Search Results

After deduplication, the initial search yielded 1,204 titles and abstracts eligible for inclusion. Following further screening, 53 articles remained, of which 40 were excluded after full text screening. An update of the search resulted in 1 additional article (Figure 1). Backward snowballing yielded no additional inclusions.

Figure 1.

Flowchart of systematic review of the literature.

ASR = Active learning for Systematic Reviews.

Study and Patient Characteristics

Table 1 displays the characteristics of the 14 included studies. Nine studies derived or validated a CDR and 5 studies validated troponin assays. Study populations ranged from 115 to 2,192 patients, with follow-up periods from 3 weeks to 6 months. An overview of the components of the included CDRs can be found in Supplemental Table 1. Supplemental Table 2 shows troponin assay characteristics and threshold values.

Quality Assessment

Five of the 14 (35.7%) studies showed a high risk of bias in at least 1 of the quality domains. High risk of bias was most common in the index test and reference standard domains.15,16,18-20 Applicability concerns were most frequent in patient selection and index test.11,12,14 A detailed overview of the risk of bias and applicability assessment can be found in Supplemental Appendix 3.

Risk Stratification Tools

A complete overview of diagnostic test characteristics can be found in Table 2.

Grijseels/Bruins Slot Rule

Grijseels et al developed an algorithm in 1995 which consists of a combination of patient and symptom characteristics and electrocardiogram (ECG) findings.11,12 Discriminative properties showed C-statistics ranging from 0.70 to 0.72. In 2011, Bruins Slot et al updated the algorithm to a point-based rule with 3 risk categories for ACS.1 Validation of the updated rule resulted in C-statistics of 0.66 to 0.72, sensitivity of 84.4% to 97.0%, and NPV of 91.7% to 98.2%. When comparing the CDR to unaided clinical judgment, GPs categorized patients with and without ACS better.1,15

Willemsen CDR

In 2019, Willemsen et al evaluated a heart-type fatty acid–binding protein (H-FABP) POCT.13 The diagnostic accuracy of H-FABP as standalone test to rule out ACS had a sensitivity of 25.8% and NPV of 91.6%. Using H-FABP with a CDR increased the sensitivity to 87.5% and NPV to 97.2%. Compared with unaided clinical judgment, fewer ACS cases were missed (1.3% vs 2.6%), but there were 13.9% more referrals.

Marburg Heart Score

The Marburg Heart Score (MHS) was developed in 2010 to rule out coronary artery disease.24 Later, it was evaluated to rule out ACS or MACE in 1 prospective and 2 retrospective studies, with different rule-out thresholds (Table 1).14,15,16 C-statistics ranged from 0.64 to 0.77, with sensitivity of 75.0% to 81.3%, and NPV of 88.0% to 98.6%. The MHS did not outperform unaided clinical judgement.14

INTERCHEST

Like the MHS, the 2017 International Chest pain prediction (INTERCHEST) score was developed to rule out coronary artery disease, and later tested to rule out ACS or MACE. The rule was derived from pooled individual patient data of 5 studies (3,099 patients total).25 Twofold retrospective validation resulted in C-statistics of 0.77 to 0.85, sensitivity of 87.5% to 88.8%, and NPV of 98.6% to 99.1%.15,16 Discriminative power was similar to unaided clinical judgement.

Gencer Rule

The Gencer rule was developed in 2010 to rule out coronary artery disease and consists of 7 components.26 It was validated to rule out MACE in 1 study with a C-statistic of 0.72, sensitivity of 84.4%, and NPV of 98.0%.15 The performance of the rule was equivalent to unaided clinical judgement.

Safety First

Wouters et al in 2022 developed a computerized risk stratification tool for the triage of acute chest pain patients in out-of-hours primary care settings.17 The tool includes 7 predictors of ACS. After derivation, an internal-external validation technique was used, resulting in a C-statistic of 0.77 to 0.79. The authors did not recommend a cut-off point; however, diagnostic accuracy for various risk thresholds (0.1% to 20%), showed a sensitivity of 46% to 98% and NPV of 93% to 99%.

HEART-Score Variants

The original HEART-score tool (History, ECG, Age, Risk factors, and Troponin), was developed to identify patients at low risk for short-term MACE among patients with acute chest pain in the emergency department.27 For use in primary care, the original HEART-score was modified.18 The first variation (simplified HEART-score) simply omits troponin. The second variation, HEART-GP, replaces troponin with the GP’s sense of alarm. The modified scores had a C-statistic of 0.86 and 0.90, respectively, sensitivity of 96.9%, and NPV of 99.7% for both versions. Compared with unaided clinical judgment, both scores improved safety, at the cost of additional referrals (simplified HEART had 26% more, and HEART-GP had 4% more).

Conventional Troponin

In a study by Planer et al in 2006, a qualitative cardiac troponin T (cTnT) test was combined with the GP’s clinical assessment.19 The combined strategy resulted in a sensitivity of 95.8% and NPV of 99.6%. The cTnT test as a standalone test showed poor sensitivity (20.8%) and did not outperform unaided clinical judgment.

Tomonaga et al in 2011 compared the use of a 3-in-1 POCT for cTnT, N-terminal pro-B-type natriuretic peptide (NT-proBNP), and D-dimer to usual care in patients with chest pain, dyspnea, or symptoms suggestive for thromboembolic events.20 The use of this combined POCT test led to more accurate working diagnoses (76% vs 60%) and fewer referrals (30% vs 55% false positives). For ACS, however, the sensitivity was 89.5% compared with 100% in the usual care group, with a NPV of 98.9%, compared with 100%. Sensitivity for troponin as standalone test to rule out ACS was 58.8%, missing 7 out of 17 (41.2%) patients with ACS.

A Swedish study by Nilsson et al in 2013 compared 3 primary care practices that already incorporated the use of cTnT-POCT to 4 practices that did not use cTnT-POCT.21 Use of cTnT-POCT by GPs reduced hospital referrals (25% vs 43%), but at the expense of sensitivity (71.4% in the practices using cTnT-POCT vs 100% in the control group). Troponin as standalone test showed a sensitivity of 28.6% and NPV of 95.9%, missing 5 out of 7 (71%) patients with ACS.

High-Sensitivity Troponin

In 2015, Andersson et al used the original (stored) samples from the study by Nilsson et al in 2013 to measure high-sensitivity (hs) cTnT and compare the diagnostic outcomes with the outcomes for conventional troponin.22 They found a sensitivity of 83.3% compared with 33.3%, missing 1 out of 6 (17%) patients with ACS. This came at the expense of specificity (76.2% vs 97.5%) and thus led to additional referrals (23% vs 2% false positives).

Lastly, in 2021, Johannessen et al validated 3 strategies to rule out AMI using hs-cTnT in a primary care emergency outpatient clinic in Norway.23 The first was a single hs-cTnT rule-out strategy. High sensitivity cTnT was measured at presentation and AMI was ruled out if hs-cTnT was below the limit of detection (ie, very low). No AMI were missed and sensitivity and NPV were 100% with a C-statistic of 0.85. The second and third strategies were the original HEART-score and a modified HEART-score with lower hs-cTnT thresholds. The modified HEART-score outperformed the original HEART-score (sensitivity 98.4% vs 91.8% and NPV 99.8% vs 99.4%), but not the single hs-cTnT rule out strategy.