Sex and race differences in the performance of the European Society of Cardiology 0/1‐h algorithm with high‐sensitivity troponin T

Abstract The diagnostic performance of the high‐sensitivity troponin T (hs‐cTnT) European Society of Cardiology (ESC) 0/1‐h algorithm in sex and race subgroups of US Emergency Department (ED) patients is unclear. A pre‐planned subgroup analysis of the STOP‐CP cohort study was conducted. Participants with 0‐ and 1‐h hs‐cTnT measures from eight US EDs (1/2017 to 9/2018) were stratified into rule‐out, observation, and rule‐in zones using the hs‐cTnT ESC 0/1 algorithm. The primary outcome was adjudicated 30‐day cardiac death or MI. The proportion with the primary outcome in each zone was compared between subgroups with Fisher's exact tests. The negative predictive value (NPV) of the ESC 0/1 rule‐out zone for 30‐day CDMI was calculated and compared between subgroups using Fisher's exact tests. Of the 1422 patients enrolled, 54.2% (770/1422) were male and 58.1% (826/1422) white with a mean age of 57.6 ± 12.8 years. At 30 days, cardiac death or myocardial infarction (MI) occurred in 12.9% (183/1422) of participants. Among patients stratified to the rule‐out zone, 30‐day cardiac death or MI occurred in 1.1% (5/436) of women versus 2.1% (8/436) of men (p = .40) and 1.2% (4/331) of non‐white patients versus 1.8% (9/490) of white patients (p = .58). The NPV for 30‐day cardiac death or MI was similar among women versus men (98.9% [95% confidence interval, CI: 97.3–99.6] vs. 97.9% [95% CI: 95.9–99.1]; p = .40) and among white versus non‐white patients (98.8% [95% CI: 96.9–99.7] vs. 98.2% [95% CI: 96.5–99.2]; p = .39). NPVs <99% in each subgroup suggest the hs‐cTnT ESC 0/1‐h algorithm may not be safe for use in US EDs. Trial Registration: High‐Sensitivity Cardiac Troponin T to Optimize Chest Pain Risk Stratification (STOP‐CP; ClinicalTrials.gov: NCT02984436; https://clinicaltrials.gov/ct2/show/NCT02984436).


| INTRODUCTION
2][3] To optimize care for these patients, accelerated diagnostic protocols (ADPs) have been developed to objectively guide patient risk stratification.6][7][8][9][10] In several studies, mostly conducted in Europe and Australasia, the hs-cTnT ESC 0/1-h algorithm has demonstrated a ≥99% negative predictive value (NPV) for 30-day myocardial infarction (MI) or cardiac death. 6,7,11,12Its performance in US patients with acute chest pain is not as well established. 12In the primary analysis of the STOP-CP multisite US cohort, the hs-cTnT ESC 0/1-h algorithm did not achieve a 99% NPV, which is the threshold commonly used for ADP safety. 13,14However, it is unclear whether diagnostic performance was consistent across key patient subgroups, such as men, women, white patients, and non-white patients and there is a paucity of prior data evaluating sex or race differences in algorithm performance.
6][17][18] In addition, it has been hypothesized that greater racial diversity of US ED patients may explain differences in the diagnostic performance of the ESC 0/1-h algorithm in the United States versus Europe and Australasia.Thus, it is plausible that the performance of the ESC 0/1-h algorithm may differ based on demographics.
The primary objective of this pre-planned STOP-CP subgroup analysis was to evaluate and compare the diagnostic performance of the hs-cTnT ESC 0/1-h algorithm for 30-day cardiac death or MI in men versus women and white versus nonwhite patients within the STOP-CP cohort.A secondary objective was to evaluate and compare diagnostic performance for 30-day major adverse cardiovascular events (MACEs; defined as cardiac death, MI, or coronary revascularization) in these subgroups.In addition, we evaluate whether combining the hs-cTnT ESC 0/1-h algorithm with the History, ECG, Age, and Risk factors (HEAR) score can improve test characteristics among patients in each subgroup. 19niversity of Utah, Salt Lake City, Utah.Institutional review board approval was obtained at all sites.Written informed consent was obtained for enrollment.STOP-CP methods are previously described. 14The Standards for Reporting of Diagnostic Accuracy Studies (STARD) guidelines helped direct the research and manuscript development processes. 20

