Electronic Health Records (EHRs) encompass patients’ diagnoses, hospitalizations, and medication histories, offering a wealth of data. Although EHR-based research, particularly in decision prediction, has made significant strides, challenges remain due to the inherently sparse and irregular nature of EHR data, which limits their direct application in time-series analysis. Physicians treating individuals with chronic illnesses must anticipate the progression of their patients' conditions, as accurate forecasts enable more informed and timely treatment decisions. The strength of prediction lies in prevention—intervening early is often more effective than attempting to reverse damage later. In this study, we present a data-driven model designed to deliver accurate and efficient predictions of disease trajectories using electronic health records (EHRs) from Veterans Affairs hospitals. Prediction of disease progression represents a fundamental challenge. EHRs contain vast volumes of frequently updated, high-dimensional, and irregularly spaced data in various formats, including numerical, textual, image, and video data. To address this complexity, we propose a new approach for predicting the progression of diabetes. This method has the potential to improve early intervention, prevent further health deterioration, and ultimately extend patients' lives. The method is based on the PM GenAI, a novel approach that significantly improves classification and regression results. The method is compared to traditional techniques such as ARIMA, LTMS, and RF showing significant improvement in disease progression evaluation. The method is demonstrated on diabetes data.
Philip de Melo (Mon,) studied this question.