{"paper_id":"afbfbda3-1819-474b-8c5a-024205bd832e","body_text":"Vol.:(0123456789)\nImproved Utilization of Serial Testing Without Increased \nAdmissions after Implementation of High‑Sensitivity \nTroponin I: a Controlled Retrospective Cohort Study\nLaura Warren, MD, PhD1, Brett G. Fischer, MD2, Amos Shemesh, MD3, Jean Scofi, MD3, \nNekee Pandya, MD2, Robert J. Kim, MD2, Caroline Andy, MS4, Sophie Rand, MPH1, Jim Yee, BS1, \nStacia Semple, MD1, Amy Chadburn, MD1, He S. Yang, PhD1, Peter A. D. Steel, MD3, and \nZhen Zhao, PhD1 \n1Department of Pathology and Laboratory Medicine, Weill Cornell Medicine, New York, NY, USA; 2Weill Department of Medicine, \nWeill Cornell Medicine, New York, NY, USA; 3Department of Emergency Medicine, Weill Cornell Medicine, New York, NY, USA; 4Division \nof Biostatistics, Weill Cornell Medicine, New York, NY, USA\nABSTRACT\nBACKGROUND: Guidelines recommend high-sensitiv-\nity cardiac troponin (hs-cTn) for diagnosis of myocardial \ninfarction. Use of hs-cTn is increasing across the U.S., \nbut questions remain regarding clinical and operational \nimpact. Prior studies have had methodologic limitations \nand yielded conflicting results.\nOBJECTIVE:  To evaluate the impact of transitioning \nfrom conventional cardiac troponin (cTn) to hs-cTn on \ntest and resource utilization, operational efficiency, and \npatient safety.\nDESIGN: Retrospective cohort study in two New York \nCity hospitals during the months before and after tran-\nsition from conventional cTn to hs-cTn at Hospital 1. \nHospital 2 served as a control.\nPARTICIPANTS: Consecutive emergency department \n(ED) patients with at least one cTn test resulted.\nINTERVENTION: Multifaceted hs-cTn intervention bun-\ndle, including a 0/2-h diagnostic algorithm for non-ST-\nelevation myocardial infarction, an educational bundle, \nenhancements to the electronic medical record, and nurs-\ning interventions to facilitate timed sample collection.\nMAIN MEASURES:  Primary outcomes included serial \ncTn test utilization, probability of hospital admis -\nsion, ED length of stay (LOS), and among discharged \npatients, probability of ED revisit within 72  h result -\ning in hospital admission. Multivariable regression \nmodels adjusted for age, sex, temporal trends, and \ninterhospital differences.\nKEY RESULTS: The intervention was associated with \nincreased use of serial cTn testing (adjusted risk differ -\nence: 48 percentage points, 95% CI: 45–50, P < 0.001) and \nED LOS (adjusted geometric mean difference: 50  min, \n95% CI: 50–51, P < 0.001). There was no significant asso-\nciation between the intervention and probability of admis-\nsion (adjusted relative risk [aRR]: 0.99, 95% CI: 0.89–1.1, \nP = 0.81) or probability of ED revisit within 72 h resulting \nin admission (aRR: 1.1, 95% CI: 0.44–2.9, P = 0.81).\nCONCLUSIONS:  Implementation of a hs-cTn inter -\nvention bundle was associated with an improvement \nin serial cTn testing, a neutral effect on probability of \nhospital admission, and a modest increase in ED LOS.\nKEY WORDS:  high sensitivity cardiac troponin; test utilization; \nemergency department; patient outcomes; length of stay; admission; \nmyocardial infarction\nJ Gen Intern Med  \nDOI: 10.1007/s11606-023-08535-3 \n© The Author(s), under exclusive licence to Society of General Internal \nMedicine 2023\nINTRODUCTION\nThe 5th generation cardiac troponin T (hs-cTnT) and high-\nsensitivity cardiac troponin I (hs-cTnI) assays have been \napproved for clinical use by the U.S. Food and Drug Admin-\nistration since 2017 and 2018, respectively. High-sensitivity \ncardiac troponin (hs-cTn) assays are more precise at low \nconcentrations than conventional assays and allow shortened \ntime interval to repeat assessment. Recent guidelines and rec-\nommendations endorse using hs-cTn with rapid diagnostic \nalgorithms to evaluate patients undergoing workup for non-\nST-elevation myocardial infarction (NSTEMI) in the emer-\ngency department (ED).1–4 Rapid diagnostics may allow for \nimproved risk stratification and faster disposition decisions, \npotentially leading to improved quality and efficiency of care.\nHospitals across the U.S. are transitioning from con -\nventional cardiac troponin (cTn) to hs-cTn, 5 but questions \nremain concerning the impact of such a change on clinical \nand operational outcomes. While there is hope that switch -\ning to hs-cTn will shorten time to ED disposition decision, \nthere also has been concern that implementing hs-cTn will \nincrease unnecessary admissions.5–7 Although European and \nAustralian studies have shown reduced ED length of stay \n(LOS) and a variable impact on admissions,7–15 these studies \nmay not be applicable in the U.S. given the vastly different \nhealthcare systems. Furthermore, results from U.S. studies \nexamining these outcomes have been less consistent. 5, 16–20 \nLaura Warren and Brett G. Fischer contributed equally to this work.