Prediction of cardiac arrest in patients with heart failure in Sweden: a registry study with development of a machine learning model

Figures

• 

  Figure 1. Flowchart of the three possible patient outcomes for patients discharged from hospital after first hospitalisation for heart failure. The primary outcome in this study is cardiac arrest registered in the Swedish Registry for Cardiopulmonary Resuscitation (SRCR). Death without resuscitation is a competing event for cardiac arrest, these are patients who died but were not registered in the SRCR. Censored are the patients who survived to the final follow-up date, 15 November 2022, without any of the aforementioned events. CRT-d, cardiac resynchronisation therapy with defibrillator; ICD, implantable cardioverter defibrillator.

• 

  Figure 2. Cumulative incidence for cardiac arrest and death without resuscitation in the study population of patients discharged after first hospitalisation for heart failure. Cumulative incidence functions were estimated using the Aalen-Johansen estimator and are displayed as step functions reflecting observed event times. Including a zoomed inset plot of the cumulative incidence of cardiac arrest. CPR, cardiopulmonary resuscitation.

• 

  Figure 3. Evaluation of performance of Random Survival Forest for competing risk predicting cardiac arrest versus death without resuscitation or no event with 82 predictors. (A) Brier score over time. (B) Observed cumulative incidence and model predicted cumulative incidence of cardiac arrest over time. (C) Receiver operating characteristics graph of Random Forest model predicting cardiac arrest with time-dependent area under the curve at 1, 3, 5 and 10 years after inclusion. AUC, area under the curve; RF model: Random Forest model.

• 

  Figure 4. Variable importance of the 10 most predictive variables from the Random Survival Forest model for competing risk predicting cardiac arrest vs death without resuscitation or no event. e-GFR, estimated glomerular filtration rate; NTproBNP, N-terminal pro B-type natriuretic peptide.

• 

  Figure 5. Partial dependency plots for the top 10 variables, plotted against years lost to cardiac arrest. A, aortic valve surgery; A+M, aortic and mitral valve surgery; BNP, B type natriuretic peptide; CABG, coronary artery bypass graft surgery; e-GFR, estimated glomerular filtration rate; M, mitral valve surgery; No, no valve surgery; NTproBNP, N-terminal pro-B-type natriuretic peptide; Other, other valve surgery; PCI, percutaneous coronary intervention; Yes, unspecified valve surgery.