Confounding by severity and indication in observational studies of antidepressant effectiveness.

BACKGROUND: It has been suggested that antidepressants worsen the course of major depressive disorder. Epidemiological data have sometimes been cited in support of this idea, but such estimates are vulnerable to confounding. The objective of this study was to assess episode incidence and recovery in relation to antidepressant use, adjusting for symptom severity.
METHODS: Random digit dialing was used to select a sample of n=3304 community residents. Each respondent was then assessed with a baseline interview followed by a series of six subsequent interviews spaced two weeks apart. The brief Patient Health Questionnaire (PHQ-9) was used to detect depressive episodes during follow-up and to provide ratings of symptom severity. Grouped time proportional hazards models were used to assess confounding by producing estimates of the association between antidepressant use and major depression incidence and prognosis adjusted for baseline symptom severity.
RESULTS: Antidepressant use in initially non-depressed respondents was associated with a markedly higher incidence of depression (Hazard Ratio, HR = 3.9, 95% CI 1.8 â 8.5). With adjustment for the depression severity score in the two weeks preceding the emergence of a new episode, this effect diminished markedly and was no longer statistically significant (HR = 1.2, 95% CI 0.6 â 2.7, p = 0.57). Antidepressant use was also associated with a lower rate of recovery from major depression (HR = 0.8, 95% CI 0.5 â 1.2, p = 0.27), but this effect also moved towards the null value after adjustment for baseline severity (HR = 0.9, 95% CI 0.6 â 1.5).
CONCLUSIONS: Antidepressant medication use is confounded with symptom severity. Observational studies seeming to show harmful effects of antidepressants are subject to bias as a result. Key words: Antidepressive agent; longitudinal studies; epidemiology; methods.