Risk of myocardial infarction at specific troponin T levels using the parameter predictive value among lookalikes (PAL)

- A biomarker's clinical usefulness is often described as the fraction patients with or without disease above or below a biomarker level.
- Common examples are sensitivity, specificity and predictive values, called binary classifications.
- A clinician often wants to know the risk of disease at the patient biomarker level, not above, the result in binary classifications.
- Predictive value among lookalikes (PAL) result in risk of disease at the patients biomarker level and could complement decision limits.

BackgroundMyocardial infarction is more likely if the heart damage biomarker cardiac troponin T (cTnT) is elevated in a blood sample, indicating that cardiac damage has occurred. No method allows the clinician to estimate the risk of myocardial infarction at a specific cTnT level in a given patient.MethodsPredictive value among lookalikes (PAL) uses pre-test prevalence, sensitivity and specificity at adjacent cTnT limits based on percentiles. PAL is the pre-test prevalence-adjusted probability of disease between two adjacent cTnT limits. If a chest pain patient's cTnT level is between these limits, the risk of myocardial infarction can be estimated.ResultsThe PAL based on percentiles had an acceptable sampling error when using 100 bootstrapped data of 18 different biomarkers from 38,945 authentic lab measurements. A PAL analysis of an emergency room cohort (n = 11,020) revealed that the diagnostic precision of a high-sensitive cTnT assay was similar among chest pain patients at different ages. The higher incidence of false positive results due to non-specific increases in cTnT in the high-age group was counterbalanced by a higher pre-test prevalence of myocardial infarction among older patients, a finding that was missed when using a conventional ROC plot analysis.ConclusionsThe PAL was able to calculate the risk of myocardial infarction at specific cTnT levels and could complement decision limits.

79