https://orcid.org/0000-0001-6402-5221
Abstract
The clinical usefulness of historical concentrations of high-sensitivity cardiac troponin T (hs-cTnT) is unknown. This study investigated the ability to rule out myocardial infarction (MI) with the use of historical hs-cTnT concentrations among patients with chest pain in the emergency department (ED).
The derivation cohort consisted of patients presenting with chest pain to nine different EDs (n = 60 071), where we included those with ≥1 hs-cTnT analysed at the index visit and ≥1 hs-cTnT results prior to the visit. We developed an algorithm to rule out MI within 30 days with a pre-specified target negative predictive value (NPV) of ≥99.5%. The performance was then validated in a separate cohort of ED chest pain patients (n = 10 994). A historical hs-cTnT < 12 ng/L and a < 3 ng/L absolute change between the historical and the index visit hs-cTnT had the best performance and ruled out 24 862 (41%) patients in the derivation cohort. In the validation cohort, these criteria identified 4764 (43%) low-risk patients in whom 18 (0.4%) MIs within 30 days occurred, and had an NPV for MI of 99.6% (99.4–99.8), a sensitivity of 96.9% (95.2–.2), and an LR− of 0.11 (0.07–0.14).
Combining a historical hs-cTnT with a single new hs-cTnT may safely rule out MI and thereby reduce the need for serial hs-cTnT measurements in ED patients with chest pain.
Introduction
High-sensitivity cardiac troponin (hs-cTn) assays were introduced into clinical practice approximately 10 years ago in Europe.1 Today, a large proportion of patients who present to the emergency department (ED) with symptoms suggestive of a myocardial infarction (MI) have available information on hs-cTn concentrations from prior ED and/or hospital visits.2 However, the clinical value of historical hs-cTn concentrations is largely unexplored. Several prospectively validated hs-cTn-based algorithms for a safe rule out of MI exist, which typically involve serial measurement of hs-cTn concentrations.3–5 Recent studies indicate that the change in hs-cTnT between historical and admission tests may be clinically useful.2 If historical hs-cTnT levels could be combined with a single hs-cTnT in the ED to safely rule out MI, this could potentially reduce serial hs-cTnT measurements, ED length of stay and help reduce ED crowding.6,7
In this large observational study, we investigated the ability to safely rule out MI using historical hs-cTnT results and the change between historical and new hs-cTnT tests at the ED.
Materials and methods
Study population
Derivation cohort
We identified all visits in patients >35 years of age with a primary complaint of chest pain at 9 different EDs in Stockholm and Gothenburg, Sweden, from 1 May 2012, through 31 December 2016 (n = 186 621; Figure 1) After excluding visits with a final diagnosis of ST-segment elevation myocardial infarction (STEMI), we included all visits that met the following inclusion criteria: (i) ≥1 hs-cTnT analysed at the ED visit and (ii) ≥1 hs-cTnT test at any time before the ED visit (n = 60 071).
Selection of the study population. hs-cTnT, high-sensitivity cardiac troponin T; MI, myocardial infarction; STEMI, ST-segment elevation myocardial infarction.
Validation cohort
The validation cohort consisted of patients included in a multicentre implementation study with registry-based follow-up (ESC-TROP: Effectiveness and Safety of the European Society of Cardiology 0-/1-h Troponin Rule-Out Protocol; NCT03421873), and the methods have been described in detail elsewhere.8 Briefly, all patients presenting to one of five EDs in southern Sweden with a primary complaint of chest pain of non-traumatic origin between the 1st of February and the 30th of November 2017 and between the 1st of February and the 30th of November 2018 were included. In contrast to the derivation cohort, patients in the validation cohort were included only at their first ED visit and not at subsequent visits. Full inclusion and exclusion criteria are provided in the Supplementary material online.
Derivation cohort data sources
In the derivation cohort, patient visit data were identified from the hospitals’ local administrative databases, which contain information regarding all ED visits, including information on presenting complaint. Laboratory data were obtained from each hospital’s IT department. At all sites, including the validation cohort, hs-cTnT concentrations were measured using the Elecsys 2010 system (Roche Diagnostics GmbH, Mannheim, Germany) The assay has a limit of blank of 3 ng/L, a recommended limit of detection of 5 ng/L, a coefficient of variation of <10% at 13 ng/L, and a 99th-percentile cut-off value of 14 ng/L.1 A cut-off at the 99th percentile value for the hs-cTnT assay of 14 ng/L was used at all sites throughout the study period. In both the validation and derivation cohort, hs-cTnT samples were obtained during routine clinical practice where immediate blood sampling at ED presentation was routinely performed at all sites in patients in whom a 0-h hs-cTnT analysis was motivated. The timing of additional blood samples was decided by the treating physician in most centres.
Data on comorbidities, current medications, and outcomes were obtained from the National Patient Register (NPR) the Cause of Death Register and the Prescribed Drug Register.10,11 Comorbidities were defined as all discharge diagnoses registered in the NPR from hospital contacts preceding the index visit. Ongoing use of medication was defined as ≥2 dispensed prescriptions during the year before the index date in the derivation cohort.
Data collection methods for the validation cohort are provided in the Supplementary material online.
The study complied with the guidelines of the Declaration of Helsinki and was approved by the regional ethics review board in Stockholm.
Outcomes
The primary outcome was MI within 30 days including the index visit. Data on the primary outcome in the derivation cohort were based on discharge diagnoses coded in the NPR (I21 or I22 registered in the primary position) according to the 10th revision of the international classification of disease (ICD-10). The secondary outcome was all-cause mortality within 30 days. Data on deaths and dates of death were obtained from the Swedish population register providing complete nationwide coverage. Cardiovascular death was defined as a cause of death in the I-chapter, or R960-R961, in ICD-10.
The validation cohort used a registry-based follow-up where the diagnosis of MI was based on a discharge diagnosis obtained from the SWEDEHEART (Swedish Web-system for Enhancement and Development of Evidence-based care in Heart disease Evaluated According to Recommended Therapies) registry, a nationwide registry that collects information on all patients admitted to a coronary care unit in Sweden.11 Both NPR and SWEDEHEART are national registries with excellent nationwide coverage and accuracy for MI events.9,11 To capture outcome data on patients not admitted to a coronary care unit, diagnoses were also obtained from the regional electronic health records from all hospitals in the region. As to not misclassify patients potentially missed during the index visit, events in patients discharged from the ED were adjudicated by two independent cardiologists in accordance with the fourth universal definition of MI.12 In case of disagreement, cases were reviewed by an adjudication committee and resolved by a majority vote.
