Time-Stratified HEART Score Discrimination and Calibration in Emergency Chest Pain: A Retrospective Cohort Study

Time-Stratified HEART Score Discrimination and Calibration in Emergency Chest Pain: A Retrospective Cohort Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Time-Stratified HEART Score Discrimination and Calibration in Emergency Chest Pain: A Retrospective Cohort Study Sebnem Zeynep Eke Kurt, Ugur Durmus, Erdem Kurt, Suphi Bahadırlı This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9430666/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Aims We compared HEART score discriminative performance and calibration for 30-day major adverse cardiac events (MACE) between off-hours and on-hours emergency department chest pain presentations. Methods A single-center retrospective cohort study was conducted at Istanbul Medipol Mega University Hospital from January 2020 through December 2025, enrolling 2,800 patients (off-hours n = 2,133; on-hours n = 667). Off-hours comprised evenings, nights, weekends, and public holidays. The primary outcome was 30-day MACE, defined as acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death. Discrimination was compared by the DeLong method (equivalence: ΔAUC < 0.08); calibration was assessed by the Hosmer-Lemeshow test, observed-to-expected ratio, and calibration slope. Results 30-day MACE occurred in 14.4% of 2,800 patients. The HEART score AUC was 0.830 (95% CI 0.805–0.855) in off-hours and 0.821 (95% CI 0.778–0.865) in on-hours patients; ΔAUC was + 0.009 (95% CI -0.041 to + 0.058; p = 0.736), within the pre-specified equivalence boundary. At the rule-out threshold (HEART ≥ 4), off-hours sensitivity was 95.7% (95% CI 92.8–97.5%) and negative predictive value (NPV) was 97.7% (95% CI 96.2–98.7%). No differential shift effect was detected (interaction OR 1.039; 95% CI 0.867–1.247; p = 0.676). Both groups demonstrated systematic overestimation of absolute MACE risk relative to Backus 2013 probabilities (off-hours calibration intercept − 0.352; 95% CI -0.495 to -0.208; p < 0.001), with no between-group difference (p = 0.998). Conclusion The HEART score demonstrated clinically equivalent discrimination for 30-day MACE regardless of presentation time, supporting unmodified use across all shifts. Local recalibration of absolute risk estimates is warranted before probability-guided triage protocols are implemented. Chest Pain Emergency Service Hospital Risk Assessment Figures Figure 1 Figure 2 Figure 3 1. Introduction Acute chest pain accounts for approximately 5–10% of all adult emergency department visits and represents one of the most demanding diagnostic challenges in emergency medicine, as approximately 10% of these patients ultimately receive a diagnosis of acute coronary syndrome while the majority are attributed to non-cardiac causes.( 1 ) Identifying that minority accurately at triage without subjecting every patient to prolonged observation or invasive workup requires a standardized, rapidly applicable risk stratification instrument. The HEART score, integrating History, Electrocardiogram, Age, Risk Factors, and Troponin into a 10-point composite, was developed for this purpose ( 2 ) and subsequently validated in a multicenter retrospective cohort of 880 patients, demonstrating that scores of 0–3 were associated with a MACE rate below 1%. ( 3 ) Subsequent prospective validation across 2,440 patients in 10 Dutch emergency departments confirmed a C-statistic of 0.83, with 30-day MACE rates below 2% in the low-risk stratum and exceeding 50% in the high-risk stratum. ( 3 ) These risk thresholds have since anchored clinical disposition decisions in emergency departments across multiple countries. ( 4 ) External validation studies have confirmed HEART score discrimination in Asian emergency populations, with a C-statistic of 0.83 across 14 hospitals and 2,906 patients, and in retrospective military emergency cohorts, where AUCs of 0.885–0.898 were reported. ( 5 , 6 ) A systematic review and meta-analysis incorporating 25 validation studies and more than 25,000 patients demonstrated a pooled sensitivity of 0.96 and negative predictive value of 0.99 for low-risk scores, confirming a consistent safety profile for rule-out. ( 4 ) However, to the best of our knowledge, no published validation study has stratified HEART score performance by time of emergency presentation. This gap is clinically relevant: off-hours admissions for acute coronary syndrome are associated with a small but persistent increase in in-hospital mortality, longer door-to-treatment intervals, and reduced specialist availability.( 7 , 8 ) Whether these operational differences translate into measurable differences in HEART score discriminative performance or calibration has not been formally examined in the published literature. This study aimed to evaluate and compare the discriminative performance and calibration of the HEART score for 30-day MACE prediction in adult patients presenting with acute chest pain during off-hours versus standard working hours at a tertiary care emergency department. Secondary objectives included assessment of HEART score calibration relative to published validation cohort reference probabilities within each shift group, comparison of threshold-based rule-out and rule-in performance at pre-specified clinical cut points, evaluation of discrimination across five granular time strata, and decision curve analysis across a clinically relevant threshold probability range. 2. Methods 2.1. Study Design and Setting This single-center retrospective cohort study was conducted in the emergency department of Istanbul Medipol Mega University Hospital, a tertiary care academic center in Istanbul, Turkey, between January 2020 and December 2025. The study constituted an external validation of the HEART score for 30-day MACE prediction and was classified as TRIPOD Type 2b/3. The study was approved by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee (approval number: E-10840098-202.3.02-470; decision number: 90). Individual patient consent was waived in accordance with the retrospective design. A systematic search of PubMed and Embase (January 2000 through December 2024) using the terms 'HEART score', 'off-hours', 'night shift', 'weekend', and 'time of presentation' identified no prior study that stratified HEART score validation by time of emergency presentation, supporting the originality of the research question. 2.2. Participants Adult patients aged 18 years or older presenting to the emergency department with acute non-traumatic chest pain as the primary complaint were eligible for inclusion, provided that a first-visit triage timestamp was documented. Patients were excluded if HEART score documentation was incomplete (n = 298), if they presented with ST-elevation myocardial infarction requiring immediate primary percutaneous coronary intervention bypassing standard triage stratification (n = 1163), if they were transferred from another institution with an established cardiac diagnosis (n = 45), or if 30-day follow-up could not be ascertained (n = 167), yielding an analytic cohort of 2,800 patients from 4473 screened. The unit of analysis was the individual emergency presentation; the first-visit flag confirmed a first-presentation design with no repeated-measures clustering. The derivation of the analytic cohort from the screened population is illustrated in Figure S1 . 2.3.Variables Primary exposure. The triage timestamp was used to classify each presentation as off-hours or on-hours. Off-hours encompassed evenings (18:00–23:59), nights (00:00–07:59), weekends (Saturday and Sunday, all hours), and public holidays. On-hours encompassed regular weekday working hours (08:00–17:59, Monday through Friday, excluding public holidays). Under this definition, off-hours encompassed approximately 75% of total weekly hours, reflecting the predominance of evenings, nights, weekends, and public holidays in the calendar; the resulting 76.2% off-hours enrollment fraction is therefore consistent with the expected distribution of presentations across this time window. Index test. The HEART score was calculated from five components, each scored 0–2: History, Electrocardiogram, Age, Risk Factors, and Troponin, yielding a total score of 0–10. ( 2 ) Patients were classified as low risk (score 0–3), intermediate risk (score 4–6), or high risk (score 7–10). Predicted probabilities for calibration analyses were derived from the Backus 2013 prospective validation cohort reference values (hereafter "the reference cohort"), low-risk group 0.017, intermediate-risk group 0.120, and high-risk group 0.652. ( 9 ) Logit-scale values of these probabilities were pre-computed for offset regression analyses. Primary outcome. The primary outcome was 30-day MACE, defined as a composite of acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death occurring within 30 days of the index emergency presentation, ascertained through hospital readmission records, cardiology outpatient records, and institutional mortality registries. Elective revascularization procedures were excluded from the composite endpoint. Covariates. The first troponin measurement (Troponin-1, ng/L) was recorded at triage. Door-to-diagnosis time was defined as the interval in minutes from triage timestamp to documented clinical diagnosis. Troponin assay type (high-sensitivity troponin I versus conventional troponin I) and ECG interpreter seniority (resident, specialist, or cardiologist) were recorded as secondary variables. Comorbidities (hypertension, diabetes mellitus, hyperlipidemia, current smoking, obesity) were ascertained from the electronic medical record. 2.4. Data Sources and Measurement Clinical data were extracted from the institutional electronic medical record system using a standardized, pre-specified data collection protocol. HEART score components were documented by the treating emergency physician at the time of presentation and abstracted verbatim from the medical record. The troponin assay transitioned from conventional troponin I to high-sensitivity troponin I during the study period; assay type was recorded per patient and addressed in a pre-specified sensitivity analysis. ECG interpretation was performed by the treating clinician and the interpreter's seniority level was recorded. The 30-day follow-up was ascertained from institutional records without active patient contact. 2.5. Bias Four principal sources of bias were addressed prospectively. Information bias arising from retrospective medical record abstraction was minimized through a standardized extraction protocol applied uniformly across both shift groups; residual observer-dependent variability in the History component of the HEART score cannot be fully quantified and may have attenuated that component's contribution to overall score accuracy, with an expected non-differential direction of effect. Temporal confounding from the mid-study transition between troponin assay platforms affects the Troponin component of the HEART score and was addressed in Sensitivity Analysis 2 (SA2), which stratified the primary discrimination and calibration analyses by assay type; overlapping confidence intervals across assay subgroups indicate that this transition did not materially bias the primary estimate. Selection bias from differential serial troponin completion between shift groups (off-hours 78.8%, on-hours 91.3%) represents the most consequential source of potential outcome misclassification; Sensitivity Analysis 1 (SA1) restricted the primary analysis to patients with complete original serial troponin measurements to characterize the direction and magnitude of this effect. Outcome ascertainment bias arising from the use of institutional records without active patient contact represents a fourth potential source of error; patients experiencing MACE at external facilities would not be captured, biasing the observed event rate toward underestimation and the negative predictive value toward overestimation, with the magnitude of this effect unquantifiable from available data. 2.6. Study Size As this was a retrospective study, no prospective power calculation was performed. All patients meeting eligibility criteria during the study period were included. The final analytic cohort comprised 2,800 patients, of whom 402 experienced the primary outcome (30-day MACE rate 14.4%). The off-hours group included 2,133 patients with 305 events and the on-hours group included 667 patients with 97 events. For the primary secondary estimand, the DeLong confidence interval for the AUC difference spanned − 0.041 to + 0.058, confirming adequate precision for the pre-specified clinical equivalence boundary of ± 0.08. 2.7. Statistical Analysis Normality assessment. For samples with n ≥ 50, normality was assessed using a combined multi-criterion protocol (pre-specified multi-criterion normality protocol): a variable was classified as non-normal if at least three of five criteria indicated departure from normality, comprising the Kolmogorov-Smirnov test with Lilliefors correction (p ≤ 0.05), skewness z-score ≥ 1.96, kurtosis z-score ≥ 1.96, Q-Q plot visual inspection, and histogram shape assessment. This multi-criterion approach was defined a priori by the investigators; no single published guideline mandates this specific combination of criteria. For samples with n < 50, the Shapiro-Wilk test was applied. All primary continuous variables (age, HEART score, troponin-1, door-to-diagnosis time) met the threshold for non-normal distribution under this protocol. Descriptive statistics and group comparisons. Continuous variables are reported as median (interquartile range) and compared between groups using the Mann-Whitney U test. Categorical variables are reported as n (%) and compared using the chi-square test with continuity correction for 2×2 tables or Fisher's exact test where any expected cell count was below 5. For the three-level HEART risk group variable, a 3×2 chi-square test was applied with the global p-value reported. Standardized mean differences (SMD) were computed for all Table 1 variables as a p-value-independent measure of group balance; an SMD exceeding 0.10 was pre-specified as indicating clinically meaningful imbalance. Table 1 Baseline characteristics by shift group. Variable Off-hours (n = 2,133) On-hours (n = 667) p SMD Continuous variables Age, median (IQR), years 59 (49–70) 59 (49–67) 0.067 0.089 HEART total score, median (IQR) 5 ( 3 – 6 ) 4 ( 3 – 6 ) 0.604 0.014 Troponin-1, median (IQR), ng/L 18.4 (7.5–54.6) 17.7 (6.6–48.4) 0.345 0.024 Door-to-diagnosis time, median (IQR), min 168 (127–202) 146 (104–184) < 0.001 0.351* Categorical variables Male sex, n (%) 1,220 (57.2) 380 (57.0) 0.954 0.005 MACE at 30 days, n (%) 305 (14.3) 97 (14.5) 0.926 0.007 Acute myocardial infarction 228 (10.7) 74 (11.1) 0.824 0.013 Revascularisation 157 (7.4) 45 (6.7) 0.653 0.024 Cardiac arrest 38 (1.8) 7 (1.0) 0.256 0.062 Death 39 (1.8) 13 (1.9) 0.970 0.009 MACE at 6 weeks, n (%) 332 (15.6) 104 (15.6) 1.000 0.001 HEART risk group, n (%) 0.301 0.024 Low risk (score 0–3) 574 (26.9) 196 (29.4) Intermediate risk (score 4–6) 1,223 (57.3) 360 (54.0) High risk (score 7–10) 336 (15.8) 111 (16.6) Comorbidities Hypertension 960 (45.0) 275 (41.2) 0.095 0.076 Diabetes mellitus 513 (24.1) 161 (24.1) 1.000 0.002 Hyperlipidaemia 826 (38.7) 262 (39.3) 0.833 0.011 Current smoking 559 (26.2) 205 (30.7) 0.025 0.100* Obesity 539 (25.3) 150 (22.5) 0.160 0.065 Process indicators Original serial troponin complete, n (%) 1,681 (78.8) 609 (91.3) < 0.001 0.356* hsTnI assay used, n (%) 1,458 (68.4) 458 (68.7) 0.918 0.007 Hospitalisation, n (%) 534 (25.0) 170 (25.5) 0.854 0.010 Cardiology consultation, n (%) 606 (28.4) 184 (27.6) 0.716 0.018 IQR = interquartile range; MACE = major adverse cardiac event; hsTnI = high-sensitivity troponin I; SMD = standardised mean difference. Continuous variables compared by Mann-Whitney U test (non-normal distribution confirmed by pre-specified five-criterion normality protocol). Categorical variables compared by chi-square with continuity correction; Fisher's exact test where expected cell count 0.10 indicates clinically meaningful imbalance. Individual MACE components are not mutually exclusive; a single patient may contribute to more than one component category. Discrimination. The area under the receiver operating characteristic curve (AUC, equivalent to the C-statistic) was computed using the DeLong method, which provides exact confidence intervals without bootstrap approximation. Threshold-based performance metrics (sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio) were computed at two pre-specified clinical thresholds: HEART ≥ 4 (standard rule-out threshold) and HEART ≥ 7 (high-risk admission threshold). Wilson score confidence intervals were applied to all proportion-based metrics. The Youden index (J = sensitivity + specificity − 1) was computed across all score values to identify a data-driven optimal cutpoint, labelled exploratory and post-hoc as pre-specification. Calibration. Calibration was assessed using five complementary metrics: ( 1 ) the Hosmer-Lemeshow goodness-of-fit test, with one group per