Associate Professor of IE
Oklahoma State University
Monday, October 31, 2016
3-5pm JDT 500
Diabetes is one of the most serious chronic conditions that affects more than 29 million Americans. It leads to several complications such as kidney disease, heart disease, neurological disorders, and eye complications (retinopathy). Among them, diabetic retinopathy is the leading cause of vision loss in the US. Retinopathy is preventable and existing treatments can slow disease progress at its early stage. The traditional method for diagnosing diabetic retinopathy is a comprehensive eye examination in which a patient’s eye is dilated and an ophthalmologist takes a photo of the retina. Unfortunately, the annual diabetic retinopathy evaluation has one of the lowest rates of patient compliance because of the unpleasant eye dilation and the low availability of ophthalmologists.
In this research, we first conducted comorbidity analysis to generate association rules among major diabetic complications. Then we applied a data mining approach to develop a clinical decision support system (CDSS) to detect diabetic retinopathy. We analyzed the demographic and lab data of more than 1.4 million diabetic patients, and we developed four sets of predictive models. The first set encompasses the models that have been developed using lab and demographic data. In the second set, comorbidity data was included in addition to basic data. The third set consists of models that are built using the oversampled data by applying SMOTE method. The fourth set includes ensemble models that have been developed using the outputs of different single predictive models. The accuracy of the best model is close to 90%. Only a simple blood test is required to apply this CDSS. Hence, the CDSS can be used by primary care providers as part of routine diabetic care to solve the problem of low compliance with annual retinopathy screenings. Physicians can detect the onset of diabetic retinopathy with a high degree of certainty, thus preventing its onset and facilitating treatment at its earliest stages.
Dr. Tieming Liu is an associate professor at the School of Industrial Engineering and Management, Oklahoma State University. He received his doctoral degree in Transportation and Logistics from the Massachusetts Institute of Technology in 2005, his master’s degree in Industrial Engineering and Management Science from Northwestern University in 2001, and his master’s and bachelor’s degrees in Control Theory and Control Engineering from Tsinghua University in 2000 and 1997, respectively. His research interests include supply chain management, logistics planning, and healthcare analytics. His research has been published in major journals, including IIE Transactions, Interfaces, Production and Operations Management, Naval Research Logistics, Operations Research Letters, European Journal of Operational Research, among others.