\nReceived May 2, 2023 \nAccepted November 9, 2023\n739\n39(5):739–46\nPublished online , 2023November 22\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort Study JGIM\nAnother potential advantage of a shortened time interval \nrequired for repeat testing is increased adherence to use of \nserial cTn testing, which is the standard of care for acute \ncoronary syndrome (ACS) evaluation. 21 The use of serial \ncTn testing varies by practitioner and clinical situation, but \nseveral U.S. studies have shown an increase in use of serial \ncTn testing after implementation of hs-cTn.16, 17, 19, 20 Impor-\ntantly, to our knowledge, all prior studies have used before-\nand-after study design, which lacks a comparison (control) \ngroup, limiting the ability to account for temporal trends \nunrelated to the intervention.\nIn February 2021, NewYork-Presbyterian/Weill Cornell \nMedical Center (Hospital 1) implemented a hs-cTnI assay, \ntogether with several clinical, educational, and electronic \nmedical record (EMR) cointerventions. The aim of this study \nis to understand the impact of this multifaceted interven -\ntion on test and resource utilization, operational efficiency, \nand patient safety in the ED of an urban U.S. hospital with \na diverse patient population. To address the limitations of \nsingle-arm before-and-after study design, we used a sister \nhospital as the control—NewYork-Presbyterian/Lower Man-\nhattan Hospital (Hospital 2), which shares ED clinicians but \ndid not implement hs-cTn. Additional hospital characteris -\ntics are described in the Supplement.\nMETHODS\nPatient Population and Data Collection\nWe performed a retrospective cohort study at two urban \nmedical centers that share ED clinicians. Consecutive \npatients with at least one cTn test resulted were eligible for \ninclusion. Eligible patients were excluded if they walked \nout, were missing a disposition order to either admit or \ndischarge, or had an ED LOS  > 72 h, which was thought \nto represent inaccuracies in the dataset. Patients with \nmultiple unique ED visits with cTn results were counted \nmultiple times. Patients who were ineligible for inclusion \n(due to not having any cTn test results) were separately \nanalyzed in a post-hoc analysis. Demographic, laboratory, \nand outcomes data were extracted from the institution’s \ndata warehouse using Structured Query Language queries. \nThis study was reviewed and determined to meet exemp -\ntion requirements at HHS 45 CFR 46.104(d) by the Weill \nCornell Medicine institutional review board.\nIntervention\nAt Hospital 1, Siemens Centaur hs-cTnI was introduced on \nFebruary 22, 2021, to replace Siemens Centaur conventional \ncardiac troponin I (cTnI) (Siemens Healthineers, Erlangen, \nGermany). At Hospital 2, Abbott Architect conventional \ncTnI (Abbott Laboratories, IL, U.S.) was used throughout \nthe study. Period 1 was defined as November 1, 2020, to \nFebruary 21, 2021, and Period 2 was defined as February \n22, 2021, to May 31, 2021 (Fig.  1). Prior to the transition at \nHospital 1, and throughout the study at Hospital 2, a clinical \npathway for the evaluation of chest pain and ACS recom -\nmended cTn testing at 0 and 4 h for patients presenting with \nsuspicion for NSTEMI.\nFor about 6 months prior to the transition, a multidis -\nciplinary team that included stakeholders from emergency \nmedicine, hospital medicine, cardiology, pathology and \nlaboratory medicine, nursing, and information technology \ngroups met at least monthly. These meetings led to the fol -\nlowing interventions: (i) development of a 0/2-h diagnostic \nalgorithm for NSTEMI (Figure S1); 22 (ii) development and \nlaunch of an educational bundle to those who order, collect, \nor interpret cTn in multiple practice settings; (iii) enhance -\nments to the EMR, including creation of a serial cTn order \nset, automatic calculation of the 2-h delta cTn value, and \nhyperlink to the diagnostic algorithm included in cTn result \nreports; and (iv) additional nursing interventions to facilitate \nappropriate timing of 2-h sample collection, including mes-\nsaging alerts and restrictions on early label printing. These \nwill be referred to as the hs-cTn intervention bundle  in this \nstudy.\nhs-cTn intervention\nBundle*i mplemented\non Feb2 2, 2021 \nHospital 1\nSiemensC entaur\nconventional cTnI\nPeriod 1\nNov1 ,2 020, to Feb2 1, 2021\nPeriod 2\nFeb2 2, 2021,t oM ay 31,2 021\nSiemensC entaur\nhs-cTnI\nNo interventionHospital 2\nAbbott Architect\nconventional cTnI\nAbbott Architect\nconventional cTnI\nFigure 1  Study design. cTnI = cardiac troponin I; hs-cTn = high-sensitivity cardiac troponin; hs-cTnI = high-sensitivity cardiac troponin I. \n*The hs-cTn intervention bundle included a 0/2-h diagnostic algorithm for non-ST-elevation myocardial infarction (shown in Figure S1), an \neducational bundle, enhancements to the electronic medical record, and nursing interventions to facilitate timed sample collection.