Risk Prediction Methods Based on Electronic Medical Records and Social Surveys for Improving Patient Outcomes and Enabling Targeted Care Services

Abstract

This paper focuses on effectiveness of methods for improving patient quality (e.g. improving treatment adherence, reducing adverse events) outcomes and targeted interventions based on psychosocial and clinical risk factors embedded structured and unstructured elements in medical records. Current methods on outcomes analysis such as adherence to treatment regimen largely rely on survey instruments, and provide lagging indicators that inhibits timely intervention and care services. In this paper we present a novel early-warning method that can predict patients at risk of non-adherence based on clinical rules, natural language processing techniques and predictive algorithms applied to risk factor information embedded in electronic medical records. We conducted studies on the effectiveness of our risk estimation methods across 2.5 million patient-visit records from a community cancer clinic that spans a 14 year time-horizon. We identified 2 distinct patient groups, between 26 and 38 (mean risk score, r=0.77, s=0.22), and 75 and 90 (r=0.81, s=0.19) years of age respectively, who exhibited a strong likelihood of non-adherence to treatment regimen. We obtained a reasonably high C-statistic (> 0.77) on predicting outcomes based on the risk factors. The dominant risk-factors, not surprisingly, included psychosocial (e.g. depression and lack of support), medical (e.g. side-effects) and financial (e.g. co-pay). We finally discuss the effectiveness of the methods for targeted and improved health care services.

Keywords: Electronic Medical Records, Medical Adherence, Cancer, Predictive Analytics, Text Mining

DOI: 10.54941/ahfe100297