HEART score value (11 groups, score 0 through 10; df = 9 under the H-L formula df = g − 2); ( 2 ) the observed-to-expected (O/E) ratio overall and within each risk stratum, with Byar approximation for 95% confidence intervals; ( 3 ) the calibration intercept, estimated by offset logistic regression with the logit of reference cohort predicted probabilities as a fixed offset term ( 4 ) the calibration slope, estimated by logistic regression of the observed outcome on the logit of predicted probabilities, with bootstrap confidence intervals (1000 iterations; seed = 2024) and a two-sided z-test against the null hypothesis of slope = 1; and ( 5 ) the Brier score and scaled Brier score, quantifying overall predictive skill relative to the null model. Decision curve analysis. Net benefit was computed across a pre-specified threshold probability range of 5–40% and compared against treat-all and treat-none reference strategies. Separate decision curves were produced for off-hours and on-hours groups. Primary comparative analysis. The primary secondary estimand was the difference in AUC between independent off-hours and on-hours groups, computed using the DeLong variance estimator for independent samples. Clinical equivalence was defined a priori as ΔAUC < 0.08, pre-specified before data acces. A bootstrap confidence interval for ΔAUC was generated as a supplementary check (1,000 iterations; seed = 2024; percentile method). A formal interaction test was conducted using logistic regression including the mean-centered HEART score, off-hours status, and their product term; the interaction p-value served as the pre-specified test of differential discrimination between groups. Within each HEART risk group, 30-day MACE rates were compared between shift groups using chi-square with continuity correction; the Mantel-Haenszel pooled odds ratio was computed with HEART risk group as the stratification variable. Time strata analysis. Five time strata were defined: on-hours (reference), evening, night, weekend daytime, and public holidays. An AUC was estimated within each stratum using the DeLong method. Heterogeneity across strata was assessed using a weighted Cochran Q statistic; pairwise comparisons against the on-hours reference were planned only in the event of a significant global test (p < 0.05), applying Benjamini-Hochberg false discovery rate correction. Sensitivity analyses. Six pre-specified sensitivity analyses were performed as pre-specified. SA1 restricted the analysis to patients with complete original serial troponin measurements, isolating the potential bias introduced by imputed or simulated second troponin values. SA2 stratified the off-hours cohort by troponin assay type (high-sensitivity troponin I versus conventional troponin I) and repeated the primary discrimination and calibration analyses within each subgroup, addressing the temporal confounding effect of assay migration on the T-component of HEART. SA3 stratified by the seniority of the clinician who interpreted the ECG (resident, specialist, or cardiologist), testing whether off-hours degradation in ECG interpretation quality affects E-component scoring and downstream model performance. SA4 replaced the 30-day MACE endpoint with the 6-week MACE outcome, aligning the follow-up window with the reference cohort and testing sensitivity of findings to outcome ascertainment duration. SA5 served as a data quality verification step confirming the first-visit-only design; by construction, all 2,800 patients carried a first-presentation flag, rendering this analysis identical to the primary. SA6 used MICE-imputed HEART scores in place of original scores (20 imputation sets, 50 iterations, seed = 2024, Rubin's rules applied); missing data were confined to components with original documentation gaps and no outcome imputation was performed. Full results for all sensitivity analyses are presented in Supplementary Table S1 . All analyses were conducted in Python 3.11 (scipy 1.11+, statsmodels 0.14+, scikit-learn 1.3+). Statistical significance was set at p < 0.05 (two-tailed) for the primary outcome. All point estimates are accompanied by 95% confidence intervals. 2.8. Missing Data Primary analysis variables (HEART total score, 30-day MACE, first troponin, and reference cohort predicted probabilities) contained no missing values. A complete-case flag identified 2,080 of 2,800 patients (74.3%) with complete data across all secondary variables. Patients with incomplete secondary variable data were excluded from the specific analyses requiring those variables; denominators are reported explicitly for each analysis step. Multiple imputation was applied only in SA6, using pre-computed MICE scores from the dataset. The 720 incomplete cases were driven by three variables: original serial troponin documentation (n = 510; 18.2%), family history (n = 180; 6.4%), and original ECG interpreter records (n = 86; 3.1%). Serial troponin missingness was differential by shift (off-hours 21.2% versus on-hours 8.7%), consistent with missing at random conditional on shift assignment and directly addressed by SA1; family history missingness was non-differential across shift groups (6.4% versus 6.6%), consistent with missing completely at random. ECG interpreter seniority was fully documented for all 2,800 patients (0 missing values), confirming that the SA3 subgroup counts are not affected by secondary variable incompleteness.. 2.9. Outcomes The primary outcome was 30-day MACE, defined as a composite of acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death within 30 days of the index presentation, pre-specified before data analysis. Secondary outcomes included MACE at 6 weeks, individual MACE components (reported separately), and HEART score discrimination stratified by time stratum. The primary secondary estimand was the difference in AUC between off-hours and on-hours groups. All analyses were pre-specified before data access; no post-hoc modifications to any estimand were made after data access, and all other analyses are reported as exploratory. 3. Results 3.1 Study cohort The analytic cohort comprised 2,800 patients presenting with acute chest pain (Figure S1 ) , of whom 2,133 (76.2%) presented during off-hours and 667 (23.8%) during on-hours. Overall, 402 patients experienced a 30-day MACE (14.4%). All entries in the dataset carried a first-visit flag, confirming the absence of repeated-measure clustering. 3.2 Baseline characteristics Table 1 presents baseline characteristics by off-hours and on-hours status. The two groups were well-matched on age (both median 59 years), MACE rate (14.3% versus 14.5%, p = 0.926), sex distribution (57.2% versus 57.0% male), HEART score (median 5 versus 4, p = 0.604), first troponin (18.4 versus 17.7 ng/L, p = 0.345), and comorbidity profiles. Risk group distribution was also comparable (chi-square p = 0.301). Assay type was balanced: hsTnI was used in 68.4% of off-hours and 68.7% of on-hours patients (p = 0.918). Three variables showed clinically meaningful imbalance (SMD > 0.10). Door-to-diagnosis time was longer in off-hours patients (median 168 versus 146 minutes, p < 0.001, SMD = 0.351), reflecting reduced specialist availability outside working hours. Serial troponin completion was lower in off-hours patients (78.8% versus 91.3%, p < 0.001, SMD = 0.356). Smoking prevalence also showed clinically meaningful imbalance (26.2% versus 30.7%, p = 0.025, SMD = 0.100). These imbalances are addressed in the relevant sensitivity analyses. 3.3. Primary validation in the off-hours cohort 3.3.1. Discrimination In the off-hours group (n = 2,133; 305 MACE events, 14.3%), the HEART score achieved an AUC of 0.830 (95% CI: 0.805–0.855), confirming good discriminative performance (Fig. 1 , left panel; Table 2 ). Table 2 Discrimination metrics at pre-specified and exploratory thresholds (off-hours cohort, n = 2,133). Metric HEART ≥ 4 (standard) HEART ≥ 7 (high-risk) HEART ≥ 6 (Youden) 95% CI method AUC (C-statistic) 0.830 (0.805–0.855) Sensitivity 95.7% (92.8–97.5) 55.4% (49.8–60.9) 75.4% (70.3–79.9) Wilson Specificity 30.7% (28.6–32.8) 90.9% (89.5–92.1) 77.5% (75.5–79.4) Wilson PPV 18.7% (16.9–20.7) 50.3% (45.0-55.6) 35.9% (32.3–39.7) Wilson NPV 97.7% (96.2–98.7) 92.4% (91.1–93.6) 95.0% (93.7–96.0) Wilson LR+ 1.38 6.07 3.35 LR- 0.139 0.491 0.317 TP / FP / TN / FN 292/1267/561/13 169/167/1661/136 230/411/1417/75 N = 2,133 off-hours patients; 305 MACE events (14.3%). Threshold ≥ 4: standard clinical rule-out. Threshold ≥ 7: high-risk admission rule-in. Threshold ≥ 6: Youden-index optimal (post hoc, data-driven; labelled exploratory, as pre-specified). AUC computed by DeLong method. PPV = positive predictive value; NPV = negative predictive value; LR = likelihood ratio. At the standard clinical rule-out threshold (HEART ≥ 4), sensitivity was 95.7% (95% CI: 92.8–97.5%) and NPV was 97.7% (95% CI: 96.2–98.7%), with 13 missed events among 574 patients below the threshold. At the high-risk admission threshold (HEART ≥ 7), specificity was 90.9% (95% CI: 89.5–92.1%) and PPV was 50.3% (95% CI: 45.0-55.6%). The positive likelihood ratio at this threshold was 6.07, indicating a clinically meaningful shift in post-test probability. The Youden-index optimal threshold was HEART ≥ 6 (J = 0.529), with sensitivity 75.4% (95% CI: 70.3–79.9%) and specificity 77.5% (95% CI: 75.5–79.4%). This threshold was identified post hoc and is labelled exploratory, as pre-specified. 3.3.2. Calibration Table 3 presents the complete calibration profile. The Hosmer-Lemeshow test was highly significant (χ²=104.32, df = 9, p < 0.001), indicating systematic departure from the reference cohort probabilities. The overall O/E ratio of 0.812 indicates that the Backus model predicted approximately 23% more MACE events (predicted N = 375.6) than were observed (N = 305). This pattern of overestimation was consistent across all three risk strata, with O/E values of 1.33 in the low-risk group (confidence interval crossing 1.0, reflecting sparse events), 0.84 in the intermediate-risk group, and 0.77 in the high-risk group. Table 3 Calibration metrics for off-hours and on-hours groups. Calibration metric Off-hours (n = 2,133) On-hours (n = 667) Difference p Hosmer-Lemeshow χ² (df = 9) 104.32 16.83 < 0.001 / 0.052 Hosmer-Lemeshow p-value < 0.001 0.052 O/E ratio, overall (Byar 95% CI) 0.812 0.724 to 0.908) 0.791 (0.642 to 0.965) Low-risk group (HEART 0–3) 1.332 (0.665–2.278) 0.625 (0.121–1.821) Intermediate-risk group (HEART 4–6) 0.838 (0.693-1.000) 0.860 (0.629–1.141) High-risk group (HEART 7–10) 0.771 (0.657–0.897) 0.747 (0.560–0.977) Calibration intercept (95% CI) -0.352 (-0.495 to -0.208) -0.351 (-0.608 to -0.094) -0.000 0.998 Calibration slope (bootstrap 95% CI) 0.828 (0.732–0.916) 0.758 (0.624–0.910) 0.070 0.426 Brier score 0.101 0.109 Scaled Brier score 0.176 0.125 χ ² p < 0.001 indicates systematic miscalibration relative to reference cohort probabilities. O/E < 1 indicates the model overestimates absolute MACE risk in this population. Calibration intercept tested H₀: intercept = 0; slope tested H₀: slope = 1. Difference and p-value for intercept and slope based on z-test using bootstrap standard errors; all other metrics descriptive only. O/E = observed-to-expected ratio. The calibration intercept was − 0.352 (95% CI: -0.495 to -0.208; p < 0.001), indicating that the reference model systematically overestimates absolute MACE risk in this cohort, consistent with a lower baseline event rate than the reference population. The calibration slope was 0.828 (bootstrap 95% CI: 0.732–0.916; p < 0.001 for deviation from 1.0), indicating miscalibration in the extremes: the reference model overestimates risk in higher-scoring patients and underestimates it in lower-scoring patients relative to what is observed in this cohort. The calibration curve (Fig. 1 , right panel) illustrates both the downward shift and the reduced steepness of observed versus predicted risk across score values. The Brier score was 0.101 and the scaled Brier score was 0.176, reflecting meaningful predictive skill above the null (scaled Brier 0 = no skill; 1 = perfect). Recalibration is indicated by the significant intercept deviation and will be reported separately. 3.3.3. Decision curve analysis Figure 2 shows decision curves for off-hours and on-hours groups across threshold probabilities of 5–40%. The HEART score model delivered positive net benefit exceeding both the treat-all and treat-none strategies across the full prespecified threshold range in both groups. At a threshold of 10%, net benefit was 0.071 in off-hours patients versus a treat-all net benefit of 0.048, a net advantage of 0.023 per patient evaluated, equivalent to avoiding approximately 23 unnecessary hospitalizations per 1,000 patients assessed. The HEART score thus provides clinical utility across the range of plausible clinical decision thresholds, with no evidence of differential performance between shift groups. 3.4. Comparative analysis 3.4.1. AUC comparison between groups The primary secondary estimand, the difference in AUC between off-hours and on-hours groups, was ΔAUC = + 0.009 (95% CI: -0.041 to + 0.058; p = 0.736) in favour of off-hours patients (Table 4 ). The bootstrap-derived 95% CI was − 0.040 to + 0.057. Both the point estimate and the entire confidence interval fell well within the pre-specified clinical equivalence boundary of ± 0.08, providing clear evidence that HEART score discrimination does not meaningfully differ by shift assignment. The AUC in the on-hours group was 0.821 (95% CI: 0.778–0.865), consistent with the off-hours estimate. Table 4 AUC comparison between off-hours and on-hours groups. Analysis Off-hours AUC (95% CI) On-hours AUC (95% CI) ΔAUC (95% CI) p Primary (30-day MACE) 0.830 (0.805–0.855) 0.821 (0.778–0.865) + 0.009 (-0.041 to + 0.058) 0.736 Bootstrap ΔAUC (1,000 iterations; seed = 2024) (-0.040 to + 0.057) 6-week MACE (SA4) 0.818 (0.793–0.843) 0.811 (0.767–0.855) + 0.007 (-0.044 to + 0.058) 0.790 ΔAUC = off-hours minus on-hours; positive values favour off-hours. DeLong method (independent groups) for primary comparison. Clinical equivalence boundary: ΔAUC < 0.08 (pre-specified before data access). MACE = major adverse cardiac event. 3.4.2. Interaction analysis The formal interaction model confirmed the absence of differential HEART score effects by shift. The interaction term (mean-centred HEART score × off-hours indicator) had OR = 1.039 (95% CI: 0.867–1.247; p = 0.676), providing no evidence that the HEART score's log-odds slope for MACE differs between off-hours and on-hours patients. The main effect of the HEART score was OR = 2.166 per unit increase (95% CI: 1.851–2.533; p < 0.001), consistent with the overall discriminative performance. The off-hours main effect was OR = 0.955 (95% CI: 0.655–1.393; p = 0.811), confirming no independent effect of shift on MACE risk after accounting for HEART score. 3.4.3. MACE rates by risk group Supplementary Table S3 presents 30-day MACE rates within each HEART risk group by shift. Rates were 2.3% versus 1.5% (p = 0.740) in the low-risk group, 10.1% versus 11.9% (p = 0.353) in the intermediate-risk group, and 50.3% versus 45.9% (p = 0.493) in the high-risk group. No statistically significant difference was observed in any stratum. The Mantel-Haenszel pooled OR, adjusted for HEART risk group, was 0.989 (95% CI: 0.752–1.302; p = 0.995), confirming no significant difference in MACE risk between off-hours and on-hours patients across strata. MACE component breakdown by risk group and time stratum is shown in Supplementary Figure S2. 3.4.4. Time strata analysis Supplementary Table S2 and Supplementary Figure S3 present AUC estimates across the five-time strata. AUC ranged from 0.799 (95% CI: 0.747–0.850) in weekend daytime presentations to 0.855 (95% CI: 0.810–0.899) in evening presentations. The Cochran Q statistic was 2.991 (df = 4, p = 0.559), providing no evidence of significant heterogeneity across strata. Per the pre-specified decision rule, pairwise comparisons were not performed. The HEART score maintained good discrimination across all time categories including public holidays (AUC = 0.835, 95% CI: 0.779–0.891). 3.4.5. Calibration comparison Calibration metrics were statistically indistinguishable between groups. The intercept difference was − 0.0003 (bootstrap 95% CI: -0.350 to + 0.327; p = 0.998), and the slope difference was 0.070 (p = 0.426). Both groups exhibited comparable systematic miscalibration relative to Backus 2013, with significant intercept deviations (off-hours p < 0.001; on-hours p = 0.007) and significant slope departures from 1.0 (both p < 0.001). This indicates that the recalibration need identified in the primary analysis applies equally to both shift groups. 