Category CS P06 Optimizing Health Care Resources Allocation Using Data Mining

Approaches

Abstract The goal of my project is to use a systematic approach to segment patients

based on predictive morbidity indices such as the Charlson Index.

Individuals do not know their exact score on the Charlson Index and

determining this score exactly can be very time consuming and resource

intensive. Through data analysis and data mining techniques, I was able to

determine some of the factors that can very accurately predict the

patient’s score. The benefit is that scarce medical resources, such as

vaccines that are in short supply and attention can be directed towards

people with higher scores, which will ensure that such resources get

applied effectively in preventing spread of epidemics.



To perform data mining, I used a tool called SQL Server on data that was

made available by a prime insurance company. The medical claims data

gave information regarding member profiles, health indicators of members,

and the member’s severity of illness at that point. All the medical conditions

as well as their demographic profiles were used as features to predict

the Charlson Index.



Different data mining techniques were used in this process. After inputting

the different features that determined the severity of illness, the data

mining tool created a decision tree which predicted one or more discrete

variables, based on the other attributes in the dataset. There were two

different analyses done: one, using only health indicators to predict the

severity of illness and another, using member demographic details, such

as age and gender, as well as health indicators to predict the severity of

illness. For each analysis a dependency network of attributes was

created and decision tree evaluation was done. Accuracy charts and lift

charts were created for validation.



My project proves that a small number of specific health related data is a

better predictor of the Charlson Score and that health care providers

should use these elements rather that traditional demographic measures

such as age and gender to make decisions regarding allocation of scarce

health care resources.

Bibliography
First Previous Next Last