Predicting Unplanned Intubations in Rib Fracture Patients: An Interpretable Machine Learning Approach

Area of Science:

• Trauma surgery
• Critical care medicine
• Machine learning in healthcare

Background:

• Traumatic rib fractures frequently lead to respiratory complications, including unplanned intubations.
• Predictors for unplanned intubations in these patients are not well-defined.
• Early identification of at-risk patients is crucial for timely intervention.

Purpose of the Study:

• To develop and validate an interpretable machine learning model to predict unplanned intubations in patients with rib fractures.
• To identify key predictors of unplanned intubations using SHapley Additive exPlanations (SHAP).

Main Methods:

• Utilized data from the TQIP 2017-2022 database for adult patients with rib fractures.
• Developed an XGBoost model to predict unplanned intubation based on admission variables.
• Employed SHAP for model interpretability and to identify influential predictors.

Main Results:

• The model achieved an AUROC of 0.83, identifying 2.3% of patients requiring unplanned intubation.
• Most influential predictors included admission location (ICU), Injury Severity Score, age, and absence of comorbidities.
• Patients identified by the model had a 7.4-fold increased risk of unplanned intubation.

Conclusions:

• Interpretable machine learning can effectively predict unplanned intubations in rib fracture patients.
• Key factors like admission location and injury severity can guide risk stratification.
• This approach facilitates closer monitoring and early interventions for high-risk individuals.