| Study population
We prospectively enrolled ED patients ≥21 years of age with serial troponins ordered for the evaluation of possible acute coronary syndrome (ACS).Exclusion criteria included ST-elevation myocardial infarction, systolic blood pressure <90 mmHg, life expectancy <90 days, a noncardiac illness requiring admission, inability to provide consent or be contacted for follow-up, non-English speaking, pregnancy, being a prisoner, or previous enrollment in the study.

| Data collection
Serial blood samples were collected for hs-cTnT measurement at baseline (<1 h from the first clinical blood draw) and 1-h later in lithium heparin tubes.hs-cTnT was quantified with the Gen 5 STAT assay on the Cobas e 601 analyzer (Roche Diagnostics).The assay has a range of 3-10,000 ng/L, limit of quantification at 6 ng/L, and a 99th percentile upper reference limit (URL) of 19 ng/L in the United States with a coefficient of variation of <10%. 21

| ESC 0/1-h algorithm
In each patient, hs-cTnT measures were used to stratify patients into rule-out, observation, and rule-in zones using established assayspecific cut-points. 4,5,14However, the hs-cTnT ESC 0/1-h algorithm's 0-h rule-out cut-point of 5 ng/L (the limit of detection) was modified to 6 ng/L (the limit of quantification), because the U.S. Food and Drug Administration does not allow reporting below the limit of quantification.8]22 In addition, we evaluated the combination of the hs-cTnT ESC 0/1-h algorithm with a HEAR score within each subgroup.HEAR scores were determined prospectively by the patient's treating provider.Consistent with prior studies, a score of 0-3 was considered low-risk, 4-6 was moderate-risk, and scores ≥7 were high-risk. 19

| Outcomes
The primary outcome was 30-day cardiac death or MI, inclusive of index visit events.Secondary outcomes included: (1) 30-day MACE, (2) the individual MACE components (cardiac death, MI, and coronary revascularization) at index and from index through 30 days, and (3) efficacy, defined as the proportion of patients classified into the ruleout zone. 23,24Medical record review and telephone follow-up through 30 days were completed to determine outcomes.Expert reviewers adjudicated any patient who experienced death, had a clinical diagnosis of MI, or had an elevated contemporary cTn.Expert adjudicators (M.H.i.V., M.M., J.P.S., and J.M.) classified deaths as cardiac or noncardiac based on the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial definition, except for death due to stroke which was classified as a noncardiac death. 25If the cause of death could not be determined, it was considered cardiac.MI was determined by the Fourth Universal Definition of MI: rise and fall of troponin (with at least one value ˃99th percentile URL) with symptoms of ischemia, ECG evidence of ischemia, imaging evidence of new nonviable myocardium, a new regional wall motion abnormality, or evidence of thrombus on angiography. 26Adjudicators had access to all clinical data, including the local clinical contemporary troponin assay results (Table S1), but were blinded to the hs-cTnT results.Any discrepancies between adjudicators were resolved through review by a third adjudicator.

