Machine Learning Diagnostic Model for Early Stage NSTEMI: Using hs-cTnI 1/2h Changes and Multiple Cardiovascular...

Area of Science:

• Cardiology
• Biomarker Analysis
• Machine Learning in Medicine

Background:

• Cardiovascular biomarker distribution differs between non-ST-segment elevation myocardial infarction (NSTEMI) and unstable angina (UA).
• Machine learning models can predict NSTEMI vs. UA using biomarkers measured at admission and 1-2 hours post-admission.

Purpose of the Study:

• To explore the diagnostic value of changes in high-sensitivity cardiac troponin I (hs-cTnI) levels.
• To differentiate between NSTEMI and UA in patients with suspected acute coronary syndrome (ACS).

Main Methods:

• Analysis of hs-cTnI, creatine kinase-MB (CK-MB), and Myoglobin (Myo) in 267 patients.
• Application of machine learning techniques to assess hs-cTnI level changes for NSTEMI diagnosis.
• Comparison of diagnostic predictive models using admission and 1-2 hour post-admission biomarker data.

Main Results:

• NSTEMI patients exhibited significantly higher levels of CK-MB, Myo, hs-cTnI, and NT-proBNP compared to UA patients (p < 0.001).
• Strong positive correlations were observed between hs-cTnI and CK-MB (R=0.72), and hs-cTnI and Myo (R=0.51, R=0.60).
• The optimal model (Hybiome_1/2h) achieved an F1-Score of 0.74, AUROC of 0.96, and AP of 0.89.

Conclusions:

• hs-cTnI is a valuable and sensitive marker for myocardial injury in NSTEMI diagnosis.
• Continuous hs-cTnI monitoring enhances the accuracy of distinguishing NSTEMI from UA.
• Incorporating hs-cTnI measurements from 1-2 hours post-admission significantly improves diagnostic model effectiveness.