Algorithm development and evaluation
The 0 h hs-cTnT was defined as the first hs-cTnT result at the index visit, and the historical hs-cTnT as the most recent hs-cTnT, analysed at any health care visit >7 days before the 0 h hs-cTnT. Historical hs-cTnT levels <7 days before the index test was not used so as to avoid potential associations with events at the index visit.
The delta hs-cTnT change was defined as the absolute change in hs-cTnT concentration between the historical hs-cTnT and the 0 h hs-cTnT. Hs-cTnT concentrations <5 ng/L were not reported as absolute values at hospitals in the derivation or the validation cohorts and were therefore assigned a level of 4 ng/L to calculate delta values.
A historical hs-cTnT-based algorithm for rule out of MI was developed in the derivation cohort, which incorporated the historical hs-cTnT results and the delta hs-cTnT change between the historical and index test results. We then identified rule-out thresholds for combinations of historical hs-cTnT results and delta hs-cTnT change to achieve a negative predictive value (NPV) for MI of ≥99.5%, a threshold commonly used for safe rule out in chest pain studies.13 The diagnostic accuracy of this algorithm was then evaluated in the validation cohort.
In the derivation cohort, we compared the performance of the historical hs-cTnT-based algorithm with a European Society of Cardiology (ESC)-based algorithm in patients who had a second hs-cTnT measured between 45 min and 3.5 h after the 0 h hs-cTnT. The rule-out criteria for the ESC-based algorithm were defined as a 0 h hs-cTnT level of <12 ng/L and a delta hs-cTnT between the 0 h hs-cTnT and a second hs-cTnT of <3 ng/L, in accordance with the thresholds for these measures in the ESC 0-/1-h algorithm for triage toward rule out of MI.3
Statistical analysis
We calculated absolute risk, sensitivity, NPV, and negative likelihood ratio (LR−s) for MI and all-cause mortality within 30 days of the index visit in patients ruled out by the historical hs-cTnT-based algorithm in both cohorts. A secondary analysis was performed to evaluate the performance for rule out of MI during the index visit. In an additional analysis, we categorized patients according to the time of the historical hs-cTnT test in relation to the 0 h hs-cTnT; <1 or >1 year before.
Results
Study population
There were 60 071 patient visits included in the derivation cohort, and 10 994 patients in the validation cohort (Figure 1, Table 1). A total of 3581 MIs occurred, with a similar incidence in the two cohorts (5.0% and 5.3%, respectively; Table 1). Patients with MI were older and had more cardiovascular comorbidities than in patients without MI. Both historical hs-cTnT and 0 h hs-cTnT levels were higher in MI patients than in patients without MI.
Baseline characteristics
| Derivation cohort | Validation cohort | |||||
|---|---|---|---|---|---|---|
| All visits | MI | No MI | All visits | MI | No MI | |
| Number of visits | 60 071 (100) | 2999 (5.0) | 57 072 (95) | 10 994 (100) | 582 (5.3) | 10 412 (95) |
| Age (years) | 68 (56–79) | 76 (67–86) | 68 (56–79) | 68 (54.0–79.0) | 75 (67–85) | 68 (53–78) |
| Women | 28 476 (47) | 1164 (2) | 27 312 (46) | 5436 (49) | 242 (42) | 5194 (50) |
| Comorbidities | ||||||
| Prior stroke | 6012 (10) | 470 (16) | 5542 (10) | 1514 (14) | 111 (19) | 1403 (14) |
| Prior MIa | 17 223 (29) | 1692 (56) | 15 531 (27) | 2449 (22) | 267 (46) | 2182 (21) |
| COPD | 7463 (12) | 403 (13) | 7060 (12) | 938 (8.5) | 49 (8.4) | 889 (8.5) |
| Heart failure | 8576 (14) | 583 (19) | 7993 (14) | 1563 (14) | 125 (22) | 1438 (14) |
| Diabetes | 10 263 (17) | 939 (31) | 9324 (16) | 2430 (22) | 210 (36) | 2220 (21) |
| Chronic kidney disease (eGFR < 60 mL/min/1.73 m2) | 4069 (6.8) | 464 (16) | 3605 (6.3) | 770 (7.0) | 84 (14) | 686 (6.6) |
| Current medications | ||||||
| Aspirin | 23 571 (39) | 1837 (61) | 21 734 (38) | 2889 (26) | 258 (44) | 2631 (25) |