\n740\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort StudyJGIM\nOutcome Measures\nThere were four primary outcomes: probability of serial \n(≥  2) cTn testing, defined as having a second cTn resulted \nprior to disposition decision (either admit or discharge) \nbeing entered into the EMR by the ED clinician; probability \nof hospital admission for any reason; ED LOS, defined as \ntime from ED arrival to time of disposition decision; and \namong discharged patients, probability of ED revisit within \n72 h resulting in admission for any reason (primary safety \noutcome), which was measured using a separate dataset that \nenabled capture of ED revisits to any of the 8 hospitals in \nthe NewYork-Presbyterian health system. For probability of \nserial cTn testing, a sensitivity analysis redefined the out -\ncome as having a second cTn resulted within 8 h of the first, \nregardless of the time of disposition decision, to account for \npatients for whom a second cTn was ordered by the inpatient \nteam. Secondary outcomes included test turnaround time \n(time from when a sample was received by the laboratory \nto when the result was reported in the EMR) and time inter-\nval between collection of the first and second tests (among \nthose who had serial testing). Post-hoc outcome measures \nincluding temporal trends in SARS-CoV-2 test results and \noverall ED volume were added to evaluate the potential for \nresidual confounding related to the coronavirus disease 2019 \n(COVID-19) pandemic.\nTroponin Assays\nThe limit of detection (LOD), limit of quantification (LOQ), \nand the  99th percentile upper reference limit (URL) of the \nSiemens Centaur conventional cTnI assay are 0.006 µg/L \n(6 ng/L), 0.03 µg/L (30 ng/L) and 0.04 µg/L (40 ng/L), \nrespectively. Values < 0.03 µg/L were reported as < 0.03 µg/L \nin the EMR. The LOQ of the Siemens Centaur hs-cTnI assay \nis 2.5 ng/L, and the  99th percentile URLs are 58 ng/L for \nmales and 39 ng/L for females. For Abbott Architect con -\nventional cTnI, Abbott did not perform the Limit of Detec -\ntion (LoD) and Limit of Quantitation (LoQ) 99% confi -\ndence interval; instead, sensitivity testing was performed to \ndescribe the values at the low end of the curve. The analyti -\ncal sensitivity was calculated to the 95% level of confidence \nas < 0.01 ng/mL (<  10 ng/L).\nStatistical Analysis\nAll analyses were performed using R version 4.2.1. Descrip-\ntive statistics were used to summarize baseline characteris -\ntics and raw outcomes data. Median and IQR were reported \nfor continuous variables, and percentages were reported for \ncategorical variables.\nFor probability of serial cTn testing, probability of \nadmission, and probability of ED revisit within 72 h result -\ning in admission, multiple Poisson regression modeling \nwas employed. The regression models contained ED site \n(Hospital 1 vs. Hospital 2), time (Period 2 vs. Period 1), \nsite-time interaction, age, and sex as covariates. The interac-\ntion term coefficient for Hospital 1, Period 2, was interpreted \nas the independent effect of implementation of the hs-cTn \nintervention bundle when adjusting for age, sex, temporal \ntrends, and interhospital differences. Results are reported as \nadjusted relative risk (aRR) and 95% CI. If the result for the \nvariable of interest was statistically significant, an additional \nstep was added to the models to produce the adjusted risk \ndifference (aRD) and 95% CI.\nFor ED LOS, multiple linear regression modeling was \nperformed using the same five covariates as the Poisson \nmodels. The LOS outcome was log-transformed to reduce \nright skewness and better satisfy the modeling assumption \nof linearity. Coefficients were then exponentiated to back-\ntransform the results onto an untransformed scale. Adjusted \ngeometric mean ratio (aGMR) with 95% CI is reported.16 If \nthe result for the variable of interest was statistically signifi-\ncant, an additional step was added to the model to produce \nthe adjusted geometric mean difference (aGMD) and 95% \nCI. Given concern that the association between interven -\ntion and ED LOS may be modified by number of cTn tests \nand ED disposition, two exploratory subgroup analyses were \nperformed based on the number of cTn tests resulted prior \nto disposition decision (1 vs. ≥  2) and disposition (admit vs. \ndischarge). To test for interaction, linear regression models \nwere constructed containing a three-way interaction between \nED site, time, and the subgroup variable.\nAdditional post-hoc analytical methods, including assess-\nment of probability of admission at Hospital 1 only (without \na control group), comparison of temporal trends in SARS-\nCoV-2 test results and overall ED volume between hospitals, \nand assessment of probability of admission and ED LOS \namong patients without cTn testing, are described in the \nSupplement.\nFor the main analysis of primary outcomes and subgroup \ntests for interaction only, P values are reported and rep -\nresent the probability of finding an effect at least as large \nas that observed if there were truly no difference between \ngroups. By convention, P < 0.05 was considered statistically \nsignificant.