David Czerwinski, Dimitris Bertsimas and Michael Kane

ORC, MIT

The purpose of this project is to determine whether medical insurance claims data can be used to measure the quality of care that a patient has received. Particularly, we are interested in identifying patients who have received poor care so that an intervention can be arranged to improve their care.

We focused our study on 100 diabetes patients. A physician, Dr. Michael Kane of MIT medical, studied the claims records for each of the patients and assessed the quality of care they were receiving. He rated the care on a three-point scale: poor, average, and very good. He also rated his
confidence in his assessment on a two-point scale: confident or not confident. In addition, he wrote a brief paragraph about each patient's care and noted aspects of it that influenced his rating. This informationwas useful for constructing variables to be included in the statistical model.

We constructed a variety of statistical models and trained them on his ratings. The techniques employed included logistic regression, support vector machines, the Lasso, random forests, and an integer-optimization based technique we developed ourselves.

In this poster session I will present results on the performance of the different models and discuss some of the interesting challenges involved.