Machine learning in perioperative medicine: a systematic review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Science in Medicine

Background:

• Risk stratification is crucial for anesthetic evaluation.
• Big Data and machine learning (ML) offer significant advantages for analyzing complex health data.
• This systematic review explores ML's role in developing predictive models for surgical outcomes and risk stratification.

Purpose of the Study:

• To systematically review the literature on machine learning applications in surgical risk prediction.
• To assess the effectiveness of ML algorithms in developing predictive models for post-surgical outcomes.
• To understand the current landscape of ML-driven risk stratification in perioperative care.

Main Methods:

• Systematic literature search conducted from January 2015 to March 2021 across major databases (PubMed, Scopus, etc.).
• Keywords included "risk prediction," "surgery," "machine learning," "intensive care unit (ICU)," and "anesthesia."
• 36 eligible studies were identified and evaluated for reporting quality using the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) checklist.

Main Results:

• Commonly predicted outcomes include mortality, systemic complications (e.g., AKI), ICU admission, and length of stay.
• Gradient boosting and random forest algorithms demonstrated the highest performance, with AUC > 0.90.
• Reporting quality was generally acceptable, with 75% of studies adhering to TRIPOD guidelines above 60%.

Conclusions:

• Machine learning holds significant potential for improving medical predictions.
• ML algorithms can predict outcomes more accurately than traditional methods and validated scores.
• An interdisciplinary approach is encouraged to evaluate AI's impact on perioperative risk assessment and healthcare.