Predicting treatment resistance from first-episode psychosis using routinely collected clinical information
Osimo EF., Perry BI., Mallikarjun P., Pritchard M., Lewis J., Katunda A., Murray GK., Perez J., Jones PB., Cardinal RN., Howes OD., Upthegrove R., Khandaker GM.
Around a quarter of people who experience a first episode of psychosis (FEP) will develop treatment-resistant schizophrenia, but there are currently no established clinically useful methods to predict this from baseline. We aimed to explore the predictive potential for clozapine use as a proxy for treatment-resistant schizophrenia of routinely collected, objective biomedical predictors at FEP onset, and to validate the model externally in a separate clinical sample of people with FEP. We developed and externally validated a forced-entry logistic regression risk prediction model for clozapine treatment, or MOZART, to predict up to 8-year risk of clozapine use from FEP using routinely recorded information including age, sex, ethnicity, triglycerides, alkaline phosphatase levels and lymphocyte counts. We also produced a least-absolute shrinkage and selection operator (LASSO) based model, additionally including neutrophil count, smoking status, body mass index and random glucose levels. The models were developed using data from two United Kingdom (UK) psychosis early intervention services and externally validated in another UK early intervention service. Model performance was assessed by discrimination and calibration. We developed the models in 785 patients and validated them externally in 1,110 patients. Both models predicted clozapine use well during internal validation (MOZART: C statistic, 0.70 (95% confidence interval, 0.63–0.76); LASSO: 0.69 (0.63–0.77)). At external validation, discrimination performance reduced (MOZART: 0.63 (0.58–0.69); LASSO: 0.64 (0.58–0.69)) but recovered after re-estimation of the lymphocyte predictor (0.67 (0.62–0.73)). Calibration plots showed good agreement between observed and predicted risk in the forced-entry model. We also present a decision-curve analysis and an online data visualization tool. The use of routinely collected clinical information including blood-based biomarkers taken at FEP onset can help to predict the individual risk of clozapine use, and should be considered equally alongside other potentially useful information such as symptom scores in large-scale efforts to predict psychiatric outcomes.