Dissertation Defense :: Bayesian Classification Methods in Determining Health Burdens
Leanne Metcalfe - Advisors: Dr. Brani Vidakovic, Committee: Dr. Paul Griffin, Dr. Charles Kemp, Dr. Stephen Sprigle, Dr. Arun Villavallam
DATE: Thursday, June 12, 2008
TIME: 1:30 PM
LOCATION: 828 W. Peachtree St, NW - 207A Exec Boardroom
There has been an almost 60 percent increase in health care expenditures in the US in the past seven years. Employer-sponsored health coverage premiums have increased significantly (87 percent) in this same period. Besides the cost of care for chronic conditions such as migraine, arthritis and diabetes, absenteeism linked to these diseases also adds financial strain. Current health financial models focus on past spending instead of modeling based on current health burdens and future trends. This approach leads to ineffective health maintenance and cost management.
Identifying the diseases which affect the most employees and are also the most costly (in terms of productivity, work-loss-days, treatment etc) is necessary, since this allows the employer to identify which combination of policies may best address the health burdens. Previous attempts at modeling health burdens have been limited as they ignore incomplete data sets, and also model each disease as a separate entity. This research investigated the hypothesis that using Bayesian methodology we will be able to create a precise predictive model of the health burdens being faced by corporations, allowing for better policies to be designed. It was also hypothesized that these models would be able to create means for more comprehensive disease and risk factor analysis.
The first specific aim was to identify which diseases were the most costly to employers both directly and indirectly, and the pathogenesis of these diseases. Co-morbidity of diseases was also taken into account as in many cases these diseases are not treated independently. This information was taken into account when designing the models as the inference was disease specific.
The Bayesian models were able to estimate the predicted disease burdens for corporations, including predicting the percentage of individuals with multiple diseases. The model was also comparable to, or better than current estimators on the market with limited input. The outputs of the model were also able to give further insight into the disease interactions which creates an avenue for further research in disease management.