Robust methods using graph and PCA for detection of anomalies in medical records

Abstract

Wellbeing in basic words is normalcy in health of human body and disease is unusual condition that influences typical working of human body with no outer wounds. Health care is about the aversion of maladies by finding and treatment. There are several methods and models developed in health care to predict and classify the chronic diseases, but what if the health care record contains some anomalies. A well designed model also would provide wrong results, if the input data contains anomalies. Any wrong decision in health care management would cost the life of the patient. In this work we have modeled Graph method and PCA method to detect anomalies in health care records by the means of frequency of incidences of disease codes in graph and correlation among the disease codes. We have used CMS medi-claim dataset in which diseases are expressed in terms of International disease code (ICD) and Hierarchical Condition Category (HCC) code to indicate the patient health condition. Game theory approach is used for evaluation of the model. The results of this work have been proved to be promising when compared with existing techniques. Since this work is related life of a patient’s, the results should be re looked by the domain experts before taking any decisions.