We appreciate Nebhinani & Grover's interest in our study ^{Reference Girgis, Phillips, Li, Li, Jiang and Wu1} as well as the opportunity to respond to the six comments. First, our study was analysed using the intent-to-treat principle. Implicit in the intent-to-treat principle is that the outcome is not the effect of treatment per se, but rather the effect of initial assignment irrespective of treatment(s) received. ^{Reference Lachin2} Second, we agree that there are controversies as to the definition of first-episode psychosis. ^{Reference Breitborde, Srihari and Woods3} As reported by Breitborde et al, ‘duration of psychosis’ possesses the most construct validity, followed by other criteria, such as ‘duration of antipsychotic medication use’ and ‘first treatment contact’. ^{Reference Breitborde, Srihari and Woods3} We conservatively identified individuals with first-episode schizophrenia using both duration of psychosis and duration of antipsychotic medication use as two of our criteria. Furthermore, we included a maximum age criterion (i.e. 40 years old at the time when symptoms began) and symptom criteria to further narrow and restrict our study participants to those who are most likely to have first-episode psychosis. Third, our conclusions and main outcomes used the intent-to-treat principle and were based on the entire sample, rather than primarily based on the 29 individuals who remained on their originally assigned medication after 9 years. We described characteristics of this smaller group, without an intent to generalise, owing to the obvious lack of representativeness in this subgroup of patients. Furthermore, it is important to note that the generalisability of a clinical finding is determined by the representativeness of the sample observed, rather than the sample size observed.

Fourth, as described in the article, we did not have any missing baseline data for weight for those participants whose weights were included in our metabolic analyses. In addition, we disagree that we indicated that there were no differences in side-effects between the two groups. Rather, we descriptively reported differences in tardive dyskinesia and agranulocytosis between the two treatment groups. Finally, we did not claim that the metabolic findings in this study are generalisable, but we do agree with Nebhinani & Grover that it would have been valuable to report on additional metabolic indices (e.g. lipids and blood pressure). Unfortunately, these data were not available.

Fifth, we reported the average percentages of time that participants in each group took other antipsychotic medications or resumed the original medication during the follow-up period, rather than the percentages of individuals in each group on any treatment regimen at 9-year follow-up. Therefore, the results (55% for the chlorpromazine group and 73% for the clozapine group) represent averages over the entire 7-year period, rather than cross-sectional results at the 9-year follow-up time point. The clinical status of these patients over the entire 9 years of the study were reported in detail in the article, both in terms of symptom measures, functional status, global clinical status, medication status, side-effects, remission status and status in the study (i.e. still in the study or dropped out).

Finally, we agree that individuals with schizophreniform disorder are likely to have better outcomes than individuals with schizophrenia, by definition. However, all diagnoses were randomly and equally assigned to the two treatment groups. Therefore, including this diagnosis is unlikely to have affected the between-group outcomes of this study.