Abstract
Background: Two methods are used to determine cardiac troponin (cTn) thresholds for high-risk stratification for myocardial infarction (MI), namely the assay’s healthy population 99th percentiles and a threshold that has a positive predictive value (PPV) of about 70%. We hypothesized that these methods may produce inconsistent prediction of MI risk. Methods: We compared the positive likelihood ratio (LR+), sensitivity, and specificity at 99th percentiles and published high-risk thresholds for a high-sensitivity (hs) cTnI (Abbott) and cTnT (Roche) assay. We simulated the high-risk threshold derivation methodology that derives thresholds with 70% PPV as a function of prevalence and sample size. We derived and validated thresholds based on LR+ = 20 and specificity = 95%. Results: In 2584 patients there was an almost 2-fold difference in LR+ between males and females at the sex-specific 99th percentiles, and a 2- to 3-fold difference between assays. Thresholds giving 70% PPV were much higher in lower-prevalence cohorts for both assays, being 3- to 10-fold lower at 5% than 20% prevalence. Thresholds derived using a LR+ or specificity had similar performance in a validation cohort between sexes but had a 2-fold difference in LR+ in a validation cohort between assays. Conclusions: In patients presenting to the emergency department with possible MI, sex-specific 99th percentile thresholds convey lower risk of MI for females than males. Risks also differed between assays. Risk thresholds derived using the PPV vary considerably with the prevalence and size of cohort. Thresholds derived with specificity or positive likelihood ratio had more consistent risk-prediction.
Funding Information
  • Christchurch Heart Institute
  • Emergency Care Foundation
  • British Heart Foundation (PG/24/12136)
  • Health Research Council of New Zealand
  • Upstream Medical Technologies

This publication has 40 references indexed in Scilit: