Objective There are a few scoring systems in emergency departments (ED) to establish critically ill patients quickly and properly and to predict hospitalisation. We aim to compare the efficacy of Modified Early Warning Score (MEWS) and Rapid Emergency Medicine Score (REMS) on in-hospital mortality, and as predictor of hospitalisation in general medical and surgical patients admitted to ED.
Methods This is a prospective, multicentre and observational cohort study. The study included general medical and surgical patients admitted to the EDs of three education and research hospitals during a period of 6 months. The primary outcome of the study is the admission of the patient to a ward/an intensive care unit (ICU)/high dependency unit (HDU) and in-hospital mortality. Receiver operating characteristics (ROC) curve analysis was performed to evaluate and compare the performances of two scores.
Results Total patients were 2000 (51.95% male, 48.05% female). The mean age was 61.41±18.92. Median MEWS and REMS values of the patients admitted to the ICU/HDU from ED were 1 and 6, respectively; and there was a significant difference in terms of REMS values, compared with patients discharged from ED. REMS (area under the curve (AUC): 0.642) was found to have a better predictive strength than MEWS (AUC: 0.568) in discriminating in-patients and discharged patients. Additionally, REMS (0.707) was superior to MEWS (AUC 0.630) in terms of predicting in-hospital mortality of patients presenting to ED.
Conclusions The efficiency of REMS was found to be superior to MEWS as a predictor of in-hospital mortality and hospitalisation in medical and surgical patients admitted to ED.
- emergency care systems, emergency departments
- management, risk management
Statistics from Altmetric.com
Several scoring systems for the assessment of the severity of an illness have been presented during the recent decades.1 Severity of illness scoring systems is essential for risk stratification, proper triage and ultimate disposition of patients.2 The majority of those scoring systems have been explored extensively in the intensive care unit (ICU) setting.3 Acute Physiology and Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA), Mortality Probability Models (MPM), and Simplified Acute Physiology Score (SAPS) are only some of them.2–4 These scores have been employed in determining long-term and short-term prognosis of critically ill patients, and in predicting their mortality.
The initial place where critically ill patients present is the emergency department (ED). In all EDs, especially in overcrowded and understaffed ones, it is crucial to establish critically ill patients as quickly and properly as possible, and to transfer them to wards or ICU/HDUs (High Dependency Unit). Although risk stratification methods have been developed for selected groups of emergency patients,5 ,6 there have been few attempts to develop a generic risk adjustment score for ED patients.7 As a result, there are a number of physiologic scoring systems designed for the ED setting.3 The Mortality in ED Sepsis Score (MEDS),5 The Rapid Acute Physiology Score (RAPS),8 and Rapid Emergency Medicine Score (REMS)1 ,6 are the most commonly known ones.
The Modified Early Warning Score (MEWS) is a tool for bedside evaluation, and was developed to be used for patients in wards and ICU/HDUs. In several studies carried out afterwards, it was employed to predict the mortality and admission of ED patients to wards or ICUs9–11 and, accordingly, it is not a disease-specific score. On the other hand, REMS, which is also not disease-specific, and is developed by Olsson et al, was studied for evaluating non-surgical patients in the ED.1 ,6 ,12 Few studies are available on REMS in the literature. To the best of our knowledge, in the literature, there are no studies comparing MEWS and REMS in terms of general medical and surgical patients admitted to the ED.
The aim of the current study is to compare the efficacy of MEWS and REMS on in-hospital mortality, and their strength as a predictor of hospitalisation in general medical and surgical patients admitted to the ED.
Material and methods
This is a prospective, multicentre and observational cohort study including general medical and surgical patients presenting to the EDs of three training and research hospitals during a period of 6 months. The study was approved by Uludag University Faculty of Medicine Ethics Committee (sanction numbers 2011-18/7).
Study setting and population
Between the period of October 2011 and April 2012, the data from 2000 patients were obtained prospectively. The study was administered in the EDs of Bursa Sevket Yilmaz Education and Research Hospital (ERH), Kayseri ERH, and Trabzon Numune ERH. Bursa Sevket Yilmaz ERH (located in Marmara Region in Turkey) is a 850-bed hospital receiving around 900–1000 ED admissions in a period of 24 h (ED patient number in 2011: 340 055). Kayseri ERH (located in Central Anatolia Region) has 1633 beds capacity and receives around 1200–1300 ED admissions in a period of 24 h (ED patient number in 2011: 447 762). Trabzon Numune ERH (located in the Black Sea Region) is a 550-bed hospital and receives nearly 700–800 ED admissions in a period of 24 h (ED patient number in 2011: 237 323).
