Accuracy of 0/1-hour algorithm for diagnosis of MI in the elderly: mono-dimensional optimization of troponin cut-offs for individual confounders or precision medicine?

This editorial refers to ‘Impact of age on the performance of the ESC 0/1h-algorithms for early diagnosis of myocardial infarction’†, by J. Boeddinghaus et al., on page 3780. Currently, the 99th percentile value of a healthy reference population is used to define normal detectable cardiac troponin (cTn) concentrations due to physiological myocardial cell turnover vs. abnormal concentrations that signal excessive myocardial injury or an acute myocardial infarction (MI).1–3 The calculation of the 99th percentile of a healthy reference population is already challenging because the definition of health is variable and because gender, age, and ethnicity may influence the 99th percentile value independently of co-morbidities, thus questioning the appropriateness of a general 99th percentile cut-off.4 With implementation of high sensitivity cardiac troponin (hsTn) assays, previously undetectable differences between healthy female and male hearts were revealed.5 As a consequence, the universal MI definition2 and the recent AACC/IFCC3 consensus started to recommend the use of gender-specific cut-offs instead of a single 99th percentile value, a proposal that is being debated in the light of sparse and controversial data.6 Accurate early diagnosis of MI using hsTn assays, particularly rule-out or rule-in within 3 h, expedites decisions for admission or discharge, particularly in crowded emergency departments (EDs), facilitates the decision and timing for an invasive strategy, and helps to tailor individual pharmacotherapy. Therefore, the 2015 European Society of Cardiology (ESC) guidelines on non-ST-elevation acute coronary syndrome (NSTE-ACS) recommend the use of a 0/1-h algorithm for earlier diagnosis of MI, if hsTn assays implementing validated algorithms are available.2 Several strategies for selecting optimal cut-offs in diagnostic tests have been proposed in the literature, yet all of these have been based on optimizing the preceding accuracy measures. In the Advantageous Predictors of Acute Coronary Syndrome Evaluation (APACE) and hs-cTnT assay for RAPID rule out of Acute Myocardial Infarction (TRAPID) studies,7 , 8 optimal cut-offs and concentration changes were derived from a CART analysis, whereas the Biomarkers in Acute Cardiac Care (BACC) study9 used a method to optimize cut-offs selecting the point(s) at which the positive and negative predictive values were optimal. In both strategies, the 99th percentile and concentration changes within 3 h, as proposed by the ESC and an expert consensus, served as the reference.1 , 10 In the TRAPID study,8 the new optimal biomarker criteria for rule-out and rule-in were very similar to ESC guideline criteria and were also biologically plausible because the optimal baseline value for rule-out was very close to the 99th percentile value of hsTnT (12 vs. 14 ng/L). In the BACC study,9 decision cut-offs for diagnosis were considerably higher (for rule-in) or lower (for rule-out) than recommended sex-specific 99th percentile values for the Abbott Architect hsTnI assay, as cut-offs were chosen to maximize sensitivities and negative predictive values (NPVs) as well as specificities and positive predictive values (PPVs). In the present issue of the European Heart Journal,11 Boeddinghaus and co-workers evaluate the performance of a secondary modification of the already optimized decision cut-offs for the 0/1-h algorithm in three independent cohorts, and separately for the hsTnT (Roche Diagnostics) and hsTnI (Abbott Architect) assays in order to comply with the challenges of increasing age on hsTn performance. It is already known that hsTn increases with age, either due to physiological ageing or due to accumulating prevalence of structural heart disease or co-morbidities, or both. Hammersten et al. 12 had nicely demonstrated that prevalence of normal baseline hsTn declines disproportionately as age increases, thereby reducing clinical specificities and PPVs, while numbers of patients above 65 years old eligible for rule-out decreased dramatically. Previous substudies from APACE13 and TRAPID14 confirmed this decrease of specificity and overall performance of baseline hsTn in the elderly.13 , 14 It was also found that C-statistics improve with the use of higher cut-offs,13 as did the number of prognostically relevant diagnostic reclassifications.14 Consistently, similar issues with hsTn performance and interpretation of baseline hsTn results have been reported for several other clinically relevant confounders, including chronic kidney disease,15 heart failure,16 and atrial fibrillation, that are frequently