Risk Stratification for Early Detection of Diabetes and Hypertension in Resource-Limited Settings: Machine Learning Analysis

• DOI: 10.2196/20123
• Publication Date: 2021-01-21T15:32:41Z
• Authors: Justin J Boutilier; Timothy C Y Chan; Manish Ranjan; Sarang Deo
• Keywords: Risk Stratification

Background The impending scale up of noncommunicable disease screening programs in low- and middle-income countries coupled with limited health resources require that such be as accurate possible at identifying patients high risk. Objective aim this study was to develop machine learning–based risk stratification algorithms for diabetes hypertension are tailored the at-risk population served by community-based low-resource settings. Methods We trained tested our models using data from 2278 collected community workers through door-to-door camp-based screenings urban slums Hyderabad, India between July 14, 2015 April 21, 2018. determined best predicting short-term (2-month) (a model a hypertension) compared these previously developed scores United States Kingdom prediction accuracy characterized area under receiver operating characteristic curve (AUC) number false negatives. Results found based on random forest had highest both diseases were able outperform US UK terms AUC 35.5% (improvement 0.239 0.671 0.910) 13.5% 0.094 0.698 0.792). For fixed specificity 0.9, reduce expected negatives 620 per 1000 220 hypertension. This improvement reduces cost incorrect $1.99 (or 35%) $1.60 21%) Conclusions In next decade, systems many planning spend significant demonstrates learning can leveraged effectively utilize improving stratification.