In EDs in Turkey, triage has been exercised in accordance with the triage classification stated in the declaration by Republic of Turkey Ministry of Health on procedures and principles regarding the exercise of emergency services in healthcare facilities with beds dated October 2009 and numbering 27 378. Triage has been performed by nurses pursuant to National Triage Algorithm. Following the triage evaluation, the patients who were aged 16 years and over, who are considered in red (life-threatening but treatable injuries requiring rapid medical attention) and yellow categories (potentially life-threatening injuries, risk of organ loss, and cases with important rate of morbidity) were included in the study. The patients with trauma, under 16 years of age, in green category (outpatients with a stable general condition and with minor health problems), and who were transported to the ED with cardiac arrest were excluded. Also, the patients with missing information were omitted.
Patient age, gender, major complaints upon admission to ED, blood pressure, pulse rate, respiratory rate, temperature, AVPU (A: Alert, V: Verbal, P: Pain, U: Unresponsive), Glasgow Coma Scale (GCS) and peripheral O2 saturation, were recorded. The admission of the patient to a ward or an ICU/HDU (HDU also includes Coronary Care Unit) and in-hospital mortality were considered as the primary outcomes.
MEWS is based on regular assessment of five basic physiologic parameters (systolic blood pressure, pulse rate, respiratory rate, temperature and AVPU score) (table 1). A number ranging from 0 to 3 is assigned to each of the parameters. As in the previous studies, 5 and over is evaluated as ‘critical score’. Critical score is associated with increased mortality rate or admission to ICU.9 ,13 The patients were classified as high risk (≥5) or low risk (<5) according to MEWS.
REMS is a scoring system based on five physiologic parameters (mean arterial pressure, respiratory rate, blood pressure, peripheral O2 saturation and GCS score) and age (table 2). Except for the age (0–6 points), each parameter is graded from 0 to 4, and the maximum score is 26.6 ,12 According to REMS, the patients were classified as high risk (>13), intermediate risk (6–13), and low risk (<6).
Shapiro–Wilk test was used to test the normality of variables. Continuous variables were presented as mean±SD for normally distributed variables; otherwise median (minimum value–maximum value) were given. Kruskal–Wallis test was performed for comparing more than two groups for continuous variables. Mann–Whitney U test was performed for comparing two groups after the Kruskal–Wallis test in order to detect significances. Categorical variables were expressed by counts and percentages. Comparisons between the groups were measured with Pearson χ2 test for categorical variables. Risk factors were also evaluated with stepwise logistic regression analysis. Receiver operating characteristics (ROC) curve analysis was performed to evaluate and compare the performances of two scores. Correlations between the variables were assessed with Spearman correlation coefficient. Significance level was taken as α=0.05. Statistical analyses were performed with IBM SPSS Statistics V.20.0 (IBM, USA).
Total patient number was 2000; 52% male (1039) and 48% female (961). The mean age was 61.41±18.92. The physiological parameters were recorded on admission. The general characteristics of patients are summarised in table 3.
When the patients were evaluated in terms of ED diagnosis, cardiovascular diseases were seen in 28.2% of the patients, while respiratory system diseases and cerebrovascular events were seen in 17.1% and 15.2% of the patients, respectively. While 40.8% of the patients were hospitalised into a ward, 29.8% were admitted to the ICU/HDU, and 29.2% were discharged. When the four groups (admitted to a ward, admitted to the ICU/HDU, discharged from ED and patients died in the ED) were compared in terms of MEWS and REMS, there were statistically significant differences between the groups (p<0.001 for both). When the pair-wise comparisons were performed, it was found that median MEWS and REMS values of the patients admitted to a ward from ED was 1 (0–7) and 6 (0–13), respectively. Median MEWS and REMS values of the patients discharged from ED was 1 (0–9) and 3 (0–16), respectively. There was a significant difference between the patients admitted to a ward from ED and discharged from ED, in terms of MEWS and REMS (p<0.001 for both). Also, median MEWS and REMS values of the patients admitted to the ICU/HDU from ED was 1 (0–9) and 6 (0–17), respectively; and there was a significant difference in terms of REMS values, compared with patients discharged from ED (p<0.001) (table 4).
Total in-hospital mortality was 7.7% (n=153). While the median MEWS and REMS values for survivals were 1 (min–max=0–9) and 5 (min–max=0–17), it was 2 (min–max=0–8) and 7 (min–max=0–17) in non-survivals, respectively. On the other hand, MEWS and REMS values were significantly higher among non-survivors than among survivors. Also, there was a significant difference between the survivals and non-survivals in terms of age, mean pulse rate, oxygen saturation and GCS (table 5).
