Respiratory National Early Warning Score for 28-day mortality prediction in suspected sepsis patients in the emergency department Authors: Kiattichai Daorattanachai, Sasimanee Maithong, Pariwat Phungoen, Sinee Weschawalit, Winchana Srivilaithon Published: 2025 ABSTRACT BackgroundMost early warning scores were derived in heterogeneous acutely ill populations and are not specifically tailored to patients with suspected sepsis in the emergency department (ED). In sepsis, respiratory rate (RR) is frequently elevated as part of the pathophysiologic response, suggesting that sepsis-focused tools may require recalibrated RR thresholds. We therefore aimed to develop and evaluate the Respiratory National Early Warning Score (R-NEWS), a sepsis-oriented modification emphasizing respiratory parameters, to improve prediction of 28- and 7-day mortality among ED patients with suspected sepsis.MethodsWe conducted a prospective cohort study in the ED of Thammasat University Hospital, enrolling adults with suspected sepsis and excluding those with cardiac arrest or do-not-resuscitate orders. Multivariable backward stepwise logistic regression identified significant predictors of 28-day mortality, which were then used to construct the R-NEWS scoring system. Predictive performance was evaluated and compared with NEWS, qSOFA, and SIRS using the area under the receiver operating characteristic curve (AuROC).ResultsA total of 1,348 patients were analyzed; 28-day mortality was 13.4%. Five independent predictors were identified: respiratory rate, need for supplemental oxygen, systolic blood pressure, heart rate, and Glasgow Coma Scale. R-NEWS demonstrated good calibration (Hosmer–Lemeshow p = 0.474) and higher discrimination than NEWS, qSOFA, and SIRS. The AuROC of R-NEWS was 0.72 (95% CI, 0.68, 0.76) for 28-day mortality and 0.76 (95% CI, 0.71, 0.81) for 7-day mortality. Internal validation using bootstrapping yielded consistent results. High-risk R-NEWS (≥ 7) was associated with a 28-day mortality of 28.9%, compared with 3.9% in the low-risk group (≤ 3).ConclusionR-NEWS, a simplified five-parameter score emphasizing respiratory function, demonstrated favorable performance for predicting short- and intermediate-term mortality in ED patients with suspected sepsis. Used alongside clinical judgment, R-NEWS may aid early risk stratification and sepsis management. External, multicenter validation and prospective evaluation are warranted before widespread implementation.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12873-025-01443-1. Introduction Sepsis is a serious condition frequently encountered in the emergency department (ED). The diagnosis and management of sepsis remain a significant challenge in the ED, with early and accurate identification and prediction of which patients will have poor outcomes being crucial for improving results [1, 2]. Several early warning scores (EWSs) have been developed and implemented to identify high-risk septic patients who could benefit from early intervention. Numerous sepsis screening tools are both utilized and evaluated in the ED, yet each has its limitations. The Sequential Organ Failure Assessment (SOFA) score possesses potential for diagnosing and predicting mortality, however, its complex calculation presents a challenge, resulting in difficulties for practical application within the ED setting [3]. Similarly, the Quick Sequential Organ Failure Assessment (qSOFA) has shown poor sensitivity for sepsis screening, although it is known for its superior specificity in predicting mortality outside the intensive care unit (ICU) [4, 5]. The National Early Warning Score (NEWS) is an established tool for the early detection of clinical deterioration and prediction of mortality in patients with general critical illness, commonly encountered in the ED and other hospital settings [6]. However, NEWS has demonstrated inconsistency in predicting mortality among patients suspected of sepsis [7, 8]. This inconsistency highlights a significant knowledge gap in the reliable early identification of high-risk sepsis patients in the ED. Further studies are recommended to assess the integration of other factors to improve its specificity [9]. A significant observation in sepsis patients is the marked increase in respiratory rate, which is a critical indicator of physiological distress and often precedes other signs of clinical deterioration. The increase in respiratory rate and the rising risk of respiratory failure in sepsis patients reflect the body’s attempt to compensate for the underlying infection and related systemic inflammation, highlighting the need for modified scoring systems that emphasize respiratory parameters [10, 11]. Sepsis triggers the release of proinflammatory cytokines (e.g., TNF-α, IL-1) that can directly stimulate medullary respiratory centers, producing tachypnea. In addition, respiratory alkalosis commonly develops as a compensatory response to metabolic acidosis arising from tissue hypoperfusion. Infection-related lung injury may cause hypoxemia, further augmenting respiratory drive. Moreover, fever and systemic hyperinflammation increase metabolic rate and oxygen demand, both of which elevate respiratory rate. Collectively, these pathophysiological processes justify the dominant role of respiratory parameters in assessing sepsis severity [3, 4, 10]. Traditional scoring systems may not adequately capture the severity of respiratory compromise in sepsis, suggesting the necessity to refine these tools