Volume 22, No. 1 - 2001 

[Table of Contents]

The Duration of Major Depressive Episodes in the Canadian General Population

Scott B Patten

Abstract

The National Population Health Survey (NPHS) has provided a wealth of new data concerning major depression in the Canadian general population. The NPHS included a brief predictor of major depression, and also two questions (only one of which was asked of each subject) concerned with the duration of episodes in the preceding year. A striking finding was that many of the episodes identified were very brief. In this paper the NPHS data were examined from a different perspective in order to derive a complementary perspective on the episode duration data. Data from the 1994/95 and 1996/97 cycles of the NPHS were used in the analysis. The longitudinal data were used to generate approximations of age and gender-specific incidence for members of the population over the age of 12 years. An estimate of prevalence was made from the 1996/97 cross-sectional file. A basic expression relating prevalence to incidence and mean duration of illness was then applied within age and gender categories. Taken together, the incidence and prevalence data from the NPHS suggest a longer duration than was indicated by the NPHS interview duration item. A probable explanation is that the NPHS duration question had an upper limit of 52 weeks, whereas some episodes of major depression last longer than this. Particularly long episodes could have a large impact on mean duration in the population. Nevertheless, these data confirm the heterogenous nature of this condition; many people with the syndrome of major depression may have quite brief episodes.

Key words: depressive disorder, epidemiology, prognosis, prevalence, incidence

Introduction

A variety of prevalence estimates for major depression have been reported in various countries.^1-7 Recent estimates of the 12-month period prevalence of major depression in the Canadian population derive from the Canadian National Population Health Survey (NPHS) ^8,9 and the Mental Health Supplement of the Ontario Health Survey.¹⁰ Estimates from several additional Canadian surveys have been published.^11,12 Since the NPHS is a longitudinal study, it is possible to approximate incidence using the NPHS data,¹³ with the proviso that the follow-up interval for the longitudinal component of the NPHS is two years and the measurement instrument employed identifies episodes occurring only in the 12-month period preceding the interview. The proportion of non-depressed subjects in 1994/95 who were subsequently found to have major depression in 1996/97 may overestimate the annual incidence proportion, since it may include some persons with an onset of major depression in the year following the 1994/95 interview. This would occur if these episodes did not resolve until two weeks or more into the year preceding the 1996/97 interview. These subjects would appear in the numerator of the incidence expression, inflating the estimated incidence proportion.

The NPHS is an ongoing longitudinal community study conducted by Statistics Canada. A national sample consisting of more than 17,000 subjects was interviewed in 1994/95, then re-interviewed in 1996/97 in the longitudinal component. The 1996/97 data collection cycle also included many "buy-ins" from specific provinces, such that the cross-sectional data from the 1996/97 cycle included over 80,000 subjects (a minority of whom were subjects being followed up two years after the initial data collection). The NPHS interview incorporated a short form version of the Composite International Diagnostic Interview (CIDI), called the Short Form for Major Depression.¹⁴ This brief predictive instrument assigns a probability of major depression using a set of questions adapted from the CIDI by Kessler et al.¹⁴ Subjects reporting at least five of nine symptoms constituting the "A" diagnostic criteria for major depression in DSM-IV, when at least one of these symptoms is depressed mood or loss of interest, are assigned a predictive probability of 90%.

The CIDI short form includes two branched series of questions. Neither branch is followed when a subject does not report a two-week period of depressed mood or a two-week period of loss of interest. Subjects reporting the former symptom follow one branch and subjects reporting the latter symptom (but not the former) follow the other branch. Each branch includes questions about other depressive symptoms requisite to the DSM-IV definition of major depression. Each branch also contains a question about the duration of depression, asking those subjects whose responses indicated an episode of depression to report the total number of weeks in the preceding year that they felt "this way" (the question was preceded by a summarizing statement referring to the occurrence of an episode of depression in the preceding 12 months, and incorporating key phrases for reported symptoms and their duration). Surprisingly brief durations were reported; they are presented in Figure 1.

FIGURE 1
Number of weeks depressed in past 52 weeks*

* Adapted from: Statistics Canada's Internet site <http://www.statcan.ca/english/Pgdb/People/Health/health35.htm>.
Date of extraction: November 1, 2000

Potentially, the brief mean duration of depressive episodes has important implications for public health. In clinical settings, the major depressive syndrome is usually regarded as an indication for treatment. Public health efforts, such as the American Depression Screening Day,¹⁵ tend to focus on encouraging people with symptoms suggestive of these disorders to come forward for treatment. Such secondary and tertiary level prevention is an obvious strategy in the absence of feasible approaches to primary prevention. However, both pharmacological and non-pharmacological approaches to treatment generally take at least 4-8 weeks, and the NPHS duration data appear to imply that the average episode would resolve during that time.

The objective of this project was to estimate the mean duration of major depressive episodes using an alternative approach: by integrating the available incidence and prevalence data. This provides an alternative perspective on the duration of these episodes.

Material and Methods

A retrospective cohort analysis was used to estimate age- and gender-specific incidence proportions. Using data collected in 1994/95, all subjects who were under the age of 12 and all subjects with major depression (according to the CIDI-SFMD) were excluded. The remaining subjects were assessed in the 1996/97 data set in order to estimate the proportion of these subjects who had newly developed major depression. Standard errors for these estimates were calculated using a bootstrap method recommended by Statistics Canada. The bootstrap variance estimates were calculated using an SAS¹⁶ macro developed by Statistics Canada for this purpose. All estimates used sampling weights to adjust for unequal selection probabilities due to the complex survey sampling strategies employed by Statistics Canada in this survey.

