External prognostic validation of the EHMRG score for risk stratification and disposition decisions in acute heart failure

Abstract

Background

Risk stratification in patients with Acute Heart Failure (AHF) is critical for optimizing disposition decisions in the Emergency Department (ED). While clinical judgment is the cornerstone of management, “borderline” patients often present a challenge. The Emergency Heart Failure Mortality Risk Grade (EHMRG) score was developed to objectively predict short-term mortality. This study aims to prospectively validate the EHMRG score for predicting 30-day all-cause mortality and to evaluate its performance for 7-day mortality as a secondary outcome, as well as its concordance with clinician disposition decisions.

Methods

This prospective, single-center, observational external prognostic validation study enrolled adult patients presenting to the ED with AHF between August 2022 and January 2023. The EHMRG score was calculated for all patients but blinded to treating physicians to prevent influence on decision-making. The primary outcome was 30-day all-cause mortality. Seven-day all-cause mortality was evaluated as a secondary outcome due to the limited number of early events. Discriminatory performance for short-term mortality was assessed using Receiver Operating Characteristic (ROC) curve analysis.

Results

A total of 237 patients were included (median age 76 years; 53.6% female). The overall mortality rate was 16.9% at 30 days and 5.9% at 7 days. The EHMRG score demonstrated moderate discriminative ability with an Area Under the Curve (AUC) of 0.755 (95% CI: 0.67–0.84) for 30-day mortality and 0.741 (95% CI: 0.61–0.87) for 7-day mortality. Notably, no deaths occurred in the very low- or low-risk groups (Negative Predictive Value > 97%). However, discordance was observed between clinical judgment and risk scores; 27% of patients were discharged despite being classified as moderate to very high risk by EHMRG. Notably, 12.5% of the 30-day non-survivors originated from this discharged group.

Conclusions

The EHMRG score demonstrated acceptable discrimination for predicting 30-day mortality in this cohort of emergency department patients with acute heart failure. When used alongside clinical judgment, the score may assist clinicians in identifying patients who could potentially be considered for safe discharge.

Trial registration

ClinicalTrials.gov, NCT05710302. Registered January 24, 2023.

Background

Heart failure remains a major global public health concern due to its high rates of morbidity and mortality [1–4]. A significant proportion of patients with heart failure first present to healthcare services through emergency departments (EDs). Despite improvements in prognostic and therapeutic strategies, patient outcomes remain suboptimal, with recent studies reporting one-year mortality rates of up to 13.5% in advanced stages [3, 5].

ED physicians face the complex challenge of determining appropriate disposition: identifying low-risk patients suitable for safe discharge versus high-risk patients requiring hospitalization. While several risk stratification tools exist (e.g., ADHERE [Acute Decompensated Heart Failure National Registry] and GWTG-HF [Get With The Guidelines–Heart Failure]), many were derived from inpatient registries and may not fully capture the physiological instability unique to the ED presentation. The Emergency Heart Failure Mortality Risk Grade (EHMRG), developed by Lee et al., addresses this gap by incorporating ED-specific parameters—including mode of arrival and initial vital signs—to predict early mortality. The EHMRG score incorporates several emergency department–specific variables, including age, mode of arrival, initial vital signs, renal function parameters, troponin levels, active malignancy, and outpatient diuretic therapy. Based on the total score, patients are stratified into predefined mortality risk categories (very low, low, moderate, high, and very high risk), which have been evaluated in subsequent external validation studies [6–8].

Although the EHMRG score has shown promise in derivation cohorts, prospective validation in diverse healthcare settings is essential to establish its external validity. More importantly, the interaction between objective risk scores and subjective clinical judgment remains understudied. The primary aim of this study is to prospectively validate the EHMRG score’s discriminatory performance for predicting 30-day and 7-day mortality and to evaluate its potential role as a decision-support adjunct for emergency department disposition decisions.