Besides, 2 logistic regression models were conducted to predict in-hospital mortality. In the first model, REMS, gender and age were included as independent variables, whereas it was MEWS, gender and age in the second one. Both the models were statistically significant (p<0.001). As a result of the forward stepwise logistic regression analysis, in the first model, REMS (6–13) increased the risk of death 2.923 (95% CI 0.026 to 4.217, p<0.001) times relative to REMS<6; REMS>13 increased the risk of death 14.564 (95% CI 4.573 to 46.573, p<0.001) times relative to REMS<6. In the second model, MEWS≥5 increased the risk of death 3.837 (95% CI2.358 to 6.243, p<0.001) times relative to MEWS<5. Also in the second model, age was found as a significant risk factor in predicting mortality (p<0.001) (table 6).
When the correlation between the MEWS and REMS values was evaluated, it revealed a significant positive correlation between the two scores (r=0.422, p<0.001).
In distinguishing the patients who were discharged or hospitalised, ROC analyses were performed to evaluate the performances of MEWS and REMS. AUC (area under the curve) was found to be 0.568 (95% CI 0.546 to 0.590) (p<0.001) for MEWS and 0.642 (95% CI 0.621 to 0.663) (p<0.001) for REMS. The performance of REMS was significantly higher than the MEWS (p<0.001) (figure 1).
When the performances of 2 scores in predicting in-hospital mortality were evaluated through ROC analysis, the AUC for the MEWS was 0.630 (95% CI 0.608 to 0.651) (p<0.001), while it was 0.707 (95% CI 0.686 to 0.727) (p<0.001) for the REMS. In this respect, the performance of REMS in discriminating between survivors and non-survivors was significantly higher than the MEWS (p<0.001) (figure 2). ROC curves for prediction of admission to ICU/HDU were also performed. AUC for the MEWS was 0.538 (95% CI 0.516 to 0.560) (p=0.009), while it was 0.589 (95% CI 0.567 to 0.611) (p<0.001) for the REMS. The performance of REMS in discriminating between the patients admitted to the ICU and to others was significantly higher than the MEWS (p<0.001).
Our study results indicated that REMS has a better predictive strength in in-hospital mortality of patients presenting to ED than MEWS has. MEWS showed a lower AUC (0.63) than REMS (0.70). In a study comparing RAPS with a modified form of REMS and conducted by Olsonn et al,1 REMS was found superior to RAPS in predicting in-hospital mortality (AUC 0.65 for RAPS, AUC 0.85 for REMS) of non-surgical ED patients. When compared with our study, the reason for higher AUC value for REMS in the mentioned study could be its administration in non-surgical ED patients as a single-centre study. In another study comparing RAPS with REMS which was carried out by Goodacre et al,7 in terms of in-hospital mortality prediction of ED patients, REMS was identified as surpassing (AUC 0.64 for RAPS, AUC 0.74 for REMS). Similarly, higher AUC value for REMS in comparison with our study results is likely due to the inclusion of patients as a population only who came to ED by ambulance and were hospitalised or died in ED. In the current study, the patients with minor or major trauma, and those who were in green category (ambulatory patients, the patients with a stable general condition and requiring minor/simple treatment) were excluded. Instead, all surgical and general medical patients with intermediate level or serious diseases were studied.
The REMS score was developed by Olsson et al1 based on a population of non-surgical ED patients. They developed REMS from APACHE II and investigated its power as a predictor in short-term and long-term mortality. They classified the patients as low risk (REMS<6), intermediate risk (REMS 6–13), and high risk (REMS>13). They demonstrated that when it was REMS>13, the mortality increased from 7.8% to 17.1%. However, they did not compare the patients discharged from ED with those who were hospitalised.1 ,6 ,12 In another similar study including patients with infection diagnosis and presenting to the ED, the relationship between mortality rates and mREMS, MEDS and CURB-65 was examined via logistic regression analysis. The results showed that the mortality stratified by the REMS score was as follows: 0–2 points, 0.6% mREMS (95% CI 0 to 1.2%), and 12–14 points 20.0% (95% CI 12.5 to 27.5%).2 In the current study, the findings showed that mortality risk of the patients in the intermediate-risk (REMS 6–13) group increased 2.923 (95% CI 0.026 to 4.217) times in comparison with those who were in the low-risk (REMS<6) group, whereas mortality risk of the patients in the high-risk group (REMS>13) rose 14.564 (95% CI 4.573 to 46.573) times when compared with the low-risk (REMS<6) group. In addition, median REMS values of the patients transferred to a ward or ICU/HDU through ED was established as significantly higher than those who were discharged. To our knowledge, as there is no published data on a comparison between the patients transferred to a ward or ICU/HDU through ED and those who were discharged concerning REMS in the literature, we could not compare our results with others.