| P2Y12 inhibitors | 8269 (14) | 746 (25) | 7523 (13) | 777 (7.1) | 84 (14) | 693 (6.7) |
| Beta-blockers | 33 037 (55) | 2079 (69) | 30 958 (54) | 2399 (22) | 171 (29) | 2228 (21) |
| ACE/ARB | 28 918 (48) | 1830 (61) | 27 088 (48) | 4012 (37) | 291 (50) | 3721 (36) |
| Statins | 24 664 (41) | 1662 (55) | 23 002 (40) | 3927 (36) | 288 (50) | 3639 (35) |
| NOACb | 2774 (4.6) | 112 (3.7) | 2662 (4.7) | 1072 (9.8) | 50 (8.6) | 1022 (9.8) |
| Warfarin | 8133 (14) | 363 (12) | 7770 (14) | 335 (3.0) | 24 (4.1) | 311 (3.0) |
| OACc | 10 677 (18) | 461 (15) | 10 216 (18) | 1407 (13) | 74 (13) | 1333 (13) |
| Laboratory data | ||||||
| 0 h hs-cTnT level (ng/L) | 10 (5–20) | 44 (21–108) | 9 (5–19) | 9 (5–20) | 50 (23–124) | 9 (4–18) |
| Historical hs-cTnT level (ng/L) | 10 (5–23) | 25 (12–69) | 10 (5–21) | 9 (4.–21) | 20 (10–64) | 9 (4–20) |
| Delta hs-cTnT changed (ng/L) | 2 (1–7) | 33 (9–145) | 2 (1–6) | 2 (1–8) | 42 (11–162) | 2 (0–7) |
| Time from historical hs-cTnT to 0 h hs-cTnT concentration (days) | 147 (40–384) | 152 (41–407) | 146 (40–383) | 465 (165–894) | 503 (147–942) | 463 (166–892) |
| Derivation cohort | Validation cohort | |||||
|---|---|---|---|---|---|---|
| All visits | MI | No MI | All visits | MI | No MI | |
| Number of visits | 60 071 (100) | 2999 (5.0) | 57 072 (95) | 10 994 (100) | 582 (5.3) | 10 412 (95) |
| Age (years) | 68 (56–79) | 76 (67–86) | 68 (56–79) | 68 (54.0–79.0) | 75 (67–85) | 68 (53–78) |
| Women | 28 476 (47) | 1164 (2) | 27 312 (46) | 5436 (49) | 242 (42) | 5194 (50) |
| Comorbidities | ||||||
| Prior stroke | 6012 (10) | 470 (16) | 5542 (10) | 1514 (14) | 111 (19) | 1403 (14) |
| Prior MIa | 17 223 (29) | 1692 (56) | 15 531 (27) | 2449 (22) | 267 (46) | 2182 (21) |
| COPD | 7463 (12) | 403 (13) | 7060 (12) | 938 (8.5) | 49 (8.4) | 889 (8.5) |
| Heart failure | 8576 (14) | 583 (19) | 7993 (14) | 1563 (14) | 125 (22) | 1438 (14) |
| Diabetes | 10 263 (17) | 939 (31) | 9324 (16) | 2430 (22) | 210 (36) | 2220 (21) |
| Chronic kidney disease (eGFR < 60 mL/min/1.73 m2) | 4069 (6.8) | 464 (16) | 3605 (6.3) | 770 (7.0) | 84 (14) | 686 (6.6) |
| Current medications | ||||||
| Aspirin | 23 571 (39) | 1837 (61) | 21 734 (38) | 2889 (26) | 258 (44) | 2631 (25) |
| P2Y12 inhibitors | 8269 (14) | 746 (25) | 7523 (13) | 777 (7.1) | 84 (14) | 693 (6.7) |
| Beta-blockers | 33 037 (55) | 2079 (69) | 30 958 (54) | 2399 (22) | 171 (29) | 2228 (21) |
| ACE/ARB | 28 918 (48) | 1830 (61) | 27 088 (48) | 4012 (37) | 291 (50) | 3721 (36) |
| Statins | 24 664 (41) | 1662 (55) | 23 002 (40) | 3927 (36) | 288 (50) | 3639 (35) |
| NOACb | 2774 (4.6) | 112 (3.7) | 2662 (4.7) | 1072 (9.8) | 50 (8.6) | 1022 (9.8) |
| Warfarin | 8133 (14) | 363 (12) | 7770 (14) | 335 (3.0) | 24 (4.1) | 311 (3.0) |
| OACc | 10 677 (18) | 461 (15) | 10 216 (18) | 1407 (13) | 74 (13) | 1333 (13) |
| Laboratory data | ||||||
| 0 h hs-cTnT level (ng/L) | 10 (5–20) | 44 (21–108) | 9 (5–19) | 9 (5–20) | 50 (23–124) | 9 (4–18) |
| Historical hs-cTnT level (ng/L) | 10 (5–23) | 25 (12–69) | 10 (5–21) | 9 (4.–21) | 20 (10–64) | 9 (4–20) |
| Delta hs-cTnT changed (ng/L) | 2 (1–7) | 33 (9–145) | 2 (1–6) | 2 (1–8) | 42 (11–162) | 2 (0–7) |
| Time from historical hs-cTnT to 0 h hs-cTnT concentration (days) | 147 (40–384) | 152 (41–407) | 146 (40–383) | 465 (165–894) | 503 (147–942) | 463 (166–892) |
Data are presented as n (%), mean ± SD, or median with IQR.
Includes only MI as a primary diagnosis.
Includes treatment with Apixaban, Rivaroxaban, Dabigatran, or Epixaban.
Includes treatment with NOAC or warfarin.
The absolute change of hs-cTnT between the historical hs-cTnT and the 0 h cTnT concentration.
ACE/ARB, angiotensin converting enzyme inhibitor/angiotensin receptor blocker; COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; hs-cTnT, high-sensitivity cardiac troponin T; MI, myocardial infarction; NOAC, new oral anticoagulants; OAC, oral anticoagulants.
