ObjectiveTo develop a simple score to help assess the presence or absence of infection in critically ill patients using routinely available variables.DesignObservational study of a prospective cohort of patients divided into a developmental set (n = 353) and a validation set (n = 140).SettingDepartment of intensive care at an academic tertiary care center.PatientsFour hundred and ninety-three adult patients admitted to the intensive care unit for ≥24 hrs.InterventionsNone.Measurements and Main ResultsThe presence of infection was defined using the Centers for Disease Control definitions. Body temperature, heart rate, respiratory rate, white blood cell count, and C-reactive protein concentrations were measured, and the Sequential Organ Failure Assessment score was calculated throughout the intensive care unit stay. Infection was documented in 92 of the 353 patients (26%) in the developmental set and in 41 of the 140 patients (29%) in the validation set. Univariate logistic regression was used to select significant predictors for infection. Each continuous predictor was transformed in a categorical variable using a robust locally weighted least square regression between infection and the continuous variable of interest. When more than two cate-gories were created, the variable was separated into iso-weighted dummy variables. A multiple logistic regression model predicting infection was calculated with all the variables coded 1 or 0 allowing for relative scoring of the different predictors. The resulting Infection Probability Score consisted of six different variables and ranged from 0 to 26 points (0–2 for temperature, 0–12 for heart rate, 0–1 for respiratory rate, 0–3 for white blood cell count, 0–6 for C-reactive protein, 0–2 for Sequential Organ Failure Assessment score).The best predictors for infection were heart rate and C-reactive protein, whereas respiratory rate was found to have the poorest predictive value. The cutoff value for the Infection Probability Score was 14 points, with a positive predictive value of 53.6% and a negative predictive value of 89.5%. Model performance was very good (Hosmer-Lemeshow statistic,p= .918), and the areas under receiver operating characteristic curves were 0.820 for the developmental set and 0.873 for the validation set.ConclusionsThe Infection Probability Score is a simple score that can help assess the probability of infection in critically ill patients. The variables used are simple, routinely available, and familiar to clinicians. Patients with a score <14 points have only a 10% risk of infection.