Fig. 3
Transferrin saturation is a SOFA-independent survival predictor. a Predictive power of age- and sex-corrected serum iron concentration, ferritin, transferrin (TF), TF saturation (TF-Sat), hematocrit (HT), and SOFA. Correlation with ICU mortality for each of the investigated parameters was analyzed with a separate logistic regression model with inclusion of age and sex as confounders. Significance for regression β estimates was determined with Wald Z test. Results are presented as a forest plot. Points represent exponentiated β estimates, and whiskers represent 95% CI. Points are labeled with exponentiated estimate, CI, and p values. Survivors: n = 41, deceased: n = 20.
b Comparison of predictive power for SOFA and compound SOFA-iron parameter models. A family of logistic regression models correlating SOFA, 0–3 combinations of the iron/hematological parameters linked in a to survival as well as of age and sex as confounders with ICU mortality was generated. For each model, estimate p-values (determined with Wald Z test), Akaike Information Criterion (AIC) and concordance index (C-index) was calculated. Best performing models, i.e., those with significant non-confounder estimates and better AIC and C-index than the SOFA-alone model, are labeled with non-confounder variable names. Survivors: n = 41, deceased: n = 20.
c Estimates for non-confounder parameters of the best-performing SOFA-compound model identified in (B, SOFA + TF-Sat) presented as a forest plot. Significance for regression β estimates was determined with Wald Z test. Points represent exponentiated β estimates, whiskers represent 95% CI. Points are labeled with exponentiated estimate, CI, and p values. Significance of the inclusion of the TF-Sat term was assessed with likelihood ratio test (pLRT, compound model vs SOFA-alone model). Survivors: n = 41, deceased: n = 20