Patients who have an initial psychotic episode generally have good response rates to antipsychotic injection treatments, according to a 2015 study. Treatment success odds correlate with the time lapse between the onset of psychotic symptoms and treatment. The less time that passes before treatment begins the better the success rate. There are at least 15 appropriate injectable medications for first-episode schizophrenia. Choosing the best medication in light of the patient’s medical history and potential drug interactions can take time and delay the start of treatment.
Researchers at National Taiwan University Hospital & College of Medicine used a machine-learning system to develop and validate an Individualized Treatment Rule (ITR) for first-episode schizophrenia. The results of the study were published on JAMA Network.
The NTU researchers used data from 32,777 patients with a mean age was 36.7 years and 48.8% were male. The training set for the study consisted of 70% of the full data population. The team ran 121 data elements for each patient in the training set through a series of machine learning models to develop the ITR. The ITR was then applied to the remaining 30% of the sample as a validation. The validation group showed 51.7% success using the ITR. This rate is a statistically significant increase of the 44.5% success rate of the observed group (the 70% training group). By using the ITR developed with machine learning, the NTU team increased the treatment success rate by just over 16%.
Next steps for the National Taiwan University researchers will focus on further training and testing to not only duplicate but to hopefully improve the ITR-based treatment success rate for patients who have first-episode psychosis.
Generalized applications of the Taiwan ITR model will depend on access to extensive patient records. Hopefully developing a highly successful improvement in first-episode schizophrenia treatment will demonstrate a big win for the value of both electronic health records (EHR) and machine learning in determining treatment plans.