Materials and methods

Study design and setting

This prospective, single-center external prognostic validation study was conducted in the Emergency Department of Marmara University Pendik Training and Research Hospital, Istanbul, from August 2022 to January 2023. The hospital serves a population of approximately 6 million as a major regional referral center. The study protocol was approved by the institutional ethics committee, and written informed consent was obtained from all participants.

Study population

Consecutive adult patients (≥ 18 years) presenting to the ED with a confirmed diagnosis of Acute Heart Failure (AHF) were eligible. Diagnosis was established according to the 2021 ESC guidelines [1], utilizing clinical evaluation, lung ultrasound/echocardiography, and N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels. Exclusion criteria included lack of follow-up, withdrawal of consent, need for immediate resuscitation, pregnancy, and dialysis dependency. Lack of follow-up was defined as inability to ascertain 30-day mortality status from hospital records or national death registry data. The need for immediate resuscitation was defined as the requirement for endotracheal intubation and/or ongoing cardiopulmonary resuscitation in the ED. Pregnancy was excluded because physiological cardiovascular changes during pregnancy may affect the applicability of the EHMRG risk model. Of the initial 250 assessed patients, 7 were excluded solely due to dialysis dependency; no patients met the other exclusion criteria such as pregnancy, need for immediate resuscitation, or lack of follow-up.

Data collection

Prospective data collection included demographic information (age, sex), comorbidities, current medications, vital signs at presentation, mode of arrival (ambulance vs. walk-in), laboratory values, and ECG findings. Ultrasonographic evaluation was performed as part of routine clinical assessment when deemed clinically necessary. Point-of-care ultrasound (POCUS) examinations were performed by emergency physicians trained in emergency ultrasound according to standardized bedside ultrasound protocols, primarily to assess pulmonary congestion and basic cardiac function. When a more comprehensive evaluation was required, transthoracic echocardiography was performed by cardiologists using standard echocardiographic criteria according to institutional practice. These ultrasonographic assessments were used as part of routine clinical evaluation and were not predefined variables used in the calculation of the EHMRG score. Clinical outcomes such as admission to the general ward, ICU transfer, or discharge were recorded. Of 243 eligible patients, 6 were excluded from the final analysis due to missing data for variables required to calculate the EHMRG score, resulting in a final analytic cohort of 237 patients.

Blinding and researcher involvement

Physicians were blinded to the EHMRG scores at the time of patient management to prevent any influence on decision-making. The EHMRG score was calculated retrospectively by researchers in a separate data environment. All clinical decisions—including discharge, admission, or referral—were made solely by the treating emergency physicians based on standard institutional protocols. In our institution, disposition decisions for AHF are guided by clinical stability after ED treatment, including improvement in dyspnea, stable vital signs, and the absence of features suggesting end-organ hypoperfusion or need for advanced monitoring. Patients considered for discharge are typically those with adequate response to therapy, ability to tolerate oral medications, and availability of outpatient follow-up, whereas patients with persistent hypotension, hypoxemia, or significant comorbidity burden are preferentially admitted to wards or ICUs. These protocols were not influenced by the EHMRG score.

Laboratory methods

Blood samples were collected in standard biochemistry tubes (BD Vacutainer^® SST™ II Advance). Biochemical parameters were analyzed using the Roche Cobas c 702 analyzer, and hormonal assays (including troponin and NT-proBNP) were performed on the Roche Cobas e 602 analyzer. High-sensitivity troponin T (hs-TnT) values > 51 ng/L were considered elevated.

Index test (EHMRG Score)

The Emergency Heart Failure Mortality Risk Grade (EHMRG) score was the index test evaluated. The score incorporates multiple variables: age, mode of arrival (ambulance), systolic blood pressure, heart rate, oxygen saturation, serum potassium, serum creatinine, elevated troponin, presence of active malignancy, and use of metolazone. All EHMRG variables were derived from the first documented vital signs and the first available laboratory results obtained at ED presentation. No post-treatment or repeat measurements were used for score calculation.

