Promoting Algorithmic Equity in In-Hospital Mortality Prediction

Models to predict death in the hospital are used in clinical medicine and research; but existing models perform differently across different groups, which really matters for patients hospitalized with acute respiratory failure (breathing problems) or sepsis (infection with an abnormal immune system response) because (1) such patients face high risks of worse model performance due to diagnostic uncertainty and high death risk; (2) racial and ethnic minority patients have increased death risk with acute respiratory failure or sepsis; and (3) models to predict death in the hospital are frequently used for these populations. Therefore, we are planning to create and test a new model to predict death in the hospital that is as fair as possible across key groups defined by race, ethnicity, sex, age, primary language, insurance status, and social vulnerability.

Kaiser Permanente Northern California
MedStar Health

National Institutes of Health/National Heart, Lung, and Blood Institute (NIH/NHLBI)