Analysis of the Effects of High-Sensitivity Cardiac Troponin Confounding Factors on AMI Diagnosis

Abstract

Abstract High-sensitivity cardiac troponin (hs-cTn) plays an essential role in facilitating the diagnosis of acute myocardial infarction (AMI). However, there are several known confounding factors affecting hs-cTn concentrations, including sex, age and renal dysfunction. Women have significantly lower hs-cTn concentrations compared to men in presumably healthy populations and in patients presenting to the emergency department (ED). Increasing age leads to elevated troponin levels; and decreasing estimated glomerular filtration rate (eGFR) associates with higher troponin concentrations. How to interpret hs-cTn results in the presence of these confounding factors remains an open question. The aim of this study is to assess whether an integrated prediction model that incorporates hs-cTnT and its confounding factors performs better than hs-cTnT delta threshold alone and to delineate the effect of each confounding factor on the AMI diagnosis. This retrospective cohort study included 17,842 ED patients with serial hs-cTnT measured using a 0h/3h algorithm and eGFR calculated by CKD-EPI creatinine equation at a US medical center. The primary outcome was AMI diagnosis at discharge. Model selection was performed using the logistic regression models of AMI diagnosis with hs-cTnT absolute delta changes and different sets of covariates, which included age, sex, eGFR, and time delta. Model with the smallest Aikaike Information Criteria (AIC) was chosen as the final model, and two-fold cross validation was performed to evaluate model goodness of fit. The area under curve (AUC) of the best fitted model was 0.95. Empirical receiving operating characteristics analysis derived the best probability cutoff of 0.03 for AMI prediction based on a 90% specificity benchmark. The diagnostic performance of this cutoff was compared to that of the hs-cTnT 0h/3h sex-specific delta thresholds we published earlier. The agreement between the prediction model cutoff and the sex-specific delta thresholds was 95.8% (Kappa coefficient 0.80). The agreement to true diagnosis was comparable with hs-cTnT sex-specific delta thresholds alone (90.3% agreement, Kappa coefficient 0.32) and with the prediction model cutoff (90.5% agreement, Kappa coefficient 0.33). To further delineate the effects of the confounders, mediation analysis and spearman correlation analysis showed that hs-cTnT delta change has a much stronger correlation with AMI diagnosis than its confounders. Among the confounders, sex is more predictive of AMI diagnosis (logistic coefficient 0.53) than age (logistic coefficient 0.02) and eGFR (logistic coefficient 0.02). While most of the effects of age and sex on AMI were mediated through delta troponin, the effect of eGFR on AMI diagnosis was almost entirely canceled by mediation through troponin. In summary, the confounding factors play a weaker role than hs-cTn delta change in AMI diagnosis. The integrated prediction model incorporating all confounding factors does not perform better than hs-cTnT delta threshold alone. Sex-specific hs-cTnT delta thresholds remain to yield the highest diagnostic accuracy.