| Statistical analysis
Counts, percentages, means, and standard deviations, or medians and interquartile (IQR) ranges were used to describe the study population.
To evaluate the performance of the ESC 0/1-h algorithm, sensitivity, specificity, NPV and PPV, and negative and positive likelihood ratios (−LR and +LR) for 30-day cardiac death or MI and 30-day MACE were calculated.For efficacy, sensitivity, specificity, NPV, and PPV, exact 95% confidence intervals (95% CI) were computed, and Fisher's exact tests were used to compare by sex (men vs. women) and race (white vs. nonwhite patients).Likelihood ratios were calculated and reported with 95% CIs using the method of Simel et al. 27 The asymptotic hypothesis test developed by Luts et al. was used to compare LRs by sex and race. 28Consistent with prior studies, sensitivity, NPV, and −LR were calculated for the rule-out zone (i.e., rule-out vs rule-in or observation) and specificity, PPV, and +LR were calculated for the rulein zone (i.e., rule-in vs. observation or rule-out). 6,7,14,19,29Fisher's exact tests were used to compare cardiac death or MI and MACE at index and from index though 30-days among sexes and races.
To assess the association of sex or race with index and 30-day cardiac death or MI and 30-day MACE, multivariable logistic regression was performed.Models were adjusted for age, sex, race (white vs. nonwhite patients), hypertension, diabetes, hyperlipidemia, obesity (BMI ≥ 30 kg/m 2 ), current smoking, prior stroke, peripheral vascular disease, and end-stage renal disease.These variables were selected due to their relevance and inclusion in previous cardiovascular risk stratification work. 29Ethnicity was not included in this analysis.For each outcome, we computed two separate adjusted models, one containing an interaction term between ESC 0/1-h zone and sex, and another containing an interaction term between ESC 0/1-h zone and race.Separate, unadjusted logistic regression models were fit to evaluate the association between sex and race with the primary and secondary outcomes within each ESC 0/1-h zone.
Unadjusted or adjusted odds ratios (aORs) with corresponding 95% CIs were calculated appropriately for each logistic model.
In addition, we evaluated the diagnostic performance of the combination of the hs-cTnT ESC 0/1-h algorithm with a HEAR score.
For this combination, patients were classified to the rule-out zone only if they met both the ESC 0/1-h algorithm rule-out cut-points and had a low-risk HEAR score of 0-3.Patients with a HEAR score of ≥7 were classified to the rule-in zone regardless of hs-cTnT measures.
Patients meeting the ESC 0/1-h algorithm's rule-out criteria, who had a HEAR score of 4-6, were reclassified to the observation zone.S2.Event rates using all-cause death instead of cardiac death are available in Table S3.
The interaction between the ESC 0/1-h algorithm and sex was significant for 30-day cardiac death or MI (p = .04)and 30-day MACE (p = .021).The adjusted odds of 30-day cardiac death or MI and 30-day MACE were higher for men in the rule-out and observation zones based on point estimates, but higher for women in the rule-in zone.Figure S2 and Table S6 show the aORs for index and 30-day cardiac death or MI and MACE (Figure 1).
The overall interaction of the ESC 0/1-h algorithm with race was not significant for outcomes of 30-day cardiac death or MI (p = .951)or 30-day MACE (p = .861).Adjusted odds ratio point estimates remained greater than one for white patients compared to nonwhite patients in each ESC 0/1-h algorithm zone.Figure S2 and Table S6 show the aORs for index and 30-day cardiac death or MI and MACE.