• Adaptation: As metolazone is not available in Turkey, thiazide-like diuretics were used as a substitute in the score calculation, consistent with previous European validation studies [9].

• Risk Stratification: Based on the total score calculated via MDCalc, patients were stratified into five mortality risk groups: very low, low, moderate, high, and very high.

Outcomes

The primary outcome was 30-day all-cause mortality after ED presentation. Seven-day all-cause mortality was assessed as a secondary/exploratory outcome due to the limited number of early events. Mortality status was ascertained through hospital records, the national death notification system, and telephone follow-up.

Sample size estimation

Sample size was estimated using MedCalc software based on a priori ROC curve analysis methodology as described by Obuchowski [10]. Calculations were based on a priori ROC curve analysis testing the Area Under the Curve (AUC) against the null value of 0.50. Assuming an expected AUC of 0.75 based on prior studies, an alpha of 0.05, a statistical power of 0.95, and an anticipated event rate of 16% (derived specifically from the 7-day observation subgroup in the validation study by Falsetti et al.), the minimum required sample size was determined to be approximately 212 participants (34 events and 178 non-events) [11]. Therefore, our final analytic cohort of 237 patients provided sufficient statistical power for the primary analyses.

Statistical analysis

Continuous variables were summarized as medians with interquartile ranges (IQR) and compared using the Mann–Whitney U test. Categorical variables were presented as frequencies and percentages and compared using the chi-square test or Fisher’s exact test, as appropriate.Risk-category analyses were conducted using the established EHMRG cut-offs (very low, low, moderate, high, and very high) as defined in the original derivation study and subsequent validation cohorts. These pre-specified categories constituted the primary validation framework for evaluating 30-day all-cause mortality, which was the primary endpoint of this study. Trends in mortality across ordered EHMRG risk categories were assessed using the Cochran–Armitage test for trend.Discriminatory performance for 30-day mortality was further evaluated using receiver operating characteristic (ROC) curve analysis, and the area under the curve (AUC) was reported with 95% confidence intervals. Exploratory ROC analyses were performed to identify cohort-derived optimal thresholds for 30-day mortality using the Youden index (sensitivity + specificity − 1).

These data-driven cut-offs were considered hypothesis-generating and were not intended to redefine the established EHMRG risk boundaries. Secondary analyses were conducted for 7-day mortality using both the established risk categories and exploratory ROC-derived thresholds. A two-sided p-value < 0.05 was considered statistically significant.

Results

Patient characteristics

A total of 237 patients were included in the final analysis (Fig. 1). The median age was 76 years (IQR: 67–84), and 53.6% were female. The majority of patients (76.7%) arrived by ambulance. The baseline characteristics of the study population, stratified by 30-day and 7-day mortality status, are presented in Table 1. Because the EHMRG score was originally developed to predict short-term mortality, the primary analyses in this study focused on mortality outcomes. Regarding outcomes, 40 patients (16.9%) died within 30 days and 14 patients (5.9%) died within 7 days.

Fig. 1.

Study flow diagram showing patient screening, exclusions, and the final analytic cohort used for external validation of the EHMRG score

Table 1.

Baseline characteristics and clinical data stratified by 30-day and 7-day mortality outcomes

