Machine learning or traditional statistical methods for predictive modelling in perioperative medicine: A narrative...

Area of Science:

• Perioperative Medicine
• Medical Informatics
• Machine Learning Applications

Background:

• Accurate prediction of perioperative outcomes is essential for clinical decision-making and risk communication.
• Machine learning (ML) is increasingly explored for enhancing predictive accuracy compared to traditional statistical models.
• Concerns exist regarding the quality of ML studies and the clinical meaningfulness of performance gains.

Purpose of the Study:

• To review and appraise studies developing perioperative predictive ML models.
• To compare the predictive performance of ML models against traditional statistical models.
• To identify factors influencing the successful application of ML in perioperative prediction.

Main Methods:

• A narrative review was conducted due to heterogeneity in study populations and outcomes.
• Data were extracted from 37 studies identified through systematic search, screening, and full-text review.
• Studies focused on the development and validation of perioperative predictive models.

Main Results:

• Several studies demonstrate that ML can enhance perioperative prediction models.
• The performance improvement offered by ML is not universal and remains context-dependent.
• Traditional statistical models also continue to be developed and show relevant performance.

Conclusions:

• ML models show potential to augment traditional methods for perioperative outcome prediction.
• Clinical utility depends on factors like patient-centered outcomes, interpretability, and external validation.
• High standards in reporting and methodological transparency are vital for advancing ML in clinical practice.