DCM-PROGRESS: predicting end-stage heart failure in non-ischemic dilated cardiomyopathy patients

Figures

• 

  Figure 1. Discriminative performance and model weights of a lasso model predicting end-stage heart failure (a composite of death, HTx, and VAD) in patients with DCM N.b. heart transplant: HTx, VAD: ventricular assist device, DCM: non-ischemic dilated cardiomyopathy. The c-statistics and 95% confidence intervals are provided as figure annotations, and were estimated using a 1999 iteration bias corrected and accelerated bootstrap. The model weights represent log odds ratio per SD or per class increase for hypertension and NYHA class.

• 

  Figure 2. Calibration and decision curve analysis comparing DCM-PROGRESS to the MAGGIC score N.b. All presented analyses were conducted in the external validation dataset from ICRC consisting of 235 DCM patients of whom 27 developed end-stage HF at 5 years. Panels A and D present the predicted risk from DCM-PROGRESS and MAGGIC, respectively. Panels B and E, present depict the model calibration for DCM-PROGRESS and MAGGIC. Note that the DCM-PROGRESS model was first recalibrated (see Supplementary note) using the UMCU training data addressing bias introduced by the lasso algorithm (see methods). For reference performance of the uncalibrated model is included in panel C. Based on the predicted risk, participants were grouped into 6 equally size bins, where the x-axis point represents the average bin-specific predicted risk and the y-axis point the mean incidence of end-stage HF. The confidence intervals for the bin specific observed end-stage HF risk were calculated using a beta distribution. Calibration slope and calibration-in-the-large estimates were calculated using the entire dataset without creating bins, with 95% confidence intervals included in brackets. Here a perfectly calibrated…

• 

  Figure 3. A risk communication illustration comparing a patients estimated risk against a reference population (left panel) and the proportion each clinical characteristics contributed to this risk (right panel) N.b. The left panel plots a hypothetical patients risk as a vertical dashed line on top of a simulated reference population based on the ICRC data (Figure 2A) with their “predicted risk” and “population percentile” annotated. The right panel depicts the relative contribution of the hypothetical patient characteristics to the estimated risk. The risk was estimated based on the following inputs: SBP of 70, r-axis of −5, TAPSE of 25, height of 182 and NYHA class of 1.