Challenges in Estimating Mortality Risk From Antipsychotics in People With Alzheimer’s Disease

To the Editor: I read with interest the article by Lopez et al. (1) in the September 2013 issue on the long-term effects of antipsychotics in people with Alzheimer’s disease, which suggested that after controlling for psychiatric symptoms, neither typical nor atypical antipsychotics were associated with time to death. I would like to expand on one of the limitations noted by the authors, as I believe it is critical to interpreting the results.

Previous research suggests that the excess mortality associated with antipsychotics is greatest early in treatment, with the relative risk trending toward no mortality effect of drug as the duration of antipsychotic use increases (2). The Lopez et al. study assessed medications every 6 months, without tracking interim changes. Periodic assessments such as this create challenges when interpreting results that are related to adverse events for which the greatest excess risk occurs early in treatment. This is a particularly important issue in a dementia sample, where death is not a rare occurrence. That is, if someone both started an antipsychotic and died during the time between two assessments, there would be no record that the individual received an antipsychotic before they died, and only those who survived the early treatment period would be recorded as receiving an antipsychotic. Thus, data are lost from individuals who may be particularly vulnerable to the adverse event of interest, in this case death. This is termed “depletion of susceptibles” and may lead to bias in effect estimates, as those who survive treatment through an initial high-risk period are often less susceptible to an adverse effect. This is one reason that new-user designs are generally preferred in pharmacoepidemiology studies of adverse drug effects (3).

Lopez et al. have presented a compelling case that psychotic symptoms are a prognostic marker for increased mortality in people with dementia and an influential unmeasured confounder in observational studies of antipsychotics and mortality. However, the design limitations make it difficult to fully discern the role of the psychiatric symptoms relative to that of antipsychotics that were received but not recorded. I admire the authors’ efforts to address this important question but do not place higher weight on these results than on those of randomized controlled trials that identify an increased risk of mortality associated with antipsychotic use in dementia. Randomized trials have their own limitations, such as highly selected samples, shorter follow-up, and small numbers of events in any given trial. Yet Schneider et al.’s meta-analysis of these trials (4) showing greater mortality with antipsychotics compared with placebo included more than 10 times as many antipsychotic users as were in the Lopez et al. study. I hope the null mortality result in this study is not interpreted as an indication that caution and restraint are not necessary when considering the use of antipsychotics in this vulnerable patient population.