Rapid Emergency Medicine Score (REMS)

Clinical background

In the emergency department, the clinician needs to distinguish quickly between patients at high risk of death and those who can be managed at a lower intensity. That judgement often rests on clinical intuition, which varies between assessors and is difficult to communicate onwards. REMS was developed precisely to provide an objective, bedside risk marker for in-hospital mortality in non-surgical emergency patients. The score rests on physiological variables that are measured routinely at triage or initial assessment anyway, and requires no laboratory tests. This makes it quick to calculate, but also sets the limits of what it can capture: the score reflects the patient's physiological status at a point in time, not diagnostic specificity or chronic disease burden beyond what the age variable captures.

Calculating the Rapid Emergency Medicine Score

REMS is an extension of the earlier Rapid Acute Physiology Score (RAPS), which in turn took its variables from APACHE II. Olsson and co-workers at Uppsala University Hospital added peripheral oxygen saturation and age to the four RAPS variables (mean arterial pressure, heart rate, respiratory rate and the Glasgow Coma Scale) and called the result REMS [1]. The score is calculated as the sum of attribute points for five physiological variables plus a consciousness score:

$$\text{REMS} = \text{AP}$${\text{age}} + \text{AP}{\text{MAP}} + \text{AP}{\text{HR}} + \text{AP}{\text{RR}} + \text{AP}_{\text{SpO}_2} + (15 - \text{GCS})

where $\text{AP}$ denotes attribute points on an APACHE-style interval scheme for age, mean arterial pressure (mmHg), heart rate (beats/min), respiratory rate (/min) and oxygen saturation (%). Each variable scores from 0 (normal range) up to 4 (markedly abnormal). The consciousness score $(15 - \text{GCS})$ gives 0 points at full alertness (GCS 15) and up to 12 points in deep coma (GCS 3). The total scale ranges from 0 to 26.

The derivation cohort consisted of 12,006 non-surgical patients presenting to an adult emergency department at a university hospital over 12 consecutive months [1]. The modelled outcome was in-hospital mortality. In this cohort, REMS showed an AUC of 0.852 (SEM ±0.014) for predicting in-hospital mortality, compared with 0.652 for RAPS. Each 1-point increase on the 26-point scale was associated with an odds ratio of 1.40 (95% CI 1.36 to 1.45) for in-hospital death [1].

Interpretation in practice

REMS has no generally accepted fixed interpretation bands in the original derivation study. Olsson reported the odds ratio per point but defined no explicit thresholds for clinical action [1]. Later validation studies have proposed different cut-offs depending on the population and outcome:

These thresholds come from a specific COVID-19 population and should not be transferred uncritically to a general non-surgical emergency population. In clinical practice, REMS works best as an aid to triage decisions: a low score (about 0 to 5) argues against a need for immediate intensive care and can support a more expectant approach, while a high score (about 8 or more) justifies early intensified monitoring, prompt medical review and consideration of an ICU bed. The score should never be the sole basis for an admission decision, but should complement clinical assessment.

Validation and performance

The original AUC of 0.852 in the derivation cohort has not been reproduced in external validations. In an Iranian prospective cohort of 2,205 medical emergency patients (ESI levels 1 to 3) at Emam Reza Hospital in Mashhad, REMS achieved an optimism-corrected AUC of 0.678 for in-hospital mortality [2]. Calibration was relatively good in this study, but discrimination was lower than for the Simple Clinical Score (0.714), the Worthing Physiological Score (0.727) and MEWS (0.698). REMS showed the highest sensitivity but the lowest specificity of the models compared [2].

In a Thai retrospective study at Siriraj Hospital of 1,622 patients with suspected sepsis, the REMS AUC for in-hospital mortality was 0.62 (95% CI 0.59 to 0.65) [3]. This was significantly better than qSOFA (0.58) and SIRS (0.52), but not significantly superior to NEWS (0.61). Calibration was acceptable for in-hospital mortality, but REMS underestimated the risk of 7-day mortality at high scores. On decision curve analysis, REMS gave the highest net benefit of the four scoring systems compared [3].

In a US prospective cohort of 227 critically ill patients admitted to the ICU directly from the ED (UCSF), the REMS AUC for 60-day mortality was 0.698 to 0.709 [5]. ICU-based scoring systems performed significantly better: APACHE III reached an AUC of 0.799. The differences between the ED-based scores (REMS, MEWS, PEDS, Seymour) were not significant among themselves. All the scoring systems showed acceptable calibration by the Hosmer-Lemeshow test [5].

A prehospital study of 13,830 COVID-19 patients found that REMS calculated on the first prehospital values had an AUC of 0.79 for death in the ED and 0.72 for in-hospital mortality [4]. Predictive ability declined for outcomes further removed in time from the measurement: the AUC for a hospital stay of ≥3 days was only 0.62 [4].

In summary, discrimination in external cohorts typically falls in the range 0.62 to 0.72, considerably lower than the 0.852 of the derivation. This pattern is typical of scores derived in a large, broad cohort and then validated in more selected populations.

Limitations

REMS was derived for non-surgical emergency patients and does not apply to trauma patients or to patients requiring immediate surgical intervention. Several studies have nonetheless applied REMS to trauma populations, with variable and generally poorer results. In suspected sepsis, REMS performs better than qSOFA and SIRS but no better than NEWS, and an AUC of around 0.62 is too low for the score alone to guide treatment decisions [3].

The score contains no variables for chronic disease burden, laboratory values or diagnostic information. Two patients with an identical REMS may have radically different prognoses depending on the underlying diagnosis. This is a structural weakness shared with most physiology-based early warning scores, but it becomes particularly evident when the score is used as the sole risk marker.

REMS is measured at a single point in time and does not reflect changes in the patient's condition. A patient may have a low score on arrival and deteriorate rapidly thereafter. Serial measurement has been proposed as an improvement but lacks a validated methodology [4].

The high AUC in the derivation cohort (0.852) has not been reproduced externally. If REMS is used with the expectation of the original performance, the score risks being overvalued in clinical practice. It is important to interpret the score in the light of the population in which it is actually to be used, and to be aware that discrimination in a general medical emergency population lies closer to 0.68 to 0.72 [2, 5].

References

1. Olsson T, Terent A, Lind L. Rapid Emergency Medicine score: a new prognostic tool for in-hospital mortality in nonsurgical emergency department patients. J Intern Med. 2004;255(5):579-587. PMID: 15078500
2. Rahmatinejad Z, Tohidinezhad F, Rahmatinejad F, et al. Internal validation and comparison of the prognostic performance of models based on six emergency scoring systems to predict in-hospital mortality in the emergency department. BMC Emerg Med. 2021;21:68. PMID: 34112088
3. Ruangsomboon O, Boonmee P, Limsuwat C, et al. The utility of the rapid emergency medicine score (REMS) compared with SIRS, qSOFA and NEWS for predicting in-hospital mortality among patients with suspicion of sepsis in an emergency department. BMC Emerg Med. 2021;21:2. PMID: 33413139
4. Bourn SS, Crowe RP, Fernandez AR, et al. Initial prehospital Rapid Emergency Medicine Score (REMS) to predict outcomes for COVID-19 patients. J Am Coll Emerg Physicians Open. 2021;2(4):e12483. PMID: 34223444
5. Moseson EM, Zhuo H, Chu J, et al. Intensive care unit scoring systems outperform emergency department scoring systems for mortality prediction in critically ill patients: a prospective cohort study. J Intensive Care. 2014;2:40. PMID: 25960880