Dr Isaac Galatzer-Levy Explains How a Predictive Algorithm Uses Machine Learning to Predict PTSD Risk
Isaac Galatzer-Levy, PhD, assistant professor in psychiatry and bioinformatics, NYU School of Medicine, and vice president of clinical and computational neuroscience, AiCure, describes how his team built a predictive algorithm using machine learning to predict the development of posttraumatic stress disorder (PTSD).
How does the predictive algorithm constructed by your lab use machine learning to predict PTSD risk?
So, PTSD risk is really dependent on multiple factors. There’s many different things that can cause a deleterious outcome after trauma. And even the outcomes that you have vary between individuals, whether it’s PTSD, depression, problems with sleep, problems with memory, problems with arousal. So we constructed an algorithm that uses electronic medical records and simple information about the patient’s objective distress to predict how the patient is going to do over the next 12 months.
We built it and validated it in 2 independent data sets. One, the Grady trauma project from Emory Hospital, roughly 220 subjects, and then we validated it in our own sample collected at Bellevue Hospital in New York City. We got a predictive accuracy of roughly 90% for predicting chronic PTSD, and we replicated that independently in the Bellevue data set.
The sources of data include information on immune response, metabolic response, all the early biological changes that are associated with the onset of PTSD and depression that cause changes in things like sleep patterns, memory, arousal.