Predicting postoperative pulmonary infection risk in patients with diabetes using machine learning

Area of Science:

• Medical Informatics
• Surgical Outcomes Research
• Diabetes Complications

Background:

• Diabetic patients have a higher risk of postoperative pulmonary infection (PPI).
• Existing predictive models for PPI are not specific to diabetic populations.

Purpose of the Study:

• To develop and validate a machine learning model for predicting PPI risk in patients with diabetes.
• To identify key clinical factors for PPI prediction in this cohort.

Main Methods:

• Retrospective study of 1,269 diabetic patients undergoing elective non-cardiac, non-neurological surgery.
• Development and comparison of nine machine learning algorithms.
• Feature selection using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression.

Main Results:

• The Ada Boost (ADA) classifier achieved the highest performance (AUC 0.901).
• Key predictors identified: ICU transfer, Age, ASA score, COPD, surgical department, and surgery duration.
• High accuracy (0.91) and specificity (0.98) were observed.

Conclusions:

• A robust machine learning model for PPI prediction in diabetic patients was developed.
• The model utilizes six significant clinical features.
• This offers a valuable tool for clinical decision-making and personalized PPI prevention.