Surgical MAchine learning model for predicting Risk of postoperative infecTions in general elective surgery (SMART):...

Area of Science:

• Medical Informatics
• Surgical Oncology
• Machine Learning in Healthcare

Background:

• Postoperative infections are a significant surgical complication, often poorly predicted by existing machine learning (ML) models.
• Previous ML models for infection prediction lacked representation of surgical patient populations.
• A novel ML model, SMART, was developed and validated for predicting postoperative infection risk in elective general abdominal surgery.

Purpose of the Study:

• To develop and internally validate a machine learning model (SMART) for predicting postoperative infection likelihood.
• To improve the accuracy of infection risk prediction in patients undergoing elective general abdominal surgery.
• To identify modifiable predictors for clinical application in preventing postoperative infections.

Main Methods:

• Retrospective data from 2,716 elective general surgical patients were utilized from the UK Health Data Research (UKHDR) Hub for Acute Care (PIONEER).
• Predictors were selected using integrated ML methods (recursive feature elimination) and expert input.
• An ensemble model combining the top three performing algorithms, including a gradient boosting classifier, was developed and validated.

Main Results:

• Nineteen predictors were identified, including demographics, comorbidities, microbiology data (e.g., MDR-infections), and C-reactive protein levels.
• The ensemble model achieved high performance during internal validation: 96.9% sensitivity, 74.1% specificity, and an Area Under the Curve (AUC) of 0.86.
• The Gradient Boosting Classifier demonstrated the best individual model performance.

Conclusions:

• The SMART model demonstrates high performance in predicting postoperative infections for elective surgery patients.
• The model's use of modifiable predictors facilitates clinical application for early intervention.
• Further external validation is crucial for successful clinical implementation and reducing antimicrobial resistance risks.