Fig. 3

The value of DRC model (dyspnea, respiration rates > 20/min, and C-reactive protein > 20 mg/l) for pneumonia prediction in external validation population. A Venn diagrams showed the overlaps among patients with dyspnea, respiration rates > 20/min, and C-reactive protein > 20 mg/l. B ROC curve of DRC model showed the predictive accuracy was highest when choosing at least one positive item (1 score) as cut-off point. C The numbers of AFRI patients with different DRC scores were summarized in different pneumonia-severity groups. The average DRC scores in each severity group increased with the elevation of patients’ PSI classes. D The numbers of infected lung lobes increased with the rise of DRC scores. Data were presented as mean ± SD. *P < 0.01. E The numbers of AFRI patients with different DRC scores were summarized according to different respiratory pathogens. F ROC curves of DRC model and its simplified form, DR model (dyspnea and respiration rates > 20/min), for pneumonia prediction. For bacterial pneumonia, the AUC of the DR model was significantly less than that of the DRC model (P < 0.01). However, for viral pneumonia, no significant difference was found in the AUCs between two models. RR, respiration rates; CRP, C-reactive protein; ROC, receiver operating characteristic; AFRI, acute febrile respiratory illness; PSI, pneumonia severity index; AUC, area under curve