BACKGROUND: We aimed to derive and validate a parsimonious and pragmatic clinical prediction rule using the concepts of Predisposition, Infection, Response, and Organ Dysfunction to predict in-hospital mortality; and to compare it with other prediction rules, as well as with conventional biomarkers for evaluating the mortality risk of patients with suspected sepsis in the emergency department (ED). METHODS: We conducted a pragmatic cohort study with consecutive ED patients aged 18 or older with documented diagnostic codes of infection and two sets of blood culture ordered by physicians between 2010 and 2012 in a tertiary teaching hospital. RESULTS: 7011 and 12,110 patients were included in the derivation cohort and the validation cohort for the final analysis. There were 479 deaths (7%) in the derivation cohort and 1145 deaths (9%) in the validation cohort. Independent predictors of death were absence of Chills (odds ratio: 2.28, 95% confidence interval: 1.75-2.97), Hypothermia (2.12, 1.57-2.85), Anemia (2.45, 1.97-3.04), wide Red cell Distribution Width (RDW) (3.27, 2.63-4.05) and history of Malignancy (2.00, 1.63-2.46). This novel clinical prediction rule (CHARM) performed well for stratifying patients into mortality risk groups (sensitivity: 99.4%, negative predictive value 99.7%, receiver operating characteristic area 0.77). The CHARM score also outperformed the other scores or biomarkers such as PIRO, SIRS, MEDS, CURB-65, C-reactive protein, procalcitonin and lactate (all p<.05). CONCLUSIONS: In patients with suspected sepsis, this parsimonious and pragmatic model could be utilized to stratify the mortality risk of patients in the early stage of sepsis.