Baseline characteristics
| Derivation cohort | Validation cohort | |||||
|---|---|---|---|---|---|---|
| All visits | MI | No MI | All visits | MI | No MI | |
| Number of visits | 60 071 (100) | 2999 (5.0) | 57 072 (95) | 10 994 (100) | 582 (5.3) | 10 412 (95) |
| Age (years) | 68 (56–79) | 76 (67–86) | 68 (56–79) | 68 (54.0–79.0) | 75 (67–85) | 68 (53–78) |
| Women | 28 476 (47) | 1164 (2) | 27 312 (46) | 5436 (49) | 242 (42) | 5194 (50) |
| Comorbidities | ||||||
| Prior stroke | 6012 (10) | 470 (16) | 5542 (10) | 1514 (14) | 111 (19) | 1403 (14) |
| Prior MIa | 17 223 (29) | 1692 (56) | 15 531 (27) | 2449 (22) | 267 (46) | 2182 (21) |
| COPD | 7463 (12) | 403 (13) | 7060 (12) | 938 (8.5) | 49 (8.4) | 889 (8.5) |
| Heart failure | 8576 (14) | 583 (19) | 7993 (14) | 1563 (14) | 125 (22) | 1438 (14) |
| Diabetes | 10 263 (17) | 939 (31) | 9324 (16) | 2430 (22) | 210 (36) | 2220 (21) |
| Chronic kidney disease (eGFR < 60 mL/min/1.73 m2) | 4069 (6.8) | 464 (16) | 3605 (6.3) | 770 (7.0) | 84 (14) | 686 (6.6) |
| Current medications | ||||||
| Aspirin | 23 571 (39) | 1837 (61) | 21 734 (38) | 2889 (26) | 258 (44) | 2631 (25) |
| P2Y12 inhibitors | 8269 (14) | 746 (25) | 7523 (13) | 777 (7.1) | 84 (14) | 693 (6.7) |
| Beta-blockers | 33 037 (55) | 2079 (69) | 30 958 (54) | 2399 (22) | 171 (29) | 2228 (21) |
| ACE/ARB | 28 918 (48) | 1830 (61) | 27 088 (48) | 4012 (37) | 291 (50) | 3721 (36) |
| Statins | 24 664 (41) | 1662 (55) | 23 002 (40) | 3927 (36) | 288 (50) | 3639 (35) |
| NOACb | 2774 (4.6) | 112 (3.7) | 2662 (4.7) | 1072 (9.8) | 50 (8.6) | 1022 (9.8) |
| Warfarin | 8133 (14) | 363 (12) | 7770 (14) | 335 (3.0) | 24 (4.1) | 311 (3.0) |
| OACc | 10 677 (18) | 461 (15) | 10 216 (18) | 1407 (13) | 74 (13) | 1333 (13) |
| Laboratory data | ||||||
| 0 h hs-cTnT level (ng/L) | 10 (5–20) | 44 (21–108) | 9 (5–19) | 9 (5–20) | 50 (23–124) | 9 (4–18) |
| Historical hs-cTnT level (ng/L) | 10 (5–23) | 25 (12–69) | 10 (5–21) | 9 (4.–21) | 20 (10–64) | 9 (4–20) |
| Delta hs-cTnT changed (ng/L) | 2 (1–7) | 33 (9–145) | 2 (1–6) | 2 (1–8) | 42 (11–162) | 2 (0–7) |
| Time from historical hs-cTnT to 0 h hs-cTnT concentration (days) | 147 (40–384) | 152 (41–407) | 146 (40–383) | 465 (165–894) | 503 (147–942) | 463 (166–892) |
| Derivation cohort | Validation cohort | |||||
|---|---|---|---|---|---|---|
| All visits | MI | No MI | All visits | MI | No MI | |
| Number of visits | 60 071 (100) | 2999 (5.0) | 57 072 (95) | 10 994 (100) | 582 (5.3) | 10 412 (95) |
| Age (years) | 68 (56–79) | 76 (67–86) | 68 (56–79) | 68 (54.0–79.0) | 75 (67–85) | 68 (53–78) |
| Women | 28 476 (47) | 1164 (2) | 27 312 (46) | 5436 (49) | 242 (42) | 5194 (50) |
| Comorbidities | ||||||
| Prior stroke | 6012 (10) | 470 (16) | 5542 (10) | 1514 (14) | 111 (19) | 1403 (14) |
| Prior MIa | 17 223 (29) | 1692 (56) | 15 531 (27) | 2449 (22) | 267 (46) | 2182 (21) |
| COPD | 7463 (12) | 403 (13) | 7060 (12) | 938 (8.5) | 49 (8.4) | 889 (8.5) |
| Heart failure | 8576 (14) | 583 (19) | 7993 (14) | 1563 (14) | 125 (22) | 1438 (14) |
| Diabetes | 10 263 (17) | 939 (31) | 9324 (16) | 2430 (22) | 210 (36) | 2220 (21) |
| Chronic kidney disease (eGFR < 60 mL/min/1.73 m2) | 4069 (6.8) | 464 (16) | 3605 (6.3) | 770 (7.0) | 84 (14) | 686 (6.6) |
| Current medications | ||||||
| Aspirin | 23 571 (39) | 1837 (61) | 21 734 (38) | 2889 (26) | 258 (44) | 2631 (25) |
| P2Y12 inhibitors | 8269 (14) | 746 (25) | 7523 (13) | 777 (7.1) | 84 (14) | 693 (6.7) |
| Beta-blockers | 33 037 (55) | 2079 (69) | 30 958 (54) | 2399 (22) | 171 (29) | 2228 (21) |
| ACE/ARB | 28 918 (48) | 1830 (61) | 27 088 (48) | 4012 (37) | 291 (50) | 3721 (36) |
| Statins | 24 664 (41) | 1662 (55) | 23 002 (40) | 3927 (36) | 288 (50) | 3639 (35) |
| NOACb | 2774 (4.6) | 112 (3.7) | 2662 (4.7) | 1072 (9.8) | 50 (8.6) | 1022 (9.8) |
| Warfarin | 8133 (14) | 363 (12) | 7770 (14) | 335 (3.0) | 24 (4.1) | 311 (3.0) |
| OACc | 10 677 (18) | 461 (15) | 10 216 (18) | 1407 (13) | 74 (13) | 1333 (13) |
| Laboratory data | ||||||
| 0 h hs-cTnT level (ng/L) | 10 (5–20) | 44 (21–108) | 9 (5–19) | 9 (5–20) | 50 (23–124) | 9 (4–18) |
| Historical hs-cTnT level (ng/L) | 10 (5–23) | 25 (12–69) | 10 (5–21) | 9 (4.–21) | 20 (10–64) | 9 (4–20) |
| Delta hs-cTnT changed (ng/L) | 2 (1–7) | 33 (9–145) | 2 (1–6) | 2 (1–8) | 42 (11–162) | 2 (0–7) |
| Time from historical hs-cTnT to 0 h hs-cTnT concentration (days) | 147 (40–384) | 152 (41–407) | 146 (40–383) | 465 (165–894) | 503 (147–942) | 463 (166–892) |
Data are presented as n (%), mean ± SD, or median with IQR.
Includes only MI as a primary diagnosis.
Includes treatment with Apixaban, Rivaroxaban, Dabigatran, or Epixaban.
Includes treatment with NOAC or warfarin.
The absolute change of hs-cTnT between the historical hs-cTnT and the 0 h cTnT concentration.
ACE/ARB, angiotensin converting enzyme inhibitor/angiotensin receptor blocker; COPD, chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; hs-cTnT, high-sensitivity cardiac troponin T; MI, myocardial infarction; NOAC, new oral anticoagulants; OAC, oral anticoagulants.
Diagnostic performance of the historical high-sensitivity cardiac troponin T-based algorithm in the derivation cohort
The diagnostic performance of different thresholds for historical hs-cTnT results and hs-cTnT change to rule out MI with an NPV of ≥99.5% are illustrated in Figure 2.
