Reply

From the commentary regarding our study on antipsychotic therapy and suicide risk in patients with treatment-resistant depression, ^{Reference Tsai, Yang, Wong, Chuang, Cheng and Shen1} Dr Narita suggests that the statistical approach, which included an interaction term between exposure status and follow-up time, implies a model assuming time-dependent hazards. ^{Reference Narita2} Including an interaction term in the pooled logistic regression between exposure and time interval is typically intended to provide flexibility in statistical modelling. ^{Reference Gaber, Ghazarian, Strassle, Ribeiro, Salas and Maringe3} We appreciate the opportunity to clarify these points. Even if the effect may vary during follow-up, an overall estimate reflects the average effect of suicide risk across the study period in our nationwide database. ^{Reference Hsieh, Shao, Sung, Hsieh, Tsai and Lin4} We agree that a single risk estimate, by itself, may not fully reflect temporal changes in risk. However, the cumulative incidence curves shown in the paper (Fig. 3) ^{Reference Tsai, Yang, Wong, Chuang, Cheng and Shen1} illustrate the event trajectories over time and help readers interpret the risk at different time points. Because the majority of the readers are healthcare professionals, this approach may be more straightforward and has better clinical interpretability.