Today in our journal club, we discussed a paper by Weissman et al. (2021) titled “Effects of Neighborhood-level Data on Performance and Algorithmic Equity of a Model That Predicts 30-day Heart Failure Readmissions at an Urban Academic Medical Center,” published in the Journal of Cardiac Failure.
We began our discussion with an overview of area-level indices derived from geocode Census data–these ranged from the Neighborhood Stress Score (NSS7), Area Deprivation Index (ADI), and Social Vulnerability Index (SVI). Weissman et al. found that model performance and algorithmic equity for predicting 30-day readmission in congestive heart failure patients was not significantly improved with the addition of ADI patients within their test dataset. We nevertheless found the paper to be a valuable example of investigating subgroup-level variation in predictive model performance and incorporating both measures of algorithmic equity and area-level indices into our own analyses.
Looking forward, we talked about how it would be important to consider where the addition of area-level data might provide the most value as we aim to produce clinically useful tools and insights and reduce algorithmic inequity. Our discussion also touched on additional recent papers by Vest (2021) and Rethorn (2020).