent birth cohorts experienced markedly higher prevalence and earlier age of onset for depressive illness. This report was followed by similar findings among community samples
Received for publication May 30,1991, and in final form October 28, 1991. Abbreviation: DSM-IU, Diagnostic and Statistical Manual, Third Edition. From the Center for Health Studies, Group Health Cooperative, Seattle, WA, and the Department of Psychiatry and Behavioral Sciences, University of Washington, Seattle, WA. Reprint requests to: Dr. Gregory Simon, Center for Health Studies, 1730 Minor Avenue, Seattle, WA 98101. This research is based on the National Institute of Mental Health Epidemiotogjc Catchment Area Survey Public Use
Data Set, Wave 1 Household Sample The catchment area survey was a senes of five epidemiologic research studies performed by five independent research teams in collaboration with National Institute of Mental Health, Division of Biometry and Epidemiology. The five research studies were carried out by Yale University, The Johns Hopkins University, Washington University, Duke University, and the University of California at Los Angeles. Robert Jamieson assisted with data preparation and translation Drs. John Lee and William Barlow provided comments on earlier versions of the manuscript
1411
1412
Simon and VonKorff
in the United States (2-4), Germany (5), New Zealand (6), and Canada (7) that included over 27,000 subjects. While surveys in Korea (8) and Puerto Rico (9) have not found such an effect, the bulk of these studies suggest that lifetime risk of major depression is approximately five times as high among recent generations as among those born at the turn of the century. These investigations were typically cross-sectional surveys of nonpatient populations which relied on structured interviews to assess lifetime psychiatric history. Longitudinal studies of depression prevalence do not find such a striking cohort effect over the period when this change in risk should have occurred. These studies have used either repeated cross-sectional sampling or true cohort designs. Of the two repeat cross-sectional studies, the Stirling County Study (10) found prevalence of anxiety and depressive disorders to be unchanged from 1952 to 1970, while the US National Sample Study found an increase in symptoms of psychologic distress from 1957 to 1976 (11). Among studies using the cohort method, the Lundby Study (12) demonstrated an increase in incidence of depression between 1947 and 1972 which was confined to disorders of mild and moderate severity, while the Midtown Manhattan Study (13) found a decrease in prevalence of psychiatric disorder from 1954 to 1974. Analysis of longitudinal data on depression from the Alameda County Study (14) found some evidence for a transient increase in the prevalence of depression during the 1970s, but prevalence rates were consistently highest among the oldest cohorts studied. None of these studies reported an increase in risk of the magnitude suggested by crosssectional studies. These disparate findings suggest the possibility that some methodological factor common to cross-sectional studies may contribute to this apparent trend. Crosssectional studies which rely on retrospective assessment of lifetime history typically find evidence of dramatic increases in prevalence, while longitudinal studies relying on current symptom reports yield mixed re-
sults. Given the time and expense required for community prevalence studies, further analysis of existing cross-sectional data may be the most practical means to examine the potential contribution of methods effects. Data from the Epidemiologic Catchment Area Survey are well suited to such an examination. Wickramaratne et al. (4) have previously described a marked temporal trend in prevalence of depression in this sample. The large sample size allows relatively stable estimates for the prevalence of psychiatric diagnoses within narrow age strata. The wide range of diagnoses examined allows study of age patterns for a range of psychiatric diagnoses as well as individual psychiatric symptoms. MATERIALS AND METHODS Epidemiologic Catchment Area Survey design
These analyses utilized public use tapes of data from wave 1 of the Epidemiologic Catchment Area Survey, a nationwide survey of psychiatric morbidity and health service utilization carried out under cooperative agreement between the National Institute of Mental Health and five university sites: Yale University, New Haven, Connecticut; The Johns Hopkins University, Baltimore, Maryland; Washington University, St. Louis, Missouri; Duke University, Durham, North Carolina; and the University of California at Los Angeles, Los Angeles, California (15). Each site sampled subjects from one or more previously designated Community Mental Health Center catchment areas, with geographic population areas of 75,000-250,000. Over 3,000 community residents and 500 institutional residents were studied at each site. This analysis considers only community residents. Data collection occurred between 1980 and 1984. Each site used a multistage probability sampling procedure to select households from the eligible geographic area (16). This procedure varied among sites, with some sites directly sampling households from a list of all of those eligible and others using an intermediate selection of geographic sub-
Secular Trends in Depression
units prior to selection of households. Within each household, all residents aged 18 years or older without other usual residence were enumerated. The Baltimore site sampled one resident aged 18-64 years in each household plus all additional residents over age 65. Other sites randomly selected a single resident, with the Durham and New Haven sites oversampling elderly residents. Selection of catchment areas or sampling units allowed intentional oversampling of blacks in St. Louis and Hispanics in Los Angeles. Overall response rates ranged across sites from 68 to 79 percent. Respondents had varying probabilities of selection because of intentional oversampling of certain groups, varying procedures for household selection, and selection of one resident regardless of household size. Each respondent was assigned a sampling weight inversely related to his or her probability of selection. Data collection instruments
Psychiatric morbidity was assessed using the Diagnostic Interview Schedule, a structured interview developed by the National Institute of Mental Health and the Washington University School of Medicine (17). The interview uses a structured question sequence to assess the presence of diagnostic criteria from the Diagnostic and Statistical Manual, Third Edition (DSM-III) (18) by the American Psychiatric Association. Questioning follows a specified probe sequence which allows for minimum interviewer discretion. Respondents are first asked whether they have ever experienced some potential psychiatric symptom. For each positive response, interviewers follow a series of questions designed to determine whether the symptom was sufficiently severe and whether alternative explanations (medical illness or drug or alcohol use) might explain every symptomatic episode. Symptoms of sufficient severity for which alternative explanations have been excluded are classified as presumptive psychiatric symptoms. Computer algorithms allowed translation of symptoms into DSM-III diagnoses. The ver-
1413
sion of the Diagnostic Interview Schedule included in the catchment area study allowed diagnoses of mania, major depression, bipolar disorder, alcohol abuse or dependence, drug abuse or dependence, schizophrenia, obsessive-compulsive disorder, phobia, somatization disorder, panic disorder, and antisocial personality. Data analysis
Data tapes containing the Wave 1 Household Sample of the Epidemiologic Catchment Area Survey were obtained from National Technical Information Service, Springfield, Virginia. The original SAS data file was translated into a form readable by the SPSS-X statistical package. Analyses were conducted using SPSS-X software (SPSS Inc., Chicago, Illinois) Analyses of psychiatric diagnoses utilized diagnoses generated by computer from Diagnostic Interview Schedule scoring algorithms (19). Hierarchical diagnostic exclusions of DSM-III (e.g., diagnosis of major depression precludes diagnosis of panic disorder) were not applied. All analyses incorporated sampling weights described above. RESULTS
Analysis first replicated previously reported findings of higher lifetime prevalence and earlier age of onset for major depression among younger respondents. Figure 1 shows the. reported lifetime prevalence of major depression (and other diagnoses) according to age of respondent at time of interview. Lifetime prevalence appears dramatically lower among older respondents despite longer periods of risk. Because lifetime prevalence should accumulate with age, any apparent decrease among older respondents implies a significant temporal trend (either a period effect in which risk increases universally with time or a birth cohort effect in which those born in recent years show increased risk). Figure 2 divides respondents into seven cohorts according to year of birth. For each cohort, the prevalence of depression at 5-year age intervals was calculated as
1414
Simon and VonKorff
20%
Alcohol Phobia Dtpr«««lon Pinlc Dltordtr Schizophrenia
18-26
26-36
38-45 46-55 69-65 Ag« at Interview
66-75
76-85
FIGURE 1. Reported lifetime prevalence of selected psychiatric disorders according to the respondent's age at the time of interview, n = 18,271.
