Figure 1.

Study flow. We randomized the patients into training and testing sets with similar sample sizes in each group. We then tuned the random forest (RF) model, chose the best-fit RF model, and fitted multinomial logistic (ML) model using the training set. Using the ML and RF models, we predicted 6-category causes of death among the patients in the testing set. During the cross-validation phase, we followed a similar protocol but with swapped datasets.