Electronic Health Records (EHRs) are an emerging data source that enables researchers to employ a data-driven approach for the prediction of health outcomes and for patient risk- stratification. Machine learning methods can be used to identify underlying patterns in an individual’s EHRs, which can predict his/her future health condition. In this talk I will present two machine learning models we have developed, one for identifying patterns of associated medical conditions among patients suffering from kidney disease and the other for identifying CKD severity-stages from standard office visit records.
For the former, we apply a machine learning method, namely, topic modeling, to EHRs, to identify distinct groups of co-occurring conditions. For the latter, that is, identifying CKD severity-stages from standard office visit records, we propose a hierarchical meta-classification method, employing simple quantitative non-text features gathered from office visit records, while addressing data imbalance. Our method effectively stratifies CKD severity-levels, obtaining high average sensitivity, precision and F-measure (~93%).
Moumita Bhattacharya is a 5th year Computer Science PhD candidate working at the Computational Biomedicine and Machine Learning Lab (Advisor: Dr. Hagit Shatkay) at University of Delaware. My dissertation research is in computational medicine, and concerns prediction of disease applying machine learning models on clinical data. I focus specifically on CKD and Heart Disease. My goal is to assist healthcare providers in clinical decision making by predicting onset of disease or adverse events such as hospitalization. We have obtained a number of promising results, including predicting the risk of CKD severity-levels from standard office visit records, identifying several novel risk factors of sudden cardiac death in hypertrophic cardiomyopathy patients, and characterizing patterns of co-occurring medical conditions among patients suffering from kidney disease.