| Combination of ESC 0/1-h algorithm with the HEAR score
NPV was increased by adding the HEAR score to the hs-cTnT ESC 0/1-h algorithm in each subgroup.For the primary outcome of 30-day cardiac death or MI, the NPV of the combination was 99.6% (95% CI: 97.6%-100%) for females, 99.0% (95% CI: 96.5-99.9)for males, 99.5% (95% CI: 97.1%-100%) for nonwhites, and 99.2% (95% CI: 97.2%-99.9%)for whites.However, even when the hs-cTnT ESC algorithm was combined with the HEAR score, the NPV for 30-day MACE remained below 99.0% for all subgroups.See Table S7 for full test characteristics among subgroups.The diagnostic performance of the ESC 0/1-h algorithm with the HEAR score using all-cause death instead of cardiac death is provided in Table S5.
In this multicenter, prospective study, the hs-cTnT ESC 0/1-h algorithm failed to achieve an acceptable 13 NPV ≥99% for 30-day cardiac death or MI among any sex or race subgroup.Although efficacy was higher in women, there was no difference in safety among men versus women.While fewer women were placed in the rule-in zone, they were more likely to experience 30-day cardiac death or MI or MACE than men.Efficacy and safety were similar among white and nonwhite patients.More nonwhite patients were classified to the rule-in zone, but they had lower odds of 30-day cardiac death or MI than white patients in the rule-in zone.guidelines. 4This recommendation is based on international studies of the algorithm, which have demonstrated an NPV ≥99% for index-MI, 30,31 30-day cardiac death or MI, 32 or 30-day MACE. 6,74][35] This includes the primary analysis of the US-based STOP-CP trial. 14Data from this subgroup analysis adds to that primary analysis by T A B L E 3 Safety events among nonwhite and white patients by ESC 0/1 zone (unadjusted).algorithm in European and US cohorts. 6,7,12,35Sex-based differences in efficacy may be explained by sex-specific differences in normal troponin ranges and thresholds for MI diagnosis.Previous evidence suggests that normal troponin concentrations are lower among women compared to men, possibly due to differences in left ventricular mass. 15,36,37The Fourth Universal Definition of Myocardial Infarction recommends using sex-specific troponin cutoffs.When used, sex-specific hs-cTnT 99th percentile URL cutoffs increase the diagnosis of MI among women. 15,26Kimenai et al. found that at given levels of elevated hs-cTnT, women had a higher risk of cardiovascular events than men, which may explain why we found a higher point estimate odds of 30-day cardiac death or MI for women in the rule-in group compared to men. 38We did not observe increased rates of 30-day cardiac death or MI or MACE for women relative to men in the rule-out zone, but the study may have been underpowered to detect this difference.The higher proportion of 30-day cardiac death or MI among women in the rule-in zone compared to men suggests sex-specific rule-in hs-cTnT thresholds may be needed for chest pain diagnostic pathways and should be a point of future study.

Use of the hs-cTnT ESC
We found no significant between race and ESC 0/1-h algorithm for 30-day cardiac death or MI, with white patients in each ESC 0/1-h algorithm zone having higher odds of 30-day cardiac death or MI compared to nonwhite patients.However, despite a significantly higher proportion of nonwhite patients placed in the rule-in zone compared to white patients, nonwhite patients had significantly lower odds of 30-day cardiac death or MI compared to white patients in the rule-in zone.This suggests that the use of hs-cTnT ESC 0/1-h algorithm may lead to disproportionate over-testing among nonwhite patients.Of note, the finding of nonwhite patients having a lower rate of cardiovascular death is discordant to the rates typically found in the literature. 39,40The underlying mechanisms for this are unclear and warrant further study but may be related to unmeasured social determinants of health. 39In addition, future studies are needed to determine if integrating individualized hs-cTn cutoffs based on race, social determinants, or other variables into a 0/1-h algorithm reduces over-classification of patients into the rule-in zone.

| LIMITATIONS
Although this study was conducted at eight US EDs, these were mostly academic sites, which limits generalizability to other care settings.Informed consent was required to participate in STOP-CP, resulting in possible selection bias.Lost to follow-up rate was <4% (56 out of 1442); however, a sensitivity analysis imputing events for patients lost to follow-up detailed in the primary manuscript of this study did not change the main results. 12Sex and race were determined by patient self-reporting and medical record review which may have led to misclassification bias.Ethnicity was not included in this analysis.The 30-day cardiac death or MI and MACE rates in STOP-CP are higher than in previous US cohorts, and this increased prevalence may impact NPV and PPV. 12,19As required by the Food and Drug Administration, the lowest reportable value for the hs-cTnT assay was 6 ng/L, which is different than the ESC 0/1-h algorithm rule-out threshold hs-cTnT value of 5 ng/L used outside of the United States.Our prior analyses suggest that this change has a minimal impact on the performance of the hs-cTnT ESC 0/1-h algorithm. 14However, it may affect the generalizability of these results outside of the United States.This study used only the Roche hs-cTnT assay.Therefore, these conclusions cannot be applied to ESC 0/1-h hs-cTnI algorithm derivations.Adjudication was performed in the context of clinical use of contemporary troponin assays (Table S1).However, when adjudication was repeated using hs-cTnT there was no substantive change. 14This pre-planned subgroup analysis was designed to evaluate the performance of the existing ESC 0/1-h hs-cTnI algorithm among sex and race subgroups and not to seek optimal sex or race-specific hs-cTnI cutoffs or develop a new algorithm.Finally, this study was observational and as such, the ESC 0/1-h hs-cTnT algorithm was not used to guide patient care.