3.5. Sensitivity analyses Supplementary Table S1 presents all six pre-specified sensitivity analyses. Across all scenarios, AUC estimates in the off-hours group ranged from 0.818 (SA4, 6-week endpoint) to 0.848 (SA2b, conventional troponin assay subgroup), with the primary estimate of 0.830 consistent with all sensitivity results. In SA1 (original serial troponin complete; n = 1,681), AUC was 0.841 (95% CI: 0.815–0.867), marginally higher than the primary estimate, with a calibration intercept of -0.294 (versus − 0.352 in the full cohort). The modest improvement suggests that the 21.2% of patients with simulated or incomplete second troponin measurements did not introduce systematic bias, although their inclusion slightly attenuated discrimination. SA2 showed consistent performance across assay types: AUC was 0.821 (95% CI: 0.791–0.851) with hsTnI and 0.848 (95% CI: 0.804–0.892) with conventional troponin, with overlapping confidence intervals. SA3 demonstrated stable HEART score performance regardless of ECG interpreter seniority (AUC range 0.824–0.833), with overlapping confidence intervals across resident, specialist, and cardiologist groups. SA6 with MICE-imputed scores yielded AUC = 0.835, marginally above the primary estimate, consistent with expected slight optimism correction from imputation. 4. Discussion This study found that HEART score discrimination for 30-day MACE was clinically equivalent across all hours of emergency presentation, with the AUC difference between shift groups falling well within the pre-specified equivalence boundary and a formal interaction test confirming no differential score effect by shift assignment. These findings indicate that the operational constraints characteristic of off-hours care, including longer door-to-diagnosis intervals and lower serial troponin completion rates, did not materially impair the score's ability to classify patients by MACE risk. The HEART score integrates five independently scored components, no single one of which dominates the composite discriminative gradient. Components anchored to fixed patient characteristics, specifically Age and Risk Factors, remain unaffected by shift-related workflow variation. The History and ECG components are the most subjectively scored elements: a prospective comparison of clinician-calculated and researcher-generated HEART scores found overall agreement of 78% (kappa 0.48), with the History component achieving the lowest pairwise kappa of 0.14 and ECG kappa of 0.40. ( 10 ) Despite this component-level variability, inter-rater reliability among emergency physicians, who staff departments continuously across shifts, was shown to be substantial for low-risk classification (kappa 0.68; 84.2% agreement). ( 11 ) The composite score architecture may therefore tolerate component-level variability without proportional degradation of aggregate discriminative performance across shift conditions. The AUC of 0.830 in the off-hours cohort aligns with the Backus 2013 prospective multicenter validation cohort (AUC 0.83 across 10 Dutch centers)and the Six 2013 multinational Asia-Pacific validation of 2,906 patients (AUC 0.83). ( 5 , 9 ) A systematic review and meta-analysis of 25 validation studies and more than 25,000 patients confirmed a pooled sensitivity of 0.96 and NPV of 0.99 for low-risk scores, with a central C-statistic of approximately 0.83 across heterogeneous institutional contexts. ( 4 ) In a direct comparison among 1,748 chest pain patients across nine Dutch hospitals, the HEART score outperformed the GRACE score (AUC 0.73) and TIMI score (AUC 0.80), identifying 381 low-risk patients with only 0.8% missed MACE compared with 2.2% under GRACE at equivalent safety levels. ( 12 ) Higher AUC values have been reported in a US military emergency cohort (0.885), ( 6 ) in institutions operating under formal HEART pathway protocols (0.898), ( 13 ) and in a recent Middle Eastern tertiary validation where baseline MACE prevalence was 24.8% (AUC 0.925). ( 14 ) In contrast, a prospective Tanzanian cohort of 927 patients yielded an AUC of only 0.61, with sensitivity of 59.4% and NPV of 74.7% at the ≥ 4 threshold, attributed to limited ECG interpretation infrastructure and absence of high-sensitivity troponin assays. ( 15 ) This low-income context finding directly contradicts the performance levels seen across high-income settings, confirming that assay quality and clinical interpretation infrastructure, rather than time of presentation, are the principal environmental determinants of HEART score discrimination. Weekend effect meta-analyses in ACS consistently show higher in-hospital mortality among weekend versus weekday admissions (pooled OR 1.06; 95% CI 1.03–1.09 across 18 studies and more than 14 million ACS patients). ( 7 ) A Spanish national registry analysis confirmed 5% higher STEMI mortality and 8% higher NSTEACS mortality on weekends and public holidays after risk adjustment. ( 8 ) These mortality data appear to contradict the present finding of preserved discrimination; however, the weekend mortality effect operates through treatment access delays and reperfusion timing, not through the risk stratification accuracy of a score applied at initial triage. Pre-specified secondary analyses confirmed the primary equivalence finding across multiple comparison frameworks. The Mantel-Haenszel pooled odds ratio for 30-day MACE, with HEART risk group as the stratification variable, was 0.989, indicating near-identical risk within each risk stratum regardless of shift assignment. Calibration metrics were statistically indistinguishable between shift groups, with a calibration intercept difference of -0.0003 (p = 0.998) and a slope difference of 0.070 (p = 0.426), confirming that systematic miscalibration relative to the reference cohort probabilities reflects population-level rather than shift-specific factors. Decision curves demonstrated positive net benefit over treat-all and treat-none strategies across the full pre-specified threshold range of 5–40%, with a net advantage of 0.023 per patient at a threshold of 10% in the off-hours group. In a post-hoc exploratory analysis, the Youden-index optimal threshold was HEART ≥ 6, yielding sensitivity of 75.4% (95% CI 70.3–79.9%) and specificity of 77.5% (95% CI 75.5–79.4%); this threshold was not pre-specified and should be regarded as hypothesis-generating only. These findings carry direct operational consequences for emergency department triage at all hours of presentation. At the rule-out threshold of HEART ≥ 4, the off-hours cohort showed a negative predictive value of 97.7%, with 13 missed events among 574 low-risk patients, a 2.3% miss rate consistent with the accepted bounds for structured early discharge protocols. A stepped-wedge cluster randomized trial of HEART score implementation across nine Dutch emergency departments confirmed that HEART-guided care is safe, with six-week MACE incidence 1.3% lower in the HEART care group than in usual care, within the pre-specified noninferiority margin.( 16 ) The high-risk threshold (HEART ≥ 7) yielded a positive likelihood ratio of 6.07, indicating a clinically meaningful increase in post-test probability; this rule-in performance was preserved across all five time strata. When the HEART pathway is combined with a 0-hour/1-hour high-sensitivity troponin protocol, NPV for 30-day MACE reaches 99.8%, identifying 49.8% of patients for safe rule-out without additional testing. ( 17 ) The AUC of 0.830 at a 14.4% MACE rate is consistent with discrimination expected at this prevalence level across published external validation datasets, ( 13 , 14 ) and the absence of a shift-related AUC difference confirms that off-hours operational constraints do not attenuate the score's triage gradient. Across both shift groups, positive net clinical benefit was observed at all threshold probabilities between 5% and 40%, indicating that neither overtriage nor undertriage is differentially amplified by off-hours conditions. The retrospective single-center design limits generalizability to tertiary academic emergency departments with comparable infrastructure; settings with more severe off-hours constraints or less standardized assay platforms may show different performance. Serial troponin completion was 12.5 percentage points lower during off-hours, and restriction to patients with complete second-troponin documentation raised the off-hours AUC from 0.830 to 0.841, indicating modest downward bias in the primary estimate. The MACE composite was restricted to unplanned coronary revascularization, but procedural urgency was ascertained from administrative records; residual misclassification of elective procedures, if present, would inflate the observed MACE rate and reduce rule-out performance metrics. A mid-study transition from conventional to high-sensitivity troponin I created temporal confounding between assay type and shift assignment; assay-stratified analyses produced overlapping confidence intervals across both platforms, suggesting limited influence on the primary AUC. Observer variability in the History and ECG components is inherent to retrospective HEART score abstraction, though the seniority subgroup analysis produced AUC values of 0.824, 0.833, and 0.827 across resident, specialist, and cardiologist groups, consistent with non-differential misclassification. Thirty-day outcomes were ascertained from institutional records without active patient contact; events occurring at external facilities would not be captured, biasing the observed event rate downward and the negative predictive value upward, with magnitude unquantifiable from available data. The off-hours definition encompassed evenings, nights, weekends, and public holidays, covering approximately 75% of weekly calendar hours and producing an enrollment asymmetry (off-hours 76.2%, on-hours 23.8%); the smaller on-hours group (n = 667) widened its AUC confidence interval (95% CI 0.778–0.865), though the pre-specified equivalence boundary was met with adequate margin. The same sample size asymmetry reduces precision of the on-hours calibration summary estimate; the overall O/E ratio for on-hours (0.791; 95% CI 0.642–0.965) carries substantially wider uncertainty than the off-hours estimate (0.812; 95% CI 0.724–0.908), and between-group calibration comparisons should be interpreted with this limitation in mind. These findings suggest that the HEART score may warrant prospective evaluation as a time-agnostic chest pain triage instrument in multicenter emergency department cohorts spanning diverse institutional and geographic contexts. The stepped-wedge implementation trial confirmed that HEART-guided care is safe but identified physician nonadherence to low-risk discharge recommendations as the principal barrier to realizing its efficiency benefits, ( 16 ) a challenge that may be amplified during off-hours when decision-making operates under greater time pressure. The systematic miscalibration identified relative to the reference cohort probabilities in both shift groups argues for local recalibration of absolute risk estimates before probability-guided triage protocols are implemented. Recalibration of the HEART score troponin threshold to the limit of detection with high-sensitivity assays has been shown to increase sensitivity for rule-out from 96.1% to 98.6% while maintaining specificity, supporting the feasibility of this approach. ( 18 ) A prospective stepped-wedge or cluster-randomized implementation trial, incorporating standardized ECG adjudication by shift category and concurrent troponin assay documentation, is required to determine whether discrimination equivalence persists across centers with different off-hours staffing models. Such a trial should also assess whether locally recalibrated probability thresholds improve absolute risk communication without compromising the safety of threshold-based rule-out. 6. Conclusion In adult emergency department patients presenting with acute chest pain, the HEART score demonstrated clinically equivalent discrimination for 30-day MACE regardless of presentation time, with an AUC of 0.830 (95% CI 0.805–0.855) in off-hours and 0.821 (95% CI 0.778–0.865) in on-hours presentations, both falling within the pre-specified equivalence boundary of ± 0.08. These findings suggest that standard HEART score risk thresholds may be applied without modification across all hours of emergency presentation, including evenings, nights, weekends, and public holidays, without compromising the sensitivity or negative predictive value of the standard rule-out threshold, at tertiary academic centers with comparable operational and assay infrastructure pending prospective multicenter confirmation. Systematic overestimation of absolute MACE risk relative to the reference cohort probabilities was observed equally across shift groups, indicating that local recalibration of predicted risk is needed before probability-guided triage protocols are adopted, irrespective of presentation time. Prospective multicenter validation across diverse emergency department settings, assay types, and shift staffing models is required before these findings can be incorporated into practice guidelines for time-stratified chest pain triage. Abbreviations ACS acute coronary syndrome AUC area under the receiver operating characteristic curve CI confidence interval DCA decision curve analysis ECG electrocardiogram EDACS Emergency Department Assessment of Chest Pain Score FDR false discovery rate GRACE Global Registry of Acute Coronary Events score HEART History,Electrocardiogram,Age,Risk Factors,Troponin score hsTnI high-sensitivity troponin I IQR interquartile range LR likelihood ratio MACE major adverse cardiac event MICE multiple imputation by chained equations NPV negative predictive value NSTEACS non-ST-elevation acute coronary syndrome O/E observed-to-expected ratio OR odds ratio PPV positive predictive value ROC receiver operating characteristic curve SA sensitivity analysis SMD standardised mean difference STEMI ST-elevation myocardial infarction TIMI Thrombolysis in Myocardial Infarction score TRIPOD Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis. Declarations Ethics Approval and Consent to Participate The study was approved by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee (approval number: E-10840098-202.3.02-470; decision number: 90). Individual patient consent was waived in accordance with the retrospective design. Competing Interests: The authors declare that they have no competing interests. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data Availability The data that support the findings of this study are not publicly available due to patient privacy constraints and institutional ethics committee restrictions. De-identified aggregate data and the analysis code are available from the corresponding author upon reasonable request, subject to approval by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee. Author Contributions S.Z.E.K.: Conceptualization, Methodology, Formal analysis, Data curation, Investigation, Resources, Writing – original draft, Writing – review & editing, Visualization, Validation, Project administration. U.D.: Methodology, Software, Investigation, Writing – review & editing, Validation. E.K.: Investigation, Resources, Writing – review & editing. S.B.: Writing – review & editing, Supervision. Acknowledgements None. References Fanaroff AC, Rymer JA, Goldstein SA, Simel DL, Newby LK. Does this patient with chest pain have acute coronary syndrome? the rational clinical examination systematic review. JAMA. 2015;314(18):1955–65. 10.1001/jama.2015.12735 . Six AJ, Backus BE, Kelder JC. Chest pain in the emergency room: value of the HEART score. Neth Heart J. 2008;16(6):191–6. 10.1007/BF03086144 . PMID: 18665203. Backus BE, Six AJ, Kelder JC, et al. Chest pain in the emergency room: a multicenter validation of the HEART score. Crit Pathw Cardiol. 2010;9(3):164–9. 10.1097/HPC.0b013e3181ec36d8 . Laureano-Phillips J, Robinson RD, Aryal S, et al. HEART score risk stratification of low-risk chest pain patients in the emergency department: a systematic review and meta-analysis. Ann Emerg Med. 2019;74(2):187–203. 10.1016/j.annemergmed.2018.12.010 . Six AJ, Cullen L, Backus BE, et al. The HEART score for the assessment of patients with chest pain in the emergency department: a multinational validation study. Crit Pathw Cardiol. 2013;12(3):121–6. 10.1097/HPC.0b013e31828b327e . Streitz MJ, Oliver JJ, Hyams JM, et al. A retrospective external validation study of the HEART score among patients presenting to the emergency department with chest pain. Intern Emerg Med. 2017;13(5):727–48. 10.1007/s11739-017-1743-4 . Kwok CS, Al-Dokheal M, Aldaham S, et al. Weekend effect in acute coronary syndrome: a meta-analysis of observational studies. Eur Heart J Acute Cardiovasc Care. 2018;8(5):432–42. 10.1177/2048872618762634 . Fernández-Ortiz A, Bas Villalobos MC, García-Márquez M, et al. The effect of weekends and public holidays on the care of acute coronary syndrome in the Spanish National Health System. Rev Esp Cardiol (Engl Ed). 2022;75(9):756–62. 10.1016/j.rec.2021.10.022 . Backus BE, Six AJ, Kelder JC, et al. A prospective validation of the HEART score for chest pain patients at the emergency department. Int J Cardiol. 2013;168(3):2153–8. 10.1016/j.ijcard.2013.01.255 . Soares WE, Knee A, Gemme SR, et al. A prospective evaluation of clinical HEART score agreement, accuracy, and adherence in emergency department chest pain patients. Ann Emerg Med. 2021;78(2):231–41. 