Understanding the Risk of Treatment Failure After Discontinuation of Long-Term Antipsychotic Treatment

To the Editor: Tiihonen et al. (1) should be commended for their follow-up investigation, published in the August 2018 issue of the Journal, of discontinuation of antipsychotic treatment in schizophrenia, using a nationwide cohort and a state-of the-art sampling design. We think that some clarifications are warranted to interpret the statement that the “risk of treatment failure increases with duration of antipsychotic treatment before discontinuation” and the conclusion that long-term antipsychotic exposure “makes discontinuation more difficult when exposure has been longer.”

Examining the absolute risk of treatment failure using the Kaplan-Meier survival functions in the four antipsychotic discontinuation groups (see Figure 2 in the original article), we find that the risk fails to increase with increasing duration of antipsychotic exposures. In fact, the survival functions show that until approximately 10 years of follow-up, the absolute risk of treatment failure decreases with longer prior antipsychotic exposures. Overall, these data do not support the conclusion that discontinuation is more difficult when antipsychotic exposure has been longer.

The apparent increase of relative risk of treatment failure with increasing antipsychotic exposures before discontinuation may be partly due to the “depletion of susceptibles,” which represents a differential attrition of those subjects from the risk set who are susceptible to the outcome of interest. In Tiihonen et al.’s study, antipsychotic continuers who were susceptible to relapse (perhaps those receiving ineffective treatment) were increasingly likely to discontinue with the passage of time. Confounding due to depletion of susceptibles may have been enhanced by the study design: patients who were rehospitalized before matching were excluded. Inspection of the survival functions suggests that the relative risk increases with time were not due to increases in relapse rates among discontinuers but were mostly driven by the progressively lower risks of relapse in the control (continuers) group.

Potential long-term improvers also require consideration. About 30% of the early discontinuers were not rehospitalized. Perhaps 20%−30% of patients with schizophrenia improve and then maintain remission for sustained periods after discontinuing antipsychotics (2). Nevertheless, the analyses made no provision for the possibility that some non-rehospitalized patients may have remitted. The standard Cox proportional hazards model that was used fails to account for this. Two emerging classes of survival models, the “cure” and the “frailty” models, make provisions for individuals who will never experience the outcome (“cures”) or who will experience it only after a long time (“the less frail”). Tiihonen et al.’s overall conclusion that the risk of treatment failure or relapse after discontinuation of antipsychotics does not decrease as a function of time might be strengthened or modified by such models.