Systematic Analysis and Critical Appraisal of Predictive Models for Lung Infection Risk in ICU Patients

Abstract

Purpose: To systematically review and evaluate predictive models for assessing the risk of lung infection in intensive care unit (ICU) patients.

Methods: A comprehensive computerized search was conducted across multiple databases, including CNKI, Wanfang, VIP, SinoMed, PubMed, Web of Science, Embase, and the Cochrane Library, covering literature published up to March 2, 2024. The PRISMA guidelines were followed for data synthesis, and data extraction was performed according to the CHARMS checklist. The PROBAST tool was used to evaluate the risk of bias and the applicability of the included studies.

Results: Fourteen studies encompassing 20 predictive models were included. The area under the curve (AUC) values of these models ranged from 0.722 to 0.936. Although the models demonstrated good applicability, the risk of bias in the included studies was high. Common predictors across the models included age, length of hospital stay, mechanical ventilation, use of antimicrobial drugs or glucocorticoids, invasive procedures, and assisted ventilation.

Conclusion: Current predictive models for lung infection risk in ICU patients exhibit strong predictive performance. However, the high risk of bias highlights the need for further improvement. The main sources of bias include the neglect of handling missing data in the research, use of univariate analysis to select candidate predictors, lack of assessment of model performance, and failure to address overfitting. Future studies should expand the sample size based on the characteristics of the data and specific problems, conduct prospective studies, flexibly apply traditional regression models and machine learning, effectively combine the two, and give full play to their advantages in developing prediction models with better predictive performance and more convenient operation.