Suranga N Kasthurirathne1,2, Shaun Grannis1,3, Paul K Halverson2, Justin Morea3,4, Nir Menachemi1,2, Joshua R Vest1,2. 1. Center for Biomedical Informatics, Regenstrief Institute, 1101 W 10th St., Indianapolis, US. 2. Indiana University Richard M Fairbanks School of Public Health, Indianapolis, US. 3. Indiana University School of Medicine, 1044 W. Walnut St, R4 402D, Indianapolis, US. 4. Eskenazi Health, Indianapolis, US.
Abstract
BACKGROUND: Emerging interest in precision health and the increasing availability of varied patient and population-level datasets present considerable potential to enable analytical approaches to identify and mitigate the effects of upstream social factors. These issues are not satisfactorily addressed in typical medical care encounters and thus opportunities to improve health outcomes, reduce costs and improve care coordination are not realized. Further, methodological expertise on the use of varied patient and population-level datasets and machine learning to predict need of upstream services is limited. OBJECTIVE: To leverage a comprehensive range of clinical, behavioral, social risk, and social determinants of health (SDoH) factors in order to develop decision models capable of identifying patients in need of various "wraparound" social services. METHODS: We leveraged comprehensive patient and population-level datasets to build decision models capable of predicting need for behavioral health, dietitian, social work, or other social service referrals within a safety-net health system. We also evaluated the value of population-level SDoH datasets in improving machine learning performance. RESULTS: Decision models for each wraparound service reported performance measures ranging between 56% and 99%. These results were statistically superior to performance measures reported during a prior machine learning effort using a limited dataset. However, inclusion of additional population-level SDoH did not contribute to any performance improvements in our population of vulnerable patients seeking care at a safety net provider. CONCLUSIONS: Precision health enabled decision models that leverage a wide range of patient and population-level datasets and advanced machine learning methods are capable of predicting need of various wraparound social services with considerable performance measures.
BACKGROUND: Emerging interest in precision health and the increasing availability of varied patient and population-level datasets present considerable potential to enable analytical approaches to identify and mitigate the effects of upstream social factors. These issues are not satisfactorily addressed in typical medical care encounters and thus opportunities to improve health outcomes, reduce costs and improve care coordination are not realized. Further, methodological expertise on the use of varied patient and population-level datasets and machine learning to predict need of upstream services is limited. OBJECTIVE: To leverage a comprehensive range of clinical, behavioral, social risk, and social determinants of health (SDoH) factors in order to develop decision models capable of identifying patients in need of various "wraparound" social services. METHODS: We leveraged comprehensive patient and population-level datasets to build decision models capable of predicting need for behavioral health, dietitian, social work, or other social service referrals within a safety-net health system. We also evaluated the value of population-level SDoH datasets in improving machine learning performance. RESULTS: Decision models for each wraparound service reported performance measures ranging between 56% and 99%. These results were statistically superior to performance measures reported during a prior machine learning effort using a limited dataset. However, inclusion of additional population-level SDoH did not contribute to any performance improvements in our population of vulnerable patients seeking care at a safety net provider. CONCLUSIONS: Precision health enabled decision models that leverage a wide range of patient and population-level datasets and advanced machine learning methods are capable of predicting need of various wraparound social services with considerable performance measures.
Authors: Titus Schleyer; Linda Williams; Jonathan Gottlieb; Christopher Weaver; Michele Saysana; Jose Azar; Josh Sadowski; Chris Frederick; Siu Hui; Areeba Kara; Laura Ruppert; Sarah Zappone; Michael Bushey; Randall Grout; Peter J Embi Journal: Learn Health Syst Date: 2021-06-23