10.1016/j.annemergmed.2021.03.024 . Gershon CA, Yagapen AN, Lin A, Yanez D, Sun BC. Inter-rater reliability of the HEART score. Acad Emerg Med. 2019;26(5):552–5. 10.1111/acem.13665 . Poldervaart JM, Langedijk M, Backus BE, et al. Comparison of the GRACE, HEART and TIMI score to predict major adverse cardiac events in chest pain patients at the emergency department. Int J Cardiol. 2017;227:656–61. 10.1016/j.ijcard.2016.10.080 . Oliver JJ, Streitz MJ, Hyams JM, et al. An external validation of the HEART pathway among emergency department patients with chest pain. Intern Emerg Med. 2018;13(8):1249–55. 10.1007/s11739-018-1809-y . Nasr Isfahani M, Mohseni H, Nasri Nasrabadi E, Sarrafzadegan N. Improving chest pain risk assessment: validation of HEART, TIMI, GRACE, EDACS-ADP, and HET for MACE prediction in the emergency department. BMC Emerg Med. 2025;25:165. 10.1186/s12873-025-01327-4 . Grisel B, Adisa O, Sakita FM, et al. Evaluating the performance of the HEART score in a Tanzanian emergency department. Acad Emerg Med. 2024;31(4):361–70. 10.1111/acem.14872 . Poldervaart JM, Reitsma JB, Backus BE, et al. Effect of using the HEART score in patients with chest pain in the emergency department: a stepped-wedge, cluster randomized trial. Ann Intern Med. 2017;166(10):689–97. 10.7326/M16-1600 . Nilsson T, Johannesson E, Lundager Forberg J, Mokhtari A, Ekelund U. Diagnostic accuracy of the HEART pathway and EDACS-ADP when combined with a 0-hour/1-hour hs-cTnT protocol for assessment of acute chest pain patients. Emerg Med J. 2021;38(11):808–13. 10.1136/emermed-2020-210833 . Khand AU, Backus B, Campbell M, et al. HEART score recalibration using higher sensitivity troponin T. Ann Emerg Med. 2023;82(4):449–62. 10.1016/j.annemergmed.2023.04.024 . Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 15 May, 2026 Reviews received at journal 27 Apr, 2026 Reviewers agreed at journal 20 Apr, 2026 Reviewers invited by journal 20 Apr, 2026 Editor invited by journal 17 Apr, 2026 Editor assigned by journal 16 Apr, 2026 Submission checks completed at journal 16 Apr, 2026 First submitted to journal 15 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9430666","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":630556014,"identity":"4a6188c2-5802-4e83-949c-54804202acd8","order_by":0,"name":"Sebnem Zeynep Eke Kurt","email":"","orcid":"","institution":"Taksim Eğitim ve Araştırma Hastanesi","correspondingAuthor":false,"prefix":"","firstName":"Sebnem","middleName":"Zeynep Eke","lastName":"Kurt","suffix":""},{"id":630556015,"identity":"fbd5d91c-7276-4a2c-9301-dd95fbc5049c","order_by":1,"name":"Ugur Durmus","email":"","orcid":"","institution":"Medeniyet Üniversitesi Göztepe Eğitim ve Araştırma Hastanesi","correspondingAuthor":false,"prefix":"","firstName":"Ugur","middleName":"","lastName":"Durmus","suffix":""},{"id":630556016,"identity":"66f81886-6377-4938-8aa9-a5f0a6de6008","order_by":2,"name":"Erdem Kurt","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxUlEQVRIiWNgGAWjYHACAwaGA0CKvbnhAGlaeHgOkqxFIrGBOPX80oc3Prpx5nC+veTDxsM8FQzy/GIELJPsSys2zrlx2LJHOrHhMM8ZBsOZsxMIuOoMj5l0zofDBjxALQdntjEkGNwmoMX+DI/5b7AWyYNALf+I0GLAw2PGDHSYAY8EY8OBjw1EaJE4w1YsnXMm3YDnTGLDgQ/HJAj7hb+HeePnnGPWBuzthw9/SKixkeeXJqAFw1bSlI+CUTAKRsEowA4AP7dF06t93WgAAAAASUVORK5CYII=","orcid":"","institution":"İstanbul Eğitim ve Araştırma Hastanesi","correspondingAuthor":true,"prefix":"","firstName":"Erdem","middleName":"","lastName":"Kurt","suffix":""},{"id":630556018,"identity":"f1919da5-f9e2-42a8-a67a-79e5aa4e7a48","order_by":3,"name":"Suphi Bahadırlı","email":"","orcid":"","institution":"Istanbul Medipol University","correspondingAuthor":false,"prefix":"","firstName":"Suphi","middleName":"","lastName":"Bahadırlı","suffix":""}],"badges":[],"createdAt":"2026-04-15 19:38:36","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9430666/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9430666/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108097777,"identity":"657b249e-e54b-44e5-b4cc-e90938d31752","added_by":"auto","created_at":"2026-04-29 10:17:01","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":184160,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eROC curves and calibration plots for the HEART score against 30-day MACE. Left: Receiver operating characteristic curves for off-hours (n=2,133; solid blue) and on-hours (n=667; dashed red) groups. AUC estimated by DeLong method. Right: Calibration curves of observed versus predicted MACE probability by HEART score group; bubble size proportional to group size; dashed diagonal = perfect calibration. Predicted probabilities from the reference cohort. Both groups demonstrate systematic overestimation of absolute risk (O/E \u0026lt;1). MACE = major adverse cardiac event; AUC = area under the ROC curve.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9430666/v1/f8a744a1fc9100dab88fa624.png"},{"id":108181556,"identity":"c985613c-319e-4b2b-b8b0-fc2e2b7b0991","added_by":"auto","created_at":"2026-04-30 08:58:45","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":151516,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eDecision curves for the HEART score model compared with treat-all and treat-none reference strategies. Left: Off-hours group (n=2,133). Right: On-hours group (n=667). X-axis: clinician's threshold probability for treatment (hospitalisation/further workup); Y-axis: net benefit per patient evaluated. The HEART score model provides positive net benefit over both reference strategies across the pre-specified threshold range (5-40%). DCA = decision curve analysis.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9430666/v1/8058dfe80984cbde7df55959.png"},{"id":108097779,"identity":"e105d05f-2154-4c7c-9fb4-114d9776208e","added_by":"auto","created_at":"2026-04-29 10:17:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":144961,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eForest plot of sensitivity analyses. Primary analysis shown in dark blue; sensitivity analyses in lighter shades. Error bars represent DeLong 95% confidence intervals. The dashed vertical line marks the primary AUC estimate (0.830). All sensitivity analyses produced AUC estimates within the pre-specified clinical equivalence range, confirming robustness of primary findings. SA = sensitivity analysis; MICE = multiple imputation by chained equations; hsTnI = high-sensitivity troponin I.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9430666/v1/acbf0a318d8db7a4962f40bf.png"},{"id":108491232,"identity":"d429dd16-3a6b-4aaf-b98b-5c6eda6788d8","added_by":"auto","created_at":"2026-05-05 09:52:59","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":826055,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9430666/v1/772ac1c0-ffaa-4ad6-be6c-c1d9dbbb173d.pdf"},{"id":108182044,"identity":"bee08358-82c1-47fb-b61c-c1ee40f35080","added_by":"auto","created_at":"2026-04-30 08:59:06","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":381178,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9430666/v1/4eb1eaf12bf1db42c336342e.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Time-Stratified HEART Score Discrimination and Calibration in Emergency Chest Pain: A Retrospective Cohort Study","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAcute chest pain accounts for approximately 5\u0026ndash;10% of all adult emergency department visits and represents one of the most demanding diagnostic challenges in emergency medicine, as approximately 10% of these patients ultimately receive a diagnosis of acute coronary syndrome while the majority are attributed to non-cardiac causes.(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) Identifying that minority accurately at triage without subjecting every patient to prolonged observation or invasive workup requires a standardized, rapidly applicable risk stratification instrument. The HEART score, integrating History, Electrocardiogram, Age, Risk Factors, and Troponin into a 10-point composite, was developed for this purpose (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) and subsequently validated in a multicenter retrospective cohort of 880 patients, demonstrating that scores of 0\u0026ndash;3 were associated with a MACE rate below 1%. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) Subsequent prospective validation across 2,440 patients in 10 Dutch emergency departments confirmed a C-statistic of 0.83, with 30-day MACE rates below 2% in the low-risk stratum and exceeding 50% in the high-risk stratum. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) These risk thresholds have since anchored clinical disposition decisions in emergency departments across multiple countries. (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eExternal validation studies have confirmed HEART score discrimination in Asian emergency populations, with a C-statistic of 0.83 across 14 hospitals and 2,906 patients, and in retrospective military emergency cohorts, where AUCs of 0.885\u0026ndash;0.898 were reported. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e) A systematic review and meta-analysis incorporating 25 validation studies and more than 25,000 patients demonstrated a pooled sensitivity of 0.96 and negative predictive value of 0.99 for low-risk scores, confirming a consistent safety profile for rule-out. (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) However, to the best of our knowledge, no published validation study has stratified HEART score performance by time of emergency presentation. This gap is clinically relevant: off-hours admissions for acute coronary syndrome are associated with a small but persistent increase in in-hospital mortality, longer door-to-treatment intervals, and reduced specialist availability.(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e) Whether these operational differences translate into measurable differences in HEART score discriminative performance or calibration has not been formally examined in the published literature.\u003c/p\u003e \u003cp\u003eThis study aimed to evaluate and compare the discriminative performance and calibration of the HEART score for 30-day MACE prediction in adult patients presenting with acute chest pain during off-hours versus standard working hours at a tertiary care emergency department. Secondary objectives included assessment of HEART score calibration relative to published validation cohort reference probabilities within each shift group, comparison of threshold-based rule-out and rule-in performance at pre-specified clinical cut points, evaluation of discrimination across five granular time strata, and decision curve analysis across a clinically relevant threshold probability range.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study Design and Setting\u003c/h2\u003e \u003cp\u003e This single-center retrospective cohort study was conducted in the emergency department of Istanbul Medipol Mega University Hospital, a tertiary care academic center in Istanbul, Turkey, between January 2020 and December 2025. The study constituted an external validation of the HEART score for 30-day MACE prediction and was classified as TRIPOD Type 2b/3. The study was approved by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee (approval number: E-10840098-202.3.02-470; decision number: 90). Individual patient consent was waived in accordance with the retrospective design. A systematic search of PubMed and Embase (January 2000 through December 2024) using the terms 'HEART score', 'off-hours', 'night shift', 'weekend', and 'time of presentation' identified no prior study that stratified HEART score validation by time of emergency presentation, supporting the originality of the research question.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Participants\u003c/h2\u003e \u003cp\u003eAdult patients aged 18 years or older presenting to the emergency department with acute non-traumatic chest pain as the primary complaint were eligible for inclusion, provided that a first-visit triage timestamp was documented. Patients were excluded if HEART score documentation was incomplete (n\u0026thinsp;=\u0026thinsp;298), if they presented with ST-elevation myocardial infarction requiring immediate primary percutaneous coronary intervention bypassing standard triage stratification (n\u0026thinsp;=\u0026thinsp;1163), if they were transferred from another institution with an established cardiac diagnosis (n\u0026thinsp;=\u0026thinsp;45), or if 30-day follow-up could not be ascertained (n\u0026thinsp;=\u0026thinsp;167), yielding an analytic cohort of 2,800 patients from 4473 screened. The unit of analysis was the individual emergency presentation; the first-visit flag confirmed a first-presentation design with no repeated-measures clustering.\u003c/p\u003e \u003cp\u003eThe derivation of the analytic cohort from the screened population is illustrated in Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3.Variables\u003c/h2\u003e \u003cp\u003e \u003cb\u003ePrimary exposure.\u003c/b\u003e The triage timestamp was used to classify each presentation as off-hours or on-hours. Off-hours encompassed evenings (18:00\u0026ndash;23:59), nights (00:00\u0026ndash;07:59), weekends (Saturday and Sunday, all hours), and public holidays. On-hours encompassed regular weekday working hours (08:00\u0026ndash;17:59, Monday through Friday, excluding public holidays).\u003c/p\u003e \u003cp\u003eUnder this definition, off-hours encompassed approximately 75% of total weekly hours, reflecting the predominance of evenings, nights, weekends, and public holidays in the calendar; the resulting 76.2% off-hours enrollment fraction is therefore consistent with the expected distribution of presentations across this time window.\u003c/p\u003e \u003cp\u003e \u003cb\u003eIndex test.\u003c/b\u003e The HEART score was calculated from five components, each scored 0\u0026ndash;2: History, Electrocardiogram, Age, Risk Factors, and Troponin, yielding a total score of 0\u0026ndash;10. (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) Patients were classified as low risk (score 0\u0026ndash;3), intermediate risk (score 4\u0026ndash;6), or high risk (score 7\u0026ndash;10). Predicted probabilities for calibration analyses were derived from the Backus 2013 prospective validation cohort reference values (hereafter \"the reference cohort\"), low-risk group 0.017, intermediate-risk group 0.120, and high-risk group 0.652. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e) Logit-scale values of these probabilities were pre-computed for offset regression analyses.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePrimary outcome.\u003c/b\u003e The primary outcome was 30-day MACE, defined as a composite of acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death occurring within 30 days of the index emergency presentation, ascertained through hospital readmission records, cardiology outpatient records, and institutional mortality registries. Elective revascularization procedures were excluded from the composite endpoint.\u003c/p\u003e \u003cp\u003e \u003cb\u003eCovariates.\u003c/b\u003e The first troponin measurement (Troponin-1, ng/L) was recorded at triage. Door-to-diagnosis time was defined as the interval in minutes from triage timestamp to documented clinical diagnosis. Troponin assay type (high-sensitivity troponin I versus conventional troponin I) and ECG interpreter seniority (resident, specialist, or cardiologist) were recorded as secondary variables. Comorbidities (hypertension, diabetes mellitus, hyperlipidemia, current smoking, obesity) were ascertained from the electronic medical record.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Data Sources and Measurement\u003c/h2\u003e \u003cp\u003eClinical data were extracted from the institutional electronic medical record system using a standardized, pre-specified data collection protocol. HEART score components were documented by the treating emergency physician at the time of presentation and abstracted verbatim from the medical record. The troponin assay transitioned from conventional troponin I to high-sensitivity troponin I during the study period; assay type was recorded per patient and addressed in a pre-specified sensitivity analysis. ECG interpretation was performed by the treating clinician and the interpreter's seniority level was recorded. The 30-day follow-up was ascertained from institutional records without active patient contact.