Fig 3.

Heatmap of preoperative risk vs postoperative risk. Preoperative (X-axis) and postoperative (Y-axis) risk scores were binned by percentile, and the counts per bin visualised as a heatmap in log scale. Preoperative risk predictions were generated using the random forest model trained on the preoperative features, including laboratory times, and the surrogate-ASA status. In (a) all patients are displayed, and in (b) only the in-hospital mortalities are shown. Some 78% of patients who die and have a preoperative risk percentile below 95% have an increased postoperative risk percentile. This is substantially greater than the percent of matched patients from a null distribution who have an increased percentile.