\nRESULTS\nPatient Characteristics\nAt Hospital 1, out of 38,956 encounters, 9,757 (25%) were \nincluded, and at Hospital 2, out of 18,106 encounters, 4,321 \n(24%) were included (Fig.  2). Baseline characteristics for \nincluded patients are described in Table  1 and Table S1. \nOverall, patients at Hospital 1 were slightly older, more \nlikely to be female, more likely to identify their primary \nlanguage as English, and more likely to have commercial \ninsurance or Medicare than at Hospital 2. At Hospital 2, \n741\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort Study JGIM\npatients were more likely to identify their primary language \nas Cantonese or Mandarin and more likely to have Medic -\naid than at Hospital 1. Within each site, these demographic \nvariables were similar between time periods. Data on race \nand ethnicity were deemed unreliable due to a high degree \nof missingness.\nPrimary Outcomes: Probability of Serial cTn \nTesting, Probability of Admission, ED LOS, \nand Probability of 72‑h ED Revisit Resulting in \nAdmission (Safety Outcome)\nResults for primary outcomes are summarized in Table  2, \nand regression model results for all included variables are \nreported in Table S2.\nThe percent of encounters with  ≥  2 cTn tests resulted \nprior to ED disposition decision at Hospital 1 was 18% dur-\ning Period 1 and 65% during Period 2. In the multivariable \nmodel, the intervention was associated with an absolute \nincrease of 48 percentage points (95% CI: 45–50) in serial \ncTn utilization (P < 0.001). In the sensitivity analysis allow-\ning for a second cTn test to have resulted up to 8 h after the \nfirst, serial cTn testing generally occurred more frequently, \nbut the overall effect was similar (aRD: 49 percentage points, \n95% CI: 46–51).\nThe probability of admission at Hospital 1 was 55% dur -\ning Period 1 and 51% during Period 2. However, in the mul-\ntivariable model, the intervention had no significant effect \non probability of admission (aRR: 0.99, 95% CI: 0.89–1.1, \nP = 0.81).\nThe median ED LOS at Hospital 1 was 293 min (IQR: \n201, 409) during Period 1 and 316 min (IQR: 241, 424) \nduring Period 2. In the multivariable model, the intervention \nwas associated with an increase of 50 min (95% CI: 50–51) \nin geometric mean ED LOS ( P < 0.001). However, in the \nexploratory subgroup analysis by number of cTn tests, ED \nLOS decreased after the intervention in both patients with \nand without serial cTn testing, and there was no statistically \nsignificant interaction between subgroups  (Pinteraction = 0.59). \nIn the exploratory subgroup analysis by ED disposition, ED \nLOS increased after the intervention in both patients who \nwere admitted and discharged, and there was no statistically \nsignificant interaction between subgroups  (Pinteraction = 0.40). \n(Table S3).\nAmong those encounters that resulted in discharge from \nthe ED at Hospital 1, the probability of ED revisit within \n72 h resulting in admission was 0.8% during Period 1 and \n38,956 ED encounters\nwere screened\nHospital 1\n10,043 Had ≥1 cTnt est\nresulted\nHospital 2\n286 Were excluded\n200 Walked out aftere valuation\n78 Had missing disposition\n7 Had ED LOS> 72 hours\n1 Walked out beforee valuation\n9,757 Were included in\nfinal analysis\n28,913 Were ineligible fori nclusion\n28,913 Did not have a cTnt est resulted\n18,106 ED encounters\nwere screened\n4,460 Had ≥1 cTnt est\nresulted\n139 Were excluded\n116W alkedo ut aftere valuation\n15 Had missing disposition\n6 Had ED LOS> 72 hours\n2 Walked outb eforee valuation\n4,321W erei ncludedi n\nfinal analysis\n13,646 Were ineligible fori nclusion\n13,646 Did noth ave a cTnt est resulted\nFigure 2  Flow diagram of the study cohort at each site. ED = emergency department; cTn = cardiac troponin; LOS = length of stay.\nTable 1  Baseline Characteristics\n* Conventional cardiac troponin I was used in these groups\n† High-sensitivity cardiac troponin I was used in this group\nHospital 1 (Intervention) Hospital 2 (Control)\nCharacteristic Overall\nN = 9,757\nPeriod 1 (Pre)*\n N = 5,163\nPeriod 2 (Post)†\nN = 4,594\nOverall\nN = 4,321\nPeriod 1*\nN = 2,246\nPeriod 2*\n N = 2,075\nAge, Median (IQR) 64 (49, 77) 64 (49, 77) 64 (49, 77) 61 (47, 76) 62 (48, 77) 60 (47, 76)\nFemale sex, N (%) 5,200 (53) 2,709 (52) 2,491 (54) 2,043 (47) 1,047 (47) 996 (48)\n742\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort StudyJGIM\n1.0% during Period 2. In the multivariable model, the \nintervention had no significant effect on probability of ED \nrevisit within 72 h resulting in admission (aRR: 1.1, 95% CI: \n0.44–2.9, P = 0.81).\nSecondary Outcomes: Test Turnaround Time, \nTime Between Serial cTn Test Collection, and \nPost‑Hoc Outcomes\nTest turnaround time was similar at both hospitals during \nboth time periods (Table S4). Among those with serial cTn \ntesting, the median time between collection of the first and \nsecond sample at Hospital 1 was 3.6 h (IQR: 3.0, 4.2) dur -\ning Period 1 and 2.0 h (IQR 1.8, 2.2) during Period 2. The \nmedian time between collection of the first and second sam-\nple at Hospital 2 during both time periods was similar to \nthat of Hospital 1 during Period 1 (Figure S2). Results of \npost-hoc analyses are described in the Supplement.