Early Risk Assessment Model for ICA Timing Strategy in Unstable Angina Patients Using Multi-Modal Machine Learning

TL;DR

This study targets the timing of invasive coronary arteriography (ICA) in unstable angina (UA), a scenario lacking clear ECG markers and typical STEMI indicators. By integrating multi-modal data—demographics, clinical risk factors, biomarkers, and pre-trained ECG embeddings processed via Inception-1D—the authors train linear and non-linear models (LR and XGBoost) to predict revascularization risk. The models outperform the GRACE score (AUC ≈ 0.58–0.72 range; LR ~0.72, GBDT ~0.71) and support a selective ICA strategy, with thresholds that reduce unnecessary procedures while preserving high true-positive rates; they further translate models into explainable risk-look-up tables for clinical use. While results are promising, the study notes limitations due to dataset size and UA-only focus, underscoring the need for larger validation to confirm ECG feature contributions and to generalize the approach in practice.

Abstract

Background: Invasive coronary arteriography (ICA) is recognized as the gold standard for diagnosing cardiovascular diseases, including unstable angina (UA). The challenge lies in determining the optimal timing for ICA in UA patients, balancing the need for revascularization in high-risk patients against the potential complications in low-risk ones. Unlike myocardial infarction, UA does not have specific indicators like ST-segment deviation or cardiac enzymes, making risk assessment complex. Objectives: Our study aims to enhance the early risk assessment for UA patients by utilizing machine learning algorithms. These algorithms can potentially identify patients who would benefit most from ICA by analyzing less specific yet related indicators that are challenging for human physicians to interpret. Methods: We collected data from 640 UA patients at Shanghai General Hospital, including medical history and electrocardiograms (ECG). Machine learning algorithms were trained using multi-modal demographic characteristics including clinical risk factors, symptoms, biomarker levels, and ECG features extracted by pre-trained neural networks. The goal was to stratify patients based on their revascularization risk. Additionally, we translated our models into applicable and explainable look-up tables through discretization for practical clinical use. Results: The study achieved an Area Under the Curve (AUC) of $0.719 \pm 0.065$ in risk stratification, significantly surpassing the widely adopted GRACE score's AUC of $0.579 \pm 0.044$. Conclusions: The results suggest that machine learning can provide superior risk stratification for UA patients. This improved stratification could help in balancing the risks, costs, and complications associated with ICA, indicating a potential shift in clinical assessment practices for unstable angina.

Figures (5)

• Figure 1: The ANOVA analysis of clinical and pre-trained ECG features. We use both univariate F-test and multivariate T-test to demonstrate the contribution of each feature. For the univariate F-test, we select features according to the Benjamini-Hochberg Procedure with the false discovery rate controlled at 0.05. For the multivariate T-test, we select features with p-values under 0.05. LDL: Low-density lipoprotein Cholesterol; HDL: High-density lipoprotein Cholesterol; CKMB: Creatine Kinase-MB; BNP: B-Type natriuretic peptide, CRP C-reactive protein; TnI: Troponin I; SCR: Serum Creatinine; Diab: Diabetes; HyT: Hyper-tension; STd: ST-segment deviation; Gly: Glycerin tricaproate; HR: Heart rate; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; BMI: Body mass index; E: Pretrained ECG feature labeled from 0 to 127;
• Figure 2: The ROC curve of machine learning models and the corresponding revascularization probability for each classification threshold. Machine learning models evaluate patients' risk as scores in the range (0,1). Patients with a score higher than a specific threshold value are considered high-risk (Requiring revascularization). The ROC curve plots the model's true-positive rate against the false-positive rate at various threshold values. Both true-positive and false-negative rates are computed via nested cross-validation to prevent data leakage from the test set. The revascularization probability states the probability of revascularization given a patient with a risk score larger than the threshold. It demonstrates the discrimination power of machine learning models. Note that all machine learning models have a lower false-positive rate than the GRACE score given a true-positive rate. This means our model could reduce the false-positive rate hence avoiding unnecessary ICA.
• Figure .3: Risk scores look-up table Derived from the GBDT model
• Figure .4: Another Risk scores look-up table Derived from the GBDT model
• Figure .5: Risk scores look-up table Derived from the logistic regression model