Distress and Depression in Men Who Have Sex With Men: The Urban Men’s Health Study

A complete description of the construction of the sampling frame and overall sample demographics are described elsewhere (20). To summarize, a household-based probability sample of 2,881 men who have sex with men was obtained by identifying geographic areas for sampling (21) and then applying disproportionate and adaptive sampling methods (22, 23) to those areas. These strategies had been previously suggested for generating representative samples of hard-to-reach populations for AIDS studies (24). All data presented here have been weighted to reflect the probability of selection, adjustment for noncoverage and nonresponse, and maintenance of proportionality between cities based on each city’s estimated total population of men who have sex with men. Potential respondents could verify the nature of the study through local, national, and university contact names and telephone numbers. Telephone interviews started in late 1996 and continued until early 1998. In total, 63,783 households were screened; 3,700 households (5.8%) were found to contain at least one man who has sex with men; and 2,881 interviews (78% of eligible households) were completed. Interviews were conducted in Spanish (N=17) or in English at a time of the respondent’s choosing and lasted an average of 75 minutes. The Committee on Human Research of the University of California, San Francisco, provided human subject review and oversight, and consent procedures consistent with telephone interviewing techniques were applied.

Distress and depression were measured with the Center for Epidemiological Studies Depression Scale (CES-D Scale) (25). The CES-D Scale is a 20-item questionnaire asking the frequency of various depressive symptoms in the previous week and was originally developed in 1977 as a survey instrument measuring depression. The original cut point of 15 or greater has been shown to be inadequate as a screening examination for current major depressive disorder (26, 27). The National Institute for Occupational Safety and Health and the Centers for Disease Control and Prevention use the CES-D Scale with a cutoff of 22 to define “major depression” in their surveillance (28). Sensitivity and specificity for major depressive disorder are greater than 85% with cutoff points of 21 and 23, respectively (29, 30). Because the CES-D Scale does not request time course data, it is impossible to differentiate major depressive disorders from either dysthymic disorder or a depressive episode of bipolar disorder. Those two disorders are probably a large proportion of the false positives found by using this scale.

What characterized the respondents with CES-D Scale scores from 16 to 21? Many psychiatric disorders other than mood disorders (and some normal conditions, such as bereavement) have significant dysphoric elements, at least episodically. Anxiety disorders have been noted to confound CES-D Scale scores of 16 or greater (31). Other confounding conditions may include—but are not limited to—alcohol and drug withdrawal and adjustment disorders. Since a broad range of conditions are likely to have CES-D Scale scores in this range, this study categorizes respondents with CES-D Scale scores in the range of 16 to 21 as “distressed” rather than as having a particular diagnosis. Together with the depressed respondents, they represent much of the group at risk for needing voluntary mental health care.

A total of 2,678 (93%) of the 2,881 respondents completed the CES-D Scale. A three-category version of CES-D Scale scores was used as the dependent variable: euthymic (i.e., of normal mood), scores of 15 and below; distressed, scores from 16 to 21; and depressed, with scores of 22 and above.

Six types of independent variables were chosen: demographics, developmental issues, substance use, sexual behavior, HIV experience, and current social context variables.

The demographic variables examined were age, race or ethnicity (white, African American, Hispanic, Asian/Pacific Islander, or Native American/other), educational level (not having a college degree, having an undergraduate degree, or postgraduate studies), area of residence (gay ghetto [32] or other), and city of residence (San Francisco, New York, Los Angeles, or Chicago).

Developmental issue variables included age at first sexual experience; history of parental substance abuse (including alcohol) or attempted suicide (both categorized “yes” or “no”); number of incidents of child abuse (defined as being “beaten up” by a parent or adult in charge), anti-gay harassment before the age of 16, and witnessing a number of incidents of domestic violence between parents (0, 1, or ≥2); lifetime number of incidents of forced sex (0, 1–5, or ≥6), and, if applicable, age at first forced sex and age at first suicide attempt.

