Fig. 3

Frequency of biomarker retention in the final prognostic model. To test the robustness of the association with outcome, we performed 500 bootstraps with random resampling. For each of the iterations, backward selection using the Akaike information criterion (AIC) was used to identify the variables retained in the final model. Variables included in the base model before backward selection of each iteration were age, sex, BMI, systolic BP, diastolic BP, HbA1c, diabetes duration, low-density lipoprotein cholesterol (LDL), albumin: creatinine ratio (ACR), estimated glomerular filtration rate (eGFR), treatment, prior history of HF, prior history of CV disease, NTproBNP, smoking, and P-ETP. The proportion (%) of times that P-ETP was retained in the final model is listed at the top of the figure