Variables	Total	7-day Outcome			30-day Outcome
	(n = 237)	Non-Survivor (n = 14)	Survivor (n = 223)	p	Non-Survivor (n = 40)	Survivor (n = 197)	p
Demographics
Age (years), median (IQR)	76 (67–84)	79 (74–84)	76 (67–84)	0.393	81 (73.5–85)	75 (67–83)	0.050
Male gender, n (%)	110 (46.4)	8 (57.1)	102 (45.7)	0.410	19 (48.0)	91 (46.0)	0.880
Transport
Ambulance, n(%)	181 (76.7)	14 (100.0)	167 (75.0)	0.045	35 (88.0)	146 (74.0)	0.069
Comorbidities
Hypertension	178 (75.1)	11 (78.6)	167 (74.9)	> 0.99	29 (72.5)	149 (75.6)	0.676
Heart Failure	157 (66.2)	7 (50.0)	150 (67.3)	0.240	26 (65.0)	131 (66.5)	0.855
Diabetes Mellitus	116 (48.9)	4 (28.6)	112 (50.2)	0.120	12 (30.0)	104 (52.8)	0.009
Coronary Artery Disease	104 (43.9)	6 (42.9)	98 (43.9)	0.940	11 (27.5)	93 (47.2)	0.022
COPD	63 (26.6)	3 (21.4)	60 (26.9)	0.770	11 (27.5)	52 (26.4)	0.885
Active Cancer	22 (9.3)	3 (21.4)	19 (8.5)	0.130	11 (27.5)	11 (5.6)	< 0.001
Stroke	18 (7.6)	0 (0.0)	18 (8.1)	0.610	0 (0.0)	18 (18.2)	0.047
Vital Signs
SBP (mmHg), median	136 (116–158)	127 (93–149)	137 (116–158)	0.135	124 (96–141)	139 (118–160)	< 0.001
SpO2 (%), median	90 (86–95)	90 (82–93)	90 (86–95)	0.735	88 (83–92)	91 (86–95)	0.046
Laboratory Values
BUN (mg/dL)	25 (17–36)	46.5 (22–56)	24 (17–34)	0.006	36 (21–46)	24 (17–33)	0.003
Creatinine (mg/dL)	1.1 (0.8–1.4)	1.2 (0.9–1.8)	1.1 (0.8–1.4)	0.432	1.1 (0.9–1.8)	1.1 (0.8–1.4)	0.311
hs-Troponin (ng/L)	32.9 (23–56)	77.1 (43–205)	32.0 (22–53)	< 0.001	52.4 (30–103)	31.2 (22–51)	< 0.001
NT-proBNP (ng/L)	5917 (2331–11592)	9024 (4251– 11715)	5779 (2267– 11576)	0.113	7918 (3007–11885)	5499 (2193– 11296)	0.094
Medication History
Furosemide	152 (64.1)	7 (50.0)	145 (65.0)	0.260	23 (57.5)	129 (65.5)	0.337
Beta-blockers	143 (60.3)	9 (64.3)	134 (60.1)	0.760	22 (55.0)	121 (61.4)	0.449
ACEi / ARB	111 (46.8)	6 (42.9)	105 (47.1)	0.760	16 (40.0)	95 (48.2)	0.342

Values are presented as n (%) or median (IQR). Continuous variables were compared using the Mann–Whitney U test and categorical variables using the chi-square or Fisher’s exact test, as appropriate. Bold values indicate statistical significance (p < 0.05). Abbreviations: SBP: Systolic Blood Pressure, SpO₂: peripheral oxygen saturation, NT-proBNP: pro–B-type natriuretic peptide; COPD: Chronic Obstructive Pulmonary Disease, ACEi: Angiotensin-Converting Enzyme inhibitor, ARB: Angiotensin Receptor Blocker, BUN: Blood Urea Nitrogen

The 30-day mortality analysis revealed a significant burden of comorbidities. Non-survivors had a significantly higher prevalence of active cancer (27.5% vs. 5.6%, p < 0.001). In contrast, a prior history of stroke was more frequent among survivors than non-survivors (18.2% vs. 0.0%, p = 0.047). Regarding vital signs, systolic blood pressure (SBP) was significantly lower in non-survivors compared to survivors (median 124 vs. 139 mmHg, p < 0.001). Similar to the 7-day analysis, elevated BUN and hs-Troponin levels remained strong predictors of 30-day mortality (p = 0.003 and p < 0.001, respectively).

In contrast, in the 7-day mortality analysis, there were no statistically significant differences in demographic characteristics or comorbidities between survivors and non-survivors. However, markers of acute physiological stress and end-organ damage were prominent; non-survivors had significantly higher BUN levels (median 46.5 vs. 24 mg/dL, p = 0.006) and hs-Troponin levels (median 77.1 vs. 32.0 ng/L, p < 0.001).

