Off-Label Antipsychotic Withdrawal in People With Intellectual Disabilities: Development and Internal Validation of a Prediction Model.

The model was developed and internally validated on the same combined data, so it has not been tested in a new sample. The combined sample was relatively small, so only seven baseline variables could be included, and potentially relevant factors such as sex, previous withdrawal attempts, caregiver attitude, parkinsonism and akathisia were not in the final model. The two source studies differed in design and withdrawal outcome, with Study 1 having a higher unsuccessful withdrawal rate than Study 2. Some movement disorder variables could not be combined because Study 1 and Study 2 measured dyskinesia, parkinsonism and akathisia differently. There was a 10-year gap between the studies, so variables important in later literature were not available in Study 1. The outcome was dichotomous: withdrawal was counted unsuccessful if it was stopped before reaching zero, so partial reductions that may still be clinically useful were not captured. The model explained only 20% of the variance, so the included predictors did not account for most of the withdrawal outcome.

Do not use this model as a ready clinical decision rule. It was developed and tested only on the same combined data, not in a new sample, and it did not explain all withdrawal outcomes.