Artificial intelligence versus logistic regression statistical modelling to predict cardiac complications after...

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Traditional cardiac risk prediction models often rely on logistic regression (LR).
• Neural networks (NN), a form of artificial intelligence, offer an alternative approach simulating biological intelligence.
• Evaluating NN's applicability for predicting cardiac events post-noncardiac surgery is crucial.

Purpose of the Study:

• To compare the predictive performance of logistic regression (LR) and neural network (NN) techniques for cardiac events after noncardiac surgery.
• To assess the potential of artificial intelligence as an alternative to conventional statistical methods in preoperative risk stratification.

Main Methods:

• Developed predictive models using backward stepwise LR and NN on a training group of 200 patients.
• Utilized 17 clinical parameters and 7 quantitative indices from dipyridamole-thallium imaging.
• Validated both models on a separate group of 160 patients, recording post-operative myocardial infarctions and/or cardiac deaths.

Main Results:

• The neural network (NN) demonstrated significantly higher specificity (96%) compared to logistic regression (LR) (82%) in predicting post-operative cardiac events (p < 0.01).
• NN also showed improved positive predictive accuracy (50%) versus LR (18%).
• Both models had similar sensitivity (67%) and negative predictive accuracy (98%).

Conclusions:

• Artificial intelligence, specifically neural networks, presents a viable and potentially superior alternative to logistic regression for preoperative cardiac risk assessment.
• NNs offer enhanced accuracy in identifying patients at risk for cardiac events following noncardiac surgery.