P3‐303: Antidepressant drug use and cognitive deficits in older men: addressing confounding by indication using different methods

Abstract

Epidemiological studies have found associations between antidepressant drug use and cognitive impairments in older persons. The objective of this study was to examine whether this association may result from confounding by indication using different analytic approaches. 544 community-dwelling men aged 65 years or older with hypertension had cognitive performance measured with the Hopkins Verbal Recall (HVR) at a Veterans Administration primary care clinic. Antidepressant and other psychotropic medication use during a one-year period were ascertained using medical records. Potential confounding by indication, defined by a diagnosis of depressive or related psychiatric disorders from medical records and depression severity at baseline interview, was examined using conventional multiple linear regression (MLR), a two-stage regression (TSR), and a structural equation model (SEM), respectively. Before adjusting for indications, a one unit increase in cumulative antidepressant exposure was associated with 1.00 (95% Confidence Interval (CI): 0.06 to 1.94, p 0.05). The apparent association between antidepressant exposure and cognitive deficit in older persons without adjusting for depressive-related psychiatric disorders may be seriously biased. Among the three methods explored, SEM offers the most powerful bias reduction and conceptually appealing empirical framework to test confounding by indication hypothesis.