| CONCLUSIONS
In this multisite, prospective US cohort study, the hs-cTnT ESC 0/1-h algorithm did not achieve an NPV ≥ 99% in any sex or race subgroup.
Efficacy was significantly higher for women compared to men.More

2 | METHODS 2 . 1 |
Study design and setting This is a preplanned subgroup analysis of the High-Sensitivity Cardiac Troponin T (Gen 5 STAT assay) to Optimize Chest Pain Risk Stratification (STOP-CP; ClinicalTrials.gov: NCT02984436) prospective, multicenter cohort study.STOP-CP enrolled patients with symptoms concerning for ACS at eight US EDs from January 25, 2017, to June 9, 2018.Study sites included the University of Florida, Gainesville, Florida; Wake Forest University, Winston Salem, North Carolina; Henry Ford Health System, Detroit, Michigan; University of Maryland St.Joseph Medical Center, Towson, Maryland; University of Maryland Medical Center, Baltimore, Maryland; University of Maryland Baltimore Washington Medical Center, Glen Burnie, Maryland; University of California-Davis, Davis, California; and

F
I G U R E 1 (A, B) The ESC 0/1-h algorithm among male and female patients for 30-day cardiac death or MI, and 30-day MACE.(A) Male, (B) Female.ACS, acute coronary syndrome; CAD, coronary artery disease; ED, emergency department; ESC, European Society of Cardiology; MACE, major adverse cardiovascular event; MI, myocardial infarction; NPV, negative predictive value; PPV, positive predictive value.
demonstrating that the algorithm failed to achieve an NPV ≥99% for 30-day cardiac death or MI or 30-day MACE among any sex or race subgroup.This suggests the hs-cTnT ESC 0/1-h algorithm alone may not be able to adequately identify US ED patients with chest pain who are safe for discharge.However, when combined with the HEAR score, the hs-cTnT ESC 0/1-h algorithm demonstrated safe NPV for 30-day cardiac death or MI.Our results demonstrated that the hs-cTnT ESC 0/1-h algorithm placed more women into the rule-out zone compared to men.This is consistent with a prior subgroup analysis of the hs-cTnT 0/1-hF I G U R E 2 (A,B) The ESC 0/1-h algorithm among white and nonwhite patients for 30-day cardiac death or MI and 30-day MACE.(A) White, (B) Nonwhite.ACS, acute coronary syndrome; CAD, coronary artery disease; ED, emergency department; ESC, European Society of Cardiology; LR, likelihood ratio; MACE, major adverse cardiovascular event; MI, myocardial infarction; NPV, negative predictive value; PPV, positive predictive value.
nonwhite patients were classified into the rule-in zone, though they had significantly lower adjusted odds of 30-day cardiac death or MI compared to white patients in the rule-in zone.When the HEAR score and hs-cTnT ESC 0/1-h algorithm were combined, NPV ≥ 99% for 30-day cardiac death or MI was achieved in all subgroups.Future studies should assess the integration of individualized hs-cTnT thresholds to optimize risk stratification of ED patients with chest pain.
0/1-h algorithm for evaluation of ED patients with chest pain is a class I recommendation by the ESC T A B L E 1 T A B L E 2 Abbreviations: CI, confidence interval; ESC, European Society of Cardiology; MACE, major adverse cardiovascular event; MI, myocardial infarction.