to enhance their predictive accuracy and clinical utility in the ED setting. Given these limitations, we sought to optimize and recalibrate the NEWS for adults with suspected sepsis in the ED. Specifically, we derived a sepsis-focused variant, R-NEWS, by empirically reweighting the respiratory rate component of NEWS, and then evaluated its discrimination, calibration, and clinical utility for predicting mortality in this population. Materials and methods Study design and setting Between November 2022 and October 2023, a prospective cohort study was conducted in the ED of Thammasat University Hospital (TUH), a tertiary care facility. TUH is an 800-bed tertiary academic teaching hospital located in the northern suburbs of Bangkok, serving approximately 1.1 million people in the area. The ED at TUH accommodates 60,000 patients annually, with around 1,300 of these patients presenting with clinical signs of sepsis each year. We established a sepsis registry system for all consecutive patients who presented to the ED with suspected sepsis. Data collection was based on a protocol from international treatment guidelines for sepsis [12]. Ethical approval This study and the sepsis data registry received approval from the Human Research Ethics Committee of Thammasat University (Faculty of Medicine) (MTU-EC-EM-0-170/65). This study was conducted in accordance with the principles outlined in the Declaration of Helsinki and in accordance with the approved clinical trial protocols and regulations. Patient population Adults at least 18 years-old presenting to the ED with clinical evidence of infection and acute organ dysfunction were classified as having suspected sepsis and were eligible for inclusion. This operational definition was applied at triage and aligned with the Sepsis-3 construct; patients with infection without organ dysfunction were not included, and the cohort was not restricted to septic shock. Exclusion criteria were: (1) cardiac arrest upon presentation, and (2) presence of a do-not-resuscitate order (3) patients who were referred from other hospitals. The definition of sepsis was based on the Third International Consensus Definitions for Sepsis and Septic Shock [3]. All patients received the same standard therapy in accordance with the International Surviving Sepsis Campaign guidelines, which include respiratory support, fluid resuscitation, early administration of antimicrobials, and management of macro- and microcirculation [12]. Data collection and outcome measures All cases suspected of sepsis were recorded in the sepsis data registry. Cases were identified and reviewed against inclusion and exclusion criteria. Data were collected by research coordinators trained for involvement in this study. The reliability of data collection in cases suspected of sepsis was assessed monthly by independent investigators. Data included the patients’ baseline characteristics (sex, age), comorbid diseases, source of infection, physiological parameters at triage, level of consciousness, need for supplemental oxygen, need for ventilation assistance, use of a vasoactive agent (norepinephrine, epinephrine, or dopamine), use of an inotropic agent, initial laboratory markers in the ED, and types of antimicrobials used. NEWS uses seven parameters: respiratory rate, oxygen saturation, the need for supplemental oxygen, temperature, systolic blood pressure, pulse rate, and level of consciousness. A higher total score indicates greater physiological disturbance [13]. NEWS scores at triage were calculated for all participants. We defined the primary outcome as in-hospital 28-day mortality. Follow-up was limited to the index admission and ended at the earliest of in-hospital death (within 28 days), discharge alive, or being alive in hospital on day 28. Patients discharged alive before day 28 were not contacted after discharge and were treated as censored at discharge for the primary outcome. The secondary outcome was 7-day mortality among suspected sepsis patients in the ED. Study size estimation and statistical analysis The sample size was calculated for a two-sample comparison of proportions, with an alpha of 0.05, 80% power, and a one-sided test, to detect a difference in mortality between types of scoring. The calculation was performed using STATA version 14 software (StataCorp LLC, College Station, Texas, USA). After adjusting the calculated number by 10% to account for missing data, the study required the recruitment of 1300 patients. Study patients were divided into two groups based on the mortality outcome: the 28-day mortality group and the survival group. Baseline patient characteristics were compared using the exact probability test for categorical variables and the Student’s t-test for continuous variables. To develop a predictive model, a multivariable stepwise backward logistic regression model was constructed. Predictors identified by univariable analysis with cut-off p-values of 0.1 were included in the model and were eliminated at a significance level of 0.05. Predictors identified by multivariable analysis were assigned weighted points proportional to their beta coefficient values. After multivariable analysis, the respiratory rate was divided into several ordinal categories to optimize predictive performance. These categories were determined based on their respective beta coefficients to enhance the area under the receiver operating characteristic curve (AuROC). The final predictive model incorporated these ordinal categories, and