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Bias\u003c/h2\u003e \u003cp\u003eFour principal sources of bias were addressed prospectively. Information bias arising from retrospective medical record abstraction was minimized through a standardized extraction protocol applied uniformly across both shift groups; residual observer-dependent variability in the History component of the HEART score cannot be fully quantified and may have attenuated that component's contribution to overall score accuracy, with an expected non-differential direction of effect. Temporal confounding from the mid-study transition between troponin assay platforms affects the Troponin component of the HEART score and was addressed in Sensitivity Analysis 2 (SA2), which stratified the primary discrimination and calibration analyses by assay type; overlapping confidence intervals across assay subgroups indicate that this transition did not materially bias the primary estimate. Selection bias from differential serial troponin completion between shift groups (off-hours 78.8%, on-hours 91.3%) represents the most consequential source of potential outcome misclassification; Sensitivity Analysis 1 (SA1) restricted the primary analysis to patients with complete original serial troponin measurements to characterize the direction and magnitude of this effect. Outcome ascertainment bias arising from the use of institutional records without active patient contact represents a fourth potential source of error; patients experiencing MACE at external facilities would not be captured, biasing the observed event rate toward underestimation and the negative predictive value toward overestimation, with the magnitude of this effect unquantifiable from available data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Study Size\u003c/h2\u003e \u003cp\u003eAs this was a retrospective study, no prospective power calculation was performed. All patients meeting eligibility criteria during the study period were included. The final analytic cohort comprised 2,800 patients, of whom 402 experienced the primary outcome (30-day MACE rate 14.4%). The off-hours group included 2,133 patients with 305 events and the on-hours group included 667 patients with 97 events. For the primary secondary estimand, the DeLong confidence interval for the AUC difference spanned\u0026thinsp;\u0026minus;\u0026thinsp;0.041 to +\u0026thinsp;0.058, confirming adequate precision for the pre-specified clinical equivalence boundary of \u0026plusmn;\u0026thinsp;0.08.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Statistical Analysis\u003c/h2\u003e \u003cp\u003e\u003cb\u003eNormality assessment.\u003c/b\u003e For samples with n\u0026thinsp;\u0026ge;\u0026thinsp;50, normality was assessed using a combined multi-criterion protocol (pre-specified multi-criterion normality protocol): a variable was classified as non-normal if at least three of five criteria indicated departure from normality, comprising the Kolmogorov-Smirnov test with Lilliefors correction (p\u0026thinsp;\u0026le;\u0026thinsp;0.05), skewness z-score\u0026thinsp;\u0026ge;\u0026thinsp;1.96, kurtosis z-score\u0026thinsp;\u0026ge;\u0026thinsp;1.96, Q-Q plot visual inspection, and histogram shape assessment. This multi-criterion approach was defined a priori by the investigators; no single published guideline mandates this specific combination of criteria. For samples with n\u0026thinsp;\u0026lt;\u0026thinsp;50, the Shapiro-Wilk test was applied. All primary continuous variables (age, HEART score, troponin-1, door-to-diagnosis time) met the threshold for non-normal distribution under this protocol.\u003c/p\u003e \u003cp\u003e \u003cb\u003eDescriptive statistics and group comparisons.\u003c/b\u003e Continuous variables are reported as median (interquartile range) and compared between groups using the Mann-Whitney U test. Categorical variables are reported as n (%) and compared using the chi-square test with continuity correction for 2\u0026times;2 tables or Fisher's exact test where any expected cell count was below 5. For the three-level HEART risk group variable, a 3\u0026times;2 chi-square test was applied with the global p-value reported. Standardized mean differences (SMD) were computed for all Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e variables as a p-value-independent measure of group balance; an SMD exceeding 0.10 was pre-specified as indicating clinically meaningful imbalance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline characteristics by shift group.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOff-hours (n\u0026thinsp;=\u0026thinsp;2,133)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOn-hours (n\u0026thinsp;=\u0026thinsp;667)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSMD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003eContinuous variables\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge, median (IQR), years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e59 (49\u0026ndash;70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59 (49\u0026ndash;67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.067\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.089\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHEART total score, median (IQR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5 (\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.014\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTroponin-1, median (IQR), ng/L\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18.4 (7.5\u0026ndash;54.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17.7 (6.6\u0026ndash;48.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.345\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDoor-to-diagnosis time, median (IQR), min\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e168 (127\u0026ndash;202)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e146 (104\u0026ndash;184)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.351*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCategorical variables\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale sex, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,220 (57.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e380 (57.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.954\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMACE at 30 days, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e305 (14.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e97 (14.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcute myocardial infarction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e228 (10.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e74 (11.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.013\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRevascularisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e157 (7.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (6.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.653\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCardiac arrest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38 (1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.256\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.062\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeath\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39 (1.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13 (1.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.970\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMACE at 6 weeks, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e332 (15.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e104 (15.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHEART risk group, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.301\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow risk (score 0\u0026ndash;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e574 (26.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e196 (29.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntermediate risk (score 4\u0026ndash;6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,223 (57.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e360 (54.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh risk (score 7\u0026ndash;10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e336 (15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e111 (16.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eComorbidities\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e960 (45.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e275 (41.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.076\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes mellitus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e513 (24.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e161 (24.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHyperlipidaemia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e826 (38.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e262 (39.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCurrent smoking\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e559 (26.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e205 (30.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.100*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eObesity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e539 (25.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e150 (22.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.065\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eProcess indicators\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOriginal serial troponin complete, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,681 (78.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e609 (91.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.356*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ehsTnI assay used, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,458 (68.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e458 (68.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.007\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHospitalisation, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e534 (25.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e170 (25.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.854\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.010\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCardiology consultation, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e606 (28.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e184 (27.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.716\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.018\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eIQR\u0026thinsp;=\u0026thinsp;interquartile range; MACE\u0026thinsp;=\u0026thinsp;major adverse cardiac event; hsTnI\u0026thinsp;=\u0026thinsp;high-sensitivity troponin I; SMD\u0026thinsp;=\u0026thinsp;standardised mean difference. Continuous variables compared by Mann-Whitney U test (non-normal distribution confirmed by pre-specified five-criterion normality protocol). Categorical variables compared by chi-square with continuity correction; Fisher's exact test where expected cell count\u0026thinsp;\u0026lt;\u0026thinsp;5. * SMD\u0026thinsp;\u0026gt;\u0026thinsp;0.10 indicates clinically meaningful imbalance. Individual MACE components are not mutually exclusive; a single patient may contribute to more than one component category.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eDiscrimination.\u003c/b\u003e The area under the receiver operating characteristic curve (AUC, equivalent to the C-statistic) was computed using the DeLong method, which provides exact confidence intervals without bootstrap approximation. Threshold-based performance metrics (sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio) were computed at two pre-specified clinical thresholds: HEART\u0026thinsp;\u0026ge;\u0026thinsp;4 (standard rule-out threshold) and HEART\u0026thinsp;\u0026ge;\u0026thinsp;7 (high-risk admission threshold). Wilson score confidence intervals were applied to all proportion-based metrics. The Youden index (J\u0026thinsp;=\u0026thinsp;sensitivity\u0026thinsp;+\u0026thinsp;specificity\u0026thinsp;\u0026minus;\u0026thinsp;1) was computed across all score values to identify a data-driven optimal cutpoint, labelled exploratory and post-hoc as pre-specification.\u003c/p\u003e \u003cp\u003e\u003cb\u003eCalibration.\u003c/b\u003e Calibration was assessed using five complementary metrics: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) the Hosmer-Lemeshow goodness-of-fit test, with one group per HEART score value (11 groups, score 0 through 10; df\u0026thinsp;=\u0026thinsp;9 under the H-L formula df\u0026thinsp;=\u0026thinsp;g \u0026minus;\u0026thinsp;2); (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) the observed-to-expected (O/E) ratio overall and within each risk stratum, with Byar approximation for 95% confidence intervals; (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) the calibration intercept, estimated by offset logistic regression with the logit of reference cohort predicted probabilities as a fixed offset term (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) the calibration slope, estimated by logistic regression of the observed outcome on the logit of predicted probabilities, with bootstrap confidence intervals (1000 iterations; seed\u0026thinsp;=\u0026thinsp;2024) and a two-sided z-test against the null hypothesis of slope\u0026thinsp;=\u0026thinsp;1; and (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e) the Brier score and scaled Brier score, quantifying overall predictive skill relative to the null model.\u003c/p\u003e \u003cp\u003e \u003cb\u003eDecision curve analysis.\u003c/b\u003e Net benefit was computed across a pre-specified threshold probability range of 5\u0026ndash;40% and compared against treat-all and treat-none reference strategies. Separate decision curves were produced for off-hours and on-hours groups.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePrimary comparative analysis.\u003c/b\u003e The primary secondary estimand was the difference in AUC between independent off-hours and on-hours groups, computed using the DeLong variance estimator for independent samples. Clinical equivalence was defined a priori as ΔAUC\u0026thinsp;\u0026lt;\u0026thinsp;0.08, pre-specified before data acces. A bootstrap confidence interval for ΔAUC was generated as a supplementary check (1,000 iterations; seed\u0026thinsp;=\u0026thinsp;2024; percentile method). A formal interaction test was conducted using logistic regression including the mean-centered HEART score, off-hours status, and their product term; the interaction p-value served as the pre-specified test of differential discrimination between groups. Within each HEART risk group, 30-day MACE rates were compared between shift groups using chi-square with continuity correction; the Mantel-Haenszel pooled odds ratio was computed with HEART risk group as the stratification variable.\u003c/p\u003e \u003cp\u003e \u003cb\u003eTime strata analysis.\u003c/b\u003e Five time strata were defined: on-hours (reference), evening, night, weekend daytime, and public holidays. An AUC was estimated within each stratum using the DeLong method. Heterogeneity across strata was assessed using a weighted Cochran Q statistic; pairwise comparisons against the on-hours reference were planned only in the event of a significant global test (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), applying Benjamini-Hochberg false discovery rate correction.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSensitivity analyses.\u003c/b\u003e Six pre-specified sensitivity analyses were performed as pre-specified. SA1 restricted the analysis to patients with complete original serial troponin measurements, isolating the potential bias introduced by imputed or simulated second troponin values. SA2 stratified the off-hours cohort by troponin assay type (high-sensitivity troponin I versus conventional troponin I) and repeated the primary discrimination and calibration analyses within each subgroup, addressing the temporal confounding effect of assay migration on the T-component of HEART. SA3 stratified by the seniority of the clinician who interpreted the ECG (resident, specialist, or cardiologist), testing whether off-hours degradation in ECG interpretation quality affects E-component scoring and downstream model performance. SA4 replaced the 30-day MACE endpoint with the 6-week MACE outcome, aligning the follow-up window with the reference cohort and testing sensitivity of findings to outcome ascertainment duration. SA5 served as a data quality verification step confirming the first-visit-only design; by construction, all 2,800 patients carried a first-presentation flag, rendering this analysis identical to the primary. SA6 used MICE-imputed HEART scores in place of original scores (20 imputation sets, 50 iterations, seed\u0026thinsp;=\u0026thinsp;2024, Rubin's rules applied); missing data were confined to components with original documentation gaps and no outcome imputation was performed. Full results for all sensitivity analyses are presented in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eAll analyses were conducted in Python 3.11 (scipy 1.11+, statsmodels 0.14+, scikit-learn 1.3+). Statistical significance was set at p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 (two-tailed) for the primary outcome. All point estimates are accompanied by 95% confidence intervals.