A multidrug-resistant microorganism infection risk prediction model: development and validation in an emergency medicine population

The aim was to develop a predictive model of infection by multidrug-resistant microorganisms (MDRO). A national, retrospective cohort study was carried out including all patients attended for an infectious disease in 54 Spanish Emergency Departments (ED), in whom a microbiological isolation was available from a culture obtained during their attention in the ED. A MDRO infection prediction model was created in a derivation cohort using backward logistic regression. Those variables significant at p < 0.05 assigned an integer score proportional to the regression coefficient. The model was then internally validated by k-fold cross-validation and in the validation cohort. A total of 5460 patients were included; 1345 (24.6%) were considered to have a MDRO infection. Twelve independent risk factors were identified in the derivation cohort and were combined into an overall score, the ATM (assessment of threat for MDRO) score. The model achieved an area under the curve-receiver operating curve of 0.76 (CI 95% 0.74–0.78) in the derivation cohort and 0.72 (CI 95% 0.70–0.75) in the validation cohort (p = 0.0584). Patients were then split into 6 risk categories and had the following rates of risk: 7% (0–2 points), 16% (3–5 points), 24% (6–9 points), 33% (10–14 points), 47% (15–21 points), and 71% (> 21 points). Findings were similar in the validation cohort. Several patient-specific factors were independently associated with MDRO infection risk. When integrated into a clinical prediction rule, higher risk scores and risk classes were related to an increased risk for MDRO infection. This clinical prediction rule could be used by providers to identify patients at high risk and help to guide antibiotic strategy decisions, while accounting for clinical judgment.