Fig. 5

Refinement of the FPHR3p risk model and comparison to other mortality risk assessment strategies. Candidate variables to refine FPHR3p risk modeling were analysed in a step-wise model selection process as depicted in Additional file 1: Fig S2. In brief, 0–3 combinations of significant prognostic variables previously identified together with FPHR3p were correlated with overall survival at baseline (stage 1 model selection) and first follow-up (stage 2 model selection) using multivariate Cox proportional hazard models. At each stage, models with significant estimate sets (Wald Z-test), better survival fit (likelihood ratio test (LRT) vs. FPHR3p-alone model), and better prediction power (Akaike information criterion (AIC) and concordance index (C-index)) than the FPHR3p-alone model were selected. a, b Relationship between inverted AIC (a), C-index (b) and the number of model estimates for FPHR3p-modifying variable sets according to baseline patient risk parameters. Each point represents a single model, point colour codes for model performance in stage 1 and 2 selection steps. Stage 1 and 2 best models were labeled with names (#1 to #5). c, d Optimal modifying FPHR3p variables sets identified in previous analyses were used to calculate FPHR3p-derived scores (for used low risk-criteria cut-offs refer to this figure). The prediction power of modified FPHR3p models in comparison to other risk assessment strategies is shown according to the inverted AIC/C-index relationship for each studied model. Prediction power according to risk parameters obtained at baseline (c) and first follow-up (d) are depicted. Each point represents a single model, whereas best predicting models are found in the right upper corner. Only models with pLRT < 0.05 and regression estimate p’s < 0.05 are shown. Blue: values for comparator risk scales. Highlighted are: red—models with better performance than comparator risk scales, orange—top 10 best C-index models. Best performing test models are labelled with names and include following parameters: #1 FPHR3p + age, #2 FPHR3p + GFR, #3 FPHR3p + RAA, #4 FPHR3p + RAA + age, #5 FPHR3p + RAA + GFR. GFR = glomerular filtration rate, RAA = right atrium area, N_baseline = 130, N_{1st follow-up} = 112, N_deceased = 43