Disparate rates of new-onset depression during the menopausal transition in 2 community-based populations: real, or really wrong?

Abstract

This study took place in eastern Massachusetts and included respondents from the Harvard Study of Moods and Cycles Cohort 1, enrolled between 1995 and 1997, and the Harvard Study of Moods and Cycles Cohort 2, enrolled between 2005 and 2009. In prospectively assessing rates of new-onset depression in 2 populations of late-reproductive-aged women with no Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) lifetime history of depression, we were surprised to find far lower rates of depression in the population with greater racial diversity and lower socioeconomic status, contrary to what had been reported in the scientific literature. To better understand why these disparate results occurred, we assessed confounding and outcome misclassification as potential explanations for the discrepancy. After determining that these were unlikely explanations for the findings, we explored 2 potential sources of selection bias: one induced by self-referral of healthy participants into the study and the other induced by the design of the study itself. We concluded that both types of selection bias were likely to have occurred in this study and could account for the observed difference in rates.

Keywords: bias; depression; prospective studies.

Figure 1.

Harvard Study of Moods and Cycles Cohorts 1 (1995­–1997) and 2 (2005–2009), eastern Massachusetts. Directed acyclic graph illustrating potential measured confounding by Z and unmeasured confounding by U. X indicates the main exposure; D₁, incident depression during follow-up.

Figure 2.

Propensity scores of enrollment for Harvard Study of Moods and Cycles Cohorts 1 (1995­–1997) (left histogram) and 2 (2005–2009) (right histogram), eastern Massachusetts. Dark shading indicates overall propensity scores; light shading indicates propensity scores for incident depression cases. All variables listed in Table 1 were included in the estimation of the propensity score. Age and score on the Center for Epidemiologic Studies Depression Scale were included as continuous variables with both linear and quadratic terms.

Figure 3.

Harvard Study of Moods and Cycles Cohorts 1 (1995­–1997) and 2 (2005–2009), eastern Massachusetts. Directed acyclic graph illustrating potential for exposure misclassification of disease status, D₁, where X represents the geographic location in which the cohort resides, and D*₁ is the observed misclassified version of D₁.

Figure 4.

Harvard Study of Moods and Cycles Cohorts 1 (1995­–1997) and 2 (2005–2009), eastern Massachusetts. Directed acyclic graph illustrating selection bias. Here, D₀ is the baseline depression status, S is selection into the study, D₁ is incident depression during follow-up, U is unmeasured confounding, and X is the geographic location in which the cohort resides. The boxes around S and D₀ indicate that the Harvard Study of Moods and Cycles conditions on these variables.