the portion of respondents with lifetime history of major depression who report onset prior to that age. Here the dramatic differences in prevalence between birth cohorts are apparent. Older respondents (those born in earlier cohorts) display markedly lower lifetime prevalence and later age of onset as previously described. This dramatic temporal trend, however, was not specific to major depression. Figure 1 displays lifetime prevalence according to age at interview for other diagnoses. All diagnoses examined show a similar pattern of lower reported lifetime prevalence among older respondents. Proportional differences in lifetime prevalence by age are most marked for schizophrenia, major depression, and panic disorder. Within those three disorders, lifetime prevalence among those aged 36-45 is six to eight times as high as among those over age 65 years. Relative differences in lifetime prevalence with age are smaller for alcohol diagnoses and phobias; younger respondents report lifetime prevalence two to three times higher than those over age 65 years. As noted above, however, any downward slope in these lifetime prevalence curves implies a significant temporal effect. Figure 3 presents cumulative prevalence stratified by birth cohort (analogous to figure 2) for each of these
diagnoses other than major depression. All disorders examined show a remarkably similar pattern. More recent cohorts all show a much steeper rise to much higher peak prevalence rates despite shorter periods of risk. As with the examination of lifetime prevalence rates above, these differences are proportionally greatest for schizophrenia and panic disorder. Phobias and alcohol diagnoses show the same pattern, but relative differences between cohorts are less marked. The consistency of these unexpected findings across diagnoses raises the possibility that some common effect of study methods may contribute to this apparent trend. Two additional analyses were used to examine this possibility. The first of these examined the reported time since onset of illness among respondents with a lifetime history of major depression. This analysis considered only respondents who reported a lifetime history of major depression. Time since onset of depression was calculated as the difference between the respondents' age at the time of interview and the reported age of onset of the first depressive episode. Figure 4 divides all cases of lifetime major depression into three age strata and displays the percentage of cases reporting time since onset of 0-10 years, 11-20 years, etc. For all three strata, onset of depression appears to
Secular Trends in Depression
1415
10%
a% —
1958-1883
-+- 1948-1958 - * - 1838-1948 - 9 - 1928-1938 - * - 1918-1928 - 0 - 1908-1918 -A- 1893-1908
10
20
30
40
50
60
70
80
FIGURE 2. Cumulative prevalence of major depression according to age, with respondents stratified by year of birth. n = 18,352.
Phobic Disorder
Alcohol Abuse/Dependence
Schizophrenia
Panic Disorder 2.6%
3%
1958-1963
1948-1958
1938-1948
1918-1928
1908-1918
1893-1908
1928-1938
FIGURE 3. Cumulative prevalence of selected psychiatric disorders according to age, with respondents stratified by year of by birth, n variable = approximately 18,000.
1416
Simon and VonKorff 80%
18-34 yrs oM
>55 yrs oM
35-54 yre oM
lhara Sine* Ow»»
so
RGURE 4. Reported time since first onset of major depression, with respondents stratified by age at time of interview, n = 18,352.
cluster in the 10 years immediately prior to interview. The next analysis investigated the possibility of incomplete recall by examining individual psychiatric symptoms. As noted above for psychiatric diagnoses, lifetime prevalence of most symptoms should increase with age as new symptoms cumulate with those occurring earlier in life. We would predict that most psychiatric symptoms would show greatest increases in lifetime prevalence during early adulthood (the period of greatest incidence) followed by a more gradual rise in later years. Even if all incidence occurs during early years, lifetime prevalence should be constant across middle age, but should not decline. The sections of the Diagnostic Interview Schedule used to diagnose major depression, dysthymic disorder, phobias, panic disorder, mania, schizophrenia, and alcohol abuse/dependence routinely examine 68 individual symptoms. Not a single symptom shows the expected pattern of highest lifetime prevalence among the oldest respondents. Table 1 summarizes the pattern of lifetime prevalence for each symptom according to the age of the respondent. Peak lifetime prevalence typically occurs between ages 18 and 45 years. All but four symptoms show higher reported lifetime prevalence among those aged 25-44 years than among those aged 65
and over. This trend is most striking for symptoms of mania, most of which have peak reported lifetime prevalence among the youngest age group with striking declines after age 35 years. This trend is least marked for phobic symptoms which show the greatest variation in peak age group and the smallest decline with age. Symptoms of depression, schizophrenia, and alcohol abuse/dependence show roughly equivalent trends: peak reported lifetime prevalence between ages 25 and 44 years with rates only half as high among those over age 65. DISCUSSION
These findings suggest that trends seen in cross-sectional data may not accurately reflect changes in the epidemiology of psychiatric disorders across birth cohorts. First, the temporal trends initially described for depression occur in varying degrees across all diagnoses examined, with trends for prevalence of schizophrenia and panic disorder as great as those for depression. Lasch et al. (20) have used these data to demonstrate just as dramatic a trend in the prevalence of mania. These results seem inconsistent with available data on the prevalence of schizophrenia (21) as well as with data from prospective studies of the prevalence of depression discussed above. One must consider
Secular Trends mDepression
that the apparent trends seen for disorders other than depression may simply reflect the association between depression and most other disorders examined. This possibility was examined by recalculating the lifetime prevalence rates displayed in figure 1 after excluding all respondents with lifetime diagnosis of major depression. In such an analysis, prevalence rates decrease slightly due to the elimination of comorbid cases, but the striking age trends show little change (data not shown). Consequently, increasing prevalence of comorbid depression does not appear to explain the temporal trends seen for other disorders. Second, examination of the reported age of onset of depression yields unexpected results. Reported first episodes of major depression appear to cluster in the 10 years prior to interview regardless of respondent age. For older respondents, this places the period of greatest risk after age 50 years. This finding is inconsistent with prospective longitudinal data on the incidence of depression reported by Murphy et al. (22). This clustering of onset in recent years may represent a period effect (increased incidence confined to the prior decade) which is strong enough to overwhelm the typical age pattern of depression onset. Alternatively, this unexpected finding may be due to underreporting of remote depressive episodes or to "telescoping," a tendency to recall remote events as having occurred more recently. Third, almost every individual psychiatric symptom examined displays a similar unexpected decline in reported lifetime prevalence among older respondents. The uniformity of this finding suggests either a common effect of study methods or a marked increase in the prevalence of almost every psychiatric symptom among more recent cohorts. The appearance of a universal increase in psychiatric morbidity and clustering of morbidity in recent years might occur if crosssectional studies systematically undercount past psychiatric symptoms among the elderly. Possible causes of such a methods effect have been reviewed by Klerman and Weissman (23). First, older residents with histories of psychiatric illness may be selec-
1417
tively excluded from community samples. This might occur if psychiatric illness was associated with higher rates of mortality or placement in long-term care. Second, remote psychiatric symptoms may not be recalled or reported during a research interview. Older respondents could appear to have lower lifetime prevalence rates because of either the greater length of recall required or the greater stigma attached to psychiatric symptoms in earlier generations. Higher rates of mortality and/or institutionalization among older respondents with histories of psychiatric illness are unlikely to account for differences in prevalence of the magnitude noted above. For example, for major depression among females, reported lifetime prevalence among those aged 36-45 years (12.1 percent) is approximately eight times the lifetime prevalence in those aged 66-75 years (1.4 percent). In the general US population, females who were 40 years old in 1952 experienced an aggregate mortality of 26 percent by age 70 (24). If one makes the conservative assumption that no first episodes of major depression occur after age 40, then aggregate mortality from age 40 to age 70 among community residents with major depression would have to be approximately 92 percent (or an average mortality rate approximately eight times that in the general population) in order to account for these findings. Some investigations have noted excess mortality among community residents with a history of psychiatric disorder (25-27), while others find no difference after excluding subjects with organic brain syndromes (28, 29). Those studies that found an increase in mortality reported risk ratios of less than two (25-27). Similar calculations demonstrate that loss of elderly residents due to selective placement in longterm care is also unlikely to account for the large trends observed. Making the conservative assumptions of a doubling of mortality rate among community residents with history of major depression and no first depressive episodes after age 40 years, over 80 percent of the remaining residents with histories of depression would have to be institutionalized by age 70 to account for the
1418
Simon and VonKorff
TABLE 1. Reported lifetime prevalence of selected psychiatric symptoms by age at time of Interview* Prevalence (years) 25-44
s65
Ratio: 25-44/S65
Panic attack
Peak age group (years) 35-44
7.0
2.3
3.05
Phobias Public speaking Speaking with friends Eating In public Being atone Tunnels/bridges Crowds Closed places Heights Bugs, mice, snakes Leaving home Transportation Animals Water Storms
25-34 35-44 18-24 18-24 45-54 45-54 35-44 45-54 35-44 25-34 35-44 18-24 45-54 55-64
1.8 1.9 0.8 1.1 1.7 2.4 2.1 4.1 4.9 1.3 2.7 0.8 2.0 1.6
0.5 0.8 0.4 0.6 1.1 1.6 1.5 3.1 3.9 1.0 2.6 2.6 2.6
3.53 2.47 1.98 1.81 1.55 1.52 1.42 1.33 1.27 1.26 1.04 0.87 0.77 0.59
Mean prevalence
25-34
2.1
1.7
1.27
Depressive symptoms Hypersomnia Suicidal ideas Suicide attempt Decreased libido Guitt/worthlessness Poor concentration Weight gain Motor agitation Fatigue 2 years depressed Loss of appetite Weight loss Slowed thinking Wanted to die Motor retardation Thoughts of death 2 weeks depressed Insomnia
18-24 25-34 25-34 35-44 25-34 35-44 25-34 25-34 35-44 35-44 25-34 25-34 35-44 35-44 35-44 25-34 25-34 35-^4
10.3 13.7 3.9 7.2 13.3 13.7 16.4 7.0 19.1 10.8 11.1 8.5 8.1 5.5 23.9 31.2 19.2
2.3 4.0 1.2 2.3 4.5 5.2 6.3 3.4 10.3 3.9 6.4 6.7 5.1 5.8 4.0 18.7 24.6 17.6
4.38 3.44 3.12 3.10 2.97 2.63 2.61 2.08 1.84 1.74 1.69 1.65 1.65 1.40 1.37 1.28 1.27 1.09
Mean prevalence
25-34
12.8
7.2
1.79
Symptom
differences in the prevalence of depression described above (compared with a rate of less than 5 percent among all US females aged 66-75 years). Failure to report remote psychiatric symptoms may account for a large portion of the observed trends in lifetime prevalence. Older respondents may either fail to recall all previous episodes of psychiatric illness, leading to falsely low prevalence estimates, or report only more recent episodes, yielding a falsely high age of onset. Previous studies have ex-
6.8
0.9
amined the reliability of recall in structured interviews (30-33) and typically demonstrate good test-retest reliability both for presence of disorder and age of onset. Some findings from these reliability studies raise questions about recall over longer time periods. Bromet et al. (30) found poor reliability for temporal questions such as age of onset and number of episodes for subjects interviewed 18 months apart. In a test-retest study over a period of 3-6 years, Prusoff et al. (31) found that reliability of recall de-
Secular Trends in Depression
TABLE 1.
