Illness severity scores have demonstrated poor generalization performance in forecasting mortality, particularly for patients at highest risk. One of the drivers of this discrepancy is the preponderance of patients at low risk of death in training cohorts. We sought to examine the effect of constraining the training data case mix on the performance of mortality forecasting for high-risk patients.
Contained herein are all of the original code files used to produce this work. The notebooks are documented and their use should be self-explanatory. Please contact me at cvc@mit.edu with any questions.