Diagnostic performance of combinations of historical hs-cTnT and changes up to index visit hs-cTnT for rule out of 30-day myocardial infarction. CI, confidence interval; hs-cTnT: high-sensitivity cardiac troponin T; MI, myocardial infarction.
Among the combinations of thresholds, a historical hs-cTnT < 12 ng/L and a <3 ng/L absolute change between the historical hs-cTnT and the index visit hs-cTnT was chosen as the best combination. This was due to both optimal safety and efficacy, with the additional advantage of using the same hs-cTnT values as recommended in the ESC 0-/1-h algorithm for ruling out MI,3 which facilitates clinical use.
Patients triaged toward rule out according to the final historical hs-cTnT-based algorithm were younger and had fewer cardiovascular comorbidities and were less frequently treated with cardiovascular medication compared with patients not ruled out (see Supplementary material online, Table S1) Within the derivation cohort, the most common discharge diagnoses in ruled-out patient who were admitted to hospital were symptom diagnoses (ICD-codes R00–R09; 38%; see Supplementary material online, Table S2).
The application of the historical hs-cTnT-based algorithm in the derivation cohort triaged 24 862 (41%) patients toward rule out, in whom 117 (0.5%) MIs occurred within 30 days (Table 2). This resulted in an NPV for 30-day MI of 99.5% (99.4–99.6), an LR− of 0.11 (0.07–0.14), and a sensitivity of 96.1% (95.4–96.8). Corresponding numbers for rule out of index visit MI, which occurred in 77 (0.3%) of patients, were 99.7% (99.6–99.8), 0.07 (0.05–0.09), and 97.0 (96.3–97.6), respectively (see Supplementary material online, Table S3).
Diagnostic performance of a historical hs-cTnT-based algorithm for rule out of acute myocardial infarction and all-cause mortality within 30 days
| Derivation cohort | Validation cohort | |
|---|---|---|
| Number of visits | 60 071 (85) | 10 994 (15) |
| MI ≤30 days after the index visit | 2999 (5.0) | 582 (5.3) |
| Rule out | ||
| Number of patients ruled out | 24 862 (41) | 4764 (43) |
| Myocardial infarction | ||
| Number of events | 117 (0.5) | 18 (0.4) |
| 30-day risk of MI (95% CI) | 0.5 (0.4–0.6) | 0.4 (0.2–0.6) |
| NPV (95% CI) | 99.5 (99.4–99.6) | 99.6 (99.4–99.8) |
| LR− (95% CI) | 0.09 (0.07–0.11) | 0.07 (0.70–0.11) |
| Sensitivity (95% CI) | 96.1 (95.4–96.8) | 96.9 (95.2–98.2) |
| All-cause mortality | ||
| Number of deaths | 38 (0.2) | 2 (0.04) |
| 30-day risk of death (95% CI) | 0.2 (0.1–0.2) | 0.04 (0.005–0.15) |
| NPV (95% CI) | 99.9 (99.8–99.9) | 99.9 (99.8–100.0) |
| LR− (95% CI) | 0.11 (0.07–0.14) | 0.26 (0.08–0.91) |
| Sensitivity (95% CI) | 95.6 (94.2–97.0) | 98.8 (95.8–99.9) |
| Cardiovascular mortality | ||
| Number of cardiovascular deaths | 14 (0.06) | 0 (0) |
| 30-day risk of cardiovascular death (95% CI) | 0.06 (0.03–0.09) | n/a |
| NPV (95% CI) | 99.9 (99.9–100.0) | n/a |
| LR− (95% CI) | 0.07 (0.03–0.11) | n/a |
| Sensitivity (95% CI) | 97.1 (95.6–98.6) | n/a |
| Derivation cohort | Validation cohort | |
|---|---|---|
| Number of visits | 60 071 (85) | 10 994 (15) |
| MI ≤30 days after the index visit | 2999 (5.0) | 582 (5.3) |
| Rule out | ||
| Number of patients ruled out | 24 862 (41) | 4764 (43) |
| Myocardial infarction | ||
| Number of events | 117 (0.5) | 18 (0.4) |
| 30-day risk of MI (95% CI) | 0.5 (0.4–0.6) | 0.4 (0.2–0.6) |
| NPV (95% CI) | 99.5 (99.4–99.6) | 99.6 (99.4–99.8) |
| LR− (95% CI) | 0.09 (0.07–0.11) | 0.07 (0.70–0.11) |
| Sensitivity (95% CI) | 96.1 (95.4–96.8) | 96.9 (95.2–98.2) |
| All-cause mortality | ||
| Number of deaths | 38 (0.2) | 2 (0.04) |
| 30-day risk of death (95% CI) | 0.2 (0.1–0.2) | 0.04 (0.005–0.15) |
| NPV (95% CI) | 99.9 (99.8–99.9) | 99.9 (99.8–100.0) |
| LR− (95% CI) | 0.11 (0.07–0.14) | 0.26 (0.08–0.91) |
| Sensitivity (95% CI) | 95.6 (94.2–97.0) | 98.8 (95.8–99.9) |
| Cardiovascular mortality | ||
| Number of cardiovascular deaths | 14 (0.06) | 0 (0) |
| 30-day risk of cardiovascular death (95% CI) | 0.06 (0.03–0.09) | n/a |
| NPV (95% CI) | 99.9 (99.9–100.0) | n/a |
| LR− (95% CI) | 0.07 (0.03–0.11) | n/a |
| Sensitivity (95% CI) | 97.1 (95.6–98.6) | n/a |
Data are presented as n (%).
CI, confidence interval; LR−, negative likelihood ratio; MI, myocardial infarction; NPV, negative predictive value.
