USING ARTIFICIAL INTELLIGENCE/MACHINE LEARNING AND REAL-WORLD DATA TO IDENTIFY PATIENTS WITH SCHIZOPHRENIA FOR WHOM AN ARIPIPRAZOLE MONOHYDRATE LAI IS LIKELY TO BE A FAVORABLE TREATMENT OPTION

Background

Schizophrenia (SCZ) is a complex psychiatric condition that disrupts all aspects of life – from cognitive functioning to social relationships to professional attainment. With symptoms appearing typically in adolescence or early adulthood, identifying effective treatments earlier in the patient journey can have profound impact on reducing lifetime and disease burden.1 Currently, real-world data to support clinicians in making personalized patient treatment decisions is limited. This study aimed to leverage large US electronic health records (EHR) and claims data, and machine learning (ML) methods to identify demographic, clinical, specific treatment choice and timing factors associated with favorable treatment response to aripiprazole monohydrate long-acting injectables (LAIs) in early-stage schizophrenia.

Methods

This prognostic study used real-world 2022-2025 data from a large, integrated EMR and open claims US database in the Atropos Evidence Network™ covering all sites of care including pharmacy.2 The ML approach used targeted minimum loss–based super learning ensembles to incorporate drivers of observed treatment selection and outcomes in predicting optimal treatment selection. The model was trained on 70% of the real-world dataset and validated on the remaining 30%. To assess the impact of time to treatment initiation with antipsychotics (APs), each patient’s index treatment was defined as the last AP treatment started within 3 years of initial SCZ diagnosis and patients needed ≥1 year follow-up. The primary treatment success metric was lack of inpatient admission (any cause) and AP treatment regimen change (switch or stop). To test robustness, sensitivity analyses were performed (different datasets, study operational definitions, statistical and ML techniques).

Results

Among the 97,215 (validation sample n=28,931) patients starting AP treatment within 3 years of initial SCZ diagnosis and having ≥1 year of follow-up, mean age was 43±16 years, 57% were male, 36% had substance use, 20% used anticholinergics in the year prior to index date. Mean time from SCZ diagnosis to index AP was 1.24±0.98 years. The most common index treatment was an oral atypical AP (43%). Overall, 62% of patients met primary treatment success (no hospitalization, no regimen change at 1 year follow-up). The most important predictor of treatment selection was anticholinergic use in the 3 months prior to index date. The most relevant treatment success predictors were recent patterns of inpatient and emergency room use. The model was able to distinguish between oral therapies and aripiprazole monohydrate LAIs as well as between aripiprazole once monthly [AOM] and aripiprazole 2-month Ready to Use [Ari 2MRTU]. ML-proposed optional treatments suggested aripiprazole monohydrate LAIs should be used more often and earlier after SCZ diagnosis. Based on ML technique and parameters, the model favored Ari 2MRTU, especially among patients who were recently diagnosed and had any inpatient admission. Predicted treatment success (if patients used ML-proposed optimal treatment for their individual characteristics) was 72% (95%, CI: 71%-74%), which was significantly higher than the overall observed success (62%, p < 0.001).

Conclusion

Patients with recent SCZ diagnosis and any inpatient admission are likely to benefit from earlier initiation of AOM or Ari 2MRTU. This ML predictive approach should be refreshed in the future as more AP treatment data becomes available.