What does this research mean for the field?

A LightGBM machine learning model utilizing laboratory indicators and electrocardiogram signals accurately predicts 30-day postoperative mortality in surgical intensive care unit non-cardiac surgery patients, substantially outperforming the Revised Cardiac Risk Index. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To develop, compare, and explain machine learning and deep learning models for predicting 30-day postoperative mortality among non-cardiac surgical patients admitted to the surgical intensive care unit.

September 25, 2026Scientific ReportsOpen Access

Predicting mortality risk among non-cardiac surgical patients in the surgical intensive care unit: a retrospective study based on the MIMIC-IV database

Q: What is the clinical evidence from this study?

Study design: Cohort. Population: Non-cardiac surgery (n=4843). Intervention: Light gradient boosting machine (LGBM) model vs. Revised Cardiac Risk Index (RCRI). Primary outcome: 30-day postoperative mortality prediction (AUROC) (AUROC 0.977, 95% CI 0.972-0.982).

View Full Paper

Ask AI

Bookmark

Key result

LGBM model beats RCRI predicting 30-day mortality in non-cardiac surgical ICU patients, achieving ~0.98 AUROC.

AUROC 0.977
95% CI 0.972-0.982
n=4,843

Why the study?

Does the LGBM machine learning model improve the prediction of 30-day postoperative mortality compared to the traditional RCRI score in non-cardiac surgical patients admitted to the SICU?

Population

4,843 elective non-cardiac surgical patients admitted to the surgical intensive care unit for…

Comparison

Light gradient boosting machine model and other… vs Traditional Revised Cardiac Risk Index score.

Design

Cohort

Follow-up

30 days

Authors

MMMengke Ma JLJiatong LiuShenyang Aerospace University CLCaiyun LiRosalind Franklin University of Medicine and Science

Discussion

Member takes

Implication

ML models with ECG and labs may refine SICU mortality prediction post-noncardiac surgery; leaves open prospective validation versus RCRI.

Key Points

To develop, compare, and explain machine learning and deep learning models for predicting 30-day postoperative mortality among non-cardiac surgical patients admitted to the surgical intensive care unit.
Extracted clinical laboratory indicators and electrocardiogram signals for 4,843 non-cardiac surgical patients from the MIMIC-IV database.
Trained five machine learning models (LR, DT, LGBM, XGBoost, SVM) and two deep learning networks (BPNN, RNN), comparing their performance against the Revised Cardiac Risk Index (RCRI).
Conducted global and local model explainability analyses using Shapley additive explanations (SHAP) on the best-performing model.
Within 30 days after non-cardiac surgery, 526 of 4,843 patients (10.9%) died.
The LGBM model outperformed all other models, achieving an AUROC of 0.977 (95% CI 0.972–0.982) and an F1 score of 0.934 (95% CI 0.913–0.942), compared to an AUROC of 0.585 (95% CI 0.564–0.607) and an F1 score of 0.663 (95% CI 0.643–0.685) for the RCRI score.
Global SHAP analysis identified preoperative international normalized ratio (INR), blood urea nitrogen (BUN), serum bicarbonate, and creatinine as the four most influential predictors of 30-day mortality.

Study Design

Type

Cohort (n=4,843)

Multicenter

Structured PICO

Does the LGBM machine learning model improve the prediction of 30-day postoperative mortality compared to the traditional RCRI score in non-cardiac surgical patients admitted to the SICU?

Population

4,843 adult patients admitted to the surgical intensive care unit following elective non-cardiac surgery, evaluated for 30-day postoperative mortality.

Exposure

Light gradient boosting machine (LGBM) model and other machine learning/deep learning architectures for predicting 30-day postoperative mortality.

Comparator

Traditional Revised Cardiac Risk Index (RCRI) score.

Outcome

30-day postoperative mortalityhard clinical

Main Result

Effect estimate: AUROC 0.977 (95% CI 0.972-0.982)

Absolute Event Rate: 0.977% vs 0.585%

The LGBM machine learning model significantly outperforms the traditional RCRI score in predicting 30-day mortality among non-cardiac surgical patients in the ICU, offering a promising tool for early risk stratification.

Limitations

Single-center retrospective design using the MIMIC-IV database without external or prospective validation.
Inputs are static cross-sectional measurements obtained upon ICU admission, lacking temporal dynamics.
Exclusion of COVID-19 pandemic-era admissions limits generalizability to post-2020 cohorts.
The model cannot independently guide targeted clinical treatment decisions.
Generalizability limitations to post-2020 cohorts due to the use of pre-pandemic data (MIMIC-IV v2.0)
Requires further external validation in independent cohorts
SHAP values only reflect correlative contributions and cannot confirm causal relationships between clinical indicators and mortality

Cite This Study

Ma et al. (2026) conducted a cohort in Non-cardiac surgery (n=4,843). Light gradient boosting machine (LGBM) model vs. Revised Cardiac Risk Index (RCRI) was evaluated on 30-day postoperative mortality prediction (AUROC) (AUROC 0.977, 95% CI 0.972-0.982). The LGBM model accurately predicted 30-day mortality in non-cardiac surgical ICU patients, achieving an AUROC of 0.977 compared to 0.585 for the traditional Revised Cardiac Risk Index.

synapsesocial.com/papers/6ab60f47406bf401c1467ae8 https://doi.org/10.1038/s41598-026-72135-4

Topics

Artificial intelligence in cardiology

View Full Paper

Ask AI

Bookmark

Predicting mortality risk among non-cardiac surgical patients in the surgical intensive care unit: a retrospective study based on the MIMIC-IV database

Discussion

Member takes

Implication

Key Points

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Topics

Predicting mortality risk among non-cardiac surgical patients in the surgical intensive care unit: a retrospective study based on the MIMIC-IV database

Discussion

Member takes

Implication

Key Points

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Also Consider

Topics

Also Consider