Continued Cumntmn
oympioni
Peak age Qroup (years)
Prevalence (years) 25-44
£65
Ratio: 25-^4/^65
Mania symptoms 1 week "high" Spending sprees Hypersexuafity Distract! ble Racing thoughts Rapid speech Overactive Grandiosity Decreased steep
18-24 18-24 25-34 18-24 18-24 18-24 18-24 18-24 18-24
2.1 4.1 10.3 8.8 5.5 3.4 3.4 6.2 8.5
0.4 1.1 2.4 3.0 2.2 1.3 1.5 3.1 4.5
5.22 3.83 2.94 2.94 2.56 2.55 2.24 2.01 1.92
Mean prevalence
18-24
5.8
2.2
2.70
35-44 25-34 25-34 35-44 18-24 25-34 25-34 35-^4 35-^4 35-44 35-44 35-44
0.9 0.7 0.7 1.0 0.5 1.8 2.2 1.1 0.8 1.3 2.2 0.6
0.2 0.2 0.2 0.4 0.2 0.8 1.1 0.6 0.5 0.9 1.7 0.5
4.88 4.87 3.52 2.85 2.70 2.31 2.01 1.89 1.71 1.50 1.29 1.27
35-44
1.1
0.6
1.97
25-34 18-24 45-54 45-54 18-24 45-54 18-24 35-44 35-44 35-44 45-54 35-44 35-44 45-54 45-54
18.6 12.1 2.6 9.2 13.8 7.6 12.7 4.8 6.8 12.8 1.1 8.4 3.7 4.1 3.8
5.1 4.2 1.0 3.5 5.8 3.3 6.6 3.9 7.5 0.7 6.0 2.8 4.0 5.2
3.63 2.91 2.64 2.62 2.39 2.28 1.92 1.91 1.77 1.69 1.58 1.40 1.35 1.04 0.73
35-44
8.2
4.1
1.98
Schizophrenia symptoms Feels mind Is read Thought Insertion Feels controlled Hears others thoughts Messages through television Feels watched Auditory hallucinations Feels plotted against Olfactory haflucinatJons Tactile haludnattons visual hallucinations Feels foRowed Mean prevalence Alcohol symptoms Fifth in 1 day Fights while drinking Job or school troubles Driving problems 7 drinks once a week Arrested Family objected Had to make rules Doctor was concerned Drinking was excessive Lost job, expelled 7 drinks/day for 2 weeks Spoke with doctor Couldn't stop drinking Drink before breakfast Mean prevalence 1
1419
2.5
Numbers of subjects varies, but is approximately 18,000 persons.
creased with time and that older respondents tended to report later age of onset at the second interview. Subjects with recent depression were more likely to recall previous depressive episodes. More recent reports have described poor reliability of lifetime recall among the study populations in
which secular trends in prevalence of depression were initially described. Among subjects in the Psychobiology of Depression Study; follow-up interviews 6 years later suggest that prior episodes of depression are frequently forgotten. More than 25 percent of those who reported a history of major
1420
Simon and VonKorff
depression at baseline interview did not recall such an episode at follow-up (34). Subjects with fewer depressive symptoms and those who did not receive treatment were more likely to "lose" lifetime diagnosis of depression between interviews. Among Epidemiologic Catchment Area respondents reinterviewed 1 year later, lifetime prevalence estimates including the 1-year prospective period were up to 80 percent higher than those calculated from retrospective reports alone (35). This dramatic difference suggests that retrospective cross-sectional assessment may markedly underestimate lifetime prevalence. Among those respondents with lifetime (but not current) history of depression at the initial interview and no recurrence during the 1-year follow-up interval, only 19 percent were positive for lifetime depression at follow-up. This finding suggests that past episodes of depression which are not followed by later "reminder" episodes may be frequently forgotten. Because current prevalence of depression is lower among older respondents, they are less likely to have suffered recent reminder episodes and more likely to underreport lifetime depression. A longitudinal study by Aneshensel et al. (36) also noted an influence of recent depression on recall or prior episodes. Few data are available on the accuracy of recall over decades, the time interval most relevant to the apparent trends described above. Wittchen et al. (32) compared age of onset of psychiatric disorder ascertained by the Diagnostic Interview Schedule with "consensus" ratings based on records and clinical interview. Agreement was low for anxiety and depressive disorders. Arguing against a methods effect is the fact that not all cross-sectional studies find evidence of a secular trend. As noted above, two cross-sectional studies found no evidence of increasing depression prevalence (8, 9). In those populations, however, lifetime prevalence rates among respondents of all ages were as high as those seen among the youngest Epidemiologic Catchment Area respondents. One interpretation is that those populations have long experienced unusually high depression rates and that rates
in other populations have only recently "caught up." Another possibility is that cross-sectional methods were more accurate among the elderly in these samples. Despite any questions raised here, the precision and reliability of the Diagnostic Interview Schedule (and other structured diagnostic interviews) represent important advances over earlier research methods. Unfortunately, these structured interview methods were not available when the longitudinal studies discussed above were conducted. Consequently, attempts to assess generational changes in the risk of psychiatric disorder must rely on two imperfect (and sometimes conflicting) sources of data. Longitudinal studies have typically relied on less precise questionnaire or unstructured interview methods to determine psychiatric diagnosis. Cross-sectional studies have used more precise instruments, but are subject to the difficulties with recall of remote events discussed above. Questions about the presence or magnitude of generational changes in risk of psychiatric disorder cannot be definitively settled using data currently available. Regardless of possible methodological difficulties, a four- to eightfold increase in risk of depression would have great public health significance. Alternative research approaches may help to resolve this issue. Because some of the social forces hypothesized to contribute to increasing prevalence of depression (urbanization, geographic mobility, and changes in family structure) show significant geographic variation, regional or international comparisons of psychiatric morbidity may help to assess the influence of these factors. Recently developed techniques to improve recall of past health events (37) include "decomposition" to sharpen recall of chronic or recurring events and construction of time lines to increase the accuracy of dating. These techniques could be applied to improve cross-sectional measures of lifetime psychiatric history. Continued cross-sectional monitoring of the current prevalence of mental disorder rates will allow assessment of any ongoing changes in prevalence rates.