\nDISCUSSION\nIn a retrospective cohort study with one intervention site and \none control site, implementation of a hs-cTn intervention \nbundle was associated with a large increase in serial cTn \ntesting, neutral effect on probability of hospital admission, \nand modest increase in ED LOS. For discharged patients, \nthe intervention had a neutral effect on probability of ED \nrevisit within 72 h resulting in admission, although the effect \nestimate was imprecise due to the small number of events.\nA major strength of this study is the comparison to a con-\ntrol hospital where the intervention was not undertaken. This \nhas significant advantages over single-arm before-and-after \nstudies, most notably in controlling for potential tempo -\nral trends unrelated to the intervention. To the best of our \nknowledge, this is the first study to employ this design to \nevaluate the effect of implementation of hs-cTn on these \noutcomes.\nRecent cardiology and emergency medicine society \nguidelines recommend use of serial cTn testing in patients \npresenting with symptoms suspicious for NSTEMI. 2–4, 21 \nThus, implementation of the hs-cTn intervention bundle \nwas associated with an improvement in care consistent with \nnational guidelines. Consistent with our findings, previous \nU.S. before-and-after studies have also demonstrated large \nincreases in serial cTn testing after switching from conven -\ntional cTn to hs-cTn.16, 17, 19, 20\nWithin the hs-cTn intervention bundle, there were a \nnumber of potential contributors to the observed change \nin ordering patterns, including employment of a diagnos -\ntic algorithm that prescribes serial cTn testing in most \npatients in whom the ED clinician is considering the \ndiagnosis of NSTEMI, implementation of a serial cTn \nTable 2  Primary Outcomes\ncTn = cardiac troponin; ED = emergency department; LOS = length of stay\n* Conventional cardiac troponin I was used in these groups\n† High-sensitivity cardiac troponin I was used in this group\n‡ Independent effect of implementation of the high sensitivity cardiac troponin intervention bundle at Hospital 1 after adjusting for age, sex, tempo -\nral trends, and interhospital differences\n§ Adjusted risk difference expressed as percentage points, derived from Poisson regression model\n‖ Adjusted relative risk, derived from Poisson regression model\n¶ ED LOS was defined as the time from ED arrival to time of decision entered into electronic medical record to either admit or discharge. ED LOS \nmedian and IQR is expressed in minutes\n# Adjusted geometric mean difference, derived from linear regression model on log-transformed data\n** Only patients discharged from the ED during the index visit were eligible for this outcome: for Hospital 1, Period 1, N = 2,346; for Hospital 1, \nPeriod 2, N = 2,299; for Hospital 2, Period 1, N = 1,107; for Hospital 2, Period 2, N = 1,101. Within each hospital and period combination, the sum \nof discharged and admitted patients does not exactly equal the total N because the data used for this outcome were derived from a separate dataset \nthat enabled capture of ED revisits to any of the 8 hospitals in the NewYork-Presbyterian network, but to which exclusion criteria were not applied\nHospital 1 (Intervention) Hospital 2 (Control)\nOutcome Period 1 (Pre)*\nN = 5,163\nPeriod 2 (Post)†\nN = 4,594\nPeriod 1*\nN = 2,246\nPeriod 2*\nN = 2,075\nAdjusted  Effect‡\n(95% CI)\nP value\nSerial (≥  2) cTn tests,\n  N (%)\n916 (18) 2,999 (65) 302 (13) 403 (19) 48 (45–50)§  < 0.001\nHospital admission,\n  N (%)\n2,833 (55) 2,344 (51) 1,122 (50) 949 (46) 0.99 (0.89–1.1)‖ 0.81\nED  LOS¶,\n  Median (IQR)\n293 (201, 409) 316 (241, 424) 263 (187, 376) 269 (189, 369) 50 (50–51)#  < 0.001\n72-h ED revisit \nresulting in hospital \nadmission**,\n   N (%)\n41 (0.8) 47 (1.0) 11 (0.5) 11 (0.5) 1.1 (0.44–2.9)‖ 0.81\n743\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort Study JGIM\norder set in the EMR, the shorter time interval needed to \ncheck a second cTn test, the educational bundle, and general \nincreased awareness of national guidelines over time. Among \nthese hypotheses, Hospital 2 may have also been somewhat \naffected by the educational bundle and clinical practice influ-\nence from the 0/2-h diagnostic algorithm for NSTEMI (given \nthat ED clinicians rotated between both sites), as well as \nincreased guideline awareness over time. However, the fact \nthat the change at Hospital 1 remained large and statisti -\ncally significant when controlling for Hospital 2 suggests \nthat the majority of the effect was isolated to features specific \nto actual implementation of the hs-cTn intervention bundle, \nthough it is not possible to further isolate the effect to any of \nthe bundle’s individual components.