Using Machine Learning and Patient-Reported Outcomes to Identify Unnecessary Hospitalizations

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT I (Richard K. Leuchter, MD) am a UCLA Internal Medicine resident who will be joining the faculty as a clinician- scientist at UCLA in July 2022. I will practice hospital medicine and pursue health services research focused on identifying and reducing medical waste - patient care that provides no net benefit in certain clinical scenarios, and can also cause harm. I will build upon the excellent health services research training I received through the R38 StARR program, and continue my research using machine learning (ML) to identify and minimize medical waste. Unnecessary hospitalizations represent one of the single largest reservoirs of medical waste and disproportionately burden racial and ethnic minorities, but efforts to address this problem have been hindered by a lack of measures that can prospectively identify hospitalizations as unnecessary with acceptable accuracy. A critical barrier to measuring and reducing unnecessary hospitalizations is that claims data (e.g., billing information submitted to payers) lack enough clinical detail to accurately classify a hospitalization as “unnecessary.” Supplementing claims data with richer electronic health record (EHR) data offers potential to improve predictive accuracy, but EHR data do not routinely include discrete patient-reported outcomes (PROs) to quantify recovery from subjective symptoms (e.g., shortness of breath), making it difficult to adjudicate the necessity of admissions for diseases such as heart failure or pneumonia. To advance my career goals and work toward my overall aim of reducing the harms arising from wasteful medical practices (especially among disadvantaged patients), I propose a new method to identify unnecessary hospitalizations: train predictive ML models from EHR data that can identify admissions with a high likelihood of being unnecessary, and then assess model performance using a combination of clinical PROs and EHR outcomes. My overarching goal is to reduce wasteful and inequitable healthcare practices by becoming a leading principal investigator developing innovative and state of the art methods to minimize medical waste. To achieve this goal, I seek support from the NHLBI K38 Career Development Award. I will acquire skills in coding and using ML to predict health outcomes, measuring and analyzing PROs, and health/healthcare disparities research. I propose two specific research aims that align with my career development goals: 1) develop ML models that can identify Emergency Department (ED) admissions for cardiopulmonary illnesses with a high likelihood of being unnecessary, and 2) measure the prospective performance of these models using a combination of PROs and EHR data that will be collected from patients presenting to the ED. I will apply knowledge learned from my training to accomplish these aims, and plan to use the products of this research to inform an NHLBI K23 proposal for a single center pragmatic pilot trial that I plan to submit in 2023. The K38 Award would provide me with the training and skills needed to become a national leader in using emerging methods to reduce medical waste and its associated healthcare disparities.