Secular Trends in Depression
Given the data currently available, the presence or size of recent increases in the risk of depression will remain a question for scientific debate which cannot be definitively resolved. Data from the Epidemiologic Catchment Area Survey and similar studies can be interpreted as showing dramatic secular changes in depression risk. After the analyses described above, we conclude that methods effects may explain a significant portion of this apparent trend. Underreporting of remote psychiatric symptoms may reflect a normal adaptive response: bad times are best forgotten. Horace Greeley described a similar phenomenon over 100 years ago when he observed, "The illusion that times that were are better than times that are has probably pervaded all ages" (38).
REFERENCES 1. Klerman GL, Lavori PW, Rice J, et al. Birth-cohort trends in rates of major depressive disorder among relatives of patients with affective disorder. Arch Gen Psychiatry 1985,42:689-93. 2. Gershon ES, Hamovit JH, Guroff JJ, et al. Birthcohort changes in manic and depressive disorders in relatives of bipolar and schizoaffective patients. Arch Gen Psychiatry 1987;44:314-19. 3. Lavori PW, KJerman GL, Keller MB, et al. Ageperiod-cohort analysis of secular trends in onset of major depression: findings in siblings of patients with major affective disorder. J Psychiatr Res 1987; 21:23-35. 4. Wickramaratne PJ, Weissman MM, Leaf PJ, et al. Age, period, and cohort effects on the risk of major depression: results from five United States communities. J Clin Epidemiol 1989;42:333-43. 5. Wittchen HU. Contribution of epidemiological data to the classification of anxiety disorders. In: Hand I, Wittchen HU, eds. Panic phobias. Berlin, Germany: Springer-Verlag, 1986:18-27. 6. Joyce PR, Oakley-Browne MA, Wells JE, et al. Birth cohort trends in major depression: increasing rates and earlier onset in New Zealand. J Affective Disord 1990;! 8:83-9. 7. Bland R, Orn H. Family violence and psychiatric disorder. Can J Psychiatry 1986;31:129-37. 8. Lee KC, Kovak YS, Rhee H. The national epidemiological study of mental disorders in Korea. J Korean Med Sci 1987^2:19-34. 9. Canino GJ, Bird HR, Shrout PE. The prevalence of specific psychiatric disorders in Puerto Rico. Arch Gen Psychiatry 1987;44:727-35. 10. Murphy JM, Sobol AM, Neff RK, et al. Stability of prevalence: depression and anxiety disorders. Arch Gen Psychiatry 1984;41:990-7. 11. Veroff J, Donovan E, Kulka RA. The inner American: a self-portrait from 1957 to 1976. New York:
1421
Basic Books, 1981. 12. Hagnell O, Rorsman JLB, Osejo L. Are we entering an age of melancholy? Depressive illness in a prospective epidemiological study over 25 years: The Lundby Study, Sweden. Psychol Med 1982; 12: 279-89. 13. Srole L, Lagner TS, Michael ST, et al. Mental health in the metropolis: The Midtown Manhattan Study. New York: New York University Press, 1978. 14. Roberts RE, Lee ES, Roberts CR. Changes in prevalence of depressive symptoms in Alameda County. J Aging Health 1991,3:66-86. 15. Regier DA, Myers JK, Kramer M. Historical context, major objectives, and study design. In: Eaton WW, Kessler L, eds. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:3-19. 16. Holzer C, Spitznagel E, Jordan K, et al. Sampling the household population. In: Eaton WW, Kessler L. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:23-48. 17. Robins LN, Helzer JE, Croughan J, et al. The National Institute of Mental Health Diagnostic Interview Schedule: its history, characteristics, and validity. Arch Gen Psychiatry 1981 ;38:381-9. 18. Diagnostic and statistical manual of mental disorders. 3rd edition. Washington, DC: American Psychiatric Association, 1980. 19. Boyd JH, Robins LN, Holzer C, et al. Making diagnoses from DIS data. In: Eaton WW, Kessler L. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:209-31. 20. Lasch K, Weissman M, Wickramaratne P, et al. Birth-cohort changes in the rates of mania. Psychiatry Res 199O;33:31-7. 21. Torrey EF. Prevalence studies in schizophrenia. Br J Psychiatry 1987; 150:598-608. 22. Murphy JE, Oliver DC, Monson RR, et al. Incidence of depression and anxiety: The Stirling County Study. Am J Public Health 1988;78: 534-40. 23. KJerman GL, Weissman MM. Increasing rates of depression. JAMA 1989;261:2229-35. 24. National Center for Health Statistics. Vital Statistics of the United States, 1960, 1970, 1980. Hyattsville, MD: National Center for Health Statistics, 1960, 1972, 1982. 25. Murphy JM, Monson RR, Olivier DC, et al. Affective disorders and mortality: a general population survey. Arch Gen Psychiatry 1987;44:473-80. 26. Murphy JE, Monson RR, Olivier DC, et al. Mortality risk and psychiatric disorders: results of a general physician survey. Soc Psychiatry Psychiatr Epidemiol 1989;24:134-42. 27. Babigian HM, Odoroff CL. The mortality experience of a population with psychiatric illness. Am J Psychiatry 1969; 126:470-80. 28. Singer E, Garfinkel R, Cohen SM, et al. Mortality and mental health: evidence from the Midtown Manhattan Study. Soc Sci Med 1976; 10:517-25. 29. Roberts RE, Kaplan GA, Camacho TC. Psychological distress and mortality: evidence from the Alameda County Study. Soc Sci Med 1990;5:527-36. 30. Bromet EJ, Dunn LO, Connell MM, et al. Long-
1422
31.
32. 33.
34.
Simon and VonKorff
term reliability of diagnosing lifetime major depression in a community sample. Arch Gen Psychiatry l986;43:435-40. Prusoff BA, Merikangas KR, Weissman MM. Lifetime prevalence and age of onset of psychiatric disorders: recall 4 years later. J Psychiatr Res 1988; 22:107-17. Wittchen HU, Burke JD, Semler G, et al. Recall and dating of psychiatric symptoms. Arch Gen Psychiatry 1989;46:437-43. Semler G, Wittchen HU, Joschke K, et al. Testretest reliability of a standardized psychiatric interview (DIS/CIDI). Eur Arch Psychiatry Neurol Sci 1987^36:214-22. Rice JP, Endicott J, Lavori PW. Stability in relatives of depressives. Presented at the 144th Annual Meeting of the American Psychiatric Association,
New Orleans, Louisiana, May 1991. 35. Regier DA, Rae DS, Narrow W. Stability of diagnoses in the Epidemiologic Catchment Area study. Presented at the 144th Annual Meeting of the American Psychiatric Association, New Orleans, Louisiana, May 1991. 36. Aneshensel CS, Estrada AL, Hansell MJ, et al. Social psychological aspects of reporting behavior lifetime depressive episode reports. J Health Soc Behav 1987^28:232-46. 37. Means B, Nigam A, Zarrow M, et al. Autobiographical memory for health-related events. Washington, DC: Department of Health and Human Services, 1989. (Vital and health statistics, Series 6, no. 2) (DHHS publication no. (PHS 89-1077). 38. Greeley H. The American conflict. Hartford, CT: ODCase, 1864.
VoL 135, No. 12 PnntBd in U.S.A.
