For all biomarkers, the time-updated considering value was a stronger predictor than the initial value though for PCT and copeptin also the initial value of the marker remained significant in the model with both the initial and the time-updated marker (P = 0.046 and P = 0.03, respectively).Performance of multivariable statistical models in LRTI patients without CAPThe multivariable models for predicting serious complications developed in CAP patients extrapolated well if evaluated in 434 patients with presumed other LRTI in the ProHOSP trial. The AUCs for these patients and the model with all CURB65 covariates and proADM, or with all biomarkers, respectively, were both 0.80 and thus better than on the original population.

There was also no indication of serious miscalibration of these models: A total of 36 serious complications were observed in non-CAP patients compared to predicted numbers of complications of 41.2 and 40.2 patients according to the two models, respectively (P = 0.39 and P = 0.48 for X2 goodness of fit test). The model with only clinical covariates extrapolated worse with an AUC of 0.75 in non-CAP patients and some evidence of miscalibration with 49.7 predicted events (P = 0.04).DiscussionIn this large community-based sample of patients with CAP and other LRTI from a multicenter study [34], five prohormones from distinct biologic pathways were specific predictors for short term serious complications with moderate improvement of clinical risk scores. Thereby, this study validates a series of previous smaller trials demonstrating a clinical utility of prohormones for an optimized risk prediction in LRTI [8-25].

Meaningful statistical assessment of the potential clinical utility of a biomarker is challenging. In addition to classical performance measures Cilengitide like two group comparisons and ROC curves, more clinically meaningful statistical approaches have been put forward [44,48]. We performed several different statistical analyses to investigate the added value of biomarkers to clinical scores; more specifically, we assessed the addition of prohormones to PSI and CURB65 scores per se and to a multivariate regression model based on CURB65 covariates. We measured the prognostic performance of these models by several different quantities (AUC, Brier score and reclassification methods). Thereby, some prohormones, namely proADM, improved both clinical risk scores and were superior per se for serious complications prediction. The incorporation of a combination of biomarkers reflecting systemic inflammation, endothelial dysfunction, stress and cardiac function to the clinical risk scores improved their prognostic accuracy for prediction of short term complication rate and to a lesser extent mortality.