Initial severity and efficacy of antipsychotics for schizophrenia and bipolar mania: Individual participant level analyses of placebo-controlled studies.

Symptom change from baseline to 6 weeks (3/4-12) will be our main outcome. All assessment time points (from baseline to endpoint) will be used in the analysis. Symptom change will be measured by total score reduction in symptom severity scales. We aim to use the same outcome measure across all trials in the field of schizophrenia and perform the meta-analysis using one scale (e.g. PANSS); the same will be attempted for all trials in the field of bipolar mania (e.g. YMRS).
Baseline symptom severity will be the main predictor (independent variable) in our study. It will be defined as the total score in symptom severity scales such as PANSS for schizophrenia and YMRS in bipolar mania. We will investigate the relationship between baseline symptom severity and subsequent symptom change comparing antipsychotics versus placebo in the acute treatment of patients with schizophrenia or bipolar mania. The two conditions (schizophrenia and bipolar mania) will be analyzed separately in order to assess whether potential effects of baseline severity on antipsychotic efficacy are restricted to a specific psychiatric disorder or is a more general phenomenon.
Sub-scores of the symptom severity scales representing specific groups of symptoms (e.g. positive and negative symptoms in schizophrenia) will be assessed separately to examine whether baseline severity has the same effect on the subsequent symptom reduction.
In addition, patient characteristics such as age, sex, duration of illness, duration of untreated psychosis, number of hospitalizations etc. will be examined for their possible confounding effect on antipsychotic efficacy.
We will conduct individual participant data meta-analysis to examine the relationship between baseline symptom severity and the differences in change scores between the antipsychotic drugs and placebo using a mixed-effects model repeated measures (MMRM) analysis with maximum likelihood estimation. The number of levels of the MMRM analysis will account for the data structure such as the represented time, the participant, and the trial. The resulting competing models with increasing complexity will be tested unadjusted and adjusted for confounders (e.g. age, sex, duration of illness, duration of untreated psychosis, number of hospitalizations etc.). The model with the smallest Bayesian Information Criterion (BIC) will be chosen as the most parsimonious. We will report results based on the best fitting models.
Sensitivity analyses will be conducted comparing the different competing models as produced by the different levels of the MMRM analysis. To examine whether the overall results observed with the total score hold also for specific groups of symptoms such as the positive and/or negative symptoms in schizophrenia or specific single items in bipolar mania, we will run the same analyses for the positive and negative symptoms or the single items separately.
As we would like to calculate the number needed to treat (NNT) of achieving a minimally important change for different baseline severity scores, we will conduct a linking analysis between YMRS and CGI-BP scales (e.g. what do the different cutoffs in the YMRS scale mean). Results of the linking analysis are very important both for clinicians and researchers.
Finally, we will use (i) a network analysis to identify a system (network) of schizophrenia and bipolar symptoms and the most central schizophrenia and bipolar symptoms based on the YMRS and PANSS respectively; and (ii) Item Response Theory to identify a short YMRS and PANSS if possible.