Reevaluation of Secular Trends in Depression Rates
Gregory E. Simon and Michael VonKorff
Results of numerous community surveys of psychiatric illness suggest a striking change in the occurrence of depression, with younger generations experiencing higher lifetime risk and earlier age of onset. Data from the National Institute of Mental Health Epidemiologic Catchment Area Survey (a cross-sectional survey of psychiatric morbidity in five US communities conducted between 1980 and 1984) were reexamined for evidence of methods effects which might contribute to these unexpected findings. A pattern of higher lifetime risk and earlier age of onset among recent birth cohorts was observed for every psychiatric disorder examined, with schizophrenia, major depression, and panic disorder showing equally strong trends. For respondents of all ages, reported first onset of major depression clustered in the 10-year period prior to the study interview, in contrast to the expectation that older respondents would report onset in early adulthood. Examination of individual psychiatric symptoms revealed a nearly universal pattern of decreasing lifetime prevalence among older respondents, a reversal of the expected accumulation of lifetime symptoms with age. These findings suggest that effects of study methods may contribute to the apparent temporal trends in prevalence of depression and that cross-sectional surveys may underestimate lifetime psychiatric morbidity among older respondents. Generational changes in the lifetime risk of depression or other psychiatric disorders may not be reliably assessed by crosssectional survey data. Am J Epidemiol 1992;135:1411-22. affective disorders; depression; epidemiologic methods; life tables
Several recent reports have described a dramatic secular trend in the lifetime prevalence of major depression. In 1985, Klerman et al. (1) reported data on the lifetime psychiatric histories of family members studied in National Institute of Mental Health Collaborative Program on the Psychobiology of Depression. Within this group at high risk for affective disorder, the reported lifetime
prevalence of major depression showed a large and unexpected decline among older subjects. Examination of lifetime risk for depression according to year of birth appeared to demonstrate that more recent birth cohorts experienced markedly higher prevalence and earlier age of onset for depressive illness. This report was followed by similar findings among community samples
Received for publication May 30,1991, and in final form October 28, 1991. Abbreviation: DSM-IU, Diagnostic and Statistical Manual, Third Edition. From the Center for Health Studies, Group Health Cooperative, Seattle, WA, and the Department of Psychiatry and Behavioral Sciences, University of Washington, Seattle, WA. Reprint requests to: Dr. Gregory Simon, Center for Health Studies, 1730 Minor Avenue, Seattle, WA 98101. This research is based on the National Institute of Mental Health Epidemiotogjc Catchment Area Survey Public Use
Data Set, Wave 1 Household Sample The catchment area survey was a senes of five epidemiologic research studies performed by five independent research teams in collaboration with National Institute of Mental Health, Division of Biometry and Epidemiology. The five research studies were carried out by Yale University, The Johns Hopkins University, Washington University, Duke University, and the University of California at Los Angeles. Robert Jamieson assisted with data preparation and translation Drs. John Lee and William Barlow provided comments on earlier versions of the manuscript
1411
1412
Simon and VonKorff
in the United States (2-4), Germany (5), New Zealand (6), and Canada (7) that included over 27,000 subjects. While surveys in Korea (8) and Puerto Rico (9) have not found such an effect, the bulk of these studies suggest that lifetime risk of major depression is approximately five times as high among recent generations as among those born at the turn of the century. These investigations were typically cross-sectional surveys of nonpatient populations which relied on structured interviews to assess lifetime psychiatric history. Longitudinal studies of depression prevalence do not find such a striking cohort effect over the period when this change in risk should have occurred. These studies have used either repeated cross-sectional sampling or true cohort designs. Of the two repeat cross-sectional studies, the Stirling County Study (10) found prevalence of anxiety and depressive disorders to be unchanged from 1952 to 1970, while the US National Sample Study found an increase in symptoms of psychologic distress from 1957 to 1976 (11). Among studies using the cohort method, the Lundby Study (12) demonstrated an increase in incidence of depression between 1947 and 1972 which was confined to disorders of mild and moderate severity, while the Midtown Manhattan Study (13) found a decrease in prevalence of psychiatric disorder from 1954 to 1974. Analysis of longitudinal data on depression from the Alameda County Study (14) found some evidence for a transient increase in the prevalence of depression during the 1970s, but prevalence rates were consistently highest among the oldest cohorts studied. None of these studies reported an increase in risk of the magnitude suggested by crosssectional studies. These disparate findings suggest the possibility that some methodological factor common to cross-sectional studies may contribute to this apparent trend. Crosssectional studies which rely on retrospective assessment of lifetime history typically find evidence of dramatic increases in prevalence, while longitudinal studies relying on current symptom reports yield mixed re-
sults. Given the time and expense required for community prevalence studies, further analysis of existing cross-sectional data may be the most practical means to examine the potential contribution of methods effects. Data from the Epidemiologic Catchment Area Survey are well suited to such an examination. Wickramaratne et al. (4) have previously described a marked temporal trend in prevalence of depression in this sample. The large sample size allows relatively stable estimates for the prevalence of psychiatric diagnoses within narrow age strata. The wide range of diagnoses examined allows study of age patterns for a range of psychiatric diagnoses as well as individual psychiatric symptoms. MATERIALS AND METHODS Epidemiologic Catchment Area Survey design
These analyses utilized public use tapes of data from wave 1 of the Epidemiologic Catchment Area Survey, a nationwide survey of psychiatric morbidity and health service utilization carried out under cooperative agreement between the National Institute of Mental Health and five university sites: Yale University, New Haven, Connecticut; The Johns Hopkins University, Baltimore, Maryland; Washington University, St. Louis, Missouri; Duke University, Durham, North Carolina; and the University of California at Los Angeles, Los Angeles, California (15). Each site sampled subjects from one or more previously designated Community Mental Health Center catchment areas, with geographic population areas of 75,000-250,000. Over 3,000 community residents and 500 institutional residents were studied at each site. This analysis considers only community residents. Data collection occurred between 1980 and 1984. Each site used a multistage probability sampling procedure to select households from the eligible geographic area (16). This procedure varied among sites, with some sites directly sampling households from a list of all of those eligible and others using an intermediate selection of geographic sub-
Secular Trends in Depression
units prior to selection of households. Within each household, all residents aged 18 years or older without other usual residence were enumerated. The Baltimore site sampled one resident aged 18-64 years in each household plus all additional residents over age 65. Other sites randomly selected a single resident, with the Durham and New Haven sites oversampling elderly residents. Selection of catchment areas or sampling units allowed intentional oversampling of blacks in St. Louis and Hispanics in Los Angeles. Overall response rates ranged across sites from 68 to 79 percent. Respondents had varying probabilities of selection because of intentional oversampling of certain groups, varying procedures for household selection, and selection of one resident regardless of household size. Each respondent was assigned a sampling weight inversely related to his or her probability of selection. Data collection instruments
Psychiatric morbidity was assessed using the Diagnostic Interview Schedule, a structured interview developed by the National Institute of Mental Health and the Washington University School of Medicine (17). The interview uses a structured question sequence to assess the presence of diagnostic criteria from the Diagnostic and Statistical Manual, Third Edition (DSM-III) (18) by the American Psychiatric Association. Questioning follows a specified probe sequence which allows for minimum interviewer discretion. Respondents are first asked whether they have ever experienced some potential psychiatric symptom. For each positive response, interviewers follow a series of questions designed to determine whether the symptom was sufficiently severe and whether alternative explanations (medical illness or drug or alcohol use) might explain every symptomatic episode. Symptoms of sufficient severity for which alternative explanations have been excluded are classified as presumptive psychiatric symptoms. Computer algorithms allowed translation of symptoms into DSM-III diagnoses. The ver-
1413
sion of the Diagnostic Interview Schedule included in the catchment area study allowed diagnoses of mania, major depression, bipolar disorder, alcohol abuse or dependence, drug abuse or dependence, schizophrenia, obsessive-compulsive disorder, phobia, somatization disorder, panic disorder, and antisocial personality. Data analysis
Data tapes containing the Wave 1 Household Sample of the Epidemiologic Catchment Area Survey were obtained from National Technical Information Service, Springfield, Virginia. The original SAS data file was translated into a form readable by the SPSS-X statistical package. Analyses were conducted using SPSS-X software (SPSS Inc., Chicago, Illinois) Analyses of psychiatric diagnoses utilized diagnoses generated by computer from Diagnostic Interview Schedule scoring algorithms (19). Hierarchical diagnostic exclusions of DSM-III (e.g., diagnosis of major depression precludes diagnosis of panic disorder) were not applied. All analyses incorporated sampling weights described above. RESULTS
Analysis first replicated previously reported findings of higher lifetime prevalence and earlier age of onset for major depression among younger respondents. Figure 1 shows the. reported lifetime prevalence of major depression (and other diagnoses) according to age of respondent at time of interview. Lifetime prevalence appears dramatically lower among older respondents despite longer periods of risk. Because lifetime prevalence should accumulate with age, any apparent decrease among older respondents implies a significant temporal trend (either a period effect in which risk increases universally with time or a birth cohort effect in which those born in recent years show increased risk). Figure 2 divides respondents into seven cohorts according to year of birth. For each cohort, the prevalence of depression at 5-year age intervals was calculated as
1414
Simon and VonKorff
20%
Alcohol Phobia Dtpr«««lon Pinlc Dltordtr Schizophrenia
18-26
26-36
38-45 46-55 69-65 Ag« at Interview
66-75
76-85
FIGURE 1. Reported lifetime prevalence of selected psychiatric disorders according to the respondent's age at the time of interview, n = 18,271.
