Comparison of linear and non-linear machine learning models for time-dependent readmission or mortality prediction among hospitalized heart failure patients

Figures

• 

  Fig. 1. The flow chart for patient inclusion and cohort splitting.

• 

  Fig. 2. The Kaplan-Meier survival curves and readmission or mortality prevalence curves in derivation and validation cohorts. (A) The Kaplan-Meier survival curve in derivation and validation cohort. (B) Readmission or mortality prevalence in derivation cohort. (C) Readmission or mortality prevalence in validation cohort.

• 

  Fig. 3. The flow chart for feature engineering and model derivation.

• 

  Fig. 4. Time-dependent AUC curves of derivation and validation cohorts in different survival prediction models.(A) Time-dependent AUC curves in multivariate Cox regression. (B) Time-dependent AUC curves in LASSO Cox regression. (C) Time-dependent AUC curves in random survival forest. (D) Time-dependent AUC curves in gradient boosting model.

• 

  Fig. 5. Time-dependent Brier score in different models. (A) Time-dependent Brier score in multivariate Cox regression. (B) Time-dependent Brier score in LASSO Cox regression. (C) Time-dependent Brier score in random survival forest. (D) Time-dependent Brier score in gradient boosting model.