To the best of our knowledge, this study is the first to compare MEWS and REMS. The majority of studies on MEWS in the literature include surgical patients or in-patients though there have been quite a few studies conducted on ED patients. In various studies, it has been stated that if MEWS score is ≥5, it is related with increased in-hospital mortality and ICU hospitalisation.9 ,10 ,13–15 In their study on a medical inpatient sample, Subbe et al9 found out that mortality risk of patients with MEWS≥5 (OR 5.4, 95% CI 2.8 to 10.7) was higher than those who had MEWS<5. Also, hospitalisation risk of patients with MEWS≥5 in ICU (OR 10.9, 95% CI 2.2 to 55.6) and in HDU (OR 3.3, 95% CI 1.2 to 9.2) was found to be high.9 In a prospective and observational study by Burch et al16 on medical patients presenting to ED, they found that MEWS was significantly high in patients admitted (patients with MEWS≥5, risk ratio 1.7, 95% CI 1.5 to 2.0) and those who died in hospital (patients with MEWS≥5, risk ratio 4.6, 95% CI 2.7 to 7.8). In this study, surgical, orthopaedic, gynaecologic patients, and patients with trauma were excluded.16 In our study, we established that median MEWS and REMS values of patients hospitalised in a ward through ED were significantly higher than the values of those who were discharged. However, in patients hospitalised in ICU/HDU, a statistically significant difference was found only in REMS values. Moreover, MEWS (MEWS≥5, OR 3.837, 95% CI 2.358 to 6.243) and REMS (REMS>13, OR 14.564, 95% CI 4.573 to 46.573) values were found to be significantly higher in non-survivals. In the study, discriminatory performances of MEWS and REMS for the in-patients, and those who were discharged from ED, were evaluated by ROC analysis. The AUC for MEWS was 0.56, whereas the AUC for REMS was 0.64, indicating that REMS had significantly higher score than MEWS. Since both studies mentioned above did not use ROC analysis in evaluating performance of MEWS on discriminating the patients hospitalised in ICU, we could not make a comparison with our findings.
In a prospective study by Ghanem-Zhoubi et al, disease-severity scoring systems were investigated in terms of patients with sepsis diagnosis and hospitalised in general internal medicine departments. In predicting in-hospital mortality, they established AUC values for MEWS and REMS as 0.69 and 0.77, respectively, which indicated that REMS predicted in-hospital mortality more accurately.11 Howell et al, in their study on 2132 patients with infection and admitted to the ED, concluded that MEDS, mREMS and CURB-65 were all efficient and correlated well with 28-day in-hospital mortality. The AUCs were 0.85, 0.80 and 0.79 for MEDS, mREMS and CURB-65, respectively.2 According to our results (AUC for MEWS, 0.63 and the AUC for REMS, 0.70), REMS performed with significantly higher scores than MEWS in discriminating survivals and non-survivals in predicting in-hospital mortality of general medical and surgical patients presenting to ED. The reason for lower AUC value for REMS in our study may be due to different population characteristics of both studies and employment of modified REMS (with the exception of the GCS score as a neurologic parameter) in the original study.
ED scoring and outcome prediction are innovative but relatively novel concepts.2 ,3 In EDs, especially in overcrowded ones with high circulation, it is crucial to establish critically ill patients in a short time, and to transfer them to wards or ICU/HDU. In the presence of overcrowded, understaffed EDs, the utility of efficient and bedside physiologic scoring systems can be of tremendous value to the multitasking ED physician.2 ,3 Therefore, which is the score that should be used in EDs? Scoring system to be used in ED patients should be easy, simple, quick, efficient, and have bedside availability. They each have different complexities and benefits. In a recent review study, it was expressed that none of the concerned scoring systems were perfect, and each has their own weaknesses. It was also stated that REMS has an acceptable level of discriminatory power but with a weak calibration.17
Our study has a number of limitations. First, the parameters which were recorded on admission to the ED were used in calculating MEWS and REMS scores. Throughout the period of the patients’ stay in the ED, serial measurements of MEWS and REMS scores were not performed. Second, the patients with trauma were not included in the study.
According to the current study results, both MEWS and REMS have been found to be effective in predicting hospitalisation and in-hospital mortality of medical and surgical patients presenting to the ED. Further, the REMS score was found to have stronger discriminatory power in terms of both hospitalisation and in-hospital mortality. However, future studies may include a larger and more diverse cohort of patients (medical, surgical and trauma patients) in order to achieve more accurate and generalisable results.
Contributors MB planned, conducted and wrote the work and performed the literature search. HC conducted and performed the literature search. DS performed the data analysis and wrote the work. AS, OD, AAT, SK and KU conducted the work.
Competing interests None.
Patient consent Obtained.
Ethics approval Uludag University Faculty of Medicine Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.