the portion of respondents with lifetime history of major depression who report onset prior to that age. Here the dramatic differences in prevalence between birth cohorts are apparent. Older respondents (those born in earlier cohorts) display markedly lower lifetime prevalence and later age of onset as previously described. This dramatic temporal trend, however, was not specific to major depression. Figure 1 displays lifetime prevalence according to age at interview for other diagnoses. All diagnoses examined show a similar pattern of lower reported lifetime prevalence among older respondents. Proportional differences in lifetime prevalence by age are most marked for schizophrenia, major depression, and panic disorder. Within those three disorders, lifetime prevalence among those aged 36-45 is six to eight times as high as among those over age 65 years. Relative differences in lifetime prevalence with age are smaller for alcohol diagnoses and phobias; younger respondents report lifetime prevalence two to three times higher than those over age 65 years. As noted above, however, any downward slope in these lifetime prevalence curves implies a significant temporal effect. Figure 3 presents cumulative prevalence stratified by birth cohort (analogous to figure 2) for each of these
diagnoses other than major depression. All disorders examined show a remarkably similar pattern. More recent cohorts all show a much steeper rise to much higher peak prevalence rates despite shorter periods of risk. As with the examination of lifetime prevalence rates above, these differences are proportionally greatest for schizophrenia and panic disorder. Phobias and alcohol diagnoses show the same pattern, but relative differences between cohorts are less marked. The consistency of these unexpected findings across diagnoses raises the possibility that some common effect of study methods may contribute to this apparent trend. Two additional analyses were used to examine this possibility. The first of these examined the reported time since onset of illness among respondents with a lifetime history of major depression. This analysis considered only respondents who reported a lifetime history of major depression. Time since onset of depression was calculated as the difference between the respondents' age at the time of interview and the reported age of onset of the first depressive episode. Figure 4 divides all cases of lifetime major depression into three age strata and displays the percentage of cases reporting time since onset of 0-10 years, 11-20 years, etc. For all three strata, onset of depression appears to
Secular Trends in Depression
1415
10%
a% —
1958-1883
-+- 1948-1958 - * - 1838-1948 - 9 - 1928-1938 - * - 1918-1928 - 0 - 1908-1918 -A- 1893-1908
10
20
30
40
50
60
70
80
FIGURE 2. Cumulative prevalence of major depression according to age, with respondents stratified by year of birth. n = 18,352.
Phobic Disorder
Alcohol Abuse/Dependence
Schizophrenia
Panic Disorder 2.6%
3%
1958-1963
1948-1958
1938-1948
1918-1928
1908-1918
1893-1908
1928-1938
FIGURE 3. Cumulative prevalence of selected psychiatric disorders according to age, with respondents stratified by year of by birth, n variable = approximately 18,000.
1416
Simon and VonKorff 80%
18-34 yrs oM
>55 yrs oM
35-54 yre oM
lhara Sine* Ow»»
so
RGURE 4. Reported time since first onset of major depression, with respondents stratified by age at time of interview, n = 18,352.
cluster in the 10 years immediately prior to interview. The next analysis investigated the possibility of incomplete recall by examining individual psychiatric symptoms. As noted above for psychiatric diagnoses, lifetime prevalence of most symptoms should increase with age as new symptoms cumulate with those occurring earlier in life. We would predict that most psychiatric symptoms would show greatest increases in lifetime prevalence during early adulthood (the period of greatest incidence) followed by a more gradual rise in later years. Even if all incidence occurs during early years, lifetime prevalence should be constant across middle age, but should not decline. The sections of the Diagnostic Interview Schedule used to diagnose major depression, dysthymic disorder, phobias, panic disorder, mania, schizophrenia, and alcohol abuse/dependence routinely examine 68 individual symptoms. Not a single symptom shows the expected pattern of highest lifetime prevalence among the oldest respondents. Table 1 summarizes the pattern of lifetime prevalence for each symptom according to the age of the respondent. Peak lifetime prevalence typically occurs between ages 18 and 45 years. All but four symptoms show higher reported lifetime prevalence among those aged 25-44 years than among those aged 65
and over. This trend is most striking for symptoms of mania, most of which have peak reported lifetime prevalence among the youngest age group with striking declines after age 35 years. This trend is least marked for phobic symptoms which show the greatest variation in peak age group and the smallest decline with age. Symptoms of depression, schizophrenia, and alcohol abuse/dependence show roughly equivalent trends: peak reported lifetime prevalence between ages 25 and 44 years with rates only half as high among those over age 65. DISCUSSION
These findings suggest that trends seen in cross-sectional data may not accurately reflect changes in the epidemiology of psychiatric disorders across birth cohorts. First, the temporal trends initially described for depression occur in varying degrees across all diagnoses examined, with trends for prevalence of schizophrenia and panic disorder as great as those for depression. Lasch et al. (20) have used these data to demonstrate just as dramatic a trend in the prevalence of mania. These results seem inconsistent with available data on the prevalence of schizophrenia (21) as well as with data from prospective studies of the prevalence of depression discussed above. One must consider
Secular Trends mDepression
that the apparent trends seen for disorders other than depression may simply reflect the association between depression and most other disorders examined. This possibility was examined by recalculating the lifetime prevalence rates displayed in figure 1 after excluding all respondents with lifetime diagnosis of major depression. In such an analysis, prevalence rates decrease slightly due to the elimination of comorbid cases, but the striking age trends show little change (data not shown). Consequently, increasing prevalence of comorbid depression does not appear to explain the temporal trends seen for other disorders. Second, examination of the reported age of onset of depression yields unexpected results. Reported first episodes of major depression appear to cluster in the 10 years prior to interview regardless of respondent age. For older respondents, this places the period of greatest risk after age 50 years. This finding is inconsistent with prospective longitudinal data on the incidence of depression reported by Murphy et al. (22). This clustering of onset in recent years may represent a period effect (increased incidence confined to the prior decade) which is strong enough to overwhelm the typical age pattern of depression onset. Alternatively, this unexpected finding may be due to underreporting of remote depressive episodes or to "telescoping," a tendency to recall remote events as having occurred more recently. Third, almost every individual psychiatric symptom examined displays a similar unexpected decline in reported lifetime prevalence among older respondents. The uniformity of this finding suggests either a common effect of study methods or a marked increase in the prevalence of almost every psychiatric symptom among more recent cohorts. The appearance of a universal increase in psychiatric morbidity and clustering of morbidity in recent years might occur if crosssectional studies systematically undercount past psychiatric symptoms among the elderly. Possible causes of such a methods effect have been reviewed by Klerman and Weissman (23). First, older residents with histories of psychiatric illness may be selec-
1417
tively excluded from community samples. This might occur if psychiatric illness was associated with higher rates of mortality or placement in long-term care. Second, remote psychiatric symptoms may not be recalled or reported during a research interview. Older respondents could appear to have lower lifetime prevalence rates because of either the greater length of recall required or the greater stigma attached to psychiatric symptoms in earlier generations. Higher rates of mortality and/or institutionalization among older respondents with histories of psychiatric illness are unlikely to account for differences in prevalence of the magnitude noted above. For example, for major depression among females, reported lifetime prevalence among those aged 36-45 years (12.1 percent) is approximately eight times the lifetime prevalence in those aged 66-75 years (1.4 percent). In the general US population, females who were 40 years old in 1952 experienced an aggregate mortality of 26 percent by age 70 (24). If one makes the conservative assumption that no first episodes of major depression occur after age 40, then aggregate mortality from age 40 to age 70 among community residents with major depression would have to be approximately 92 percent (or an average mortality rate approximately eight times that in the general population) in order to account for these findings. Some investigations have noted excess mortality among community residents with a history of psychiatric disorder (25-27), while others find no difference after excluding subjects with organic brain syndromes (28, 29). Those studies that found an increase in mortality reported risk ratios of less than two (25-27). Similar calculations demonstrate that loss of elderly residents due to selective placement in longterm care is also unlikely to account for the large trends observed. Making the conservative assumptions of a doubling of mortality rate among community residents with history of major depression and no first depressive episodes after age 40 years, over 80 percent of the remaining residents with histories of depression would have to be institutionalized by age 70 to account for the