associated with increasing age. Among other interesting results, Boeddinghaus et al. 11 now report on three key findings. First, NPVs associated with low, very low, or undetectable baseline hsTn were ∼100%, but the low prevalence of patients with normal hsTnT or hsTnI, eligible for rule-out, reduced triage efficacy. Secondly, the numbers of patients with elevated baseline hsTn increased with advancing age, regardless of whether the reason was non-ST segment elevation myocardial infarction (NSTEMI) or myocardial injury due to co-morbidities, resulting in a further reduction of specificities and PPVs for MI and an increase of patients to be allocated to the observational zone. In the derivation and in both validation cohorts, use of slightly higher cut-off concentrations increased specificity, maintained sensitivity of rule-out for hsTnT and hsTnI, but had opposing effects on triage efficacy for hsTnT without lowering efficacy of hsTnI. While the numbers of patients qualifying for rule-in using the 0/1-h algorithm decreased for hsTnT, no reduction was observed using hsTnI. The present study has several strengths. Most importantly, the authors have to be congratulated for the quality of their analysis which supports the practicability and clinical safety of the 0/1-h algorithm. Secondly, findings on optimized baseline cut-offs and concentration changes were validated in two independent cohorts, providing evidence for both currently available validated hsTn assays for the 0/1-h algorithm. There are, however, shortcomings that need to be addressed as well. First, the new optimal cut-offs were not validated to rule-out or rule-in NSTEMI prospectively but were tested for their performance relative to another optimized algorithm in two retrospectively adjudicated cohorts. In addition, the studies enrolled patients with chest pain or angina but did not enrol patients with a lower pre-test probability for an ACS, i.e. patients presenting with dyspnoea, diaphoresis, weakness, or other atypical symptoms that occur more frequently with increasing age. Secondly, use of age-specific cut-offs for patients above 65 years reduced numbers of NSTEMI and translated into a better risk stratification for ACS patients. Conversely, use of higher cut-offs in the elderly carries the risk of underestimating underlying pathologies and the adverse prognosis associated with cTn elevations due to non-ACS-related conditions. Thirdly, a truly individualized diagnosis requires the adjustment not only for age but also for other individual co-morbidities. To improve precision on a patient level (Take home figure), not only one, but multiple confounders need to be considered. This may also need dedicated software applications to assist the ED physician. Modified algorithms adjusting for mono-dimensional parameters such as gender, age, and renal function are important first steps in the right direction. However, further modifications beyond already established reference cut-offs cannot be recommended for routine care for two reasons. First, there are no prospective data showing improved outcome by implementation of modified algorithms in representative real-world cohorts. Secondly, consideration of more confounders in the single patient will require dedicated software algorithms, which are not available yet in the ED setting. In clinical practice, diagnosis of ACS is based on a few fundamental diagnostic tools including ECG and user-friendly biomarker algorithms. Uni- vs. multi-dimensional optimization. Displayed are presumed specificities for the diagnosis of acute myocardial infarction using high sensitivity troponin (hsTn). The left panels show the effect of increasing baseline hsTn values as multiples of normal (mon) on specificities for rule-in of MI in atrial fibrillation (A) and dyspnoea (B) in suspected ACS. The panels on the right side show the specificity of hsTn for rule-in of MI according to renal function (C) and increasing age (D); (D) is plotted from data provided in Boeddinghaus et al. 11 In conclusion, findings of the present analysis corroborate the excellent performance of the ESC 0/1-h algorithm1 and confirm its value despite the co-existence of potential confounders. The decreasing performance of baseline hsTn associated with increasing age and the proposed adjustment of cut-offs accordingly are reasonable strategies to maintain precision. However, there are more confounders than age, such as chronic kidney disease, chronic heart failure, atrial fibrillation, etc. that also need to be incorporated. Such a multi-variable adjustment requires not only dedicated software but also prospective clinical trials to show its superiority vs. the poor man’s single cut-off combined with careful clinical assessment. Conflicts of interest: M.