Prognostic performance of EHMRG

Across the established EHMRG risk categories, a clear stepwise increase in mortality was observed (Fig. 2). For 30-day mortality, rates increased from 0% in the very low- and low-risk groups to 32.6% in the very high-risk group (p < 0.001 for trend). A similar gradient was observed for 7-day mortality, which increased to 12.4% in the very high-risk group (p = 0.003 for trend). No deaths occurred in the very low- or low-risk groups.

Fig. 2.

Observed 7-day and 30-day mortality across EHMRG risk categories. Mortality increased progressively across ordered EHMRG risk strata from very low to very high risk. Bars represent the proportion of deaths within each category, and error bars indicate exact binomial 95% confidence intervals (Clopper–Pearson). P values represent the Cochran–Armitage test for trend across ordered EHMRG categories

When the very low and low risk groups were combined, 30-day mortality remained 0%, reinforcing the potential role of the established EHMRG categorization as a tool to support discharge risk assessment. ROC analysis demonstrated moderate discriminative power (Fig. 3).

Fig. 3.

Receiver operating characteristic (ROC) curves illustrating the discriminatory performance of the EHMRG score for predicting 7-day and 30-day all-cause mortality in patients presenting to the emergency department with acute heart failure. The area under the curve (AUC) for 30-day mortality was 0.755 (95% CI 0.67–0.84), while the AUC for 7-day mortality was 0.741 (95% CI 0.61–0.87). The 7-day analysis should be interpreted cautiously because of the limited number of early outcome events

Discriminatory performance for the primary endpoint of 30-day mortality yielded an AUC of 0.755 (95% CI: 0.67 − 0.84, Table 2). In exploratory analyses using the Youden index, a cohort-derived threshold of > 56.8 demonstrated a sensitivity of 92.5%, specificity of 53.8%, positive likelihood ratio of 2.00, negative likelihood ratio of 0.139, positive predictive value of 28.9%, and negative predictive value of 97.2% (Table 2). These data-driven thresholds should be interpreted cautiously and require confirmation in independent cohorts before being considered for clinical decision thresholds.

Table 2.

Exploratory ROC-Derived Performance Metrics of the EHMRG Score

Outcome	AUC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	+LR (95% CI)	-LR (95% CI)	Accuracy (95% CI)
7-day mortality	0.741 (0.61–0.87)	78.6% (49.2–95.3)	65.0% (58.4–71.3)	12.4% (6.3–21.0)	98.0% (94.2–99.6)	2.25 (1.62–3.11)	0.33 (0.12–0.90)	65.8% (59.4–71.8)
30-day mortality	0.755 (0.67–0.84)	92.5% (79.6–98.4)	53.8% (46.6–60.9)	28.9% (21.2–37.6)	97.2% (92.2–99.4)	2.00 (1.68–2.38)	0.14 (0.04–0.41)	60.3% (53.8–66.6)

Cut-off values were derived using the Youden index in the present cohort and are considered exploratory. Abbreviations: AUC: Area Under the Curve, CI: Confidence Interval, PPV: Positive Predictive Value, NPV: Negative Predictive Value, +LR: Positive Likelihood Ratio, -LR: Negative Likelihood Ratio

For 7-day mortality (secondary endpoint), the EHMRG score demonstrated an AUC of 0.741; however, given the limited number of early events (n = 14), these findings should be interpreted cautiously and considered exploratory. The Youden-derived threshold of > 89.4 yielded a sensitivity of 78.6%, specificity of 65%, positive likelihood ratio of 2.25, negative likelihood ratio of 0.330, positive predictive value of 12.4%, and negative predictive value of 98% (Table 2).

Concordance with clinical decisions

Overall, clinicians admitted 57.0% (135/237) of patients to hospital wards or the intensive care unit, while 43.0% (102/237) were discharged from the emergency department. There was a strong and statistically significant association between EHMRG risk stratification and clinician disposition decisions (χ² test, p < 0.001).

