Facial pressure injury is a common complication in patients with prone position ventilation. Local exposure of the trauma can increase the risk of systemic infection, and affect the therapeutic effect of prone position ventilation, and even cause permanent functional damage to local tissues. Exploring the risk factors and constructing a prediction model are of great clinical significance for the prevention of prone position ventilation related facial pressure injuries.

To investigate the risk factors for prone position ventilation-related facial pressure injuries and its optimal modeling methods.

A total of 159 patients who were admitted to the Department of Critical Care Medicine of the First Affiliated Hospital of Xinjiang Medical University from June 2020 to March 2023 and received prone position ventilation were selected and divided into the pressure injury group (n=22) and non-pressure injury group (n=137) according to whether facial pressure injuries occurred or not. General information, disease diagnosis, therapeutic measures, and laboratory test results were collected. Stepwise Logistic regression, multivariate Logistic regression, and Lasso-Logistic regression were used to screen risk factors for facial pressure injuries and develop predictive models, respectively. The area under receiver operating characteristic curve (AUC) was plotted to evaluate the model discrimination. The Akaike Information Criterion (AIC) , Bayesian Information Criterion (BIC) , and calibration curve were applied to evaluate the calibration of the model. Decision curves were applied to evaluate the clinical application value of the models. The optimal modeling method was selected by comparing the predictive efficacy and clinical application differences of the three logistic regression models.

The results of stepwise Logistic regression model showed that the influencing factors of facial pressure injuries were age (OR=39.041) , diabetes mellitus (OR=7.256) , and duration of a single-prone ventilation session (OR=6.705) . The results of the multivariate Logistic regression model showed that the factors influencing facial pressure injuries were age (OR=26.882) , diabetes mellitus (OR=1.770) , length of stay in the ICU (OR=2.610) , and duration of a single-prone ventilation session (OR=5.340) . The results of Lasso-Logistic regression showed that the factors influencing facial pressure injuries were age (OR=38.256) , diabetes mellitus (OR=1.094) , duration of single prone ventilation (OR=5.738) , and RASS score (OR=1.179) . The AUC, sensitivity and specificity of the Lasso-Logistic regression model for predicting prone position ventilation-related facial pressure injuries were 0.855, 0.959 and 0.750, respectively, which were better than those of the stepwise and multivariate Logistic regression models. The AIC and BIC were 44.634 and 55.745, respectively, which were lower than the stepwise and multivariate Logistic regression models. The calibration curves showed that the Lasso-Logistic regression model predicted probabilities fitted the actual probabilities best. The decision curve showed that the Lasso-Logistic regression model obtained clinical benefits corresponding to risk thresholds of 0.01 to 0.98, which was better than the stepwise and multivariate Logistic regression models.

Age, diabetes mellitus, length of a single prone ventilation session, and Richmond Agitation Sedation Score are risk factors for ventilation-related facial pressure injuries. The Lasso-Logistic regression model has better predictive efficacy and clinical application value than stepwise and multivariate Logistic regression models, making it the best modeling method.