a new scoring system was developed to maximize the AuROC. A risk score was calculated for each patient. NEWS was also assigned to each patient to compare its predictive ability with that of the developed predictive model. The discriminative ability of the developed predictive models was quantified using the AuROC with 95% confidence intervals (CI). Comparisons of AuROC between models were performed using DeLong’s method; two-sided p < 0.05 was considered statistically significant. The predictive model was stratified into three risk groups (low, moderate, and high) based on the probability of mortality. All data analyses were performed using STATA software (version 14.0, StataCorp, College Station, TX). Results Model construction During the 12-month study period, 1,562 patients presented to the ED with suspected sepsis. After excluding patients according to the predefined exclusion criteria, a total of 1,348 patients were included in the analysis (Fig. 1). One hundred eighty-one patients (13.4%) were enrolled in the 28-day mortality group. Table 1 summarizes the patients’ baseline characteristics and results of the univariable analysis between the 28-day mortality group and the survival group. Fig. 1Flow of the study Flow of the study Table 1Baseline characteristics between patients who died within 28 days (28-day mortality) and those who survived in patients suspected of sepsis in the emergency departmentCharacteristic28-day mortality (N = 181)n (%)Survived(N = 1167)n (%)p-valueSex0.27 Male96 (53.0)568 (48.7) Female85 (47.0)599 (51.3)Age, mean (± SD) (years)75.1 ± 15.372.3 ± 15.50.02*Body mass index, mean (± SD) (kg/m2)20.9 ± 4.922.3 ± 5.30.001*Comorbid disease Hypertension100 (55.2)696 (59.6)0.26 Diabetes mellitus62 (34.3)438 (37.5)0.39 Chronic obstructive pulmonary disease8 (4.4)39 (3.3)0.46 Chronic liver diseases7 (3.9)24 (2.1)0.13 Chronic kidney diseases22 (12.2)103 (8.8)0.35 Coronary artery disease14 (7.7)98 (8.4)0.76 Stroke41 (22.7)270 (23.1)0.87 Malignancy with metastasis21 (11.6)42 (3.6)< 0.001*Initial physiologic parameter Heart rate, mean (± SD) (bpm)108.1 ± 27.9102.1 ± 22.60.001* Systolic blood pressure, mean (± SD) (mmHg)112.7 ± 30.5123.3 ± 31.2< 0.001* Diastolic blood pressure, mean (± SD) (mmHg)67.6 ± 19.771.6 ± 18.40.001* Respiratory rate, mean (± SD) (/min)28.4 ± 6.226.1 ± 5.2< 0.001* Body temperature, mean (± SD) (°C)37.1 ± 0.937.6 ± 1.2< 0.001* Oxygen saturation, mean (± SD) (%)89.1 ± 8.792.6 ± 6.3< 0.001*Initial Glasgow coma scale< 0.001* 1574 (40.9)747 (64.3) 13–1413 (7.2)81 (7.0) ≤1294 (51.9)334 (28.7)Initial suspected sepsis with hypotension43 (23.9)187 (16.0)0.009*Early warning score NEWS score, mean (± SD)10.2 ± 3.28.1 ± 3.5< 0.001* qSOFA score, mean (± SD)1.8 ± 0.81.4 ± 0.8< 0.001* SIRS, mean (± SD)2.5 ± 0.92.4 ± 1.10.26Source of suspected infection Respiratory tract infection106 (58.6)596 (51.1)0.06 Genitourinary tract infection41 (22.7)346 (29.6)0.05 Abdominal infection23 (12.7)129 (11.1)0.51 Central nervous system infection2 (1.1)5 (0.4)0.24 Skin and soft tissue infection5 (2.8)43 (3.7)0.53 Bloodstream/catheter-related infection3 (1.7)44 (3.8)0.15 Undefined site of infection12 (6.6)74 (6.3)0.88Initial investigation Blood lactate, median (IQR) (mmol/L)3.4 (2.3, 6.4)2.1 (1.5, 3.3)< 0.001* Hemoglobin, median (IQR) (g/dL)10 (8.4, 11.8)11 (9.2, 12.7)< 0.001* White blood cell count, median (IQR) (cell/mm3)10795 (6945, 15700)10438.5 (7058, 14587)0.78 Creatinine, median (IQR) (mg/dL)1.4 (0.8, 2.4)1.0 (0.7, 1.7)0.01*Treatment Vasoactive agent requirement21 (11.6)70 (6.0)0.001* Need supplementary oxygen158 (87.3)769 (65.9)< 0.001* Low flow oxygen requirement77 (42.5)577 (49.4)0.08 High flow oxygen requirement24 (13.3)61 (5.2)< 0.001* Using non-invasive ventilation1 (0.6)6 (0.5)0.95 Using invasive mechanical ventilation57 (31.5)122 (10.5)< 0.001*Timing from sepsis recognition to antibiotic delivered, median (IQR) (minute)40 (28, 60)40 (25, 60)0.79Hospital length of stay, median (IQR) (day)6 (2, 13)10 (6, 17)< 0.001*Dead from admission, median (IQR) (day)6.5 (2, 14)42 (29.5, 50)< 0.001*NEWS; National Early Warning Score, qSOFA; quick sequential organ failure assessment, SIRS; systemic inflammatory response syndrome Table 1Baseline characteristics between patients who died within 28 days (28-day mortality) and those who survived in patients suspected of sepsis in the emergency departmentCharacteristic28-day mortality (N = 181)n (%)Survived(N = 1167)n (%)p-valueSex0.27 Male96 (53.0)568 (48.7) Female85 (47.0)599 (51.3)Age, mean (± SD) (years)75.1 ± 15.372.3 ± 15.50.02*Body mass index, mean (± SD) (kg/m2)20.9 ± 4.922.3 ± 5.30.001*Comorbid disease Hypertension100 (55.2)696 (59.6)0.26 Diabetes mellitus62 (34.3)438 (37.5)0.39 Chronic obstructive pulmonary disease8 (4.4)39 (3.3)0.46 Chronic liver diseases7 (3.9)24 (2.1)0.13 Chronic kidney diseases22 (12.2)103 (8.8)0.35 Coronary artery disease14 (7.7)98 (8.4)0.76 Stroke41 (22.7)270 (23.1)0.87 Malignancy with metastasis21 (11.6)42 (3.6)< 0.001*Initial physiologic parameter Heart rate, mean (± SD) (bpm)108.1 ± 27.9102.1 ± 22.60.001* Systolic blood pressure, mean (± SD) (mmHg)112.7 ± 30.5123.3 ± 31.2< 0.001* Diastolic blood pressure, mean (± SD) (mmHg)67.6 ± 19.771.6 ± 18.40.001* Respiratory rate, mean (± SD) (/min)28.4 ± 6.226.1 ± 5.2< 0.001* Body temperature, mean (± SD) (°C)37.1 ± 0.937.6 ± 1.2< 0.001* Oxygen saturation, mean (± SD) (%)89.1 ± 8.792.6 ± 6.3< 0.001*Initial Glasgow coma scale< 0.001* 1574 (40.9)747 (64.3) 13–1413 (7.2)81 (7.0) ≤1294 (51.9)334 (28.7)Initial