Prediction of Diagnostic Codes of Chronic Condition for Preventive Care

Abstract

The advent of big data technology and Machine Learning, in clinical and healthcare sectors, it has become possible for detailed analysis of healthcare records to achieve early detection of chronic illness and making it possible for member patients to take preventive care. Nonetheless, the investigation exactness is decreased when the nature of clinical information is inadequate. Also, various regions show novel attributes of certain provincial sicknesses, which may debilitate the prediction of the outbreak of the diseases.  The purpose of this work is to predict whether the member patient is suffering from chronic ailment or not using the medi-claim data. The medi-claim data is considered for the study because of its validity, authenticity and volume. The claim data of the member patient also provides history of ailments that the patient had suffered from. The claim data is pre-processed and transformed into Term Document Matrix (TDM) document. Five chronic disease such as heart disease, liver disease, kidney disease, cancer disease and diabetes disease are studied in this work. The experiment is conducted on TDM data and subspace data using individual prediction technique and ensemble approaches. Mean square error (MSE) is used for prediction model evaluation. Random forest regression technique produced promising result with average MSE of 0.06 after the outliers in TDM data were filtered. From this model it possible to draw significant clinical implications for preventive care of chronicle diseases.