A Model for Health Response Assessment (HRA)

Document Type


Lead Author Type

MBI Masters Student


Dr. Guenter Tusch, tuschg@gvsu.edu

Embargo Period



The goal of my Capstone project is to develop an extended model for Health Response Assessment (HRA) for a local insurance agency (Priority Health). The project is mainly based on targeting diverse segments of consumers with the most relevant products and services. The main objective of this project is to predict members with higher cost on the basis of a health questionnaire. So we could find consumers before they might be having higher cost and we could provide them good care, which would help reduce the company’s overall cost. In this project I used the UPMC model as a starting point on our members’ dataset to evaluate whether it is at all applicable to our members or to what extent. Another objective was, if the first approach was not successful, to develop a new model to predict members with higher cost.

The project is based on data of the initial Health Questionnaire from the year 2013. It contains a total of 388 questions and 8968 observations. The data types are only are categorical and numeric. I used a frequency distribution graph to identify members in the dataset that potentially can generate high cost expenditures in the future. According to this dataset I decided to have the top 10% of all expenditures defined as high cost and rest as low cost. The UPMC model, developed by the University of Pittsburgh Medical Center to predict high cost members, is able to predict a total of 505 members and out of those 160 members are true positive. The final model was developed using a different set of questions. It is able to predict a total of 328 members and out of that 146 members are true positive. To develop this model linear regression, multiple regression, and a decision tree algorithm was used. Only those questions whose impact is significant on the total cost were selected from the Health Questionnaire. Generalization of the developed model was assessed by a validation technique.

Because the developed model is based only on Priority Health data, it is not necessarily generalizable to other insurers or health agencies, while the UPMC model is considered a universal model that can be used in the entire US.

This document is currently not available here.