When physicians’ choice of antidepressant dosage is affected by risk of poor prognosis, confounding by indication may occur (2) and lead to results suggesting that high dosages of antidepressants are safe. Even before the FDA warning, low dosages of antidepressants were commonly prescribed for frail or elderly patients, who are more likely to develop adverse outcomes. Thus, the decreased or absent risk associated with a high-dosage prescription may result from a failure to consider various important unmeasured confounders (e.g., increased cardiac output) and mediators (e.g., smoking cessation) in a healthier study population. This bias can worsen when a measurement of person-time is used, because combining the number of persons and the duration of their time in the study may increase the diversity of interactions, allowing the impact of omitted variables to become more pervasive in regression analysis. Taken together, if the risk for different dosage groups is not assessed with equipoise, the results cannot substitute for results obtained from the crossover study submitted to the FDA.