Abstracts Category : Other

An analytics approach to problems in healthcare

by Jerry Lai Kung

Health care expenditures in the United States havebeen increasing at unsustainable rates for more than thirty yearswith no signs of abating. Decisions to accept or rejectdeceased-donor kidneys offered to patients on the kidneytransplantation wait-list currently rely on physician experienceand intuition. Scoring rules to determine which end-stage liverdisease patients are in most dire need of immediate transplantationhave been haphazardly designed and reactively modified in anattempt to decrease waitlist mortality and increase fairness forcancer patients. For each of the above problem settings, we proposea framework that takes real-world data as input and draws uponmodern data analytics methods ranging from mixed integer linearoptimization to predictive machine learning to yield actionableinsights that can add a significant edge over current practice. Wedescribe an approach that, given insurance claims data, leadsconservatively to a 10% reduction in health care costs in a studyinvolving a large private US employer. Using historical data forpatients on the kidney waitlist and organ match runs, we build amodel that achieves an out-of-sample AUC of 0.87 when predictingwhether or not a patient will receive a kidney of a particularquality within three, six, or twelve months. Given historical datafor patients on the liver waitlist, we create a unified model thatis capable of averting an additional 25% of adverse events insimulation compared to current practice without disadvantagingcancer patients.Advisors/Committee Members: Dimitris Bertsimas (advisor).