Every year, more than 11,000 homecare agencies across the United States provide care to more than 5 million older adults. Currently, about one in three homecare patients are hospitalized or visit an emergency department (ED) during the 30-60 day homecare episode. Up to 40% of these events are preventable with appropriate and timely care. In our pilot work, we developed a risk prediction model (called Homecare- CONCERN) that accurately identified patients at risk for hospital admission and ED visits solely from homecare clinical notes using NLP. This study brings together an interdisciplinary team of experts in homecare, data science, nursing and risk model development to explore whether cutting-edge data science approaches can improve timely identification of patients at risk in homecare.
Our specific aims are to: 1. Further develop and validate a preventable hospitalization or ED visit risk prediction model (Homecare- CONCERN). We will apply traditional (time varying Cox regression) and cutting-edge time-sensitive analytical methods (Deep Survival Analysis and Long-Short Term Memory Neural Network) for risk model development. 2. Prepare Homecare-CONCERN for clinical trial via pilot testing. We will apply user centered design to develop Homecare-CONCERN clinical decision support tool and pilot test the tool for clinical validity and acceptability. 3. Inform the future implementation of Homecare-CONCERN clinical decision support tool in the homecare setting. We will examine if all risk elements can be mapped to a data standard (Fast Healthcare Interoperability Resources - FHIR) and conduct interviews with key informants across the US about current readiness, barriers and facilitators, and implementation strategies for adopting such tools in homecare setting. This proposal addresses the AHRQ program announcement (PA-18-795) to harness data to improve healthcare quality and patient outcomes. The study will build a first-of-a-kind clinical decision support system to trigger timely and personalized alerts about concerning patient trends that activate appropriate and timely care to prevent avoidable hospitalizations and ED visits from homecare.
Our previous work has shown that clinician documentation patterns and content are proxy for patient risk. ?Concerning? documentation patterns can be used to identify patients who are deteriorating so clinical team can intervene before it is too late. Although several previous studies attempted to create models predicting patient?s risk in homecare, no studies to date used all available data (clinical notes and electronic health record data). This study aims to use all available clinical data on homecare patients to create personalized models of risk for preventable hospitalization and emergency department visit. We will also explore the feasibility and readiness of homecare agencies to adopt such predictive tools. Study results will catalyze a paradigm shift in homecare by making it possible to develop a first-of-its-kind data-driven clinical decision support system.
