Abstract
Hospital-acquired pneumonia (HAP) is prevalent in the neuro-intensive care unit (NICU), significantly increasing susceptibility to infections with multidrug-resistant organisms (MDROs), which result in high mortality rates and substantial healthcare burdens. Recognition and intervention are paramount. This study aimed to build a prediction model for MDRO infections among NICU patients with HAP. Clinical and laboratory data were collected from the NICU of a grade-A tertiary hospital. Five machine learning models (logistic regression, classification tree, support vector machine, random forest, and K-nearest neighbor) were evaluated based on sensitivity, specificity, accuracy, and receiver operating characteristic curves. A nomogram was developed using the model that performed best in MDRO infection prediction. The performance and clinical utility were assessed using the calibration curve, Brier score, and decision curve analysis. Among 791 neurocritical care patients with HAP, 172 (21.7%) were diagnosed with MDRO infections. The prediction model established by logistic regression exhibited the best performance, with an area under the curve of 0.805. Length of NICU stay (odds ratio [OR] 1.078; 95% confidence interval [CI], 1.070-1.141; P < 0.000), number of antibiotics used (OR 1.391; 95% CI, 1.138-1.700; P = 0.001), diabetes (OR 1.775; 95% CI, 1.006-3.133; P = 0.048), and carbamide (OR 1.038; 95% CI, 1.003-1.074; P = 0.035) were significantly correlated with MDRO infections and incorporated into the nomogram. The model demonstrated good calibration (Brier score 0.137). This model can provide clinicians with a tool for prevention and management of MDRO infections in NICU patients with HAP.IMPORTANCEPatients with hospital-acquired pneumonia (HAP) in the neuro-intensive care unit (NICU) are at a high risk of developing multidrug-resistant organism (MDRO) infections owing to complex conditions, critical illness, and frequent invasive procedures. However, studies focused on constructing prediction models for assessing the risk of MDRO infections in neurocritically ill patients with HAP are lacking at present. Therefore, this study aims to establish a reliable and easy-to-use nomogram for predicting the risk of MDRO infections in patients with HAP admitted to the NICU. Four easily accessed variables were included in the model, including length of NICU stay, number of antibiotics used, diabetes, and carbamide. This nomogram might help in the prediction and implementation of targeted interventions against infections with MDRO among patients with HAP in the NICU.