\nFindings from previous U.S. studies assessing the impact \nof hs-cTn implementation on probability of admission and \nED LOS have been less consistent.5, 16–20 For example, Suh, \net al., showed no change in median provider-to-disposition \ndecision time (defined similarly to our ED LOS outcome) \nand no increase in hospital admissions, 17 while Ford, et al., \nreported increased ED LOS and decreased admissions. 16 In \nthe largest of these studies, McCarthy, et al. demonstrated a \nstatistically significant decrease in ED LOS, but the clinical \nand operational importance of this finding was unclear, with \na difference in median ED LOS of  < 30 min. 5 Many factors \nmay have contributed to these discrepant observations, such \nas differences in clinical and EMR pathways employed, turn-\naround time of serial cTn results, patient population, the edu-\ncation and training provided to clinical teams surrounding the \ntransition, and study design and statistical methods, including \nthe challenges in correcting for multifactorial operational effi-\nciencies contributing to variations in ED LOS.\nAs the first two-arm cohort study with a control group to \nevaluate the effect of implementation of hs-cTn on the prob-\nability of admission and ED LOS among patients with cTn \ntesting, our findings greatly enhance the quality of the over-\nall body of evidence. The post-hoc analysis using before-\nand-after design without a control group was performed \nto demonstrate this point: had the present study followed \nthe same design as prior studies, an incorrect conclusion—\ndecreased probability of admission after the intervention—\ncould have been drawn. While our controlled analysis did not \ndemonstrate a reduction in probability of admission after the \nintervention, given the present study’s strengths, the lack of \neffect should mitigate concerns that implementing hs-cTn \nmight increase hospital admissions. 5–7\nOn the other hand, ED LOS increased after the implemen-\ntation of the hs-cTn intervention bundle, despite a reduction \nof 1.6 h in median time between serial cTn test collection \nat Hospital 1 and a stable test turnaround time. Post-hoc \nanalysis demonstrated that ED LOS was stable in patients \nwho did not have cTn testing. The clinical and operational \nimportance of the increased ED LOS remains unclear, with \na difference in geometric mean ED LOS of  < 1 h.\nThe most plausible explanation for the observed increase \nin ED LOS is the increase in serial cTn testing instead of \nsingle cTn testing, suggesting that the observed increase in \nED LOS may have been an unintended but worthwhile cost. \nOur hospital implemented a 0/2-h algorithm, but the finding \nof increased ED LOS may warrant reconsideration of using \na 0/1-h algorithm. Interestingly, when analyzed by subgroup \nbased on number of cTn tests, ED LOS appeared to decrease \nin both groups with and without serial testing. However, this \nexploratory subgroup analysis should be interpreted with \ncaution due to concern for confounding by other patient fac-\ntors, given the drastic shift in the proportion of patients with \nserial cTn testing between pre- and post-intervention periods \nat Hospital 1. Furthermore, number of cTn tests is likely a \nmediator rather than a confounder in the relationship between \nthe intervention and ED LOS, and stratification by mediators \nmay introduce collider bias.23\nAs with many similar studies, our study has several addi-\ntional limitations. First, we conducted this study in a single \nhospital system, which may limit generalizability. Further -\nmore, interpretation of our data on race and ethnicity is lim-\nited by a large degree of missingness, a previously described \nproblem that is not unique to our dataset.24\nSecond, as with any observational study, we cannot rule \nout residual confounding. While Hospital 2 shares the same \nED clinicians and is within the same geographic location \nas Hospital 1 (Manhattan, New York), the patient popula -\ntion may vary in terms of demographics, diagnoses, and \nacuity. For example, more patients at Hospital 2 identified \nCantonese or Mandarin as their primary language (which \nwe included as a complement to race and ethnicity data). \nWhile these differences may represent potential causes of \nresidual confounding, we did attempt to control for inter -\nhospital differences in our statistical models. In addition, \nwhile inclusion of Hospital 2 was intended to control for \npotential temporal trends unrelated to the intervention, \nthere could have been variation in temporal trends between \nsites. However, we are reassured that, at least with ED vol -\nume and COVID-19 infections, this did not seem to be the \ncase.\nThird, we were not able to include patient-level clinical \ndata, limiting our ability to restrict inclusion to patients with \nsuspicion for ACS, assess for cardiovascular risk factors, \nevaluate the cause of 72-h ED revisits resulting in admission, \nand to assess clinical outcomes.