Table 1.

Comparison of representative modeling approaches for perioperative blood pressure dynamics.

Study	Modeling focus	Method type	Explainability	Subgroup analysis	Key limitation	Contribution of present study
Wiórek et al.	BPV → mortality	Traditional regression	No	No	No predictive model; poor generalizability	Enables individualized BPV prediction
Maheshwari et al.	BPV + waveform → IOH	Regression + BPV	No	No	No real-time prediction	Combines waveform & demographics under ML
Chen & Zhang	Induction IOH prediction	ML (XGBoost, RF, etc.)	No	No	No BPV modeling or interpretability	Extends ML to BPV with interpretability
Hatib et al.	Real-time IOH prediction (HPI)	Proprietary ML (HPI)	No	No	Black-box model; lacks transparency	Provides transparent alternative to HPI
Kouz et al.	IOH phenotype clustering	Unsupervised clustering	No	No	Not predictive; pattern only	Translates phenotypes into predictions
Dai et al.	Time-series BP → AKI (TA-AAD)	Explainable ML (XGBoost + SHAP)	Yes	No	Cardiac surgery specific; no BPV index	Applies explainable ML beyond cardiac field
Present Study	HIBPV detection	Explainable ML (XGBoost + SHAP)	Yes	Yes	—	First interpretable HIBPV model with subgrouping

BPV: blood pressure variability; HIBPV: high intraoperative blood pressure variability; IOH: intraoperative hypotension; ML: machine learning; RF: random forest; XGBoost: extreme gradient boosting; SHAP: Shapley additive explanations; HPI: Hypotension Prediction Index; TFT: temporal fusion transformer; ASA: American Society of Anesthesiologists; AKI: acute kidney injury; TA-AAD: type A acute aortic dissection.