Comparison among random forest, logistic regression, and existing clinical risk scores for predicting outcomes in patients with atrial fibrillation: A report from the J‐RHYTHM registry

Figures

• 

  FIGURE 1. C‐statistics of the outcomes. The receiver operating characteristic curves for (A) thromboembolisms, (B) major bleeding, and (C) all‐cause mortality are shown in the upper figures. The C‐statistic and 95% confidence intervals are presented in the lower table. The C‐statistics of the RF were compared to that for the LR and clinical risk scores. RF: random forest, LR: stepwise logistic regression. The other abbreviations of the risk scores are shown in the supplementary file. Compared to RF: * <0.05, † < 0.01, # <0.001

• 

  FIGURE 2. Permutation importance of the random forest model. Permutation importance for the classification of (A) thromboembolisms, (B) major bleeding, and (C) all‐cause mortality. The mean decrease in the AUC is a measure of the permutation importance. It shows how much a prediction made by the random forest model is degraded if a particular variable is shuffled (effectively removed). By inference, a variable with a larger decrease in the AUC must be contributing more to the model's predictive ability. AUC: area under the curve. The definitions of abnormal renal or hepatic function and other abbreviations are as in Table 1

• 

  FIGURE 3. Independent predictors of the stepwise logistic regression model. The independent predictors and their odds ratio and 95% confidence interval of (A) thromboembolisms, (B) major bleeding, and (C) all‐cause mortality are shown. The definitions of abnormal renal or hepatic function and other abbreviations are as in Table 1

• 

  FIGURE 4. Calibration plots. The plots comparing the predicted event rates (horizontal axis) and observed event rates (vertical axis) for thromboembolisms (A, B, C), major bleeding (D, E, F), and all‐cause mortality (G, H, I) are shown. The blue line in the RF and LR indicates the trend for the calibration. When the intersect of the observed and expected event rates is below the dotted line, this indicates an overestimation of the event risks for that quintile. (C, F, I) The plots comparing the categorical score (horizontal axis) and observed event rates (vertical axis) are shown. The high‐risk event rate cutoff values were 3.0%, 4.1%, and 7.8% for thromboembolisms, major bleeding, and all‐cause mortality, respectively (red shaded area). The abbreviations and categorical grouping are shown in Table 1 and the Supplementary File (Appendix S1 and Table S1). RF: random forest, LR: stepwise logistic regression