\nFourth, we did not evaluate utilization of additional down-\nstream testing, and we did not assess financial impact for the \nhospital or patients.\nFifth, the use of ED revisit within 72 h resulting in acute \nhospital admission as a surrogate quality measurement, \nwhile well recognized in ED literature, is limited by the \ndataset not including patients who returned to other institu -\ntions, which existing research suggests occurs in 11–24% of \nall ED return visits. 25–27 Our study utilized data to capture \n744\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort StudyJGIM\nreturn visits across 8 New York City EDs, but this may have \nprovided incomplete mitigation.\nThese limitations will serve as areas of further research.\nCONCLUSIONS\nIn this retrospective observational cohort study with one \nintervention site and one control site, implementation of a \nhs-cTn intervention bundle when switching from conven -\ntional cTnI to hs-cTnI was associated with increased adher-\nence to guideline-recommended serial cTn testing without \nsignificant effect on probability of admission or 72-h ED \nrevisit leading to admission, and a modest increase in ED \nLOS. Taken together with the evidence from prior U.S. \nstudies in real-world ED settings, transitions to hs-cTn are \nconsistently associated with improvements in adherence \nto guideline-recommended serial cTn testing, and while \nthe impact on hospital admissions and ED LOS remains \nunclear, it seems most likely to be small.\nSupplementary Information The online version contains supplementary \nmaterial available at https:// doi. org/ 10. 1007/ s11606- 023- 08535-3.\nAcknowledgements:  We thank Arthur T. Evans, MD, MPH (Weill \nCornell Medicine) for his invaluable guidance in developing the sta -\ntistical analysis plan and Patrick Rumble, BS (Weill Cornell Medi -\ncine), for his instrumental assistance in data acquisition.\nCorresponding Author:  He S. Yang, PhDPeter  A. D. Steel, MDZ -\nhen Zhao, PhD; Department of Pathology and Laboratory Medicine \nDepartment of Emergency Medicine, Weill Cornell Medicine Weill \nCornell Medicine, New York, NY, USA, New York, NY, USA (e-mail: \nhey9012@med.cornell.edu pes9027@med.cornell.edu zhz9010@\nmed.cornell.edu).\nData Availability The datasets generated and analyzed during the \ncurrent study are available from the corresponding authors on  rea-\nsonable request.\nDeclarations: \nConflict of Interest: LW is pending a patent for “Methods for detect-\ning and treating endometriosis” Publication number: 20210096137. \nJS had a registration fee waived as a speaker at XGM Conference May \n2023 and is the Co-Chair of the didactics Sub-committee for the Society \nof Academic Emergency Medicine 2022–2023 (unpaid). RJK has stock \noptions in Cleerly Health, Inc. AC has received consulting fees from Lei-\nca Biosystems (Immunohistochemistry) and Boehringer-Ingelheim (he-\nmatopoietic neoplasms) and lecture fees at McGill University, Medical \nCollege of Wisconsin, and Northshore University Hospital (Evanston, IL). \nHSY received a speaker fee from Siemens Healthineers. P ADS received \nconsulting fees for ET health, speaker fees for the American College of \nEmergency Physicians Directors Academy, and is on the David Lynch \nFoundation National Advisory Board. ZZ has sponsored research sup-\nported by Novartis, Waters, Siemens Healthineers, Polymedco, Waters, \nRoche and ET Healthcare and has received consulting/speaker fee \nfrom Siemens Healthineers, Roche and ET Healthcare.\nREFERENCES\n 1. Kontos MC, de Lemos JA, Deitelzweig SB, et al. 2022 ACC Expert Con-\nsensus Decision Pathway on the Evaluation and Disposition of Acute \nChest Pain in the Emergency Department: A Report of the American \nCollege of Cardiology Solution Set Oversight Committee. J Am Coll Cardiol. \n2022;80(20):1925-1960.\n 2. Collet JP, Thiele H, Barbato E, et a l. 2020 ESC Guidelines for \nthe management of acute coronary syndromes in patients pre -\nsenting without persistent ST-segment elevation . Eur Heart J. \n2021;42:1289-1367.\n 3. Sandoval Y, Apple FS, Mahler SA, Body R, Collinson PO, Jaffe AS . \nHigh-Sensitivity Cardiac Troponin and the 2021 AHA/ACC/ASE/\nCHEST/SAEM/SCCT/SCMR Guidelines for the Evaluation and Diag -\nnosis of Acute Chest Pain. Circulation. 2022;146:569-581.\n 4. Gulati M, Levy PD, Mukherjee D, et  al. 2021 AHA/ACC/ASE/\nCHEST/SAEM/SCCT/SCMR Guideline for the Evaluation and Diag -\nnosis of Chest Pain: A Report of the American College of Cardiology/\nAmerican Heart Association Joint Committee on Clinical Practice \nGuidelines . Circulation. 2021;144:e368-e454.\n 5. McCarthy C, Li S, Wang TY, et  al. Implementation of High-Sensitiv -\nity Cardiac Troponin Assays in the United States . J Am Coll Cardiol. \n2023;81(3):207-219.\n 6. Yang HS, Shemesh A, Li J, et  al. No increase in the incidence of car -\ndiac troponin I concentration above the 99th percentile by Siemens \nCentaur high-sensitivity compared to the contemporary assay . Clin \nBiochem. 2021;89:77-80.