Prevalence estimates were made from the 1996/97 NPHS data, since the sample size for this data collection cycle was much greater than that of the 1994/95 survey. Again, the data were weighted, this time using a different set of sampling weights produced specifically for the relevant data file. Bootstrap methods were used to calculate confidence intervals for the prevalence estimates.

In order to integrate the incidence data into a broader epidemiological context, an incidence-prevalence model was used. This model represented the relationship between incidence and prevalence in an assumed stable population (and in the absence of migration) as an equality between the inflow of new cases into a prevalence pool and the outflow from the prevalence pool. The outflow includes all terminations of disease, including recovery and death. This model was described by Rothman and Greenland,¹⁷ as follows. In a stable population, the inflow of new cases into the pool of cases (prevalence pool) equals the outflow from the prevalence pool:

(Equation 1)

Where P represents the number of persons in the population with a disorder, I represents the incidence, N the total number of persons in the population, an interval of time and the mean duration of the disorder. This expression may be simplified to:¹⁷

(Equation 2)

Here, the ratio on the left side of the equation is the prevalence odds, a parameter that approximates prevalence when a disorder is rare.

For the purpose of this analysis, point prevalence was considered equivalent to one-month period prevalence. One-month period prevalence is often considered equivalent to point prevalence in psychiatric epidemiology because definitions of mental disorders include a requirement not only for the presence of signs and symptoms, but also for the persistence of these signs and symptoms over several weeks. For major depression, the symptoms must persist over at least two weeks. To perform the calculations using months as the time unit, it was necessary to denominate the incidence as a compatible person-month rate by estimating a monthly incidence rate that would be expected to result in the observed incidence proportion over a one-year period. For the purpose of generating this approximation, the "exponential formula" was used:

(Equation 3)

Here, IP is the approximation of the incidence proportion from the longitudinal NPHS data and I_k is a monthly incidence rate that would result in this incidence proportion over a 12-month follow-up interval. The latter estimate is the one suitable for substitution into Equation 2. If episode duration is measured in months, then the I_k rate (which has the units months^-1) will result in a dimensionless prevalence odds.

The NPHS survey used a 12-month predictor of major depression prevalence, whereas current prevalence is the parameter of most relevance to the model under development. It was, therefore, necessary to estimate current prevalence by combining the NPHS data with supplementary information from the literature. Fortunately, the literature in this area was found to be strikingly consistent. Kessler et al.,¹⁸ using data from 15-24 year-olds participating in the National Comorbidity Survey (NCS), reported that the ratio of annual to current (30-day) major depression was 12.4% to 5.8%, or approximately 2:1, this ratio being similar in male (9.0% to 3.8%) and female (16.1% to 8.0%) subjects. The overall 12-month period prevalence of major depression in the NCS was 10.3%,¹⁹ compared to the 4.9% one-month period prevalence,¹ a ratio also approximating 2:1. The ratio of annual to current cases in the NCS was similarly comparable among male (7.7% and 3.8%, respectively) and female respondents (12.9% and 4.9%, respectively).

Since the NCS is an American study, the ratio of annual to current major depression was also examined using published data from an earlier survey conducted in Edmonton. This study utilized methods resembling those employed in the Epidemiological Catchment Area (ECA) studies in the United States. Here, an annual prevalence of 4.6% and a one-month prevalence of 2.3% were reported, a 2:1 ratio.¹¹ The ECA study itself reported a one-year prevalence of (4.2%)²⁰ and a one-month period prevalence of 2.2%,⁴ also closely approximating 2:1. Based on these very consistent findings, the ratio of annual to current major depression in this analysis was taken to be 2:1 for both men and women. Hence, prior to estimation using equation 2, the annual prevalence from the NPHS was multiplied by 0.5 to generate an approximation of point (30-day) prevalence.


Results

The 1994/95 NPHS had a sample size of 17,626. The longitudinal data file included 15,670 subjects providing follow-up data. The current analysis excluded subjects who were under the age of 12 (n = 1908), subjects who had major depression at baseline (n = 781) or who did not provide valid data on the major depression predictor either at the baseline interview, the follow-up interview or both (n = 691). As such, the current analysis of incidence was based on 12,290 subjects. The prevalence estimates were based on 70,538 subjects over the age of 12 in the 1996/97 cross-sectional component of the NPHS. The sample included 73,402 subjects within this age group, but 2,864 (3.9%) who did not provide a valid rating on the CIDI Short Form were excluded.

The age- and gender-specific incidence of major depression, along with 95% confidence intervals, are presented in Table 1. The observed pattern of incidence generally resembled that of prevalence, as depicted in Table 2. Among female subjects, incidence and prevalence tended to be higher in younger women and to decline with advancing age. In males, the incidence rates tended to be slightly higher in middle-aged than in younger subjects. In the subjects aged 75 years and older, the incidence and prevalence increased slightly in both genders.

The prevalence-incidence model embodied in equations 1 and 2 relates incidence and the prevalence proportion (approximately, prevalence) to the mean duration of disease irrespective of whether the disease is terminated by recovery or death. Completing these calculations for each age and gender group yielded the data presented in Table 3. There is no evidence of gender differences in model-based estimates of disease duration, nor is there a pronounced trend towards such differences in association with age. However, the duration of the episodes, as predicted by the incidence-prevalence model, appears to be more brief in the youngest age group.