Risk Prediction of Unplanned Extubation in Inpatients Using Random Forest and Logistic Regression Models.

Background: Unplanned extubation (UEX) represents a significant risk event in hospitalized patients and is considered one of the most serious safety concerns. Prevention and early detection of these events have become essential components of high-quality nursing care.

Objective: To compare random forest and logistic regression models for the prediction of UEX.

Methods: In total, 775 UEX events were selected from the adverse nursing events database of a hospital in Zhejiang Province between January 2021 and December 2022 as the observation group. In addition, 775 planned extubation events were included from the database of hospitalized patients during the same period through 1:1 propensity score matching across various inpatient departments. Subsequently, patients were randomly allocated in a 7:3 ratio to form the development group and the validation group. Both random forest and logistic regression models were constructed. Their performances were compared using metrics including accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and the area under the receiver operating characteristic curve (AUC).

Results: In addition, multivariate logistic regression analysis identified individuals aged 65 years and over (OR = 3.34, 95% CI: 2.43-4.59), male (OR = 1.64, 95% CI: 1.18-2.27), impaired awareness (OR = 2.56, 95% CI: 1.44-4.56), concurrent dual catheters (OR = 4.18, 95% CI: 2.77-6.32), presence of 3 or more catheters (OR = 5.55, 95% CI: 3.44-8.97), catheter indwelling time exceeding 1 week but <1 month (OR = 3.32, 95% CI: 2.04-5.41) or more than 1 month (OR = 4.51, 95% CI: 1.55-13.10), and the presence of medium-risk (OR = 0.22, 95% CI: 0.12-0.41) or high-risk catheters (OR = 0.08, 95% CI: 0.04-0.17) with secondary fixation (OR = 0.07, 95% CI: 0.04-0.12) as influential factors for UEX events in inpatients. Several variables, including catheter indwelling time, number of coexisting catheters, age, secondary fixation, and catheter grade, were selected for predicting UEX events using the random forest model. The AUC of the random forest prediction model was 0.812, while the AUC of the logistic regression prediction model was slightly lower at 0.793.

Conclusion: The random forest model outperforms the logistic regression model in predicting inpatient UEX events. However, the logistic regression model remains valuable for its ability to provide intuitive explanations of the results.