\n 7. Khan E, Lambrakis K, Nazir SA, et  al. Implementation of more sen -\nsitive cardiac troponin T assay in a state-wide health service . Int J \nCardiol. 2022;347:66-72.\n 8. Peck D, Knott J, Lefkovits J . Clinical impact of a high-sensitivity \ntroponin assay introduction on patients presenting to the emergency \ndepartment . Emerg Med Australas. 2016;28:273-8.\n 9. Greenslade JH, Parsonage W, Foran L, et  al. Widespread Introduc -\ntion of a High-Sensitivity Troponin Assay: Assessing the Impact on \nPatients and Health Services . J Clin Med. 2020;9(6):1883.\n 10. Jülicher P, Greenslade JH, Parsonage WA, Cullen L . The organi -\nsational value of diagnostic strategies using high-sensitivity tro -\nponin for patients with possible acute coronary syndromes: a trial-\nbased cost-effectiveness analysis . BMJ Open. 2017;7:e013653.\n 11. Roos A, Holzmann M J. Healthcare and Resource Use in Patients \nWith Stable High-Sensitivity Cardiac Troponin T Levels. Am J Cardiol. \n2020;128:67-74.\n 12. Anand A, Lee KK, Chapman AR, et  al. High-Sensitivity Cardiac \nTroponin on Presentation to Rule Out Myocardial Infarction: A \nStepped-Wedge Cluster Randomized Controlled Trial . Circulation. \n2021;143:2214-2224.\n 13. Ambavane A, Lindahl B, Giannitsis E, et  al. Economic evaluation \nof the one-hour rule-out and rule-in algorithm for acute myocardial \ninfarction using the high-sensitivity cardiac troponin T assay in the \nemergency department . PLoS One. 2017;12:e0187662.\n 14. Conway R, Byrne D, Cournane S, O’Riordan D, Coveney S, Silke B. \nIs there excessive troponin testing in clinical practice? Evidence from \nemergency medical admissions. Eur J Intern Med. 2021;86:48-53.\n 15. Johannessen TR, Halvorsen S, Atar D, et  al. Cost-effectiveness of a \nrule-out algorithm of acute myocardial infarction in low-risk patients: \nemergency primary care versus hospital setting . BMC Health Serv \nRes. 2022;22:1274.\n 16. Ford JS, Chaco E, Tancredi DJ, Mumma BE . Impact of high-sen -\nsitivity cardiac troponin implementation on emergency department \nlength of stay, testing, admissions, and diagnoses . Am J Emerg Med. \n2021;45:54-60.\n 17. Suh EH, Tichter AM, Ranard LS, et  al. Impact of a rapid high-\nsensitivity troponin pathway on patient flow in an urban emergency \ndepartment . J Am Coll Emerg Physicians Open. 2022;3:e12739.\n 18. Bevins NJ, Chae H, Hubbard JA, et  al . Emergency Depart -\nment Management of Chest Pain With a High-Sensitivity Tro -\nponin-Enabled 0/1-Hour Rule-Out Algorithm . Am J Clin Pathol. \n2022;157:774-780.\n 19. Ola O, Akula A, Michieli LD, et al. Clinical Impact of High-Sensitivity \nCardiac Troponin T Implementation in the Community . J Am Coll \nCardiol. 2021;77:3160-3170.\n 20. Younis A, Farooq S, Bisognano JD, et  al. Outcomes Associated with \nIntroduction of the 5(th) Generation High-Sensitivity Cardiac Tro -\nponin in Patients Presenting with Cardiovascular Disorders . J Emerg \nMed. 2022;62:657-667.\n 21. Tomaszewski CA, Nestler D, Shah KH, Sudhir A, Brown MD . Clini-\ncal Policy: Critical Issues in the Evaluation and Management  of Emer-\ngency Department Patients With Suspected  Non-ST-Elevation Acute \nCoronary Syndromes . Ann Emerg Med. 2018;72:e65-e106.\n745\n\nWarren et al.: High-Sensitivity Troponin Implementation: Controlled Cohort Study JGIM\n 22. Fischer BG, Evans AT . High-Sensitivity Cardiac Troponin Algo -\nrithms and the Value of Likelihood Ratios. J Gen Intern Med. \n2023;38(9):2189-2193.\n 23. Rohrer JM. Thinking clearly about correlations and causation: Graphi-\ncal causal models for observational data . Advances in Methods and \nPractices in Psychological Science. 2018;1:27-42.\n 24. Polubriaginof FCG, Ryan P, Salmasian H, et al. Challenges with qual-\nity of race and ethnicity data in observational databases. J Am Med \nInform Assoc. 2019;26:730-736.\n 25. Chartier LB, Ovens H, Hayes E, et  al. Improving Quality of Care \nThrough a Mandatory Provincial Audit Program: Ontario’s Emer -\ngency Department Return Visit Quality Program . Ann Emerg Med. \n2021;77:193-202.\n 26. Rising KL, Karp DN, Powell RE, Victor TW, Carr BG . Geography, \nNot Health System Affiliations, Determines Patients’ Revisits to the \nEmergency Department . Health Serv Res. 2018;53:1092-1109.\n 27. Shy BD, Loo GT, Lowry T, et al. Bouncing Back Elsewhere: Multilevel \nAnalysis of Return Visits to the Same or a Different Hospital After Initial \nEmergency Department Presentation . Ann Emerg Med. 2018;71:555-\n563.e1.\nPublisher’s Note Springer Nature remains neutral with regard to \njurisdictional claims in published maps and institutional affiliations.\nSpringer Nature or its licensor (e.g. a society or other partner) holds \nexclusive rights to this article under a publishing agreement with \nthe author(s) or other rightsholder(s); author self-archiving of the \naccepted manuscript version of this article is solely governed by the \nterms of such publishing agreement and applicable law.\n746","source_license":"public-domain-us","license_restricted":false}