To the Editor The study by Miller and colleagues1 highlights the strength of using data from a large clinical population, since it would be difficult to recruit such individuals by other means. In contrast to randomized clinical trials, however, health care data reflect real-life clinical practice in that allocation is not based on a random number but a clinical decision process. Family physicians and/or psychiatrists make decisions to change doses of drugs on the basis of clinical guidelines, medical training, and prior experience. It is hence unlikely that a clinician would prescribe higher antidepressant doses unless a person has more severe psychiatric problems than those who were prescribed a standard dose.2 Although the authors have used propensity score methods to balance characteristics of individuals receiving different treatments, this method is not a magic bullet. Unlike randomized clinical trials, such an approach fails to account for confounding by indication, such as the acute presentation of psychiatric illness. Thus, it is premature to infer that the increase in self-harm is solely attributed to high-dose antidepressants. Rather, high-dose antidepressants are markers of the severity of the condition.