Association Between High‑sensitivity Cardiac Troponin Levels and Early Prediction of Acute Myocardial Infarction

High sensitivity cardiac troponin/European society of cardiology 0/1h algorithm has shown to be a better method of immediate detection of myocardial infarction, however it has its drawbacks particularly in a sub group of patients which is confounded with other variables such as age, sex and kidney functions. It was a retrospective cohort design study which involved 1,203 patients who presented with the suspicion of acute coronary syndrome and hs-cTnI serially at presentation and one hour. When creating a machine learning model, the gradient boosting model was used that incorporated the age, sex, serial troponin concentrations, dynamic changes and renal functioning. The comparison of the models was done with the traditional ESC 0/1h algorithm with the assistance of discrimination, calibration, reclassification and analysis of decision curve to compare the models. The machine learning model proved to be more discriminative than the traditional ESC algorithm and the area of the curve of machine learning model is 0.941 as compared to the conventional ESC algorithm 0.892. Positive predictive value improved from 70.2% to 79.4%, specificity from 84.3% to 89.6%, and sensitivity from 92.1% to 94.8%. The most significant changes were observed in patients with renal impairment where changes in positive predictive value were better by 10.1 percentage points and patients aged 65 years and above where the changes in positive predictive value were better by 9.6 percentage points. The net reclassification between patients with renal impairments had an improvement of 21.1% and 26.7%. The model was also found to have good calibration in all the sub-groups where the Hosmer-Lemeshow tests were not significant and decision curve analysis found that the model had a better clinical net benefit at all the clinically relevant risk thresholds. Serial Hs-cTnI + patient specific variables AI based model proves to be more diagnostic of NSTEMI than the more traditional ESC 0/1h algorithm particularly in at risk patients. The approach deals with a more personal approach to diagnostics that addresses the key shortcomings of the existing guidelines and can be justified to be further promoted in the medical practice.