1418
Simon and VonKorff
TABLE 1. Reported lifetime prevalence of selected psychiatric symptoms by age at time of Interview* Prevalence (years) 25-44
s65
Ratio: 25-44/S65
Panic attack
Peak age group (years) 35-44
7.0
2.3
3.05
Phobias Public speaking Speaking with friends Eating In public Being atone Tunnels/bridges Crowds Closed places Heights Bugs, mice, snakes Leaving home Transportation Animals Water Storms
25-34 35-44 18-24 18-24 45-54 45-54 35-44 45-54 35-44 25-34 35-44 18-24 45-54 55-64
1.8 1.9 0.8 1.1 1.7 2.4 2.1 4.1 4.9 1.3 2.7 0.8 2.0 1.6
0.5 0.8 0.4 0.6 1.1 1.6 1.5 3.1 3.9 1.0 2.6 2.6 2.6
3.53 2.47 1.98 1.81 1.55 1.52 1.42 1.33 1.27 1.26 1.04 0.87 0.77 0.59
Mean prevalence
25-34
2.1
1.7
1.27
Depressive symptoms Hypersomnia Suicidal ideas Suicide attempt Decreased libido Guitt/worthlessness Poor concentration Weight gain Motor agitation Fatigue 2 years depressed Loss of appetite Weight loss Slowed thinking Wanted to die Motor retardation Thoughts of death 2 weeks depressed Insomnia
18-24 25-34 25-34 35-44 25-34 35-44 25-34 25-34 35-44 35-44 25-34 25-34 35-44 35-44 35-44 25-34 25-34 35-^4
10.3 13.7 3.9 7.2 13.3 13.7 16.4 7.0 19.1 10.8 11.1 8.5 8.1 5.5 23.9 31.2 19.2
2.3 4.0 1.2 2.3 4.5 5.2 6.3 3.4 10.3 3.9 6.4 6.7 5.1 5.8 4.0 18.7 24.6 17.6
4.38 3.44 3.12 3.10 2.97 2.63 2.61 2.08 1.84 1.74 1.69 1.65 1.65 1.40 1.37 1.28 1.27 1.09
Mean prevalence
25-34
12.8
7.2
1.79
Symptom
differences in the prevalence of depression described above (compared with a rate of less than 5 percent among all US females aged 66-75 years). Failure to report remote psychiatric symptoms may account for a large portion of the observed trends in lifetime prevalence. Older respondents may either fail to recall all previous episodes of psychiatric illness, leading to falsely low prevalence estimates, or report only more recent episodes, yielding a falsely high age of onset. Previous studies have ex-
6.8
0.9
amined the reliability of recall in structured interviews (30-33) and typically demonstrate good test-retest reliability both for presence of disorder and age of onset. Some findings from these reliability studies raise questions about recall over longer time periods. Bromet et al. (30) found poor reliability for temporal questions such as age of onset and number of episodes for subjects interviewed 18 months apart. In a test-retest study over a period of 3-6 years, Prusoff et al. (31) found that reliability of recall de-
Secular Trends in Depression
TABLE 1.
Continued Cumntmn
oympioni
Peak age Qroup (years)
Prevalence (years) 25-44
£65
Ratio: 25-^4/^65
Mania symptoms 1 week "high" Spending sprees Hypersexuafity Distract! ble Racing thoughts Rapid speech Overactive Grandiosity Decreased steep
18-24 18-24 25-34 18-24 18-24 18-24 18-24 18-24 18-24
2.1 4.1 10.3 8.8 5.5 3.4 3.4 6.2 8.5
0.4 1.1 2.4 3.0 2.2 1.3 1.5 3.1 4.5
5.22 3.83 2.94 2.94 2.56 2.55 2.24 2.01 1.92
Mean prevalence
18-24
5.8
2.2
2.70
35-44 25-34 25-34 35-44 18-24 25-34 25-34 35-^4 35-^4 35-44 35-44 35-44
0.9 0.7 0.7 1.0 0.5 1.8 2.2 1.1 0.8 1.3 2.2 0.6
0.2 0.2 0.2 0.4 0.2 0.8 1.1 0.6 0.5 0.9 1.7 0.5
4.88 4.87 3.52 2.85 2.70 2.31 2.01 1.89 1.71 1.50 1.29 1.27
35-44
1.1
0.6
1.97
25-34 18-24 45-54 45-54 18-24 45-54 18-24 35-44 35-44 35-44 45-54 35-44 35-44 45-54 45-54
18.6 12.1 2.6 9.2 13.8 7.6 12.7 4.8 6.8 12.8 1.1 8.4 3.7 4.1 3.8
5.1 4.2 1.0 3.5 5.8 3.3 6.6 3.9 7.5 0.7 6.0 2.8 4.0 5.2
3.63 2.91 2.64 2.62 2.39 2.28 1.92 1.91 1.77 1.69 1.58 1.40 1.35 1.04 0.73
35-44
8.2
4.1
1.98
Schizophrenia symptoms Feels mind Is read Thought Insertion Feels controlled Hears others thoughts Messages through television Feels watched Auditory hallucinations Feels plotted against Olfactory haflucinatJons Tactile haludnattons visual hallucinations Feels foRowed Mean prevalence Alcohol symptoms Fifth in 1 day Fights while drinking Job or school troubles Driving problems 7 drinks once a week Arrested Family objected Had to make rules Doctor was concerned Drinking was excessive Lost job, expelled 7 drinks/day for 2 weeks Spoke with doctor Couldn't stop drinking Drink before breakfast Mean prevalence 1
1419
2.5
Numbers of subjects varies, but is approximately 18,000 persons.
creased with time and that older respondents tended to report later age of onset at the second interview. Subjects with recent depression were more likely to recall previous depressive episodes. More recent reports have described poor reliability of lifetime recall among the study populations in
which secular trends in prevalence of depression were initially described. Among subjects in the Psychobiology of Depression Study; follow-up interviews 6 years later suggest that prior episodes of depression are frequently forgotten. More than 25 percent of those who reported a history of major
1420
Simon and VonKorff
depression at baseline interview did not recall such an episode at follow-up (34). Subjects with fewer depressive symptoms and those who did not receive treatment were more likely to "lose" lifetime diagnosis of depression between interviews. Among Epidemiologic Catchment Area respondents reinterviewed 1 year later, lifetime prevalence estimates including the 1-year prospective period were up to 80 percent higher than those calculated from retrospective reports alone (35). This dramatic difference suggests that retrospective cross-sectional assessment may markedly underestimate lifetime prevalence. Among those respondents with lifetime (but not current) history of depression at the initial interview and no recurrence during the 1-year follow-up interval, only 19 percent were positive for lifetime depression at follow-up. This finding suggests that past episodes of depression which are not followed by later "reminder" episodes may be frequently forgotten. Because current prevalence of depression is lower among older respondents, they are less likely to have suffered recent reminder episodes and more likely to underreport lifetime depression. A longitudinal study by Aneshensel et al. (36) also noted an influence of recent depression on recall or prior episodes. Few data are available on the accuracy of recall over decades, the time interval most relevant to the apparent trends described above. Wittchen et al. (32) compared age of onset of psychiatric disorder ascertained by the Diagnostic Interview Schedule with "consensus" ratings based on records and clinical interview. Agreement was low for anxiety and depressive disorders. Arguing against a methods effect is the fact that not all cross-sectional studies find evidence of a secular trend. As noted above, two cross-sectional studies found no evidence of increasing depression prevalence (8, 9). In those populations, however, lifetime prevalence rates among respondents of all ages were as high as those seen among the youngest Epidemiologic Catchment Area respondents. One interpretation is that those populations have long experienced unusually high depression rates and that rates
in other populations have only recently "caught up." Another possibility is that cross-sectional methods were more accurate among the elderly in these samples. Despite any questions raised here, the precision and reliability of the Diagnostic Interview Schedule (and other structured diagnostic interviews) represent important advances over earlier research methods. Unfortunately, these structured interview methods were not available when the longitudinal studies discussed above were conducted. Consequently, attempts to assess generational changes in the risk of psychiatric disorder must rely on two imperfect (and sometimes conflicting) sources of data. Longitudinal studies have typically relied on less precise questionnaire or unstructured interview methods to determine psychiatric diagnosis. Cross-sectional studies have used more precise instruments, but are subject to the difficulties with recall of remote events discussed above. Questions about the presence or magnitude of generational changes in risk of psychiatric disorder cannot be definitively settled using data currently available. Regardless of possible methodological difficulties, a four- to eightfold increase in risk of depression would have great public health significance. Alternative research approaches may help to resolve this issue. Because some of the social forces hypothesized to contribute to increasing prevalence of depression (urbanization, geographic mobility, and changes in family structure) show significant geographic variation, regional or international comparisons of psychiatric morbidity may help to assess the influence of these factors. Recently developed techniques to improve recall of past health events (37) include "decomposition" to sharpen recall of chronic or recurring events and construction of time lines to increase the accuracy of dating. These techniques could be applied to improve cross-sectional measures of lifetime psychiatric history. Continued cross-sectional monitoring of the current prevalence of mental disorder rates will allow assessment of any ongoing changes in prevalence rates.