suspected sepsis with hypotension43 (23.9)187 (16.0)0.009*Early warning score NEWS score, mean (± SD)10.2 ± 3.28.1 ± 3.5< 0.001* qSOFA score, mean (± SD)1.8 ± 0.81.4 ± 0.8< 0.001* SIRS, mean (± SD)2.5 ± 0.92.4 ± 1.10.26Source of suspected infection Respiratory tract infection106 (58.6)596 (51.1)0.06 Genitourinary tract infection41 (22.7)346 (29.6)0.05 Abdominal infection23 (12.7)129 (11.1)0.51 Central nervous system infection2 (1.1)5 (0.4)0.24 Skin and soft tissue infection5 (2.8)43 (3.7)0.53 Bloodstream/catheter-related infection3 (1.7)44 (3.8)0.15 Undefined site of infection12 (6.6)74 (6.3)0.88Initial investigation Blood lactate, median (IQR) (mmol/L)3.4 (2.3, 6.4)2.1 (1.5, 3.3)< 0.001* Hemoglobin, median (IQR) (g/dL)10 (8.4, 11.8)11 (9.2, 12.7)< 0.001* White blood cell count, median (IQR) (cell/mm3)10795 (6945, 15700)10438.5 (7058, 14587)0.78 Creatinine, median (IQR) (mg/dL)1.4 (0.8, 2.4)1.0 (0.7, 1.7)0.01*Treatment Vasoactive agent requirement21 (11.6)70 (6.0)0.001* Need supplementary oxygen158 (87.3)769 (65.9)< 0.001* Low flow oxygen requirement77 (42.5)577 (49.4)0.08 High flow oxygen requirement24 (13.3)61 (5.2)< 0.001* Using non-invasive ventilation1 (0.6)6 (0.5)0.95 Using invasive mechanical ventilation57 (31.5)122 (10.5)< 0.001*Timing from sepsis recognition to antibiotic delivered, median (IQR) (minute)40 (28, 60)40 (25, 60)0.79Hospital length of stay, median (IQR) (day)6 (2, 13)10 (6, 17)< 0.001*Dead from admission, median (IQR) (day)6.5 (2, 14)42 (29.5, 50)< 0.001*NEWS; National Early Warning Score, qSOFA; quick sequential organ failure assessment, SIRS; systemic inflammatory response syndrome Baseline characteristics between patients who died within 28 days (28-day mortality) and those who survived in patients suspected of sepsis in the emergency department NEWS; National Early Warning Score, qSOFA; quick sequential organ failure assessment, SIRS; systemic inflammatory response syndrome Table 2 presents the results from the multivariable stepwise backward logistic regression analysis, which included variables associated with the prediction of 28-day mortality. Variables from Table 1 with p-values below 0.1 were included in the initial model analysis. The final model identified five significant predictors of 28-day mortality with p-values below 0.05. These significant predictors, respiratory rate (RR), need for supplemental oxygen, systolic blood pressure, heart rate, and Glasgow Coma Scale (GCS), were assigned integer weights derived from their multivariable β-coefficients by dividing each β by the smallest non-zero absolute β, rounding to the nearest integer, and bounding values to 0–3 to align with the NEWS framework (0 = no added risk; 3 = highest risk). Monotonicity (i.e., greater physiologic derangement received higher points) was verified and minimally adjusted if needed. Examination of observed 28-day mortality across respiratory rate intervals demonstrated a U-shaped pattern, with the lowest mortality occurring at approximately 18–26 breaths per minute, while both lower and higher respiratory rates were associated with increasing mortality risk (Fig. 2). The applied scores and cutoff values for each predictor were refined to achieve the best possible predictive performance while maintaining clinical interpretability. Consequently, the Respiratory National Early Warning Score (R-NEWS) scoring system for suspected sepsis in the ED, with scores ranging from zero to thirteen, was created and applied to all study subjects. The Hosmer–Lemeshow goodness-of-fit test showed no statistically significant difference between observed and predicted probabilities (p = 0.474). Calibration plots for 28-day and 7-day mortality comparing R-NEWS, predicted and observed risks, are shown in Fig. 3. Table 2Odds ratio of predictors and 28-day mortality from multivariable stepwise backward elimination logistic regression analysis, and item score assignmentPredictorOdds ratio95% CIp-valueBeta coefficientStandard errorScoreRespiratory rate (breath per minute) − 18–26RefRefRefRefRef0 − 27–301.91.3, 2.70.0020.620.521 - ≥ 312.51.6, 4.0< 0.0010.921.12 - ≤ 178.92.3, 34.80.0012.196.283Need Supplemental oxygen*2.11.3, 3.30.0010.740.462Systolic blood pressure (mmHg)* − 111–219RefRefRefRefRef0 − 101–1101.30.8, 2.30.2740.290.351 − 91–1002.11.2, 3.70.0070.760.642 - ≥ 2204.50.2, 91.90.3261.516.593 - ≤ 901.91.2, 2.90.0030.630.393Heart rate (beats per minute) * − 51–90RefRefRefRefRef0 − 91–1001.20.8, 1.90.3640.210.291 − 41–500.70.1, 6.00.758-0.330.71 − 111–1301.50.9, 2.40.0920.410.362 - ≥ 13121.2, 3.50.0090.710.543 - ≤ 40N/AN/AN/AN/AN/A3Glasgow coma scale − 15RefRefRefRefRef0 − 13–141.70.8, 3.30.1280.520.561 - ≤ 12.21.6, 3.2< 0.0010.790.392Constant0.030.02, 0.05< 0.001-3.60.01N/A* The grouping and scoring for need supplemental oxygen, systolic blood pressure, and heart rate use the same values as NEWS for ease of use. Other grouping and scoring methods did not affect the AuROC of the model Table 2Odds ratio of predictors and 28-day mortality from multivariable stepwise backward elimination logistic regression analysis, and item score assignmentPredictorOdds ratio95% CIp-valueBeta coefficientStandard errorScoreRespiratory rate (breath per minute) − 18–26RefRefRefRefRef0 − 27–301.91.3, 2.70.0020.620.521 - ≥ 312.51.6, 4.0< 0.0010.921.12 - ≤ 178.92.3, 34.80.0012.196.283Need Supplemental oxygen*2.11.3, 3.30.0010.740.462Systolic blood pressure (mmHg)* − 111–219RefRefRefRefRef0 − 101–1101.30.8, 2.30.2740.290.351 − 91–1002.11.2, 3.70.0070.760.642 - ≥ 2204.50.2, 91.90.3261.516.593 - ≤ 901.91.2, 2.90.0030.630.393Heart rate (beats per minute) * − 51–90RefRefRefRefRef0 − 91–1001.20.8, 1.90.3640.210.291 − 41–500.70.1, 6.00.758-0.330.71 − 111–1301.50.9, 2.40.0920.410.362 - ≥ 13121.2, 3.50.0090.710.543 - ≤ 40N/AN/AN/AN/AN/A3Glasgow coma scale − 15RefRefRefRefRef0 − 13–141.70.8, 3.30.1280.520.561 - ≤ 12.21.6, 3.2< 0.0010.790.392Constant0.030.02, 0.05< 0.001-3.60.01N/A* The grouping and scoring for need supplemental oxygen, systolic blood pressure, and heart rate use the same values as NEWS for ease of use. Other grouping and scoring methods did not affect the AuROC of the model Odds ratio of predictors and 28-day mortality from multivariable stepwise backward elimination logistic regression analysis, and item score assignment * The grouping and scoring for need supplemental oxygen, systolic blood pressure, and heart rate use the same values as NEWS for ease of use. Other grouping and scoring methods did not affect the AuROC of the model Fig. 2U-shaped association between respiratory rate and observed 28-day mortality with 95% confidence intervals U-shaped association between respiratory rate and observed 28-day mortality with 95% confidence intervals Fig. 3Calibration plots for 28-day and 7-day mortality comparing predicted and observed risks using the R-NEWS model Calibration plots for 28-day and 7-day mortality comparing predicted and observed risks using the R-NEWS model Table 3 displays the R-NEWS scoring system for suspected sepsis in the ED, which was modified from the NEWS scoring system. The adjustments to the respiratory rate and the GCS were based on the outcomes presented in Table 2. The scores for oxygen saturation and body temperature were removed to achieve the best possible AuROC. Comparisons of AuROC between R-NEWS and variants that include or exclude temperature and oxygen saturation are presented in Supplementary Figure S1, with accompanying statistics provided in the supplementary materials. Table 3The Respiratory National Early Warning Score (R-NEWS) scoring system for suspected sepsis in the ED*Physiological parameters3210123Respiratory rate (breaths per minute)≤ 1718–2627–30≥ 31Supplemental oxygenYesNoSystolic blood pressure (mmHg)≤ 9091–100101–110111–219≥ 220Heart rate(beats per minute)≤ 4041–5051–9091–110111–130≥ 131Glasgow coma scale≤ 1213–1415*Modified from National Early Warning Score (NEWS): London RCoP. National Early Warning Score (NEWS): Standardising the assessment of acute-illness severity in the NHS. Report of a working party Royal College of Physicians. 2012 Table 3The Respiratory National Early Warning Score (R-NEWS) scoring system for suspected sepsis in the ED*Physiological parameters3210123Respiratory rate (breaths per minute)≤ 1718–2627–30≥ 31Supplemental oxygenYesNoSystolic blood pressure (mmHg)≤ 9091–100101–110111–219≥ 220Heart rate(beats per minute)≤ 4041–5051–9091–110111–130≥ 131Glasgow coma scale≤ 1213–1415*Modified from National Early Warning Score (NEWS): London RCoP. National Early Warning Score (NEWS): Standardising the assessment of acute-illness severity in the NHS. Report of a working party Royal College of Physicians. 2012 The Respiratory National Early Warning Score (R-NEWS) scoring system for suspected sepsis in the ED* Heart rate (beats per minute) *Modified from National Early Warning Score (NEWS): London RCoP. National Early Warning Score (NEWS): Standardising the assessment of acute-illness severity in the NHS. Report of a working party Royal College of Physicians. 2012 Estimates of the accuracy of the model The discriminative ability of the R-NEWS scoring system is demonstrated in Fig. 4a and b, and Table 4. The AuROC of the development dataset for predicting 28-day mortality was 0.72 (95% CI: 0.68, 0.76), and for predicting 7-day mortality, it was 0.76 (95% CI: 0.71, 0.81). The DeLong’s test revealed statistically significant differences between the AuROC of the R-NEWS and NEWS scoring systems for both outcomes (p < 0.001 and p = 0.007, respectively). The AuROC of qSOFA for predicting 28-day mortality was 0.64 (95% CI: 0.60, 0.68) and for 7-day mortality was 0.65 (95% CI: 0.60, 0.71). The DeLong’s test comparing R-NEWS and qSOFA showed a p-value of less than 0.001 for both outcomes. The AuROC of Systemic Inflammatory Response Syndrome (SIRS) for predicting 28-day mortality was 0.52 (95% CI: 0.47, 0.55) and for 7-day mortality was 0.56 (95% CI: 0.50, 0.61). Similarly, The DeLong’s test comparing R-NEWS and SIRS also demonstrated a p-value of less than 0.001 for both outcomes. Fig. 4The area under the receiver operating characteristic curve (AuROC) of the Respiratory National Early Warning Score (R-NEWS), NEWS, quick Sequential Organ Failure Assessment (qSOFA), and Systemic Inflammatory Response Syndrome (SIRS) for predicting mortality in patients suspected of sepsis in the emergency department. A AuROC of R-NEWS compared with other scoring systems for predicting 28-day mortality. B. AuROC of R-NEWS compared with other scoring systems for predicting 7-day mortality The area under the receiver operating characteristic curve (AuROC) of the Respiratory National Early Warning Score (R-NEWS), NEWS, quick Sequential Organ Failure Assessment (qSOFA), and Systemic Inflammatory Response Syndrome (SIRS) for predicting mortality in patients suspected of sepsis in the emergency department. A AuROC of R-NEWS compared with other scoring systems for predicting 28-day mortality. B. AuROC of R-NEWS compared with other scoring systems