In total, 156 patients (65.8%) were managed concordantly with EHMRG recommendations, whereas 81 patients (34.2%) were managed discordantly.

Among patients classified as very low or low risk by EHMRG, 38 patients (69.1%) were appropriately discharged, while 17 patients (30.9%) were admitted despite EHMRG recommending discharge. Conversely, among patients classified as moderate-to-very high risk, 118 patients (64.8%) were appropriately admitted, whereas 64 patients (35.2%) were discharged despite EHMRG recommending hospital admission.

Notably, this discordant discharge group represented 27.0% (64/237) of the entire cohort. Of particular concern, 12.5% (5/40) of all 30-day non-survivors belonged to this discordant discharge group, indicating a potential underestimation of risk by clinical judgment alone. The concordance matrix between clinician disposition and EHMRG risk stratification is presented in Table 3. The distribution of clinician disposition decisions across EHMRG risk categories is shown in Fig. 4. Discharge rates decreased and hospital admission rates increased with higher EHMRG risk categories. A statistically significant trend toward hospital admission was observed with increasing EHMRG risk category (p < 0.001, Chi-square test for trend).

Table 3.

Concordance Matrix Between EHMRG Risk Stratification and Clinician Disposition

EHMRG Risk Category	Discharged n (%)	Admitted n (%)	Total
Very Low / Low Risk
(generally considered suitable for discharge)	38 (69.1%) Concordant	17 (30.9%) Discordant	55
Moderate / High / Very High Risk
(generally considered for hospital admission)	64 (35.2%) Discordant	118 (64.8%) Concordant	182
Total	102 (43.0%)	135 (57.0%)	237

EHMRG: Emergency Heart Failure Mortality Risk Grade

Fig. 4.

Distribution of clinician disposition decisions (discharge vs. hospital admission) across EHMRG risk categories. Admission rates increased with higher EHMRG risk strata, while discharge was more frequent in very low and low-risk groups. Discordant cases represent patients discharged despite being classified as moderate-to-very high risk by EHMRG

EHMRG score distributions according to mortality status are presented in Table 4. Patients who died had substantially higher EHMRG scores compared with survivors for both 7-day and 30-day mortality analyses. The potential clinical implications of different EHMRG-based admission thresholds were further explored using hypothetical scenarios based on established EHMRG risk categories. The corresponding mortality rates, sensitivity, specificity, and negative predictive values are summarized in Table 5.

Table 4.

Distribution of EHMRG scores according to 7-day and 30-day mortality status

Outcome	Total Cohort (n = 237)	Alive	Dead	p
7-day mortality	62.2 (22.2–117.0)	59.2 (19.1–113.0)	124.0 (95.8–163.0)	0.001
30-day mortality	62.2 (22.2–117.0)	47.4 (13.4–104.0)	118.0 (75.0–142.0)	< 0.001

Values are presented as median (25th – 75th percentiles). P-values were calculated using the Mann-Whitney U test

Table 5.

Performance of hypothetical EHMRG-based admission thresholds for predicting 30-day mortality

Admission Threshold	Hypothetical Action	Total (n)	30-day Deaths (n)	Mortality (%) *	Sensitivity % (95% CI)	Specificity % (95% CI)	NPV % (95% CI)
≥Moderate Risk	Admit	182	40	22.0	100 (91.2–100)	27.9 (21.8–34.7)	-
(EHMRG Score > 17.9)	Discharge	55	0	0.0			100 (93.5–100)
≥High Risk	Admit	128	37	28.9	92.5 (79.6–98.4)	53.8 (46.6–60.9)	-
(EHMRG Score > 56.5)	Discharge	109	3	2.8			97.2 (92.2–99.4)
≥Very High Risk	Admit	89	29	32.6	72.5 (56.1–85.4)	69.5 (62.6–75.9)	-
(EHMRG Score > 89.3)	Discharge	148	11	7.4			92.6 (87.1–96.2)