Secular Trends in Depression
Given the data currently available, the presence or size of recent increases in the risk of depression will remain a question for scientific debate which cannot be definitively resolved. Data from the Epidemiologic Catchment Area Survey and similar studies can be interpreted as showing dramatic secular changes in depression risk. After the analyses described above, we conclude that methods effects may explain a significant portion of this apparent trend. Underreporting of remote psychiatric symptoms may reflect a normal adaptive response: bad times are best forgotten. Horace Greeley described a similar phenomenon over 100 years ago when he observed, "The illusion that times that were are better than times that are has probably pervaded all ages" (38).
REFERENCES 1. Klerman GL, Lavori PW, Rice J, et al. Birth-cohort trends in rates of major depressive disorder among relatives of patients with affective disorder. Arch Gen Psychiatry 1985,42:689-93. 2. Gershon ES, Hamovit JH, Guroff JJ, et al. Birthcohort changes in manic and depressive disorders in relatives of bipolar and schizoaffective patients. Arch Gen Psychiatry 1987;44:314-19. 3. Lavori PW, KJerman GL, Keller MB, et al. Ageperiod-cohort analysis of secular trends in onset of major depression: findings in siblings of patients with major affective disorder. J Psychiatr Res 1987; 21:23-35. 4. Wickramaratne PJ, Weissman MM, Leaf PJ, et al. Age, period, and cohort effects on the risk of major depression: results from five United States communities. J Clin Epidemiol 1989;42:333-43. 5. Wittchen HU. Contribution of epidemiological data to the classification of anxiety disorders. In: Hand I, Wittchen HU, eds. Panic phobias. Berlin, Germany: Springer-Verlag, 1986:18-27. 6. Joyce PR, Oakley-Browne MA, Wells JE, et al. Birth cohort trends in major depression: increasing rates and earlier onset in New Zealand. J Affective Disord 1990;! 8:83-9. 7. Bland R, Orn H. Family violence and psychiatric disorder. Can J Psychiatry 1986;31:129-37. 8. Lee KC, Kovak YS, Rhee H. The national epidemiological study of mental disorders in Korea. J Korean Med Sci 1987^2:19-34. 9. Canino GJ, Bird HR, Shrout PE. The prevalence of specific psychiatric disorders in Puerto Rico. Arch Gen Psychiatry 1987;44:727-35. 10. Murphy JM, Sobol AM, Neff RK, et al. Stability of prevalence: depression and anxiety disorders. Arch Gen Psychiatry 1984;41:990-7. 11. Veroff J, Donovan E, Kulka RA. The inner American: a self-portrait from 1957 to 1976. New York:
1421
Basic Books, 1981. 12. Hagnell O, Rorsman JLB, Osejo L. Are we entering an age of melancholy? Depressive illness in a prospective epidemiological study over 25 years: The Lundby Study, Sweden. Psychol Med 1982; 12: 279-89. 13. Srole L, Lagner TS, Michael ST, et al. Mental health in the metropolis: The Midtown Manhattan Study. New York: New York University Press, 1978. 14. Roberts RE, Lee ES, Roberts CR. Changes in prevalence of depressive symptoms in Alameda County. J Aging Health 1991,3:66-86. 15. Regier DA, Myers JK, Kramer M. Historical context, major objectives, and study design. In: Eaton WW, Kessler L, eds. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:3-19. 16. Holzer C, Spitznagel E, Jordan K, et al. Sampling the household population. In: Eaton WW, Kessler L. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:23-48. 17. Robins LN, Helzer JE, Croughan J, et al. The National Institute of Mental Health Diagnostic Interview Schedule: its history, characteristics, and validity. Arch Gen Psychiatry 1981 ;38:381-9. 18. Diagnostic and statistical manual of mental disorders. 3rd edition. Washington, DC: American Psychiatric Association, 1980. 19. Boyd JH, Robins LN, Holzer C, et al. Making diagnoses from DIS data. In: Eaton WW, Kessler L. Epidemiologic field methods in psychiatry. The NIMH Epidemiologic Catchment Area Program. Orlando, FL: Academic Press, Inc., 1985:209-31. 20. Lasch K, Weissman M, Wickramaratne P, et al. Birth-cohort changes in the rates of mania. Psychiatry Res 199O;33:31-7. 21. Torrey EF. Prevalence studies in schizophrenia. Br J Psychiatry 1987; 150:598-608. 22. Murphy JE, Oliver DC, Monson RR, et al. Incidence of depression and anxiety: The Stirling County Study. Am J Public Health 1988;78: 534-40. 23. KJerman GL, Weissman MM. Increasing rates of depression. JAMA 1989;261:2229-35. 24. National Center for Health Statistics. Vital Statistics of the United States, 1960, 1970, 1980. Hyattsville, MD: National Center for Health Statistics, 1960, 1972, 1982. 25. Murphy JM, Monson RR, Olivier DC, et al. Affective disorders and mortality: a general population survey. Arch Gen Psychiatry 1987;44:473-80. 26. Murphy JE, Monson RR, Olivier DC, et al. Mortality risk and psychiatric disorders: results of a general physician survey. Soc Psychiatry Psychiatr Epidemiol 1989;24:134-42. 27. Babigian HM, Odoroff CL. The mortality experience of a population with psychiatric illness. Am J Psychiatry 1969; 126:470-80. 28. Singer E, Garfinkel R, Cohen SM, et al. Mortality and mental health: evidence from the Midtown Manhattan Study. Soc Sci Med 1976; 10:517-25. 29. Roberts RE, Kaplan GA, Camacho TC. Psychological distress and mortality: evidence from the Alameda County Study. Soc Sci Med 1990;5:527-36. 30. Bromet EJ, Dunn LO, Connell MM, et al. Long-
1422
31.
32. 33.
34.
Simon and VonKorff
term reliability of diagnosing lifetime major depression in a community sample. Arch Gen Psychiatry l986;43:435-40. Prusoff BA, Merikangas KR, Weissman MM. Lifetime prevalence and age of onset of psychiatric disorders: recall 4 years later. J Psychiatr Res 1988; 22:107-17. Wittchen HU, Burke JD, Semler G, et al. Recall and dating of psychiatric symptoms. Arch Gen Psychiatry 1989;46:437-43. Semler G, Wittchen HU, Joschke K, et al. Testretest reliability of a standardized psychiatric interview (DIS/CIDI). Eur Arch Psychiatry Neurol Sci 1987^36:214-22. Rice JP, Endicott J, Lavori PW. Stability in relatives of depressives. Presented at the 144th Annual Meeting of the American Psychiatric Association,
New Orleans, Louisiana, May 1991. 35. Regier DA, Rae DS, Narrow W. Stability of diagnoses in the Epidemiologic Catchment Area study. Presented at the 144th Annual Meeting of the American Psychiatric Association, New Orleans, Louisiana, May 1991. 36. Aneshensel CS, Estrada AL, Hansell MJ, et al. Social psychological aspects of reporting behavior lifetime depressive episode reports. J Health Soc Behav 1987^28:232-46. 37. Means B, Nigam A, Zarrow M, et al. Autobiographical memory for health-related events. Washington, DC: Department of Health and Human Services, 1989. (Vital and health statistics, Series 6, no. 2) (DHHS publication no. (PHS 89-1077). 38. Greeley H. The American conflict. Hartford, CT: ODCase, 1864.
