Once selected for inclusion in the study, subjects were interviewed by telephone, using the Composite International Diagnostic Interview-Short Form for Major Depression (CIDI-SFMD) (8). The questions contained in this screener derive from the CIDI (9,10). Demographic data were collected using questions from the 1996 Canadian Census Form. Interviews were tightly scripted, using standardized introductory statements, transitional statements, and a series of  “ fall-back” statements for responding to subject queries.

Because the CIDI-SFMD is a brief predictor of major depression, an internal validation study was incorporated into the data collection. All subjects scoring positively during the first 9 months of data collection and a random sample of subjects rating negatively during this interval were invited to participate in a full administration of the CIDI mood disorders section (Version 2.1). Collection of this validation data was planned a priori to address concerns that the CIDI-SFMD might be nonspecific for major depression (11). This allowed us to determine the positive and negative predictive value of the screener and adjust the crude prevalence estimate for misclassification resulting from the use of the CIDI-SFMD screener as a diagnostic measure.

A sampling weight for each subject was calculated as described by Pothoff (7). After examination of the data, an additional adjustment to the weighting was made to accommodate an overrepresentation of women in the sample, a slight overrepresentation of subjects under the age of 35 years, and a slight overrepresentation of subjects over the age of 65 years. As a result, the estimates were weighted by age and sex to the 1996 census figures for household residents in Calgary over the age of 17 years. (Incidentally, an overrepresentation of women has been reported previously when the last birthday method has been used for subject selection [12]). All analyses used the “CSAMPLE” program in Epi Info (13), accounting for both the clustering and unequal selection probabilities resulting from the sampling procedure.

Results

A total of 15 819 randomly generated telephone numbers were called. Of these, 2849 (18.0%) were disconnected or not working, 3541 (22.4%) were businesses or fax numbers, and an additional 1730 (10.9%) could not be classified after the protocol’s 6 required callbacks. It was possible to determine that 476 numbers (3%) were residential, but no other information could be gathered, either because a household member could not be contacted or because of a language barrier with the person answering the phone. In 4083 instances (25.8%), a household was reached, but the household contact refused further participation in the survey. In the remaining 3140 households, an effort was made to randomly select 1 member for the survey. In 15 households (0.48%), however, no eligible members were present (for example, all residents were too young), and in an additional 114 (3.6%), the selected household member was away, ill, or unavailable. Eight subjects (0.25%) were excluded because they were found to be members of households already sampled, 5 (0.16%) because of protocol violations or uncertainties, and 17 (0.54%) because of language barriers. Of the remaining 2981 randomly selected household members who were successfully contacted, 2542 consented to participate in the study. The individual response rate was calculated as 2542/2981 (85.3%).

Characteristics of the study sample are presented in Table 1. The weighted proportion of study subjects positive on the CIDI-SFMD was 14.7%. The validation component of the study had 413 subjects, 277 (67.1%) of whom were screener-positive and 136 (32.9%) of whom were screener-negative. Of the 277 subjects who were positive on the screener, 69 did not meet DSM-IV criteria for major depression on the CIDI, indicating a positive predictive value for the screener of 75.1%. This suggests that a prevalence estimate using the short form would overestimate the true prevalence proportion by approximately 33%, leading to an adjusted 12-month period prevalence estimate of 11.0%. The negative predictive value of the screener was nearly perfect: 133 of 136 (97.8%) screener-negative subjects were confirmed not to have major depression by the full CIDI.