From quality measures to disparity signatures: Elucidating cancer within-hospital hospice racial disparities using clinically informed machine learning.

Abstract

179 Background: Racial disparities in hospice care are well documented for patients with cancer, but the existence, direction, and extent of disparity findings are contradictory across the literature. Current methods to identify racial disparities aggregate data to produce single-value quality measures that exclude important patient quality elements, such as timing of initiation and length of stay, and, consequently, lack information to identify actionable equity improvement insights. Our goal was to develop an explainable machine learning approach that elucidates healthcare disparities and provides more actionable quality improvement information. Methods: We infused clinical information with an engineering systems modeling approach, and data science to develop a time-by-utilization profile per patient group at each hospital. We applied this approach to US Medicare hospice utilization data for a patient cohort of beneficiaries with poor-prognosis advanced cancer that died April-December 2016. We calculated the difference between group profiles for people of color and white people to identify racial disparity signatures. Using machine learning, we clustered racial disparity signatures across hospitals and compared these clusters to classic quality measures and hospital characteristics. Results: With 45,125 patients across 362 hospitals, we identified 7 clusters; 4 clusters (n=190 hospitals) showed more hospice utilization by people of color than White people, but at varying times, 2 clusters (n=106) showed more white patient hospice utilization than people of color, and 1 cluster (n=66) showed no difference. We found no evidence that quality measures or hospital characteristics explain the shape of disparity signatures. Our approach of explicitly incorporating time ``informs" machine learning to explicitly design for an understanding of racial disparities. Our findings suggest unique and heterogeneous within-hospital racial disparity behaviors that cannot be predicted from quality measures or exogenous hospital characteristics and show how the true shape of disparities can be distorted through the lens of quality measures. Conclusions: The developed informatics approach is first to elucidate the shape of hospice racial disparities within hospitals algorithmically from administrative claims data. Racial disparity signatures that may help explain mixed literature findings and highlight the importance of developing bespoke equity solutions.