Background: Tardive dyskinesia (TD) is associated with antipsychotics. To date, predictive models assessing the clinical characteristics on TD risk have not been developed and validated in the US population.
Objective: To develop a prediction model identifying factors associated with TD occurrence.
Methods: Adult patients with schizophrenia, major depressive disorder, or bipolar disorder taking oral antipsychotics, with 6 months of data prior to the index date (date of first claim for an antipsychotic drug after a claim for the underlying disorder but before diagnosis of TD), were identified from Medicaid claims from six US states. A multivariate Cox prediction model was developed using a cross-validated version of the least absolute shrinkage and selection operator (LASSO) regression method. The predictive performance was assessed in a separate validation set via model discrimination (concordance) and calibration.
Results: 66,723 patients with bipolar disorder, 68,573 with depressive disorder, and 54,119 with schizophrenia were identified. The prediction model had a clinically meaningful concordance of 70% and was well calibrated (P=0.46, Hosmer–Lemeshow goodness-of-fit test). Age at index date (hazard ratio [HR]: 1.03), schizophrenia diagnosis (HR: 1.73), antipsychotic dosage at index date (up to 100 mg/day chlorpromazine equivalent; HR: 1.40), and bipolar and related disorders (HR: 1.16) were significantly associated with an increased risk of TD. Use of typical antipsychotics at index date was associated with a modest reduction in the risk of TD (HR=0.94).
Conclusions: This study identified a group of risk factors associated with TD, and may facilitate better monitoring of patients by physicians.