Machine Learning with the Personalized Advantage Index




Project Summary

This summer, as a recipient of PURM, I worked at the DeRubeis Lab in the Department of Psychology. A main focus of the lab is to develop methods for treatment selection in clinical psychology so that individual patients can receive the treatment that best fits their individual case, and so that available resources can be more efficiently allocated. The lab developed the Personalized Advantage Index (PAI), which predicts how end-of-treatment symptom severity differs for individual patients with Major Depressive Disorder in two treatments: antidepressant medication and cognitive behavioral therapy. The main goal of the project I worked on this summer is to see if machine learning techniques can produce prescriptive variables that yield better predictions than the ones chosen in the original study. We used machine learning techniques including Random Forest, Elastic Net, BART and Bootstep AIC and tuned the weights of the variables selected using 10-fold cross validation. The result showed that the new prescriptive variables selected were able to predict a larger difference in post-treatment symptom severity between patients who were randomly assigned to the optimal treatment group and those who were randomly assigned to the non-optimal treatment group. The result of this project can serve as a basis of comparison with other machine learning approaches and modeling in treatment selection. 

One of the most beneficial feature of PURM that makes it such a great program is that recipients get the opportunity to explore interesting fields that they may not have prior exposure to, in order to experience what academic research is like on a day-to-day basis. Treatment selection in clinical psychology is something I was not familiar with before, and learning from scratch through this hands-on experience has taught me so much more than I could have learned in a class. Throughout the project, under the patient guidance and help from Professor DeRubeis and members of the lab, I learned a great deal about data analysis, statistical modeling as well as programming in R. Attending lab meetings also gave me an opportunity to get a glimpse at the dynamics in this area of study in general, as well as learn how the lab is progressing. It was very informative and interesting to hear about other projects going on at the lab as well. Besides being an amazing learning experience, my time at the lab has also been enjoyable and fun because of the people I got to know there, and I received a lot of great advice for my future studies and beyond. I am extremely glad and grateful that I got to work at the lab through PURM, and I hope I can further develop the skills I learned this summer.