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8. Missing Data\u003c/h2\u003e \u003cp\u003ePrimary analysis variables (HEART total score, 30-day MACE, first troponin, and reference cohort predicted probabilities) contained no missing values. A complete-case flag identified 2,080 of 2,800 patients (74.3%) with complete data across all secondary variables. Patients with incomplete secondary variable data were excluded from the specific analyses requiring those variables; denominators are reported explicitly for each analysis step. Multiple imputation was applied only in SA6, using pre-computed MICE scores from the dataset. The 720 incomplete cases were driven by three variables: original serial troponin documentation (n\u0026thinsp;=\u0026thinsp;510; 18.2%), family history (n\u0026thinsp;=\u0026thinsp;180; 6.4%), and original ECG interpreter records (n\u0026thinsp;=\u0026thinsp;86; 3.1%). Serial troponin missingness was differential by shift (off-hours 21.2% versus on-hours 8.7%), consistent with missing at random conditional on shift assignment and directly addressed by SA1; family history missingness was non-differential across shift groups (6.4% versus 6.6%), consistent with missing completely at random. ECG interpreter seniority was fully documented for all 2,800 patients (0 missing values), confirming that the SA3 subgroup counts are not affected by secondary variable incompleteness..\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.9. Outcomes\u003c/h2\u003e \u003cp\u003eThe primary outcome was 30-day MACE, defined as a composite of acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death within 30 days of the index presentation, pre-specified before data analysis. Secondary outcomes included MACE at 6 weeks, individual MACE components (reported separately), and HEART score discrimination stratified by time stratum. The primary secondary estimand was the difference in AUC between off-hours and on-hours groups. All analyses were pre-specified before data access; no post-hoc modifications to any estimand were made after data access, and all other analyses are reported as exploratory.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study cohort\u003c/h2\u003e \u003cp\u003eThe analytic cohort comprised 2,800 patients presenting with acute chest pain \u003cb\u003e(Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e)\u003c/b\u003e, of whom 2,133 (76.2%) presented during off-hours and 667 (23.8%) during on-hours. Overall, 402 patients experienced a 30-day MACE (14.4%). All entries in the dataset carried a first-visit flag, confirming the absence of repeated-measure clustering.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Baseline characteristics\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents baseline characteristics by off-hours and on-hours status. The two groups were well-matched on age (both median 59 years), MACE rate (14.3% versus 14.5%, p\u0026thinsp;=\u0026thinsp;0.926), sex distribution (57.2% versus 57.0% male), HEART score (median 5 versus 4, p\u0026thinsp;=\u0026thinsp;0.604), first troponin (18.4 versus 17.7 ng/L, p\u0026thinsp;=\u0026thinsp;0.345), and comorbidity profiles. Risk group distribution was also comparable (chi-square p\u0026thinsp;=\u0026thinsp;0.301). Assay type was balanced: hsTnI was used in 68.4% of off-hours and 68.7% of on-hours patients (p\u0026thinsp;=\u0026thinsp;0.918).\u003c/p\u003e \u003cp\u003eThree variables showed clinically meaningful imbalance (SMD\u0026thinsp;\u0026gt;\u0026thinsp;0.10). Door-to-diagnosis time was longer in off-hours patients (median 168 versus 146 minutes, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, SMD\u0026thinsp;=\u0026thinsp;0.351), reflecting reduced specialist availability outside working hours. Serial troponin completion was lower in off-hours patients (78.8% versus 91.3%, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, SMD\u0026thinsp;=\u0026thinsp;0.356). Smoking prevalence also showed clinically meaningful imbalance (26.2% versus 30.7%, p\u0026thinsp;=\u0026thinsp;0.025, SMD\u0026thinsp;=\u0026thinsp;0.100). These imbalances are addressed in the relevant sensitivity analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Primary validation in the off-hours cohort\u003c/h2\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1. Discrimination\u003c/h2\u003e \u003cp\u003eIn the off-hours group (n\u0026thinsp;=\u0026thinsp;2,133; 305 MACE events, 14.3%), the HEART score achieved an AUC of 0.830 (95% CI: 0.805\u0026ndash;0.855), confirming good discriminative performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, left panel; Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDiscrimination metrics at pre-specified and exploratory thresholds (off-hours cohort, n\u0026thinsp;=\u0026thinsp;2,133).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHEART\u0026thinsp;\u0026ge;\u0026thinsp;4 (standard)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHEART\u0026thinsp;\u0026ge;\u0026thinsp;7 (high-risk)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHEART\u0026thinsp;\u0026ge;\u0026thinsp;6 (Youden)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e95% CI method\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAUC (C-statistic)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003cp\u003e(0.805\u0026ndash;0.855)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e95.7% (92.8\u0026ndash;97.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e55.4% (49.8\u0026ndash;60.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e75.4% (70.3\u0026ndash;79.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWilson\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30.7% (28.6\u0026ndash;32.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e90.9% (89.5\u0026ndash;92.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e77.5% (75.5\u0026ndash;79.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWilson\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18.7% (16.9\u0026ndash;20.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50.3% (45.0-55.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35.9% (32.3\u0026ndash;39.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWilson\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e97.7% (96.2\u0026ndash;98.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.4% (91.1\u0026ndash;93.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e95.0% (93.7\u0026ndash;96.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eWilson\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.139\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.491\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.317\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTP / FP / TN / FN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e292/1267/561/13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e169/167/1661/136\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e230/411/1417/75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eN\u0026thinsp;=\u0026thinsp;2,133 off-hours patients; 305 MACE events (14.3%). Threshold\u0026thinsp;\u0026ge;\u0026thinsp;4: standard clinical rule-out. Threshold\u0026thinsp;\u0026ge;\u0026thinsp;7: high-risk admission rule-in. Threshold\u0026thinsp;\u0026ge;\u0026thinsp;6: Youden-index optimal (post hoc, data-driven; labelled exploratory, as pre-specified). AUC computed by DeLong method. PPV\u0026thinsp;=\u0026thinsp;positive predictive value; NPV\u0026thinsp;=\u0026thinsp;negative predictive value; LR\u0026thinsp;=\u0026thinsp;likelihood ratio.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAt the standard clinical rule-out threshold (HEART\u0026thinsp;\u0026ge;\u0026thinsp;4), sensitivity was 95.7% (95% CI: 92.8\u0026ndash;97.5%) and NPV was 97.7% (95% CI: 96.2\u0026ndash;98.7%), with 13 missed events among 574 patients below the threshold. At the high-risk admission threshold (HEART\u0026thinsp;\u0026ge;\u0026thinsp;7), specificity was 90.9% (95% CI: 89.5\u0026ndash;92.1%) and PPV was 50.3% (95% CI: 45.0-55.6%). The positive likelihood ratio at this threshold was 6.07, indicating a clinically meaningful shift in post-test probability.\u003c/p\u003e \u003cp\u003eThe Youden-index optimal threshold was HEART\u0026thinsp;\u0026ge;\u0026thinsp;6 (J\u0026thinsp;=\u0026thinsp;0.529), with sensitivity 75.4% (95% CI: 70.3\u0026ndash;79.9%) and specificity 77.5% (95% CI: 75.5\u0026ndash;79.4%). This threshold was identified post hoc and is labelled exploratory, as pre-specified.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2. Calibration\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the complete calibration profile. The Hosmer-Lemeshow test was highly significant (χ\u0026sup2;=104.32, df\u0026thinsp;=\u0026thinsp;9, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), indicating systematic departure from the reference cohort probabilities. The overall O/E ratio of 0.812 indicates that the Backus model predicted approximately 23% more MACE events (predicted N\u0026thinsp;=\u0026thinsp;375.6) than were observed (N\u0026thinsp;=\u0026thinsp;305). This pattern of overestimation was consistent across all three risk strata, with O/E values of 1.33 in the low-risk group (confidence interval crossing 1.0, reflecting sparse events), 0.84 in the intermediate-risk group, and 0.77 in the high-risk group.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCalibration metrics for off-hours and on-hours groups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCalibration metric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOff-hours (n\u0026thinsp;=\u0026thinsp;2,133)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOn-hours (n\u0026thinsp;=\u0026thinsp;667)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDifference\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHosmer-Lemeshow χ\u0026sup2; (df\u0026thinsp;=\u0026thinsp;9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e104.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001 / 0.052\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHosmer-Lemeshow p-value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.052\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eO/E ratio, overall (Byar 95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.812 0.724 to 0.908)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.791 (0.642 to 0.965)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow-risk group (HEART 0\u0026ndash;3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.332 (0.665\u0026ndash;2.278)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.625 (0.121\u0026ndash;1.821)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntermediate-risk group (HEART 4\u0026ndash;6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.838 (0.693-1.000)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.860 (0.629\u0026ndash;1.141)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh-risk group (HEART 7\u0026ndash;10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.771 (0.657\u0026ndash;0.897)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.747 (0.560\u0026ndash;0.977)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCalibration intercept (95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.352 (-0.495 to -0.208)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.351 (-0.608 to -0.094)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.998\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCalibration slope (bootstrap 95% CI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.828 (0.732\u0026ndash;0.916)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.758 (0.624\u0026ndash;0.910)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.070\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.426\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBrier score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.101\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.109\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScaled Brier score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.176\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.125\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eχ\u003cem\u003e\u0026sup2; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 indicates systematic miscalibration relative to reference cohort probabilities. O/E\u0026thinsp;\u0026lt;\u0026thinsp;1 indicates the model overestimates absolute MACE risk in this population. Calibration intercept tested H₀: intercept\u0026thinsp;=\u0026thinsp;0; slope tested H₀: slope\u0026thinsp;=\u0026thinsp;1. Difference and p-value for intercept and slope based on z-test using bootstrap standard errors; all other metrics descriptive only. O/E\u0026thinsp;=\u0026thinsp;observed-to-expected ratio.\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe calibration intercept was \u0026minus;\u0026thinsp;0.352 (95% CI: -0.495 to -0.208; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), indicating that the reference model systematically overestimates absolute MACE risk in this cohort, consistent with a lower baseline event rate than the reference population. The calibration slope was 0.828 (bootstrap 95% CI: 0.732\u0026ndash;0.916; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for deviation from 1.0), indicating miscalibration in the extremes: the reference model overestimates risk in higher-scoring patients and underestimates it in lower-scoring patients relative to what is observed in this cohort. The calibration curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, right panel) illustrates both the downward shift and the reduced steepness of observed versus predicted risk across score values.\u003c/p\u003e \u003cp\u003eThe Brier score was 0.101 and the scaled Brier score was 0.176, reflecting meaningful predictive skill above the null (scaled Brier 0\u0026thinsp;=\u0026thinsp;no skill; 1\u0026thinsp;=\u0026thinsp;perfect). Recalibration is indicated by the significant intercept deviation and will be reported separately.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e3.3.3. Decision curve analysis\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows decision curves for off-hours and on-hours groups across threshold probabilities of 5\u0026ndash;40%. The HEART score model delivered positive net benefit exceeding both the treat-all and treat-none strategies across the full prespecified threshold range in both groups. At a threshold of 10%, net benefit was 0.071 in off-hours patients versus a treat-all net benefit of 0.048, a net advantage of 0.023 per patient evaluated, equivalent to avoiding approximately 23 unnecessary hospitalizations per 1,000 patients assessed. The HEART score thus provides clinical utility across the range of plausible clinical decision thresholds, with no evidence of differential performance between shift groups.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Comparative analysis\u003c/h2\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003ch2\u003e3.4.1. AUC comparison between groups\u003c/h2\u003e \u003cp\u003eThe primary secondary estimand, the difference in AUC between off-hours and on-hours groups, was ΔAUC\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.009 (95% CI: -0.041 to +\u0026thinsp;0.058; p\u0026thinsp;=\u0026thinsp;0.736) in favour of off-hours patients (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The bootstrap-derived 95% CI was \u0026minus;\u0026thinsp;0.040 to +\u0026thinsp;0.057. Both the point estimate and the entire confidence interval fell well within the pre-specified clinical equivalence boundary of \u0026plusmn;\u0026thinsp;0.08, providing clear evidence that HEART score discrimination does not meaningfully differ by shift assignment. The AUC in the on-hours group was 0.821 (95% CI: 0.778\u0026ndash;0.865), consistent with the off-hours estimate.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAUC comparison between off-hours and on-hours groups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnalysis\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOff-hours AUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOn-hours AUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eΔAUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrimary (30-day MACE)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003cp\u003e(0.805\u0026ndash;0.855)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.821\u003c/p\u003e \u003cp\u003e(0.778\u0026ndash;0.865)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.009\u003c/p\u003e \u003cp\u003e(-0.041 to +\u0026thinsp;0.058)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.736\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBootstrap ΔAUC (1,000 iterations;\u003c/p\u003e \u003cp\u003eseed\u0026thinsp;=\u0026thinsp;2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e(-0.040 to +\u0026thinsp;0.057)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6-week MACE (SA4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.818\u003c/p\u003e \u003cp\u003e(0.793\u0026ndash;0.843)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.811\u003c/p\u003e \u003cp\u003e(0.767\u0026ndash;0.855)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.007\u003c/p\u003e \u003cp\u003e(-0.044 to +\u0026thinsp;0.058)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.790\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eΔAUC\u0026thinsp;=\u0026thinsp;off-hours minus on-hours; positive values favour off-hours. DeLong method (independent groups) for primary comparison. Clinical equivalence boundary: ΔAUC\u0026thinsp;\u0026lt;\u0026thinsp;0.08 (pre-specified before data access). MACE\u0026thinsp;=\u0026thinsp;major adverse cardiac event.