Diagnostic performance of a historical hs-cTnT-based algorithm for rule out of acute myocardial infarction and all-cause mortality within 30 days
| Derivation cohort | Validation cohort | |
|---|---|---|
| Number of visits | 60 071 (85) | 10 994 (15) |
| MI ≤30 days after the index visit | 2999 (5.0) | 582 (5.3) |
| Rule out | ||
| Number of patients ruled out | 24 862 (41) | 4764 (43) |
| Myocardial infarction | ||
| Number of events | 117 (0.5) | 18 (0.4) |
| 30-day risk of MI (95% CI) | 0.5 (0.4–0.6) | 0.4 (0.2–0.6) |
| NPV (95% CI) | 99.5 (99.4–99.6) | 99.6 (99.4–99.8) |
| LR− (95% CI) | 0.09 (0.07–0.11) | 0.07 (0.70–0.11) |
| Sensitivity (95% CI) | 96.1 (95.4–96.8) | 96.9 (95.2–98.2) |
| All-cause mortality | ||
| Number of deaths | 38 (0.2) | 2 (0.04) |
| 30-day risk of death (95% CI) | 0.2 (0.1–0.2) | 0.04 (0.005–0.15) |
| NPV (95% CI) | 99.9 (99.8–99.9) | 99.9 (99.8–100.0) |
| LR− (95% CI) | 0.11 (0.07–0.14) | 0.26 (0.08–0.91) |
| Sensitivity (95% CI) | 95.6 (94.2–97.0) | 98.8 (95.8–99.9) |
| Cardiovascular mortality | ||
| Number of cardiovascular deaths | 14 (0.06) | 0 (0) |
| 30-day risk of cardiovascular death (95% CI) | 0.06 (0.03–0.09) | n/a |
| NPV (95% CI) | 99.9 (99.9–100.0) | n/a |
| LR− (95% CI) | 0.07 (0.03–0.11) | n/a |
| Sensitivity (95% CI) | 97.1 (95.6–98.6) | n/a |
| Derivation cohort | Validation cohort | |
|---|---|---|
| Number of visits | 60 071 (85) | 10 994 (15) |
| MI ≤30 days after the index visit | 2999 (5.0) | 582 (5.3) |
| Rule out | ||
| Number of patients ruled out | 24 862 (41) | 4764 (43) |
| Myocardial infarction | ||
| Number of events | 117 (0.5) | 18 (0.4) |
| 30-day risk of MI (95% CI) | 0.5 (0.4–0.6) | 0.4 (0.2–0.6) |
| NPV (95% CI) | 99.5 (99.4–99.6) | 99.6 (99.4–99.8) |
| LR− (95% CI) | 0.09 (0.07–0.11) | 0.07 (0.70–0.11) |
| Sensitivity (95% CI) | 96.1 (95.4–96.8) | 96.9 (95.2–98.2) |
| All-cause mortality | ||
| Number of deaths | 38 (0.2) | 2 (0.04) |
| 30-day risk of death (95% CI) | 0.2 (0.1–0.2) | 0.04 (0.005–0.15) |
| NPV (95% CI) | 99.9 (99.8–99.9) | 99.9 (99.8–100.0) |
| LR− (95% CI) | 0.11 (0.07–0.14) | 0.26 (0.08–0.91) |
| Sensitivity (95% CI) | 95.6 (94.2–97.0) | 98.8 (95.8–99.9) |
| Cardiovascular mortality | ||
| Number of cardiovascular deaths | 14 (0.06) | 0 (0) |
| 30-day risk of cardiovascular death (95% CI) | 0.06 (0.03–0.09) | n/a |
| NPV (95% CI) | 99.9 (99.9–100.0) | n/a |
| LR− (95% CI) | 0.07 (0.03–0.11) | n/a |
| Sensitivity (95% CI) | 97.1 (95.6–98.6) | n/a |
Data are presented as n (%).
CI, confidence interval; LR−, negative likelihood ratio; MI, myocardial infarction; NPV, negative predictive value.
A total of 38 (0.2%) ruled-out patients died within 30 days, which gave a sensitivity for the death of 95.6% (94.2–97.0) and an NPV of 99.9% (99.8–99.9). Only 14 (0.06%) cardiovascular deaths occurred in ruled-out patients, which resulted in a sensitivity and NPV of 97.1 (95.6–98.6) and 99.9 (99.9–100.0), respectively (Table 2). The diagnostic performance for rule out of MI and all-cause mortality was similar when the analysis was stratified by the time from the historical hs-cTnT to the 0 h hs-cTnT (see Supplementary material online, Table S4).
More than one-third (37%) of patients triaged toward rule out had at least two hs-cTnT analysed at the index visit, in whom the majority (61%) had a low (<14 ng/L) but detectable 0 h hs-cTnT ≥ 5 ng/L (see Supplementary material online, Table S5). Correspondingly, 57% of patients in the validation cohort in whom a second hs-cTnT analysis was performed had a low but measurable 0-h hs-cTnT concentration.
In patients who had a second hs-cTnT measured at the ED visit within 3.5 h from the 0 h hs-cTnT (Table 3), the historical hs-cTnT-based algorithm ruled out 5912 (35%) patients with a 30-day MI risk of 0.5% (0.3–0.7), and ESC-based algorithm ruled out 7819 (46%) patients with a 30-day MI risk 0.4% (0.3–0.6). The majority (71%) of patients ruled out by the ESC-based algorithm was also ruled out by the historical hs-cTnT-based algorithm.
Comparison of diagnostic performance of a historical high-sensitivity cardiac troponin T-based algorithm for rule out of myocardial infarction with the modified ESC 0/1-h algorithm
| Historical hs-cTnT-based algorithm | ||||
|---|---|---|---|---|
| Modified ESC 0/1-h algorithma | Rule out | Not rule out | Total | |
| Rule out | MI ≤30 days | 11/5559 | 21/2260 | 32/7819 |
| 30-day MI risk (95% CI) | 0.2 (0.1–0.4) | 0.9 (0.6–1.4) | 0.4 (0.3–0.6) | |
| Not rule out | MI ≤30 days | 19/353 | 1074/8876 (12) | 1093/9229 |
| 30-day MI risk (95% CI) | 5.4 (3.4–8.4) | 12.1 (11.4–12.8) | 11.8 (11.2–12.5) | |
| Total | MI ≤30 days | 30/5912 | 1095/11 136 | 1125/17 048 |
| 30-day MI risk (95% CI) | 0.5 (0.3–0.7) | 9.8 (9.3–10.4) | 6.6 (6.2–7.0) | |
| Historical hs-cTnT-based algorithm | ||||
|---|---|---|---|---|
| Modified ESC 0/1-h algorithma | Rule out | Not rule out | Total | |
| Rule out | MI ≤30 days | 11/5559 | 21/2260 | 32/7819 |
| 30-day MI risk (95% CI) | 0.2 (0.1–0.4) | 0.9 (0.6–1.4) | 0.4 (0.3–0.6) | |
| Not rule out | MI ≤30 days | 19/353 | 1074/8876 (12) | 1093/9229 |
| 30-day MI risk (95% CI) | 5.4 (3.4–8.4) | 12.1 (11.4–12.8) | 11.8 (11.2–12.5) | |
| Total | MI ≤30 days | 30/5912 | 1095/11 136 | 1125/17 048 |
| 30-day MI risk (95% CI) | 0.5 (0.3–0.7) | 9.8 (9.3–10.4) | 6.6 (6.2–7.0) | |
0 h hs-cTnT <12 ng/L and a delta hs-cTnT change between the 0 h hs-cTnT and the second hs-cTnT concentration of <3 ng/L.
