Improving diagnosis of obstructive myocardial infarction using new machine learning tools

In a recent study published in the journal natural medicineresearchers developed and tested an intelligent model for diagnosing patients with obstructive myocardial infarction (OMI).

This model helped diagnose patients without ST elevation on electrocardiogram (ECG), which clinicians and current commercial interpretation systems may miss. Using data from over 7,000 patients, the model was able to correctly reclassify her 1 out of 3 diagnostic errors committed by the traditional risk stratification system.

study: Machine learning for ECG diagnosis and risk stratification of obstructive myocardial infarction. Image credit: vchal / Shutterstock.com

Challenges in Timely Diagnosis of OMI

The ST is a segment of the cardiac scan trace and when elevated, a condition also known as ST segment elevation (STE) indicates acute coronary syndrome (ACS).

ST-segment elevation myocardial infarction (STEMI) with acute chest pain is widely considered to be one of the more severe and life-threatening types of heart attack, requiring immediate catheterization. Accurate interpretation of ECG measurements is therefore very important for the timely treatment of OMI.

Recent studies have shown that not all patients with chest pain have access to on-demand ECG. Even when an ECG report is present, 24–35% of her patients experience her OMI without STEMI for unknown reasons, resulting in a misdiagnosis.

Biomarker-based diagnosis is limited because its interpretation is highly variable and based on visual interpretation by clinicians. This cumulatively results in delayed diagnosis and treatment, which inevitably leads to increased mortality in patients with OMI.

There are also limited biomarkers that can detect OMI in the absence of STEMI. This is because it can only be detected after peak levels of OMI have been reached, i.e. after the critical period of myocardial recovery has passed. Her initial assessment report of more than 60% of patients admitted for chest pain was inconclusive, and as a result mortality in patients with ACS is estimated to increase by 14-22%.

About research

The present study builds on the authors’ previous work, which developed a prototype artificial intelligence (AI) algorithm for ECG analysis for automated prehospital ACS screening. This study presents the first observational cohort study evaluating the diagnostic accuracy of machine learning for use in STEMI diagnosis and risk assessment.

This study cohort included 7,313 patients who reported chest pain. Patients ranged in age from 43 to 75 years, 47% were female, and 5.2% were ultimately found to be OMI positive.

The study cohort was divided into two groups, a derived group containing 4,026 patients and a validation group containing 3,287 patients. Patients in both groups had similar age, sex, and 30-day cardiovascular mortality, but increased proportions of blacks and Hispanics and slightly increased prevalence of ACS and OMI in the validation cohort was selected.

The AI ​​model was trained using 12 pre-hospital reports for each of 4,026 derived patients. The model identified 554 spatio-temporal metrics, 73 of which were selected after incorporating recommendations from subject matter experts. These indices were used to differentiate between ACS and non-ACS patients and to develop her 10 classifiers to elucidate the likelihood of an ACS patient suffering from her OMI.

One of the models, the Random Forest (RF) model, was selected for testing in the validation cohort because it performed better than currently available commercial ECG systems and clinicians in preliminary tests. The final step in model development involves defining a risk index called OMI score to classify patients into low, medium and high OMI risk groups.

The model was then tested using data obtained from the validation cohort.

research result

Our model generalized and maintained high classification performance, outperforming commercial ECG systems and clinicians. ”

The OMI classification identified 74.4% of the 3,287 patients as having low OMI risk, as indicated by a score of <5. By comparison, 21% of this cohort were identified as having moderate her OMI risk with scores of 5–20, whereas 4.6% had high OMI risk with scores >20.

Using the OMI score alone, the model significantly outperformed the previous gold standard HEART index, which uses a combination of age, environmental risk factors, troponin levels, ECG data, and patient history.

The diagnostic accuracy of the model was consistent regardless of gender, comorbidities, age, race, and baseline ECG measurements, demonstrating no aggregation bias. The model also found his ECG variables that were ignored in clinical guidelines as indicative of his future development of OMI, furthering researchers’ understanding of ACS.

Conclusion

In this study, researchers developed and validated a machine learning AI model for clinical diagnosis and risk assessment of potential OMI patients. This model accurately classified the patient into her ACS and non-ACS groups, and the ACS patient was further classified into low, intermediate and high risk of her impending OMI.

This model outperforms currently available commercial indices and clinicians in OMI risk assessment, even in the absence of STEMI on the patient’s ECG report. In addition, the model identified 73 major adverse effects of OMI risk, some of which were largely ignored in clinical diagnostic recommendations.

The clinical implications of this study are numerous, as it can assist clinicians in real-time evaluation of ECG reports and reduce physician visual errors and biases.

Until now, clinicians have not had highly sensitive or highly specific tools to identify OMI very early, even in the absence of a STEMI pattern. ”

This type of model can help rapidly assess patient risk, thereby enabling timely medical intervention and, as a result, significantly reducing mortality in patients with OMI.