Can Machine Learning Improve Myocardial Infarction Diagnosis?

In a recent article published in a magazine natural medicineIn , researchers discuss a novel clinical decision support system based on machine learning (ML) models for predicting an individual’s probability of developing a myocardial infarction. This system utilizes data from the High-Sensitivity Troponin (High-STEACS) Trial Population in the Evaluation of Patients with Suspected Acute Coronary Syndrome to calculate the Collaboration for the Diagnosis and Assessment of Acute Coronary Syndrome (CoDE-ACS) score. generate.

The CoDE-ACS integrated cardiac troponin concentrations at presentation or serial examination with clinical characteristics and used as a continuous scale to calculate a score from 0 to 100. This score reflects how likely it is that an individual will later develop an acute myocardial infarction. The CoDE-ACS was trained separately, but sequentially, in patients with and without myocardial injury at presentation.

For comparison, the investigators used a sex-specific 99-based conventional diagnostic pathway.^th It also rules out or predicts the possibility of myocardial infarction in an individual.

study: Machine learning for myocardial infarction diagnosis using cardiac troponin concentration. Image credit: TippaPatt / Shutterstock.com

Background

Conventional cardiac troponin assays are highly sensitive in diagnosing acute myocardial infarction in symptomatic patients. Both national and international clinical practice guidelines recommend these assays. However, they come with certain limitations.

For example, these assays use a fixed troponin threshold that varies with age, comorbidities, and patient gender for all patients. Additionally, these assays do not consider electrocardiogram (ECG) results or symptom onset time.

Also, consistently applying specific time points for serial testing in a busy hospital emergency department (ED) is cumbersome, but a requirement when performing these assays. Additionally, 99^th Percentile diagnostic thresholds for cardiac troponin assays are inconsistent across groups based on age, sex, and comorbidities.

About research

In this study, researchers first trained an ML model-based CoDE-ACS system on derived cohort data of 10,038 patients with probable myocardial infarction. The average age of the study participants was 70, of whom 48% were women. All patients were diagnosed with type 1, 4b, or 4c myocardial infarction without her ST-segment elevation at initial admission.

Two clinicians independently reviewed all studies performed by CoDE-ACS according to the fourth universal definition of myocardial infarction. A third reviewer resolved the reported disagreements between these two clinicians.

Diagnostic performance on data from 10,286 patients was externally validated from seven cohorts. This allowed researchers to compare this data to diagnostic methods currently used in clinical practice to elucidate its clinical relevance. The prevalence of myocardial infarction across the external health care system ranged from 4% to 16%.

research result

Compared with patients with myocardial injury, patients without myocardial injury at presentation had CoDE-ACS scores <3 and >61, respectively, meeting prespecified diagnostic performance criteria.

Patients with no history of myocardial infection had a negative predictive value of 99.5 and a sensitivity of 90.2. By comparison, the positive predictive value was 80.1 and the specificity was 83.4 in those with a history of myocardial infarction. CoDE-ACS scores performed consistently across all subgroups.

CoDE-ACS successfully identified myocardial infarction as indicated by an area under the curve (AUC) value of 0.953. In addition, CoDE-ACS identified that he was 61% less likely to develop a myocardial infarction at presentation. This compares with 27% of identified patients using a fixed cardiac troponin threshold with a comparable negative predictive value.

The CoDE-ACS score also enabled researchers to identify fewer patients who were more likely to develop an acute myocardial infarction with a higher positive predictive value.

Patients with a low risk of myocardial infarction had a low risk of death after hospital discharge. Thus, less than 1 of her in 300 of these individuals experienced cardiac death 1 year after symptom onset.

Patients with moderate or high likelihood of myocardial infarction had an increased risk of cardiac death within 30 days and 1 year of onset, respectively.

Conclusion

CoDE-ACS Pathway, a novel ML model-based clinical decision support system, incorporates information on examination time, serial measurement of cardiac troponin levels at flexible time points, and time since symptom onset.

Current diagnostic methods require that patients present within 3 hours of symptom onset following an episode of myocardial ischemia for measurement of cardiac troponin. By comparison, CoDE-ACS ruled out myocardial infarction even in early-onset subjects using a single cardiac troponin test, thereby reducing the harm caused by non-adherence to measurement timing. In this study, CoDE-ACS ruled out myocardial infarction in his 71% of patients with his single examination.

In a conservative healthcare system, a low CoDE-ACS score may identify patients with a very low likelihood of myocardial infarction. The system’s false-negative rate is her 1 in 500, meaning that clinicians can discharge nearly half of patients with her one test. However, a low CoDE-ACS score can also identify patients at high risk of myocardial infarction, reducing the proportion of patients requiring observation and serial testing within the ED.

Therefore, clinicians can use CoDE-ACS to create optimal pathways for patient flow according to local clinical priorities. In the future, the system may be integrated with another ML approach, 12-lead ECG, to improve performance and reduce the proportion of patients requiring attention.

Given the flexibility of CoDE-ACS, the authors advocate the adoption of CoDE-ACS in clinical settings to reduce unnecessary hospitalizations in patients with low myocardial infarction probability. Additionally, the system may allow clinicians to shift focus to patients at increased risk of cardiac death.