EHMRG: Emergency Heart Failure Mortality Risk Grade; CI: Confidence Interval; NPV: Negative Predictive Value. *Mortality rate is calculated within the respective admitted or discharged hypothetical subgroup. Threshold values used in this table (> 17.9, > 56.5, > 89.3) represent the established, original EHMRG risk category boundaries derived from prior validation studies, distinct from our cohort-specific ROC-derived cut-offs

Discussion

This prospective external validation study evaluates the prognostic performance of the Emergency Heart Failure Mortality Risk Grade (EHMRG) score in a tertiary emergency department and demonstrates its clinical utility for short-term risk stratification. In this external validation study, EHMRG provided fair-to-moderate overall discrimination, with established risk categories demonstrating consistent stepwise increases in 30-day mortality. Importantly, the score achieved a high negative predictive value—with no 30-day deaths occurring in the very low and low-risk groups— supporting its potential role as a decision-support adjunct for discharge risk assessment. Although formal calibration analysis was not performed due to the limited number of events, the observed increase in mortality across risk strata is consistent with the expected direction of risk prediction. Consistent with prior validation studies, these findings reinforce the generalizability of the score. However, although exploratory ROC analyses identified cohort-derived thresholds similar to previously reported high-risk boundaries, these data-driven cut-offs should be interpreted cautiously and require independent validation [8, 12, 13]. Importantly, our primary analyses relied on established EHMRG risk categories. The exploratory ROC-derived thresholds were not intended to redefine these validated boundaries, but rather to characterize the score’s discriminatory behavior within this specific cohort.

A central observation of this study is the discordance between clinician disposition and EHMRG risk classification. Although most high-risk patients were appropriately admitted, 27% of patients classified as moderate-to-very high risk by EHMRG were discharged based on clinical assessment alone, and a proportion of these patients died within 30 days. This finding highlights an opportunity for improved risk recognition at the point of disposition. Because decision rationales were not prospectively captured, we cannot determine the specific factors that influenced clinician judgment in these cases. One possible explanation is that initial clinical stabilization or apparently reassuring vital signs may have contributed to discharge decisions; however, this was not directly measured. In contrast, EHMRG integrates objective indicators of physiological stress and end-organ injury (e.g., renal function, troponin, and mode of arrival) that may reflect risk not readily apparent on initial bedside assessment. Taken together, these data suggest that EHMRG could serve as a decision-support adjunct to complement, rather than replace, clinical judgment when evaluating emergency department disposition decisions [12–14].

Our dual-timeframe analysis underscores a temporal shift in prognostic determinants. Early (7-day) mortality was predominantly associated with markers of acute hemodynamic and renal stress (notably elevated blood urea nitrogen and high-sensitivity troponin), whereas mid-term (30-day) mortality was more strongly related to baseline frailty and comorbidity burden, particularly active malignancy. Active cancer was not a significant predictor of death within the first week but emerged as a strong determinant at 30 days, suggesting that while emergency care can stabilize acute threats, limited physiological reserve may constrain recovery over subsequent weeks [8, 12, 14].

Pre-hospital factors also carried prognostic weight. All patients who died within 7 days arrived by ambulance, aligning with EURODEM findings that emergency medical services (EMS) arrival is associated with higher short-term mortality [15]. This supports the inclusion of mode of arrival within EHMRG and illustrates the value of ED-specific variables that are often absent from inpatient-derived risk models.

We further observed that lower admission systolic blood pressure was associated with higher 30-day mortality. This accords with the concept of “reverse epidemiology” in heart failure, wherein higher blood pressure may reflect preserved pump function, while lower values may signal advanced myocardial dysfunction and poorer prognosis [16].

An inverse association between diabetes and mortality was observed; however, our dataset did not capture antidiabetic subclasses. Consequently, we cannot attribute this association to specific therapies (e.g., SGLT2 inhibitors). While contemporary trials demonstrate survival benefits with SGLT2 inhibitors in heart failure, any inference regarding their contribution in our cohort would be speculative [17, 18]. This observation should therefore be interpreted cautiously and warrants dedicated investigation with granular pharmacotherapy data.

