EHMRG, mortality risk grade in acute heart failure

Clinical background

Deciding whether to admit or discharge a patient with acute heart failure from the emergency department is one of the commonest and most difficult disposition decisions in emergency care. Clinical judgement alone not infrequently leads both to unnecessary admission and to unsafe discharge. EHMRG was developed to provide objective support for precisely this decision, with a short-term focus: the risk of death within 7 days. The score is not a diagnostic instrument; it presupposes that the diagnosis of acute heart failure has already been made. Nor is it intended to replace clinical judgement, but to quantify the mortality risk carried by the vital signs and laboratory values on arrival.

Calculating the EHMRG score

The calculator implements EHMRG-7S, the score-based 7-day variant. The score is calculated as:

$$\text{Score} = 12 + 2 \times \text{age} + 60 \cdot \mathbb{1}$${\text{ambulance}} - \min(\text{SBP}, 160) + \text{HR}{\text{clamped}} - 2 \times \min(\text{SpO}2, 92) + 20 \times S{\text{cr}} + K_{\text{points}} + 45 \cdot \mathbb{1}{\text{cancer}} + 60 \cdot \mathbb{1}{\text{metolazone}}

where $\text{HR}${\text{clamped}} = \max(80, \min(\text{HR}, 120))HRclamped​=max(80,min(HR,120)), $S${\text{cr}} is expressed in mg/dL, and $K_{\text{points}}$ is 5 for a potassium <4.0 mmol/L, 0 for 4.0 to 4.5 mmol/L, and 30 for >4.5 mmol/L. The systolic blood pressure is capped at 160 mmHg and the oxygen saturation at 92 per cent, which means that values above these thresholds do not lower the score further. The heart rate is clamped to the range 80 to 120 beats/min: values below 80 count as 80 and values above 120 count as 120.

The variables are: age in whole years, arrival by ambulance (yes gives +60), systolic blood pressure on arrival, heart rate on arrival, oxygen saturation on arrival, creatinine in mg/dL, potassium in three bands, active cancer (yes gives +45) and outpatient metolazone treatment (yes gives +60).

The derivation cohort consisted of 12,591 patients presenting to emergency departments at 86 hospitals in Ontario, Canada, between 2004 and 2007 [1]. The derivation cohort comprised 7,433 patients (mean age 75.4 years, 51.5 per cent men) and the validation cohort 5,158 patients (mean age 75.7 years, 51.6 per cent men). The outcome modelled was death within 7 days of arrival. The original multivariable model also included troponin, but the score-based variant implemented by this calculator (EHMRG-7S) does not include troponin. The EHMRG-ST variant instead adds ST depression and troponin to predict 30-day mortality and is a separate model.

Interpretation in practice

The score is graded into five bands with cut-points at −49, −15.8, 18 and 56.6. In the original derivation cohort, 7-day mortality rose stepwise from 0.3 per cent in the lowest quintile to 8.2 per cent in the highest decile [1]. In the external validation in Alberta, the corresponding mortality was 0.0, 0.8, 1.6, 4.0, 4.2 and 12.0 per cent across risk categories 1 to 4 and 5A and 5B [2].

The prospective validation from Turkey found no deaths at all in the two lowest risk bands, with a negative predictive value above 97 per cent for 7-day mortality [3]. This supports the use of low scores to identify patients in whom early discharge is safe, provided clinical stabilisation has been achieved.

Validation and performance

The derivation study achieved an AUC of 0.805 in the derivation cohort and 0.826 in the internal validation cohort [1]. The external validation in Alberta (n = 6,708, patients arriving by ambulance, 2012 to 2016) showed a c-statistic of 0.73 (95 per cent CI 0.71 to 0.76) for 7-day mortality and 0.71 (95 per cent CI 0.70 to 0.73) for 30-day mortality [2]. Discrimination for composite outcomes (death, admission, re-attendance within 30 days) was lower, in the range 0.61 to 0.67. Adding natriuretic peptides significantly improved prediction (NRI 0.268, P<0.001) and removing troponin worsened it, suggesting that the score-based variant without troponin performs somewhat less well than the full model [2].