Notes: Only patients who had a second hs-cTnT concentration measured at the ED visit within 3.5 h from the 0 h hs-cTnT measurement were included in the analysis.
CI, confidence interval; ESC, European Society of Cardiology; hs-cTnT, high-sensitivity cardiac troponin T; MI, myocardial infarction.
Comparison of diagnostic performance of a historical high-sensitivity cardiac troponin T-based algorithm for rule out of myocardial infarction with the modified ESC 0/1-h algorithm
| Historical hs-cTnT-based algorithm | ||||
|---|---|---|---|---|
| Modified ESC 0/1-h algorithma | Rule out | Not rule out | Total | |
| Rule out | MI ≤30 days | 11/5559 | 21/2260 | 32/7819 |
| 30-day MI risk (95% CI) | 0.2 (0.1–0.4) | 0.9 (0.6–1.4) | 0.4 (0.3–0.6) | |
| Not rule out | MI ≤30 days | 19/353 | 1074/8876 (12) | 1093/9229 |
| 30-day MI risk (95% CI) | 5.4 (3.4–8.4) | 12.1 (11.4–12.8) | 11.8 (11.2–12.5) | |
| Total | MI ≤30 days | 30/5912 | 1095/11 136 | 1125/17 048 |
| 30-day MI risk (95% CI) | 0.5 (0.3–0.7) | 9.8 (9.3–10.4) | 6.6 (6.2–7.0) | |
| Historical hs-cTnT-based algorithm | ||||
|---|---|---|---|---|
| Modified ESC 0/1-h algorithma | Rule out | Not rule out | Total | |
| Rule out | MI ≤30 days | 11/5559 | 21/2260 | 32/7819 |
| 30-day MI risk (95% CI) | 0.2 (0.1–0.4) | 0.9 (0.6–1.4) | 0.4 (0.3–0.6) | |
| Not rule out | MI ≤30 days | 19/353 | 1074/8876 (12) | 1093/9229 |
| 30-day MI risk (95% CI) | 5.4 (3.4–8.4) | 12.1 (11.4–12.8) | 11.8 (11.2–12.5) | |
| Total | MI ≤30 days | 30/5912 | 1095/11 136 | 1125/17 048 |
| 30-day MI risk (95% CI) | 0.5 (0.3–0.7) | 9.8 (9.3–10.4) | 6.6 (6.2–7.0) | |
0 h hs-cTnT <12 ng/L and a delta hs-cTnT change between the 0 h hs-cTnT and the second hs-cTnT concentration of <3 ng/L.
Notes: Only patients who had a second hs-cTnT concentration measured at the ED visit within 3.5 h from the 0 h hs-cTnT measurement were included in the analysis.
CI, confidence interval; ESC, European Society of Cardiology; hs-cTnT, high-sensitivity cardiac troponin T; MI, myocardial infarction.
Evaluation of the historical high-sensitivity cardiac troponin T-based algorithm in the validation cohort
The historical hs-cTnT-based algorithm ruled out a total of 4764 (43%) patients of whom 18 (0.4%) patients had an MI within 30 days (Table 2). The diagnostic performance for rule out was similar as in the derivation cohort, with a sensitivity of 96.9% (95.2–98.2), NPV of 99.6% (99.4–99.8), and LR− of 0.07 (0.70–0.11; Table 2, see Supplementary material online, Table S4). Altogether, 14 (0.3%) MIs occurred during the index visit among ruled-out patients, and the algorithm had a corresponding sensitivity, NPV and LR− for this outcome of 97.4% (95.7–98.6), 99.7% (99.5–99.8), and 0.06 (0.03–0.10), respectively (see Supplementary material online, Table S3). Only two deaths occurred in ruled-out patients, of which both were non-cardiovascular (Table 2).
Discussion
Using a large cohort of ED patients with a primary complaint of chest pain, we found that an algorithm combining historical hs-cTnT levels with the change up to the initial hs-cTnT at presentation would allow for a safe rule out of MI or death within 30 days. The performance of the algorithm was similar when validated in an external patient cohort. Our findings suggest that historical hs-cTnT levels may be clinically useful for rapid and safe rule out of MI and could spare repeated hs-cTnT measurements in a considerable proportion of ED patients.
hs-cTn assays can detect cTn levels approximately ten times lower than older generations of assays, which has improved the rule out accuracy of MI,14,15 and allowed for the development of strategies to identify low-risk patients suitable for early discharge.5,16,17 Current guidelines recommend algorithms that identify low-risk patients in whom discharge may be appropriate already after a second hs-cTnT 1–3 h from the first hs-cTnT at ED presentation.3 An algorithm able to safely rule out MI already by the first hs-cTnT will allow even earlier management decisions, even though a two-sample strategy may rule out somewhat more patients. In clinical practice, a large proportion of ED chest pain patients have had hs-cTnT tests at prior visits (40% in the present study), and this proportion will likely increase in the future. Knowledge about the clinical use of these historical hs-cTnT results has so far been limited. A recent study on low-risk patients with a first ED hs-cTnT <12 ng/L suggest that a combination of a historical hs-cTnT <12 ng/L and a change up to the first ED hs-cTnT of <3 ng/L may be able to safely rule out MI.2 In the present study, we confirm and extend these findings to include all-comers in a population with an overall MI incidence of 5%. Our historical hs-cTnT-based algorithm had an NPV ≥99.5% for both MI and all-cause and cardiovascular mortality within 30 days, as well as high-sensitivity and low LR−.
