Investigation of the relationship between troponin and platelet levels in emergency patients: AI-based study on acute coronary syndrome

Abstract

Aim: Advancements in Artificial Intelligence (AI) and Machine Learning (ML) have revolutionized the field of medical diagnostics, offering new possibilities for early detection and risk assessment of critical conditions. This study utilizes machine learning models to predict acute coronary syndrome (ACS) and evaluate its association with key biomarkers, specifically Troponin T, Platelet (PLT), and Mean Platelet Volume (MPV). The primary objective is to assess the efficiency of ML algorithms in identifying high-risk patients based on laboratory parameters and improving early intervention strategies.

Materials and Methods: A retrospective analysis was conducted on 4092 patients admitted to the emergency department, of whom 3640 survived and 452 deceased. The study population consisted of 2427 male and 1665 female patients. Feature importance analysis using the Random Forest algorithm identified Troponin T, PLT, and MPV as the most critical biomarkers for predicting ACS and mortality risk. To enhance predictive accuracy, three machine learning models-Gradient Boosting, XGBoost, and Decision Tree—were implemented and evaluated. The Gradient Boosting model achieved an accuracy of 73.99% and an ROC AUC score of 0.6058, demonstrating the best overall performance. The XGBoost model followed closely with an accuracy of 69.11% and an ROC AUC score of 0.6054. The Decision Tree model exhibited the highest accuracy of 77.29%, but its ROC AUC score of 0.5623 suggested weaker sensitivity in distinguishing ACS cases. Each model was assessed based on accuracy, ROC AUC scores, confusion matrices, and classification reports.

Results: The findings indicate that Gradient Boosting and XGBoost models demonstrated strong predictive capabilities, with Gradient Boosting achieving the highest classification accuracy at 73.99%. Despite having the highest accuracy, the Decision Tree model exhibited a lower ROC AUC score of 0.5623, suggesting limitations in its ability to differentiate between ACS-positive and ACS-negative cases. The results further reinforce that patients with abnormal Troponin T, MPV, and PLT levels were at a significantly higher risk of developing ACS, highlighting their potential as key diagnostic markers in cardiac risk assessment.

Conclusion: This study underscores the importance of artificial intelligence in predicting acute coronary syndrome and emphasizes the critical role of Troponin T, PLT, and MPV as biomarkers. By integrating machine learning models into emergency healthcare settings, clinicians can identify high-risk ACS patients with greater accuracy, enabling timely interventions and reducing mortality rates. Future research should focus on incorporating advanced ensemble learning techniques (Random Forest, XGBoost, Deep Learning) and additional clinical variables to enhance predictive performance further. AI-driven diagnostic models can play a vital role in improving early ACS detection, optimizing patient care, and transforming emergency medicine with data-driven decision-making.