for predicting 7-day mortality Table 4Discrimination power (area under the receiver operating characteristic curve, AuROC) of the respiratory National early warning score (R-NEWS), NEWS, quick sequential organ failure assessment (qSOFA), and systemic inflammatory response syndrome (SIRS) for predicting 28-day mortality and 7-day mortality in patients suspected of sepsis in the emergency department28-day mortalityAuROC95% confidence intervalp-value*R-NEWS0.720.68, 0.76RefNEWS0.670.63, 0.71< 0.001qSOFA0.640.60, 0.68< 0.001SIRS0.520.47, 0.55< 0.0017-day mortalityAuROC95% confidence intervalp-value*R-NEWS0.760.71, 0.81RefNEWS0.720.67, 0.770.007qSOFA0.650.60, 0.71< 0.001SIRS0.560.50, 0.61< 0.001*DeLong’s test Table 4Discrimination power (area under the receiver operating characteristic curve, AuROC) of the respiratory National early warning score (R-NEWS), NEWS, quick sequential organ failure assessment (qSOFA), and systemic inflammatory response syndrome (SIRS) for predicting 28-day mortality and 7-day mortality in patients suspected of sepsis in the emergency department28-day mortalityAuROC95% confidence intervalp-value*R-NEWS0.720.68, 0.76RefNEWS0.670.63, 0.71< 0.001qSOFA0.640.60, 0.68< 0.001SIRS0.520.47, 0.55< 0.0017-day mortalityAuROC95% confidence intervalp-value*R-NEWS0.760.71, 0.81RefNEWS0.720.67, 0.770.007qSOFA0.650.60, 0.71< 0.001SIRS0.560.50, 0.61< 0.001*DeLong’s test Discrimination power (area under the receiver operating characteristic curve, AuROC) of the respiratory National early warning score (R-NEWS), NEWS, quick sequential organ failure assessment (qSOFA), and systemic inflammatory response syndrome (SIRS) for predicting 28-day mortality and 7-day mortality in patients suspected of sepsis in the emergency department *DeLong’s test The model underwent internal validation using a bootstrap-based validation dataset constructed from two-thirds of the full cohort. A total of 890 patients were included, showing discrimination comparable to the development sample: 28-day mortality AuROC 0.71 (95% CI, 0.66, 0.76) and 7-day mortality AuROC 0.77 (95% CI, 0.71, 0.83). ROC curves from the internal validation are shown in Supplementary Figure S2. Probability categories of mortality Table 5 shows the distribution of R-NEWS scores by 28-day mortality, demonstrating a clear association whereby higher scores correspond to a higher prevalence of mortality. Patients were divided into three groups based on their R-NEWS scores: zero to three, four to six, and above six, categorized into low-risk, moderate-risk, and high-risk groups, respectively. Detailed performance metrics of R-NEWS for predicting 28-day mortality are presented in Supplementary Tables S1. Table 5Distribution of R-NEWS score categories by 28-day mortality status (death vs. survival)R-NEWS (point)Total(n = 1,337)Prevalence(%)28-day mortalitySurvived n % n %0302.213.32996.71896.70089100213710.353.713296.3322016.4135.920794.1420215.12512.417787.6520415.32512.317987.7618513.83116.815483.271067.92523.68176.48755.62026.75573.39523.91834.63465.410221.61045.51254.511100.8330.0770.01250.4240.0360.013000000 Table 5Distribution of R-NEWS score categories by 28-day mortality status (death vs. survival)R-NEWS (point)Total(n = 1,337)Prevalence(%)28-day mortalitySurvived n % n %0302.213.32996.71896.70089100213710.353.713296.3322016.4135.920794.1420215.12512.417787.6520415.32512.317987.7618513.83116.815483.271067.92523.68176.48755.62026.75573.39523.91834.63465.410221.61045.51254.511100.8330.0770.01250.4240.0360.013000000 Distribution of R-NEWS score categories by 28-day mortality status (death vs. survival) Table 6 shows the score-categorized probability groups, the likelihood ratio, and the 95% CI for 28-day and 7-day mortality. The 28-day mortality rates in the low, moderate, and high-risk groups were 3.9%, 13.7%, and 28.9%, respectively. The 7-day mortality rates in the low, moderate, and high-risk groups were 1.3%, 6.1%, and 16.7%, respectively. The distribution of NEWS scores by 28-day mortality and the score-categorized probability groups are presented in Supplementary Tables S2 and S3, respectively. Table 6R-NEWS category–specific predicted probabilities and likelihood ratios (with 95% confidence intervals) for 28-day and 7-day mortality among emergency department patients with suspected sepsisScore categorized probability groupsScore28-day mortalitySurvivedLikelihood ratio of positive95% confidence intervalp-value(n = 178)(n = 1,159)n (%)n (%)Low (n = 476)≤ 319 (3.9)457 (96.1)0.270.18, 0.42< 0.001Moderate (n = 591)4–681 (13.7)510 (86.3)1.030.87, 1.230.383High (n = 270)≥ 778 (28.9)192 (71.1)2.652.14, 3.27< 0.001Mean (± SD)6.3 (± 2.3)4.4 (± 2.3)< 0.001Score categorized probability groupsScore7-day mortalitySurvivedLikelihood ratio of positive95% confidence intervalp-value(n = 87)(n = 1,250)n (%)n (%)Low (n = 476)≤ 36 (1.3)470 (98.7)0.180.08, 0.39< 0.001Moderate (n = 591)4–636 (6.1)555 (93.9)0.930.72, 1.210.332High (n = 270)≥ 745 (16.7)225 (83.3)2.872.27, 3.63< 0.001Mean (± SD)6.7 (± 2.2)4.5 (± 2.3)< 0.001 Table 6R-NEWS category–specific predicted probabilities and likelihood ratios (with 95% confidence intervals) for 28-day and 7-day mortality among emergency department patients with suspected sepsisScore categorized probability groupsScore28-day mortalitySurvivedLikelihood ratio of positive95% confidence intervalp-value(n = 178)(n = 1,159)n (%)n (%)Low (n = 476)≤ 319 (3.9)457 (96.1)0.270.18, 0.42< 0.001Moderate (n = 591)4–681 (13.7)510 (86.3)1.030.87, 1.230.383High (n = 