AI Clin Med

AI in Clinical Medicine, ISSN 2819-7437 online, Open Access

Article copyright, the authors; Journal compilation copyright, AI Clin Med and Elmer Press Inc

Journal website https://aicm.elmerpub.com

Review

Volume 2, June 2026, e27

Machine Learning in Perioperative Medicine: A Comparative Review of Predictive, Causal, and Foundation Model Approaches in Surgical Data Science

↓ **Table 1.** Comparative Summary of ML Paradigms in Perioperative Research
Dimension	Traditional supervised ML	Causal ML/meta-learners	TabPFN
AIPW: augmented inverse probability weighting; CDT: causal distillation tree; EBM: Explainable Boosting Machine; ML: machine learning; SaMD: Software as a Medical Device; SHAP: Shapley Additive explanations; TabPFN: Tabular Prior-data Fitted Network.
Primary purpose	Outcome prediction	Causal effect estimation	Outcome prediction
Sample size requirement	Large preferred	Moderate to large	Small to large
Hyperparameter tuning	Extensive	Moderate	None required
Causal inference capacity	No	Yes	No
Missing data handling	Requires imputation	Requires imputation	Native handling
Interpretability tools	SHAP, EBM	CDT, SHAP	Moderate (SHAP)
Calibration	Variable	Good (doubly robust)	Task-dependent
External validation evidence	Moderate (multicenter studies emerging)	Limited	Early stage
Regulatory readiness (SaMD)	Moderate	Moderate	Early stage

↓ Table 1. Comparative Summary of ML Paradigms in Perioperative Research

Dimension

Traditional supervised ML

Causal ML/meta-learners

TabPFN

AIPW: augmented inverse probability weighting; CDT: causal distillation tree; EBM: Explainable Boosting Machine; ML: machine learning; SaMD: Software as a Medical Device; SHAP: Shapley Additive explanations; TabPFN: Tabular Prior-data Fitted Network.

Primary purpose

Outcome prediction

Causal effect estimation

Outcome prediction

Sample size requirement

Large preferred

Moderate to large

Small to large

Hyperparameter tuning

Extensive

Moderate

None required

Causal inference capacity

Yes

Missing data handling

Requires imputation

Native handling

Interpretability tools

SHAP, EBM

CDT, SHAP

Moderate (SHAP)

Calibration

Variable

Good (doubly robust)

Task-dependent

External validation evidence

Moderate (multicenter studies emerging)

Limited

Early stage

Regulatory readiness (SaMD)

Moderate

Early stage

↓ **Table 2.** Representative Perioperative Databases Used in Surgical ML Research
Database	Source/country	Scale	Data types	Typical ML use cases
ML: machine learning; AKI: acute kidney injury; eICU CRD: eICU Collaborative Research Database; ICU: intensive care unit; MIMIC-IV: Medical Information Mart for Intensive Care IV; MOVER: Medical Informatics Operating Room Vitals and Events Repository; NSQIP: National Surgical Quality Improvement Program; STS: Society of Thoracic Surgeons; DSSR: Danish Society for Patient Safety Registry.
MIMIC-IV	Beth Israel Deaconess Medical Center, USA	About 300,000 ICU admissions	Vitals, labs, medications, notes	ICU mortality, sepsis prediction, AKI
NSQIP	American College of Surgeons, USA	About 1 million cases/year	Preoperative risk factors, 30-day outcomes	Surgical complication risk scoring
VitalDB	Seoul National University Hospital, Korea	6,388 cases	High-resolution intraoperative signals, drugs	Vasopressor effects, intraoperative monitoring
MOVER	UC Irvine Medical Center, USA	67,134 cases	Demographics, labs, vitals, medications	Anesthetic dose-response, ICU risk
eICU CRD	Philips Healthcare, multi-site USA	About 200,000 ICU stays	Clinical assessments, vitals, labs	Critical care prediction, mortality
National registries (e.g., STS, DSSR)	Multi-institutional, various countries	Millions of records	Procedural data, outcomes, comorbidities	Cardiac surgery outcomes, risk scoring

↓ Table 2. Representative Perioperative Databases Used in Surgical ML Research

Database

Source/country

Scale

Data types

Typical ML use cases

ML: machine learning; AKI: acute kidney injury; eICU CRD: eICU Collaborative Research Database; ICU: intensive care unit; MIMIC-IV: Medical Information Mart for Intensive Care IV; MOVER: Medical Informatics Operating Room Vitals and Events Repository; NSQIP: National Surgical Quality Improvement Program; STS: Society of Thoracic Surgeons; DSSR: Danish Society for Patient Safety Registry.

MIMIC-IV

Beth Israel Deaconess Medical Center, USA

About 300,000 ICU admissions

Vitals, labs, medications, notes

ICU mortality, sepsis prediction, AKI

NSQIP

American College of Surgeons, USA

About 1 million cases/year

Preoperative risk factors, 30-day outcomes

Surgical complication risk scoring

VitalDB

Seoul National University Hospital, Korea

6,388 cases

High-resolution intraoperative signals, drugs

Vasopressor effects, intraoperative monitoring

MOVER

UC Irvine Medical Center, USA

67,134 cases

Demographics, labs, vitals, medications

Anesthetic dose-response, ICU risk

eICU CRD

Philips Healthcare, multi-site USA

About 200,000 ICU stays

Clinical assessments, vitals, labs

Critical care prediction, mortality

National registries (e.g., STS, DSSR)

Multi-institutional, various countries

Millions of records

Procedural data, outcomes, comorbidities

Cardiac surgery outcomes, risk scoring