Substance use was measured by frequent heavy alcohol use (weekly ingestion of at least five cocktail equivalents at one sitting or other), multiple drug use (use of at least three categories of drugs of abuse or other), three or more alcohol-related problems (“yes,” “no,” or “missing”), and frequent drug use (at least one drug of abuse at least once a week or other). Because of a technical problem, a high proportion (N=229) of the 2,881 respondents were missing data related to the number of alcohol-related problems. Therefore, a “marker” variable was created—“missing alcohol-related problems” (“yes” or “no”), and this variable was entered in the logistical model just before the substance use block to control for the missing data (33).

The variables for sexual behavior examined were any sexual dysfunction in the previous year (“yes” or “no”), number of sex partners in the previous year, and unsafe sex in the previous year, defined as anal or vaginal intercourse with a partner of different or unknown HIV serostatus (“yes” or “no”).

HIV exposure was assessed by self-reported—but internally validated (34)—HIV status (HIV positive or other); any care of a person with AIDS in the previous year (“yes” or “no”); responses to “Being [your serostatus] has made you feel like an outsider in the gay community,” and “You find it hard to tell other gay men [your serostatus]” (scored from 1 for “strongly disagree” to 4 for “strongly agree”); and numbers of HIV-related deaths of intimates (lovers, former lovers, close friends, or family members) and of current HIV-positive intimates. Of those who took care of any people with AIDS in the prior year, further variables studied were currently caring for a current or ex-domestic partner (“yes” or “no”), the number of people with AIDS cared for in the previous year, and the number of current care-giving relationships with persons with AIDS.

Current social context variables were being in a domestic partnership (living with a man in a committed relationship, regardless of its legal status) (“yes” or “no”), self-defined sexual identity (dichotomized as gay, queer, homosexual, or other), outness, community cohesion, and community alienation (all defined elsewhere [32]). Current variables were also numbers of incidents of anti-gay violence or threat of violence in the previous 5 years, of incidents of domestic violence in the prior 5 years, of gay-identified community organizations in which the respondent was involved, and of non-gay-identified community organizations in which the respondent was involved. Domestic partnership status and self-defined sexual identity were included in this block rather than as demographic characteristics because a respondent’s answer could easily be changed with no apparent change in behavior.

Initially, t tests were used to test statistical comparisons of continuous independent variables for significance; statistical comparisons of categorical independent variables used chi-square analysis. Subsequent simultaneous comparisons of the three CES-D Scale categories by demographic, developmental, substance use, sexual behavior, HIV experience, and social context variables used multinomial logistic regression rather than ordinal logistic regression because the proportional odds assumption could not be made, given the diagnostic heterogeneity of the distressed subjects. For this regression, continuous variables were converted to categorical variables based on quartile values of the items in question, except age (categorized as 18–29, 30–39, 40–49, 50–59, 60–69, and ≥70 years), incidents of anti-gay violence or threat of violence in the previous year, incidents of domestic violence in the prior 5 years, and incidents of non-gay-identified community organization (all recoded as 0, 1, and ≥2). Histories of child abuse and early anti-gay harassment were dichotomized (as present or not). Two skewed 4-point Likert-scale variables were divided into two categories: responses to “Being [your serostatus] has made you feel like an outsider in the gay community,” was categorized as “agree” or “disagree,” and, “You find it hard to tell other gay men [your serostatus]” was recoded as “strongly disagree” or “less than strongly disagree”). Independent variables with p<0.05 in the preliminary comparisons were entered into the multivariate model in blocks in the following order: demographic characteristics, developmental issues, missing alcohol-related problems, substance use, sexual behavior, experience with the HIV epidemic, and current social context. Variables were eliminated stepwise within each block when p≥0.05. Conditional odds ratios were calculated for the final model by STATA, using the “svymlog” procedure (35, 36). The value of the goodness of fit test was calculated by first running separate binomial logistical models through LOGITSE, an SAS macro (37), with a weighted version of the preferred Hosmer-Lemeshow decile goodness of fit test on each pair of two values of the dependent variable (i.e., euthymic versus distressed, euthymic versus depressed, and distressed versus depressed). P values for goodness of fit obtained in these regressions were corrected by a reversed Bonferroni procedure to obtain an overall goodness of fit (38).