Machine Learning to Predict the Likelihood of Acute Myocardial Infarction

Abstract

Background: Variations in cardiac troponin concentrations by age, sex, and time between samples in patients with suspected myocardial infarction are not currently accounted for in diagnostic approaches. We aimed to combine these variables through machine learning to improve the assessment of risk for individual patients. Methods: A machine learning algorithm (myocardial-ischemic-injury-index [MI 3 ]) incorporating age, sex, and paired high-sensitivity cardiac troponin I concentrations, was trained on 3013 patients and tested on 7998 patients with suspected myocardial infarction. MI 3 uses gradient boosting to compute a value (0–100) reflecting an individual’s likelihood of a diagnosis of type 1 myocardial infarction and estimates the sensitivity, negative predictive value, specificity and positive predictive value for that individual. Assessment was by calibration and area under the receiver operating characteristic curve. Secondary analysis evaluated example MI 3 thresholds from the training set that identified patients as low risk (99% sensitivity) and high risk (75% positive predictive value), and performance at these thresholds was compared in the test set to the 99th percentile and European Society of Cardiology rule-out pathways. Results: Myocardial infarction occurred in 404 (13.4%) patients in the training set and 849 (10.6%) patients in the test set. MI 3 was well calibrated with a very high area under the receiver operating characteristic curve of 0.963 [0.956–0.971] in the test set and similar performance in early and late presenters. Example MI 3 thresholds identifying low- and high-risk patients in the training set were 1.6 and 49.7, respectively. In the test set, MI 3 values were <1.6 in 69.5% with a negative predictive value of 99.7% (99.5–99.8%) and sensitivity of 97.8% (96.7–98.7%), and were ≥49.7 in 10.6% with a positive predictive value of 71.8% (68.9–75.0%) and specificity of 96.7% (96.3–97.1%). Using these thresholds, MI 3 performed better than the European Society of Cardiology 0/3-hour pathway (sensitivity, 82.5% [74.5–88.8%]; specificity, 92.2% [90.7–93.5%]) and the 99th percentile at any time point (sensitivity, 89.6% [87.4–91.6%]); specificity, 89.3% [88.6–90.0%]). Conclusions: Using machine learning, MI 3 provides an individualized and objective assessment of the likelihood of myocardial infarction, which can be used to identify low- and high-risk patients who may benefit from earlier clinical decisions. Clinical Trial Registration: URL: https://www.anzctr.org.au . Unique identifier: ACTRN12616001441404.

High-Sensitivity Cardiac Troponin I vs a Clinical Chemistry Score for Predicting All-Cause Mortality in an Emergency Department Population

“…We compared hs-cTnI ≥ 5 ng/L vs CCS ≥ 1 and hs-cTnI ≥ 5 ng/L vs CCS ≥ 2 to assess if sensitivity ≥ 99% or NPV ≥ 99.5%, 2 metrics that have been selected as necessities for a test to rule out. 22 In contrast, for high risk, we compared hs-cTnI > 26 ng/L (the overall 99th percentile) vs CCS = 5 and hs-cTnI > 26 ng/L vs CCS ≥ 4 to assess if specificity ≥ 90% or PPV ≥ 75% could be attained. 22 We also compared the rate of death between the CCS of < 1 or < 2 with that of hs-cTnI < 5 ng/L as surveys of Canadian, American, and Australasian ED physicians suggest a miss rate not exceed 2%.…”

“… 22 In contrast, for high risk, we compared hs-cTnI > 26 ng/L (the overall 99th percentile) vs CCS = 5 and hs-cTnI > 26 ng/L vs CCS ≥ 4 to assess if specificity ≥ 90% or PPV ≥ 75% could be attained. 22 We also compared the rate of death between the CCS of < 1 or < 2 with that of hs-cTnI < 5 ng/L as surveys of Canadian, American, and Australasian ED physicians suggest a miss rate not exceed 2%. 23 Kaplan–Meier survival curves for all-cause mortality over 5 years were also constructed for both the CCS categories and hs-cTnI ranges with censoring at the end of the 5-year observation window ( P value by log-rank).…”

Systematic Review of Clinical Decision Support Systems for Prehospital Acute Coronary Syndrome Identification

“… 27 A computer-based machine learning algorithm for the diagnosis of MI has been developed with a paired troponin, analyzing the rate of change of troponin along with age and sex showing strong sensitivity at 97.8% and specificity of 92.2%. 39 However, the study required a second troponin at 1–3 hours following the initial troponin measurement and therefore would not be feasible in the prehospital environment. The value of an isolated troponin in the prehospital situation maybe more apparent in combination with other components of CDSS such as patient history and suggestive ECG features.…”

Use of cardiac troponin in the early diagnosis of acute myocardial infarction

“…This concept has already been used in the detection of perioperative myocardial infarction/injury [81]. D) Information technology-based solutions may allow to integrate all known confounders and ultimately provide even more accurate estimates for the presence or absence of AMI among patients presenting with acute chest discomfort [4,82].…”