Consistent with prior literature, active malignancy and elevated high-sensitivity troponin were strong predictors of 30-day mortality. In contrast, N-terminal pro-BNP did not differ significantly between survivors and non-survivors, which may reflect uniformly elevated values in an acute cohort and suggests that, in the very short term, hemodynamic and renal markers may offer superior discriminative value compared with natriuretic peptides alone [8, 12]. The higher prevalence of prior stroke among survivors should be interpreted cautiously, as this finding may reflect the limited number of events and the small sample size rather than a true protective association.

In our cohort, the observed mortality among discharged patients classified as moderate-to-high risk appeared higher than that reported in some prior validation studies such as van Hattem et al. Several factors may explain this difference. First, our study was conducted in a single tertiary emergency department with local discharge practices that may differ from those in multicenter cohorts. Second, differences in baseline patient characteristics and comorbidity burden may have influenced post-discharge outcomes. Third, the relatively small number of events in the discharged subgroup may lead to proportionally higher mortality estimates. Finally, variations in access to early outpatient follow-up or community support systems may contribute to differences in post-discharge risk.

Overall, our findings support the external prognostic validity of EHMRG for 30-day mortality discrimination in ED patients with acute heart failure. The absence of deaths in the very low and low-risk groups highlights its potential value for identifying patients who may be suitable for discharge consideration. In contrast, analyses related to 7-day mortality and cohort-derived thresholds should be interpreted as exploratory, given the limited number of early events, and require confirmation in larger independent cohorts. Although the ROC-derived threshold (> 56.8) was numerically close to previously reported high-risk boundaries, this value should be interpreted as a cohort-specific statistical estimate rather than a clinically validated decision threshold. Therefore, EHMRG should be interpreted as a complementary decision-support tool rather than a substitute for clinical judgement in emergency department disposition decisions.

Limitations

This study has several limitations. First, its single-center design may limit generalizability to other healthcare systems and practice environments. Second, although disposition decisions were prospectively recorded, the specific clinical rationales (e.g., response to therapy, perceived social support, or clinician risk tolerance) were not systematically captured; therefore, the causes of discordance between EHMRG and clinician judgment cannot be determined. Third, metolazone is not available in our region; we substituted thiazide-like diuretics in the score calculation, consistent with prior European validations, but this adaptation may introduce minor measurement variability. Fourth, our dataset lacked granular pharmacotherapy data (including antidiabetic subclasses), precluding causal inference regarding medication effects. Fifth, the number of very early events (7-day mortality) was modest, limiting power for more granular subgroup analyses. Because of the relatively limited number of outcome events, particularly for early mortality, formal calibration analyses were not performed and should be addressed in future validation studies with larger cohorts. Finally, although we assessed prognostic accuracy and concordance, the study was not designed to evaluate whether EHMRG-guided management improves outcomes; randomized implementation studies are required to address clinical impact.

Conclusion

In this cohort, the EHMRG score demonstrated acceptable discrimination for 30-day mortality and may support risk stratification and disposition decisions in the emergency department. Its excellent negative predictive value supports its use in identifying low-risk candidates for discharge. Furthermore, its ability to detect high-risk patients—potentially missed by clinical judgment—suggests a valuable role as a decision-support adjunct. Integrating the EHMRG score into ED practice may support more standardized risk assessment and safer disposition decisions, although its effect on clinical outcomes and its role in routine implementation should be confirmed in larger independent studies.

Author contributions

BÜ and ADA conceived and designed the study. OD and ÇÖ were responsible for data collection and curation. BÜ and EA performed the statistical analysis. BÜ drafted the manuscript. ADA supervised the study and critically reviewed the manuscript. All authors read and approved the final manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Clinical Research Ethics Committee of Marmara University School of Medicine on 29.07.2022 with protocol code 09.2022.742. Written informed consent was obtained from all participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.