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e3.4.2. Interaction analysis\u003c/h2\u003e \u003cp\u003eThe formal interaction model confirmed the absence of differential HEART score effects by shift. The interaction term (mean-centred HEART score \u0026times; off-hours indicator) had OR\u0026thinsp;=\u0026thinsp;1.039 (95% CI: 0.867\u0026ndash;1.247; p\u0026thinsp;=\u0026thinsp;0.676), providing no evidence that the HEART score's log-odds slope for MACE differs between off-hours and on-hours patients. The main effect of the HEART score was OR\u0026thinsp;=\u0026thinsp;2.166 per unit increase (95% CI: 1.851\u0026ndash;2.533; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), consistent with the overall discriminative performance. The off-hours main effect was OR\u0026thinsp;=\u0026thinsp;0.955 (95% CI: 0.655\u0026ndash;1.393; p\u0026thinsp;=\u0026thinsp;0.811), confirming no independent effect of shift on MACE risk after accounting for HEART score.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section3\"\u003e \u003ch2\u003e3.4.3. MACE rates by risk group\u003c/h2\u003e \u003cp\u003eSupplementary Table S3 presents 30-day MACE rates within each HEART risk group by shift. Rates were 2.3% versus 1.5% (p\u0026thinsp;=\u0026thinsp;0.740) in the low-risk group, 10.1% versus 11.9% (p\u0026thinsp;=\u0026thinsp;0.353) in the intermediate-risk group, and 50.3% versus 45.9% (p\u0026thinsp;=\u0026thinsp;0.493) in the high-risk group. No statistically significant difference was observed in any stratum. The Mantel-Haenszel pooled OR, adjusted for HEART risk group, was 0.989 (95% CI: 0.752\u0026ndash;1.302; p\u0026thinsp;=\u0026thinsp;0.995), confirming no significant difference in MACE risk between off-hours and on-hours patients across strata. MACE component breakdown by risk group and time stratum is shown in Supplementary Figure S2.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e3.4.4. Time strata analysis\u003c/h2\u003e \u003cp\u003eSupplementary Table S2 and Supplementary Figure S3 present AUC estimates across the five-time strata. AUC ranged from 0.799 (95% CI: 0.747\u0026ndash;0.850) in weekend daytime presentations to 0.855 (95% CI: 0.810\u0026ndash;0.899) in evening presentations. The Cochran Q statistic was 2.991 (df\u0026thinsp;=\u0026thinsp;4, p\u0026thinsp;=\u0026thinsp;0.559), providing no evidence of significant heterogeneity across strata. Per the pre-specified decision rule, pairwise comparisons were not performed. The HEART score maintained good discrimination across all time categories including public holidays (AUC\u0026thinsp;=\u0026thinsp;0.835, 95% CI: 0.779\u0026ndash;0.891).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section3\"\u003e \u003ch2\u003e3.4.5. Calibration comparison\u003c/h2\u003e \u003cp\u003eCalibration metrics were statistically indistinguishable between groups. The intercept difference was \u0026minus;\u0026thinsp;0.0003 (bootstrap 95% CI: -0.350 to +\u0026thinsp;0.327; p\u0026thinsp;=\u0026thinsp;0.998), and the slope difference was 0.070 (p\u0026thinsp;=\u0026thinsp;0.426). Both groups exhibited comparable systematic miscalibration relative to Backus 2013, with significant intercept deviations (off-hours p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; on-hours p\u0026thinsp;=\u0026thinsp;0.007) and significant slope departures from 1.0 (both p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). This indicates that the recalibration need identified in the primary analysis applies equally to both shift groups.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Sensitivity analyses\u003c/h2\u003e \u003cp\u003eSupplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e presents all six pre-specified sensitivity analyses. Across all scenarios, AUC estimates in the off-hours group ranged from 0.818 (SA4, 6-week endpoint) to 0.848 (SA2b, conventional troponin assay subgroup), with the primary estimate of 0.830 consistent with all sensitivity results.\u003c/p\u003e \u003cp\u003eIn SA1 (original serial troponin complete; n\u0026thinsp;=\u0026thinsp;1,681), AUC was 0.841 (95% CI: 0.815\u0026ndash;0.867), marginally higher than the primary estimate, with a calibration intercept of -0.294 (versus \u0026minus;\u0026thinsp;0.352 in the full cohort). The modest improvement suggests that the 21.2% of patients with simulated or incomplete second troponin measurements did not introduce systematic bias, although their inclusion slightly attenuated discrimination.\u003c/p\u003e \u003cp\u003eSA2 showed consistent performance across assay types: AUC was 0.821 (95% CI: 0.791\u0026ndash;0.851) with hsTnI and 0.848 (95% CI: 0.804\u0026ndash;0.892) with conventional troponin, with overlapping confidence intervals. SA3 demonstrated stable HEART score performance regardless of ECG interpreter seniority (AUC range 0.824\u0026ndash;0.833), with overlapping confidence intervals across resident, specialist, and cardiologist groups. SA6 with MICE-imputed scores yielded AUC\u0026thinsp;=\u0026thinsp;0.835, marginally above the primary estimate, consistent with expected slight optimism correction from imputation.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study found that HEART score discrimination for 30-day MACE was clinically equivalent across all hours of emergency presentation, with the AUC difference between shift groups falling well within the pre-specified equivalence boundary and a formal interaction test confirming no differential score effect by shift assignment. These findings indicate that the operational constraints characteristic of off-hours care, including longer door-to-diagnosis intervals and lower serial troponin completion rates, did not materially impair the score's ability to classify patients by MACE risk.\u003c/p\u003e \u003cp\u003eThe HEART score integrates five independently scored components, no single one of which dominates the composite discriminative gradient. Components anchored to fixed patient characteristics, specifically Age and Risk Factors, remain unaffected by shift-related workflow variation. The History and ECG components are the most subjectively scored elements: a prospective comparison of clinician-calculated and researcher-generated HEART scores found overall agreement of 78% (kappa 0.48), with the History component achieving the lowest pairwise kappa of 0.14 and ECG kappa of 0.40. (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e) Despite this component-level variability, inter-rater reliability among emergency physicians, who staff departments continuously across shifts, was shown to be substantial for low-risk classification (kappa 0.68; 84.2% agreement). (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e) The composite score architecture may therefore tolerate component-level variability without proportional degradation of aggregate discriminative performance across shift conditions.\u003c/p\u003e \u003cp\u003eThe AUC of 0.830 in the off-hours cohort aligns with the Backus 2013 prospective multicenter validation cohort (AUC 0.83 across 10 Dutch centers)and the Six 2013 multinational Asia-Pacific validation of 2,906 patients (AUC 0.83). (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e) A systematic review and meta-analysis of 25 validation studies and more than 25,000 patients confirmed a pooled sensitivity of 0.96 and NPV of 0.99 for low-risk scores, with a central C-statistic of approximately 0.83 across heterogeneous institutional contexts. (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) In a direct comparison among 1,748 chest pain patients across nine Dutch hospitals, the HEART score outperformed the GRACE score (AUC 0.73) and TIMI score (AUC 0.80), identifying 381 low-risk patients with only 0.8% missed MACE compared with 2.2% under GRACE at equivalent safety levels. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e) Higher AUC values have been reported in a US military emergency cohort (0.885), (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e) in institutions operating under formal HEART pathway protocols (0.898), (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e) and in a recent Middle Eastern tertiary validation where baseline MACE prevalence was 24.8% (AUC 0.925). (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) In contrast, a prospective Tanzanian cohort of 927 patients yielded an AUC of only 0.61, with sensitivity of 59.4% and NPV of 74.7% at the \u0026ge;\u0026thinsp;4 threshold, attributed to limited ECG interpretation infrastructure and absence of high-sensitivity troponin assays. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e) This low-income context finding directly contradicts the performance levels seen across high-income settings, confirming that assay quality and clinical interpretation infrastructure, rather than time of presentation, are the principal environmental determinants of HEART score discrimination. Weekend effect meta-analyses in ACS consistently show higher in-hospital mortality among weekend versus weekday admissions (pooled OR 1.06; 95% CI 1.03\u0026ndash;1.09 across 18 studies and more than 14\u0026nbsp;million ACS patients). (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e) A Spanish national registry analysis confirmed 5% higher STEMI mortality and 8% higher NSTEACS mortality on weekends and public holidays after risk adjustment. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e) These mortality data appear to contradict the present finding of preserved discrimination; however, the weekend mortality effect operates through treatment access delays and reperfusion timing, not through the risk stratification accuracy of a score applied at initial triage.\u003c/p\u003e \u003cp\u003ePre-specified secondary analyses confirmed the primary equivalence finding across multiple comparison frameworks. The Mantel-Haenszel pooled odds ratio for 30-day MACE, with HEART risk group as the stratification variable, was 0.989, indicating near-identical risk within each risk stratum regardless of shift assignment. Calibration metrics were statistically indistinguishable between shift groups, with a calibration intercept difference of -0.0003 (p\u0026thinsp;=\u0026thinsp;0.998) and a slope difference of 0.070 (p\u0026thinsp;=\u0026thinsp;0.426), confirming that systematic miscalibration relative to the reference cohort probabilities reflects population-level rather than shift-specific factors. Decision curves demonstrated positive net benefit over treat-all and treat-none strategies across the full pre-specified threshold range of 5\u0026ndash;40%, with a net advantage of 0.023 per patient at a threshold of 10% in the off-hours group. In a post-hoc exploratory analysis, the Youden-index optimal threshold was HEART\u0026thinsp;\u0026ge;\u0026thinsp;6, yielding sensitivity of 75.4% (95% CI 70.3\u0026ndash;79.9%) and specificity of 77.5% (95% CI 75.5\u0026ndash;79.4%); this threshold was not pre-specified and should be regarded as hypothesis-generating only.\u003c/p\u003e \u003cp\u003eThese findings carry direct operational consequences for emergency department triage at all hours of presentation. At the rule-out threshold of HEART\u0026thinsp;\u0026ge;\u0026thinsp;4, the off-hours cohort showed a negative predictive value of 97.7%, with 13 missed events among 574 low-risk patients, a 2.3% miss rate consistent with the accepted bounds for structured early discharge protocols. A stepped-wedge cluster randomized trial of HEART score implementation across nine Dutch emergency departments confirmed that HEART-guided care is safe, with six-week MACE incidence 1.3% lower in the HEART care group than in usual care, within the pre-specified noninferiority margin.(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e) The high-risk threshold (HEART\u0026thinsp;\u0026ge;\u0026thinsp;7) yielded a positive likelihood ratio of 6.07, indicating a clinically meaningful increase in post-test probability; this rule-in performance was preserved across all five time strata. When the HEART pathway is combined with a 0-hour/1-hour high-sensitivity troponin protocol, NPV for 30-day MACE reaches 99.8%, identifying 49.8% of patients for safe rule-out without additional testing. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e) The AUC of 0.830 at a 14.4% MACE rate is consistent with discrimination expected at this prevalence level across published external validation datasets, (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) and the absence of a shift-related AUC difference confirms that off-hours operational constraints do not attenuate the score's triage gradient. Across both shift groups, positive net clinical benefit was observed at all threshold probabilities between 5% and 40%, indicating that neither overtriage nor undertriage is differentially amplified by off-hours conditions.\u003c/p\u003e \u003cp\u003eThe retrospective single-center design limits generalizability to tertiary academic emergency departments with comparable infrastructure; settings with more severe off-hours constraints or less standardized assay platforms may show different performance. Serial troponin completion was 12.5 percentage points lower during off-hours, and restriction to patients with complete second-troponin documentation raised the off-hours AUC from 0.830 to 0.841, indicating modest downward bias in the primary estimate. The MACE composite was restricted to unplanned coronary revascularization, but procedural urgency was ascertained from administrative records; residual misclassification of elective procedures, if present, would inflate the observed MACE rate and reduce rule-out performance metrics. A mid-study transition from conventional to high-sensitivity troponin I created temporal confounding between assay type and shift assignment; assay-stratified analyses produced overlapping confidence intervals across both platforms, suggesting limited influence on the primary AUC. Observer variability in the History and ECG components is inherent to retrospective HEART score abstraction, though the seniority subgroup analysis produced AUC values of 0.824, 0.833, and 0.827 across resident, specialist, and cardiologist groups, consistent with non-differential misclassification. Thirty-day outcomes were ascertained from institutional records without active patient contact; events occurring at external facilities would not be captured, biasing the observed event rate downward and the negative predictive value upward, with magnitude unquantifiable from available data. The off-hours definition encompassed evenings, nights, weekends, and public holidays, covering approximately 75% of weekly calendar hours and producing an enrollment asymmetry (off-hours 76.2%, on-hours 23.8%); the smaller on-hours group (n\u0026thinsp;=\u0026thinsp;667) widened its AUC confidence interval (95% CI 0.778\u0026ndash;0.865), though the pre-specified equivalence boundary was met with adequate margin. The same sample size asymmetry reduces precision of the on-hours calibration summary estimate; the overall O/E ratio for on-hours (0.791; 95% CI 0.642\u0026ndash;0.965) carries substantially wider uncertainty than the off-hours estimate (0.812; 95% CI 0.724\u0026ndash;0.908), and between-group calibration comparisons should be interpreted with this limitation in mind.\u003c/p\u003e \u003cp\u003eThese findings suggest that the HEART score may warrant prospective evaluation as a time-agnostic chest pain triage instrument in multicenter emergency department cohorts spanning diverse institutional and geographic contexts. The stepped-wedge implementation trial confirmed that HEART-guided care is safe but identified physician nonadherence to low-risk discharge recommendations as the principal barrier to realizing its efficiency benefits, (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e) a challenge that may be amplified during off-hours when decision-making operates under greater time pressure. The systematic miscalibration identified relative to the reference cohort probabilities in both shift groups argues for local recalibration of absolute risk estimates before probability-guided triage protocols are implemented. Recalibration of the HEART score troponin threshold to the limit of detection with high-sensitivity assays has been shown to increase sensitivity for rule-out from 96.1% to 98.6% while maintaining specificity, supporting the feasibility of this approach. (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e) A prospective stepped-wedge or cluster-randomized implementation trial, incorporating standardized ECG adjudication by shift category and concurrent troponin assay documentation, is required to determine whether discrimination equivalence persists across centers with different off-hours staffing models. Such a trial should also assess whether locally recalibrated probability thresholds improve absolute risk communication without compromising the safety of threshold-based rule-out.