A prospective validation from Turkey (n = 237, 2022 to 2023) reported an AUC of 0.741 (95 per cent CI 0.61 to 0.87) for 7-day mortality and 0.755 (95 per cent CI 0.67 to 0.84) for 30-day mortality [3]. In this cohort, 27 per cent of patients were discharged despite being classified as moderate to very high risk by EHMRG, and 12.5 per cent of those who died within 30 days came from this discharged group, illustrating the gap between the score and actual disposition.

A temporal validation in Ontario (n = 7,537, 2017 to 2019) showed that EHMRG-7S retained good discrimination over time, with a c-statistic of 0.795 (95 per cent CI 0.789 to 0.802) for 7-day mortality [4]. Calibration was good for the 7-day model, with only marginal overestimation in the highest decile. For the 30-day model (EHMRG30-ST), by contrast, systematic overestimation was observed in the upper five deciles, which would lead to conservative recommendations in the direction of admission. Performance was robust across sociodemographic subgroups, but lower for patients in residential and nursing homes (c-statistic 0.71) [4].

A comparative study from Turkey (n = 346, abstract) found that EHMRG performed better than the Ottawa Heart Failure Risk Scale for 30-day mortality (AUC 0.774 versus 0.706) but less well for predicting admission (AUC 0.682 versus 0.742) [5].

Limitations

EHMRG applies to adult patients with acute heart failure in the emergency department. The model is not validated for patients requiring immediate resuscitation, for severely unstable patients, or for patients on dialysis. Pregnant patients are excluded from the validation studies because of the physiological cardiovascular changes of pregnancy [3].

Creatinine is entered in mg/dL in the formula, which requires conversion from the µmol/L used in Swedish clinical practice. A patient with a serum creatinine of 150 µmol/L has approximately 1.7 mg/dL, which gives 34 points from the creatinine component. Confusing the units leads to large errors in calculation.

Metolazone is included as a binary variable and reflects the fact that the patient was already on metolazone as an outpatient, a marker of severe heart failure with diuretic resistance. In Sweden, metolazone is less commonly used than in North American cohorts, which may reduce the discriminating power of this variable in a Swedish population. In the Turkish validation, metolazone was replaced by thiazide-like diuretics, since metolazone was not available [3].

Left ventricular function is not included in the model, which is an explicit limitation in the derivation study [1]. Patients with a severely reduced ejection fraction may be at higher risk than the score indicates, particularly when the other parameters are relatively well preserved.

The score is a snapshot based on values on arrival. It does not capture improvement after treatment in the emergency department, and should not be used in isolation for disposition after initial treatment. A patient who arrives with a high score but stabilises quickly may nonetheless have a lower actual risk than the score suggests.

References

1. Lee DS et al. Prediction of heart failure mortality in emergent care: a cohort study. Ann Intern Med 2012. PMID: 22665814
2. Sepehrvand N et al. External Validation and Refinement of Emergency Heart Failure Mortality Risk Grade Risk Model in Patients With Heart Failure in the Emergency Department. CJC Open 2019. PMID: 32159095
3. Ünlü B et al. External prognostic validation of the EHMRG score for risk stratification and disposition decisions in acute heart failure. BMC Emerg Med 2026. PMID: 41942909
4. Abdul-Samad K et al. Evaluating the temporal and sociodemographic generalizability of the emergency heart failure mortality risk grade. ESC Heart Fail 2026. PMID: 42212505
5. Kilinc H et al. Comparison of Emergency Heart Failure Mortality Risk Grade and Ottawa Heart Failure Risk Score Scores in Predicting Mortality in Patients Presenting to the Emergency Department with Heart Failure. J Emerg Med 2025. PMID: 41138554