270)≥ 778 (28.9)192 (71.1)2.652.14, 3.27< 0.001Mean (± SD)6.3 (± 2.3)4.4 (± 2.3)< 0.001Score categorized probability groupsScore7-day mortalitySurvivedLikelihood ratio of positive95% confidence intervalp-value(n = 87)(n = 1,250)n (%)n (%)Low (n = 476)≤ 36 (1.3)470 (98.7)0.180.08, 0.39< 0.001Moderate (n = 591)4–636 (6.1)555 (93.9)0.930.72, 1.210.332High (n = 270)≥ 745 (16.7)225 (83.3)2.872.27, 3.63< 0.001Mean (± SD)6.7 (± 2.2)4.5 (± 2.3)< 0.001 R-NEWS category–specific predicted probabilities and likelihood ratios (with 95% confidence intervals) for 28-day and 7-day mortality among emergency department patients with suspected sepsis Discussion This study developed and internally validated the R-NEWS to predict 28-day and 7-day mortality among ED patients with suspected sepsis. R-NEWS demonstrated numerically higher discrimination than NEWS. High-risk R-NEWS (≥ 7) yielded high positive likelihood ratios for 28-day and 7-day mortality, whereas the low-risk category (≤ 3) showed low positive likelihood ratios. The key deviation from conventional NEWS is the RR weighting. In our sepsis cohort, the association between RR and mortality was non-linear (U-shaped), with the highest risk at lower RRs (≤ 17/min) despite this range being normal in the general population. In multivariable analysis using 18–26/min as the reference, RR ≤ 17/min was linked to a markedly increased odds of 28-day mortality (OR 8.9, p = 0.001), whereas RR 27–30 and ≥ 31/min showed more moderate risk elevations (OR 1.9 and 2.5, respectively). We therefore assigned the maximum score (3 points) to RR ≤ 17/min to reflect its empirically derived prognostic weight in this ED sepsis population. We hypothesize that RR typically rises above normal in sepsis due to physiologic stressors (fever, dehydration, compensation for metabolic acidosis); thus, a normal or low RR may signal ventilatory failure or central depression (e.g., fatigue, hypercapnia, encephalopathy) and herald impending decompensation, explaining its stronger association with mortality than mild–moderate tachypnea. Early identification of high-risk sepsis patients is crucial for prompt treatment, significantly improving outcomes. These findings suggest that R-NEWS offers a more reliable prediction of short-term and intermediate-term mortality than NEWS. The improved predictive performance highlights the potential of R-NEWS as a valuable tool for clinicians in the ED, enhancing early sepsis management and contributing to better mortality outcomes. NEWS, a well-known early warning score, has been utilized to identify deteriorating conditions and clinical deterioration in the ED [14]. The advantages of the NEWS score are: (1) it has been a standard tool in many EDs and inpatient wards, (2) it is calculated only by physiological parameters, without requiring laboratory values, and (3) it detects clinical deterioration early and helps physicians provide timely intervention. A systematic review of the NEWS score to predict 24-hour mortality and 30-day mortality in patients with infection demonstrated a high AuROC of 0.90 and 0.76, respectively [15]. The disadvantages of NEWS are: (1) the calculation of the NEWS score involves numerous physiological parameters, which complicates its practical application and increases staff workload, (2) decreased attention to signs of deterioration, and (3) low sensitivity for detecting early clinical deterioration. The predictive accuracy may vary in different settings, potentially leading to missed cases of deterioration or false alarms [16]. qSOFA is known as a tool for the early identification of sepsis patients at risk of mortality. Similar to R-NEWS, qSOFA is easy to calculate and does not require laboratory tests, making it quick to use in emergency settings. Although qSOFA has shown high specificity for predicting mortality in non-ICU settings, it has poor sensitivity, potentially missing patients who may benefit from early interventions [17, 18]. Compared to qSOFA, R-NEWS includes more parameters related to respiratory and cardiovascular function. A more comprehensive assessment of the patient’s condition leads to more accurate risk stratification, thus improving its ability to predict mortality. In the comparison between R-NEWS and the NEWS score, the R-NEWS was adapted from the original NEWS parameters to create a more simplified and practical version. Specifically, R-NEWS includes only five parameters: respiratory rate, the need for oxygen supplementation, systolic blood pressure, heart rate, and the GCS. Body temperature (BT) and oxygen saturation were removed to achieve the best performance of the score. Fever often develops from inflammation during sepsis. BT was excluded from the R-NEWS score despite its significance in sepsis for several reasons: (1) variability in measurement; BT can vary due to different measurement methods, environmental factors, medications, and improper techniques, making it less reliable as a consistent predictor, and (2) the evidence is not strong enough on prognostic value. Studies have shown mixed results regarding the association between BT and mortality in sepsis patients. A study by Lee et al. found no significant association between BT and mortality in critically ill septic patients requiring ICU care [19]. However, a study conducted by Inghammer et al. highlighted the significance of BT in predicting mortality in ED settings. Moreover, elevated BT was associated with a higher rate of sepsis bundle achievement in the ED [20]. Another