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eIn adult emergency department patients presenting with acute chest pain, the HEART score demonstrated clinically equivalent discrimination for 30-day MACE regardless of presentation time, with an AUC of 0.830 (95% CI 0.805\u0026ndash;0.855) in off-hours and 0.821 (95% CI 0.778\u0026ndash;0.865) in on-hours presentations, both falling within the pre-specified equivalence boundary of \u0026plusmn;\u0026thinsp;0.08. These findings suggest that standard HEART score risk thresholds may be applied without modification across all hours of emergency presentation, including evenings, nights, weekends, and public holidays, without compromising the sensitivity or negative predictive value of the standard rule-out threshold, at tertiary academic centers with comparable operational and assay infrastructure pending prospective multicenter confirmation. Systematic overestimation of absolute MACE risk relative to the reference cohort probabilities was observed equally across shift groups, indicating that local recalibration of predicted risk is needed before probability-guided triage protocols are adopted, irrespective of presentation time. Prospective multicenter validation across diverse emergency department settings, assay types, and shift staffing models is required before these findings can be incorporated into practice guidelines for time-stratified chest pain triage.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eACS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eacute coronary syndrome\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the receiver operating characteristic curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003econfidence interval\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003edecision curve analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eECG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eelectrocardiogram\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEDACS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEmergency Department Assessment of Chest Pain Score\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFDR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003efalse discovery rate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGRACE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGlobal Registry of Acute Coronary Events score\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHEART\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHistory,Electrocardiogram,Age,Risk Factors,Troponin score\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ehsTnI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ehigh-sensitivity troponin I\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIQR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003einterquartile range\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003elikelihood ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMACE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emajor adverse cardiac event\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMICE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emultiple imputation by chained equations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003enegative predictive value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNSTEACS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003enon-ST-elevation acute coronary syndrome\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eO/E\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eobserved-to-expected ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eodds ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epositive predictive value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ereceiver operating characteristic curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003esensitivity analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSMD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003estandardised mean difference\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSTEMI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eST-elevation myocardial infarction\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTIMI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eThrombolysis in Myocardial Infarction score\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTRIPOD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTransparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics Approval and Consent to Participate\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee (approval number: E-10840098-202.3.02-470; decision number: 90). Individual patient consent was waived in accordance with the retrospective design.\u003c/p\u003e\n\u003cp\u003eCompeting Interests:\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003eData Availability\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of this study are not publicly available due to patient privacy constraints and institutional ethics committee restrictions. De-identified aggregate data and the analysis code are available from the corresponding author upon reasonable request, subject to approval by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eS.Z.E.K.: Conceptualization, Methodology, Formal analysis, Data curation, Investigation, Resources, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing, Visualization, Validation, Project administration.\u003c/p\u003e\n\u003cp\u003eU.D.: Methodology, Software, Investigation, Writing \u0026ndash; review \u0026amp; editing, Validation.\u003c/p\u003e\n\u003cp\u003eE.K.: Investigation, Resources, Writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003eS.B.: Writing \u0026ndash; review \u0026amp; editing, Supervision.\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eNone.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eFanaroff AC, Rymer JA, Goldstein SA, Simel DL, Newby LK. Does this patient with chest pain have acute coronary syndrome? the rational clinical examination systematic review. JAMA. 2015;314(18):1955\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2015.12735\u003c/span\u003e\u003cspan address=\"10.1001/jama.2015.12735\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSix AJ, Backus BE, Kelder JC. Chest pain in the emergency room: value of the HEART score. Neth Heart J. 2008;16(6):191\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/BF03086144\u003c/span\u003e\u003cspan address=\"10.1007/BF03086144\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 18665203.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBackus BE, Six AJ, Kelder JC, et al. Chest pain in the emergency room: a multicenter validation of the HEART score. Crit Pathw Cardiol. 2010;9(3):164\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/HPC.0b013e3181ec36d8\u003c/span\u003e\u003cspan address=\"10.1097/HPC.0b013e3181ec36d8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLaureano-Phillips J, Robinson RD, Aryal S, et al. HEART score risk stratification of low-risk chest pain patients in the emergency department: a systematic review and meta-analysis. Ann Emerg Med. 2019;74(2):187\u0026ndash;203. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annemergmed.2018.12.010\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2018.12.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSix AJ, Cullen L, Backus BE, et al. The HEART score for the assessment of patients with chest pain in the emergency department: a multinational validation study. Crit Pathw Cardiol. 2013;12(3):121\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/HPC.0b013e31828b327e\u003c/span\u003e\u003cspan address=\"10.1097/HPC.0b013e31828b327e\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStreitz MJ, Oliver JJ, Hyams JM, et al. A retrospective external validation study of the HEART score among patients presenting to the emergency department with chest pain. Intern Emerg Med. 2017;13(5):727\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11739-017-1743-4\u003c/span\u003e\u003cspan address=\"10.1007/s11739-017-1743-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKwok CS, Al-Dokheal M, Aldaham S, et al. Weekend effect in acute coronary syndrome: a meta-analysis of observational studies. Eur Heart J Acute Cardiovasc Care. 2018;8(5):432\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/2048872618762634\u003c/span\u003e\u003cspan address=\"10.1177/2048872618762634\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFern\u0026aacute;ndez-Ortiz A, Bas Villalobos MC, Garc\u0026iacute;a-M\u0026aacute;rquez M, et al. The effect of weekends and public holidays on the care of acute coronary syndrome in the Spanish National Health System. Rev Esp Cardiol (Engl Ed). 2022;75(9):756\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.rec.2021.10.022\u003c/span\u003e\u003cspan address=\"10.1016/j.rec.2021.10.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBackus BE, Six AJ, Kelder JC, et al. A prospective validation of the HEART score for chest pain patients at the emergency department. Int J Cardiol. 2013;168(3):2153\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcard.2013.01.255\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2013.01.255\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSoares WE, Knee A, Gemme SR, et al. A prospective evaluation of clinical HEART score agreement, accuracy, and adherence in emergency department chest pain patients. Ann Emerg Med. 2021;78(2):231\u0026ndash;41. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annemergmed.2021.03.024\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2021.03.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGershon CA, Yagapen AN, Lin A, Yanez D, Sun BC. Inter-rater reliability of the HEART score. Acad Emerg Med. 2019;26(5):552\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/acem.13665\u003c/span\u003e\u003cspan address=\"10.1111/acem.13665\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoldervaart JM, Langedijk M, Backus BE, et al. Comparison of the GRACE, HEART and TIMI score to predict major adverse cardiac events in chest pain patients at the emergency department. Int J Cardiol. 2017;227:656\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcard.2016.10.080\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2016.10.080\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOliver JJ, Streitz MJ, Hyams JM, et al. An external validation of the HEART pathway among emergency department patients with chest pain. Intern Emerg Med. 2018;13(8):1249\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11739-018-1809-y\u003c/span\u003e\u003cspan address=\"10.1007/s11739-018-1809-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNasr Isfahani M, Mohseni H, Nasri Nasrabadi E, Sarrafzadegan N. Improving chest pain risk assessment: validation of HEART, TIMI, GRACE, EDACS-ADP, and HET for MACE prediction in the emergency department. BMC Emerg Med. 2025;25:165. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12873-025-01327-4\u003c/span\u003e\u003cspan address=\"10.1186/s12873-025-01327-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrisel B, Adisa O, Sakita FM, et al. Evaluating the performance of the HEART score in a Tanzanian emergency department. Acad Emerg Med. 2024;31(4):361\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/acem.14872\u003c/span\u003e\u003cspan address=\"10.1111/acem.14872\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoldervaart JM, Reitsma JB, Backus BE, et al. Effect of using the HEART score in patients with chest pain in the emergency department: a stepped-wedge, cluster randomized trial. Ann Intern Med. 2017;166(10):689\u0026ndash;97. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7326/M16-1600\u003c/span\u003e\u003cspan address=\"10.7326/M16-1600\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNilsson T, Johannesson E, Lundager Forberg J, Mokhtari A, Ekelund U. Diagnostic accuracy of the HEART pathway and EDACS-ADP when combined with a 0-hour/1-hour hs-cTnT protocol for assessment of acute chest pain patients. Emerg Med J. 2021;38(11):808\u0026ndash;13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/emermed-2020-210833\u003c/span\u003e\u003cspan address=\"10.1136/emermed-2020-210833\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhand AU, Backus B, Campbell M, et al. HEART score recalibration using higher sensitivity troponin T. Ann Emerg Med. 2023;82(4):449\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annemergmed.2023.04.024\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2023.04.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emmd","sideBox":"Learn more about [BMC Emergency Medicine](http://bmcemergmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/emmd","title":"BMC Emergency Medicine","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Chest Pain, Emergency Service, Hospital, Risk Assessment","lastPublishedDoi":"10.21203/rs.3.rs-9430666/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9430666/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eAims\u003c/h2\u003e \u003cp\u003eWe compared HEART score discriminative performance and calibration for 30-day major adverse cardiac events (MACE) between off-hours and on-hours emergency department chest pain presentations.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA single-center retrospective cohort study was conducted at Istanbul Medipol Mega University Hospital from January 2020 through December 2025, enrolling 2,800 patients (off-hours n\u0026thinsp;=\u0026thinsp;2,133; on-hours n\u0026thinsp;=\u0026thinsp;667). Off-hours comprised evenings, nights, weekends, and public holidays. The primary outcome was 30-day MACE, defined as acute myocardial infarction, unplanned coronary revascularization, cardiac arrest, or all-cause death. Discrimination was compared by the DeLong method (equivalence: ΔAUC\u0026thinsp;\u0026lt;\u0026thinsp;0.08); calibration was assessed by the Hosmer-Lemeshow test, observed-to-expected ratio, and calibration slope.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003e30-day MACE occurred in 14.4% of 2,800 patients. The HEART score AUC was 0.830 (95% CI 0.805\u0026ndash;0.855) in off-hours and 0.821 (95% CI 0.778\u0026ndash;0.865) in on-hours patients; ΔAUC was +\u0026thinsp;0.009 (95% CI -0.041 to +\u0026thinsp;0.058; p\u0026thinsp;=\u0026thinsp;0.736), within the pre-specified equivalence boundary. At the rule-out threshold (HEART\u0026thinsp;\u0026ge;\u0026thinsp;4), off-hours sensitivity was 95.7% (95% CI 92.8\u0026ndash;97.5%) and negative predictive value (NPV) was 97.7% (95% CI 96.2\u0026ndash;98.7%). No differential shift effect was detected (interaction OR 1.039; 95% CI 0.867\u0026ndash;1.247; p\u0026thinsp;=\u0026thinsp;0.676). Both groups demonstrated systematic overestimation of absolute MACE risk relative to Backus 2013 probabilities (off-hours calibration intercept\u0026thinsp;\u0026minus;\u0026thinsp;0.352; 95% CI -0.495 to -0.208; p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), with no between-group difference (p\u0026thinsp;=\u0026thinsp;0.998).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe HEART score demonstrated clinically equivalent discrimination for 30-day MACE regardless of presentation time, supporting unmodified use across all shifts. Local recalibration of absolute risk estimates is warranted before probability-guided triage protocols are implemented.\u003c/p\u003e","manuscriptTitle":"Time-Stratified HEART Score Discrimination and Calibration in Emergency Chest Pain: A Retrospective Cohort Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-29 10:16:57","doi":"10.21203/rs.3.rs-9430666/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"208314906050084119943695702931772405890","date":"2026-05-15T06:29:12+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-27T17:14:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"257780319750285039739979149362016006411","date":"2026-04-20T19:50:10+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-20T19:39:01+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-17T10:22:58+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-16T05:57:45+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-16T05:56:49+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Emergency Medicine","date":"2026-04-15T19:30:15+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-emergency-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"emmd","sideBox":"Learn more about [BMC Emergency Medicine](http://bmcemergmed.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/emmd","title":"BMC Emergency Medicine","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"7273b0bf-4ad6-433d-991d-b52da4fd8eee","owner":[],"postedDate":"April 29th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"208314906050084119943695702931772405890","date":"2026-05-15T06:29:12+00:00","index":45,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-29T10:16:57+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-29 10:16:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9430666","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9430666","identity":"rs-9430666","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}