Predicting Acute Endophthalmitis for Patients with Cataract Surgery Using Hierarchical and Probabilistic Representation of Clinical Codes

Abstract

Post-surgery acute endophthalmitis is a rare ocular disease that is most prevalent among patients with cataract surgery. It is difficult to identify these rare but possible diseases as the symptoms are not quite visible. In addition, endophthalmitis introduces significant burden in public health and requires patients to undergo increased physician visits and procedures such as pars plana vitrectomy. Eventually, endophthalmitis patients become impaired with reduced vision and degraded quality of life. Therefore, the early diagnosis or prediction of acute endophthalmitis would benefit a significant portion of the patients to lead healthy life after the cataract surgery. In this study, we propose Predicting Cataract surgery-based acute Endophthalmitis (PrediCatE), a prediction technique that is applied on patient’s previous visits to predict acute endophthalmitis after their cataract surgery. We specifically apply neighborhood-based language modeling and classification algorithms on the Intelligent Research in Sight (IRIS®) registry dataset. In addition, we explain the use of mixture of clinical concepts in the classification process. Our empirical results demonstrate promising outcomes that will enhance the development and evaluation process of predictive modeling approaches in ophthalmic healthcare.