study revealed a higher mortality rate in patients with low BT and delayed time to achieve the sepsis bundle [21]. Oxygen saturation is one of the parameters measured during the initial assessment of patients in the ED. A decrease in oxygen saturation serves as an indicator of hypoxemic respiratory failure. Many EWSs have oxygen saturation in their scores, like NEWS, MEWS, and REMS. The inconsistency of SpO2 measurements obtained via pulse oximetry is affected by several limitations such as poor tissue perfusion, motion artifacts, device variability, cold extremities, vasopressor use, sepsis, and septic shock [22]. In our analysis, SpO2 was also removed from the R-NEWS for the following reasons: (1) redundancy with other parameters; SpO2 often correlates with other respiratory and circulatory parameters, such as respiratory rate and the need for supplemental oxygen, making it somewhat redundant when these factors are already included in the score, and (2) simplification and practicality; removing SpO2 simplifies the scoring process, reducing the workload for ED staff while maintaining the predictive performance of the score. This makes R-NEWS easier to implement in the ED. Numerous EWSs incorporate respiratory rate as a key component, reflecting its value as a predictor of adverse outcomes [23–25]. In the R-NEWS score, the weighting assigned to RR is increased compared to NEWS. For R-NEWS, an RR of 27–30 breaths per minute is allocated 1 point, and a rate of 31 breaths per minute or more is allocated 2 points. Pneumonia is the most common source of infection leading to sepsis in the ED [26, 27]. In our study, we observed that approximately half of the patients presented with respiratory tract infections. In the dataset utilized for the development of our scoring system, patients who developed adverse outcomes exhibited higher RR. Consequently, we assigned a higher RR to the R-NEWS, reflecting its significance in predicting the outcomes. As supported by recent evidence showing that data-driven predictive models outperform clinician judgment alone in identifying patients at risk of adverse outcomes [28], the R-NEWS score provides an objective adjunct to bedside assessment and may enhance early recognition of deterioration in sepsis. Existing prognostic models, such as the SOFA and its modified version (mSOFA), have demonstrated strong predictive value for ICU mortality but are often impractical in emergency settings due to their reliance on laboratory parameters [29]. The R-NEWS model aims to address this limitation by utilizing readily available physiological variables at triage. Furthermore, comparable studies have shown that general-purpose scoring systems such as NEWS and qSOFA demonstrate only moderate discrimination for mortality prediction (AUC 0.65, 0.72) [30]. These findings underscore the limitations of universal tools and highlight the need for sepsis-specific adaptations such as R-NEWS, while also emphasizing the importance of further validation across diverse cohorts. Limitations The limitations of our study include: (1) It was conducted at a single center, potentially limiting the generalizability of the findings to other healthcare settings with different patient populations, resources, and clinical practices. (2) The scoring system utilized the GCS instead of the Alert, Voice, Pain, Unresponsive (AVPU) scale for assessing neurological function. While GCS provides detailed neurological assessment, it may reduce ease of use and rapid applicability, especially by non-specialized medical staff, potentially impacting practicality in emergency settings. (3) Temperature did not demonstrate a significant association with 28-day mortality, possibly due to low variability in temperature measurements or environmental influences before ED arrival, which may limit the generalizability of this finding. (4) Due to the observational design, inherent biases, including selection and information bias, may have influenced the results. The analyses were conducted using conventional regression methods; however, recent ED prediction-model research increasingly compares such approaches with machine learning techniques, which may offer improved predictive performance in some contexts [31]. (5) Because follow-up was limited to the index hospitalization, deaths after discharge but within 28 days could not be captured, potentially leading to outcome misclassification and bias. Therefore, the reported 28-day in-hospital mortality should be interpreted with caution, and future studies with post-discharge follow-up are warranted. Finally, although R-NEWS showed favorable predictive performance in our study setting, further validation across diverse clinical environments, patient demographics, and healthcare systems is required before recommending widespread adoption. Conclusion In this single-center study, the Respiratory National Early Warning Score (R-NEWS) demonstrated utility for predicting 28- and 7-day mortality among ED patients with suspected sepsis. Because it uses fewer parameters than NEWS, R-NEWS may offer a more pragmatic option for busy ED workflows without evidence of compromised discrimination on internal validation. Used alongside clinical judgment, R-NEWS could support timely risk stratification and early sepsis management. Nevertheless, external, multicenter validation and prospective evaluation of clinical impact are needed before recommending widespread implementation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1