Clinical prediction models for hospital-acquired infection of multi-drug-resistant organism in intensive care units: a systematic review.

Multi-drug-resistant organism (MDRO) infections pose a significant threat to patient safety in intensive care units (ICUs). Risk prediction models (RPMs) are promising tools for early identification, but their stability and generalizability remain uncertain. This systematic review aimed to evaluate the status, methodological quality, and performance of RPMs for MDRO infection in adult ICU patients. We searched five databases for studies published up to 30th July, 2025. Studies that developed or validated a prediction model were included. Data on characteristics, predictors, methods, and performance were extracted. Quality was assessed using the Prediction model Risk Of Bias Assessment Tool (PROBAST). Sixty-two studies, comprising 100 prediction models, were included in the analysis. Most were single-centre, retrospective studies from mainland China. Predictors were categorized into ten domains, with antibiotic use, comorbidities, and invasive procedures being the most frequent. Logistic regression was the most common technique. Model validation was insufficient: 37 studies did not report detailed validation, and only 17 performed external validation. PROBAST indicated a high risk of bias in 87.1% of studies, primarily due to analytical shortcomings like inappropriate predictor handling, suboptimal variable selection, and lack of proper validation. The systematic review reveals that existing RPMs have methodological limitations and constrained generalizability, hindering clinical application. Future efforts should focus on integrating dynamic predictors, conducting rigorous external validation, and developing models based on large-scale, prospective, multi-centre data.