Fig. 2

Calibration plot for the re-estimated LASSO prediction model. A least absolute shrinkage and selection operator (LASSO) prediction model was used to identify risk factors for TD. The model was developed with data in the modeling set and validated and re-estimated with the validation data set. The risk of TD at 2 years after the index date as predicted by the model was compared with actual TD observed, within the validation set (one-third of the data set). Concordance was 70.6%, Hosmer–Lemeshow goodness-of-fit test, P = 0.32. TD, tardive dyskinesia