Quality of care and drug surveillance : a data-driven perspective

Abstract

In this thesis, we describe the use of medical insurance claims data in three important areas of medicine. First, we develop expert- trained statistical models of quality of care based on variables derived from insurance claims. Such models can be used to identify patients who are receiving poor care so that interventions can be arranged to improve their care. Second, we develop an algorithm that utilizes claims data to perform post-marketing surveillance of drugs to detect previously unknown side effects. The algorithm performed strongly in several realistic simulation tests, detecting side effects a large fraction of the time while controlling the false detection rate. Lastly, we use insurance claims data to improve our understanding of the costs of care for patients who suffer from depression and a chronic disease.