Construction and validation of a risk prediction model for hospital-associated infection caused by carbapenem-resistant gram-negative organism in intensive care unit

OBJECTIVE To explore and analyze the risk factors for hospital-associated infection caused by carbapenem-resistant gram-negative organism (CRO) in patients in the intensive care unit (ICU)， and to establish and validate a risk prediction model.

METHODS A retrospective collection was conducted on 338 patients with hospital-associated infection caused by gram-negative organism admitted to the adult ICU of Peking University Third Hospital from Jan. to Jun. 2024， serving as the modeling group. Additionally， 141 patients with hospital-associated infection caused by gram-negative organism admitted from Jul. to Dec. 2024 were included as the validation group. Based on the data from the modeling group， a lasso logistic regression model was established. Risk factors were assigned scores based on β values， and a risk prediction model was developed. The model′s performance was validated by plotting the receiver operating characteristic (ROC) curve for predicting CRO hospital-associated infection risk in ICU patients.

RESULTS The logistic regression analysis indicated that mechanical ventilation， emergency admission， Acute Physiology and Chronic Health Evaluation II (APACHE II) score ≥15， C-reactive protein level ≥10mg/L， use of enzyme inhibitor compound preparations and use of carbapenems were risk factors for the development of CRO hospital-associated infection in patients in the ICU in the modeling group (P＜0.05). In the corresponding risk prediction model， these factors were assigned scores of 2， 1， 1， 1， 1 and 2， respectively. The area under the ROC curve (AUC) of the risk model in the modeling group was 0.895 (95%CI: 0.860-0.930， P＜0.001)， with a specificity of 81.02% and a sensitivity of 83.84%. In the validation group， the AUC was 0.764 (95%CI: 0.686-0.842， P＜0.001)， with a specificity of 59.13% and a sensitivity of 77.30%.

CONCLUSIONS The risk assessment system constructed in this study demonstrates good predictive performance for all inpatients in the ICU， including those in the modeling group and validation group. It allows for the early identification of potential high-risk groups and facilitates timely intervention.