The NPV for MI of ≥99.5% means that the algorithm identified a group of patients with a post-test MI rate <0.5%, which seems to be an acceptable level of risk to most ED clinicians.18 Although the sensitivity of the algorithm was less than 99%, it was comparable to other strategies, such as the approach of using a hs-cTnI <5 ng/L at ED presentation which has been shown to have a sensitivity for MI of 94.5–97.1% in different populations.5,19 Prospective evaluations, including randomized controlled trials however support the suitability for routine clinical application of these strategies.20
Biomarker-based rule-out pathways should always be used in conjunction with history, clinical examination, and ECG to assess the risk of MI and other acute cardiovascular events. Some patients assigned to the rule-out group may still have had ischaemic symptoms and/or ECG signs indicative of unstable angina or an evolving MI, in whom further clinical evaluation and hospitalization may be indicated. Indeed, studies have shown that the addition of clinical information improves the rule-out safety compared to strategies based on hs-cTn alone.21–25 It therefore seems reasonable to believe that the addition of clinical information such as risk scores or ECG would further improve the performance of the historical hs-cTnT-based algorithm, but this needs further study.
Most patients with ≥2 hs-cTnT tests at the ED who would have been ruled out by the ESC 0-/1-h algorithm were also ruled out with the historical hs-cTnT-based algorithm, and the MI risk was comparable in the ruled-out patients. This further supports the ability of the historical hs-cTnT-based algorithm to identify patients in whom serial hs-cTnT tests may not be needed at the ED. Moreover, 60% of all ruled-out patients in whom a second hs-cTnT analysis was performed had a low but measurable 0-h hs-cTnT concentration, which indicate that the algorithm could affect the decision to rule out MI in a considerable number of patients with admission hs-cTnT concentrations in the lower range.
Further studies are needed to help develop clinical guidelines for safe and effective use of historical hs-cTnT results, and these should ideally evaluate cut-offs for the change between hs-cTnT tests in relation to the time between them. Prior studies indicate that increases of hs-cTnT over time are associated with an increased risk of cardiovascular events.26–28 However, it is unknown whether historical hs-cTnT tests could also improve the rule in of MI at the ED, and this merits further attention.
Strengths
The large sample size with a relatively large incidence of MI allowed us to evaluate the diagnostic performance of several combinations of historical and 0 h hs-cTnT results for the ability to rule out MI. We retrieved data on background characteristics and outcomes from validated health care registers with complete nationwide coverage.10,11
The study was conducted at several hospital sites across different regions in Sweden and the results were validated in a separate large cohort from yet another region. The developed algorithm had high external validity, and we believe that results can be generalized to national and international health care settings with a comparable MI incidence.
Limitations
Patients were not managed according to our derived algorithm. Consequently, the safety and efficacy of this algorithm in clinical practice are unknown. Some patients ruled out by our algorithm could still have a high risk of acute medical conditions, and thus need further evaluation or hospital admission. In the present study, three of four (73%) patients within the derivation cohort who were ruled out by the historical hs-cTnT-based algorithm were in reality discharged from the ED, and the majority of the remaining (admitted) patients had an unspecific symptom diagnosis at hospital discharge. This indicates that a considerable proportion of ruled-out patients may have been suitable for early discharge, without the need of hospitalization, although the ‘true’ additive number of patients in whom early discharge from the ED may be appropriate remains unknown.
Although the performance of the algorithm was promising when evaluated in both cohorts, 3–4% of all MIs would still have been ‘missed’ among patients triaged toward rule out. Nonetheless, the sensitivity was similar to that reported for existing rule-out pathways for MI in observational studies.5,19
Diagnoses in the derivation cohort were based on the National Patient Register and were not adjudicated. Therefore, some misclassification of the outcome may have been present, although the MI diagnosis in the NPR has been reported to have positive predictive value of >99% according to adjudication of medical records.9,29
Due to the risk of false-negative results, we do not however recommend the use of 0 h hs-cTn strategies in patients presenting very early after symptom onset. We did not have information on the time from symptom onset to ED presentation at the index visit to perform a subgroup analysis. Very early presenters may therefore have incorrectly been ruled out by the historical hs-cTnT-based algorithm, contributing to a lower sensitivity and NPV than would be seen if used in clinical practice. Further studies are thus needed to evaluate the performance of the algorithm in early presenters.
The MI incidence observed in both cohorts was lower than what has been reported by others.30–32 However, the incidence of MI varies between prospective chest pain cohorts and is highly dependent on the selection of patients with hs-cTn testing.33 A liberal approach to troponin testing among all-comers with chest pain at the EDs included in this study is likely the main reason for the low overall incidence of MI. However, we believe that while this may limit the comparability with chest pain cohorts in which specific criteria for troponin testing are applied, it will allow a generalization of the results to ‘real-world’ clinical contexts where hs-cTnT assays are used today.
Since many European countries have been using hs-cTnT now for many years, historical hs-cTnT is in most places readily available. Our suggested approach will thereby be able to be easily implemented. We do acknowledge however that in a setting where prior hs-cTn data are scarce, the applicability of this approach will be limited.
Lastly, the results of the present study need to be confirmed in prospective studies before implementation in routine care, preferably at multiple sites with varying MI incidence.
Conclusion
A combination of a historical hs-cTnT and the change up to the first hs-cTnT at the ED safely ruled out MI or death within 30 days in two large cohorts of acute chest pain patients. The use of historical hs-cTnT results has the potential to reduce the need for serial hs-cTnT tests at the ED and to expedite the evaluation of these patients.
Supplementary material
Supplementary material is available at European Heart Journal: Acute Cardiovascular Care online.
Data availability statement
The data that support the findings of this study are available from the corresponding author, upon reasonable request.
Funding
No specific funding was obtained for this study. M.J.H. holds a research position funded by the Swedish Heart-Lung Foundation (Grant number: 20170804) and the Stockholm County Council (Grant number: 20170686). A.R. holds a research position funded by the Stockholm County Council (Grant number: 20200935). The ESC-TROP study was partly supported by an ALF research grant, the Swedish Heart-Lung Foundation, VINNOVA, Bundy Academy, and the Swedish Research Council. The sponsors had no role in the design or conduct of this study.
Conflicts of interest: M.J.H. received consultancy honoraria from Idorsia and Pfizer. The remaining authors declare no conflicts of interest.
References
Author notes
Deceased 5 June 2021.
Comments