Machine Learning-Based Prediction of Postoperative Pneumonia Among Super-Aged Patients With Hip Fracture

Area of Science:

• Geriatric Medicine
• Surgical Outcomes
• Artificial Intelligence in Healthcare

Background:

• Hip fractures pose significant health risks for super-aged patients (≥80 years).
• These patients are highly susceptible to postoperative pneumonia due to frailty and comorbidities.
• Accurate prediction of postoperative pneumonia is crucial for this demographic.

Purpose of the Study:

• To develop and validate a predictive model for postoperative pneumonia in super-aged hip fracture patients.
• To identify key factors contributing to pneumonia risk post-hip fracture surgery.
• To provide a tool for early identification and management of pneumonia in this vulnerable population.

Main Methods:

• Utilized data from 555 super-aged hip fracture patients from the PLAGH Hip Fracture Cohort Study.
• Employed machine learning algorithms including decision tree, random forest, extreme gradient boosting machine (eXGBM), support vector machine, neural network, and logistic regression.
• Randomly split data into training (70%) and validation (30%) sets for model development and testing.

Main Results:

• The extreme gradient boosting machine (eXGBM) model achieved the highest predictive performance (AUC: 0.929).
• Other models like random forest (AUC: 0.916) and logistic regression (AUC: 0.720) were also evaluated.
• The eXGBM model demonstrated superior accuracy (0.858), precision (0.870), and F1 score (0.855).

Conclusions:

• A reliable machine learning-based model (eXGBM) was developed and validated for predicting postoperative pneumonia in super-aged hip fracture patients.
• This model can effectively identify patients at high risk for pneumonia.
• The validated model can guide clinical strategies for improved patient outcomes.