190518 Machine Learning Approaches for Personalized Treatment Selection in Psychiatry

Title:
Machine Learning Approaches for Personalized Treatment Selection in Psychiatry

Speaker:
吳其炘 (Chi-Shin Wu), MD, PhD, National Taiwan University Hospital / Harvard

Time:
05/18 (Sat.) 5 pm PDT, 6 pm MDT, 7 pm CDT, 8 pm EDT
05/19 (Sun.) 8 am Taiwan

Keywords:
Medicine, Epidemiology, Pharmacoepidemiology

Abstract:

Identifying the best regimen for individual patient remains challenging, clinicians need to rely on trial and error. The effect of using machine learning to select personalized treatments needs to be assessed and validated.

This study utilized Taiwan's National Health Insurance Research Database. We focused on patients with schizophrenia, which is one of the most severe mental disorders. Prediction models based on Super Learner algorithms were used. We identified the top 3 antipsychotic regimens with the highest probabilities of treatment success. In the test dataset, we compared the rates of treatment success between patients treated with machine-selected regimens and those treated with non-selected regimens.

The results showed patients treated with machine-selected regimens had 43.0% treatment success, while those treated with nonselected regimens have 27.7% success. In conclusions, that the machine learning-based prediction models had acceptable prediction accuracies, which suggested that the use of machine-selected regimens will increase the treatment success rate for patients with schizophrenia.