Article Impact Level: HIGH Data Quality: STRONG Summary of Nature Medicine, 29(5), 1201–1210. https://doi.org/10.1038/s41591-023-02325-4 Dr. Dimitrios Doudesis et al.
Points
- Machine learning models incorporating cardiac troponin concentrations, clinical features, and the CoDE-ACS score were developed to enhance the diagnosis of myocardial infarction.
- The models were trained on data from 10,038 patients (48% women) and validated externally using data from 10,286 patients (35% women) across seven cohorts.
- CoDE-ACS demonstrated excellent discrimination for myocardial infarction, with an area under the curve of 0.953.
- CoDE-ACS identified more patients at presentation with a low probability of myocardial infarction (61%) compared to fixed cardiac troponin thresholds (27%) while maintaining a similar negative predictive value.
- Patients identified as having a low probability of myocardial infarction using CoDE-ACS had significantly lower rates of cardiac death at both 30 days (0.1%) and one year (0.3%) compared to those with intermediate or high probability (30-day rates: 0.5% and 1.8%; 1-year rates: 2.8% and 4.2%).
Summary
This research paper introduces machine learning models that integrate cardiac troponin concentrations, clinical features, and the Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome (CoDE-ACS) score to enhance the diagnosis of myocardial infarction. The models were trained on a dataset of 10,038 patients, 48% women. External validation was conducted on data from an additional 10,286 patients, with 35% women, obtained from seven cohorts.
The CoDE-ACS scoring system exhibited excellent discrimination for myocardial infarction, as indicated by the area under the curve of 0.953 (95% confidence interval: 0.947–0.958). It displayed consistent performance across various subgroups. Notably, at presentation, the CoDE-ACS identified a higher proportion of patients with a low probability of myocardial infarction compared to fixed cardiac troponin thresholds (61% versus 27%). This was achieved while maintaining a similar negative predictive value. Furthermore, the CoDE-ACS identified fewer patients with a high probability of myocardial infarction (10% versus 16%) but with a greater positive predictive value.
Patients classified as having a low probability of myocardial infarction using the CoDE-ACS system experienced significantly lower cardiac death rates than those categorized as having intermediate or high probability. At 30 days and one year from patient presentation, the cardiac death rates for the low probability group were 0.1% and 0.3%, respectively. In contrast, the rates for the intermediate and high probability groups were considerably higher, at 0.5% and 1.8% for 30 days and 2.8% and 4.2% for one year (P < 0.001 for both).
The findings highlight the potential of utilizing CoDE-ACS as a clinical decision support system, which could reduce hospital admissions and significantly benefit patients and healthcare providers. By incorporating machine learning techniques and considering individual factors such as troponin concentrations and clinical features, this approach offers an improved and personalized diagnostic tool for myocardial infarction, leading to enhanced patient outcomes and more efficient resource allocation in the healthcare domain.
Link to the article: https://www.nature.com/articles/s41591-023-02325-4
References Doudesis, D., Lee, K. K., Boeddinghaus, J., Bularga, A., Ferry, A. V., Tuck, C., Lowry, M. T. H., Lopez-Ayala, P., Nestelberger, T., Koechlin, L., Bernabeu, M. O., Neubeck, L., Anand, A., Schulz, K., Apple, F. S., Parsonage, W., Greenslade, J. H., Cullen, L., Pickering, J. W., … Smith, S. W. (2023). Machine learning for diagnosis of myocardial infarction using cardiac troponin concentrations. Nature Medicine, 29(5), 1201–1210. https://doi.org/10.1038/s41591-023-02325-4
