Prediction of postoperative infection in elderly using deep learning-based analysis: an observational cohort study

Area of Science:

• Medical Informatics
• Geriatric Surgery
• Infectious Disease Epidemiology

Background:

• Elderly patients face higher risks of postoperative infections and mortality.
• Perioperative factors influencing these infections require better predictive tools for improved patient outcomes.

Purpose of the Study:

• To develop and validate deep learning (DL) models for predicting postoperative infections in elderly patients undergoing elective surgery.
• To compare the predictive performance of DL models against conventional methods.

Main Methods:

• An observational cohort study involving 2014 elderly patients from 28 Chinese hospitals.
• Development of DL models using a training dataset (1510 patients) and validation using a separate dataset (504 patients).
• Comparison of DL models (incorporating baseline clinical characteristics and surgical factors) with a conventional predictive model.

Main Results:

• A conventional model showed an area under the curve (AUC) of 0.728.
• A DL model incorporating baseline characteristics had an AUC of 0.641.
• A DL model including baseline variables and surgical factors achieved a higher AUC of 0.763, with improved specificity.

Conclusions:

• Deep learning models can be developed to predict postoperative infections in the elderly.
• The feasibility of using DL for risk prediction in this demographic is demonstrated, warranting further clinical validation.