- Email: [email protected]
Short National Early Warning Score — Developing a Modified Early Warning Score
G Model
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
Australian Critical Care xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Australian Critical Care journal homepage: www.elsevier.com/locate/aucc
Research paper
Short National Early Warning Score — Developing a Modified Early Warning Score Leandro Luís a Carla Nunes b,∗ a
Centro Hospitalar Lisboa Central, Escola de Nacional de Saúde Pública — Universidade Nova de Lisboa, Portugal Centro de Investigac¸ão em Saúde Pública, Escola de Nacional de Saúde Pública — Universidade Nova de Lisboa, Avenida Padre Cruz, 1600 560 Lisboa, Portugal b
article information Article history: Received 16 May 2017 Received in revised form 7 November 2017 Accepted 12 November 2017 Keywords: NEWS Hospital Clinical derangement Vital signs Patient safety Statistic models
a b s t r a c t Introduction: Early Warning Score (EWS) systems have been developed for detecting hospital patients clinical deterioration. Many studies show that a National Early Warning Score (NEWS) performs well in discriminating survival from death in acute medical and surgical hospital wards. NEWS is validated for Portugal and is available for use. A simpler EWS system may help to reduce the risk of error, as well as increase clinician compliance with the tool. Objectives: The aim of the study was to evaluate whether a simplified NEWS model will improve use and data collection. Methods: We evaluated the ability of single and aggregated parameters from the NEWS model to detect patients’ clinical deterioration in the 24 h prior to an outcome. There were 2 possible outcomes: Survival vs Unanticipated intensive care unit admission or death. We used binary logistic regression models and Receiver Operating Characteristic Curves (ROC) to evaluate the parameters’ performance in discriminating among the outcomes for a sample of patients from 6 Portuguese hospital wards. Results: NEWS presented an excellent discriminating capability (Area under the Curve of ROC (AUCROC) = 0.944). Temperature and systolic blood pressure (SBP) parameters did not contribute significantly to the model. We developed two different models, one without temperature, and the other by removing temperature and SBP (M2). Both models had an excellent discriminating capability (AUCROC: 0.965; 0.903, respectively) and a good predictive power in the optimum threshold of the ROC curve. Conclusions: The 3 models revealed similar discriminant capabilities. Although the use of SBP is not clearly evident in the identification of clinical deterioration, it is recognized as an important vital sign. We recommend the use of the first new model, as its simplicity may help to improve adherence and use by health care workers. © 2017 Australian College of Critical Care Nurses Ltd. Published by Elsevier Ltd. All rights reserved.
1. Introduction Patients who suffer from an adverse event on the ward, such as cardiac arrest or death, often have a physiological deterioration several hours before this event.1–3 Many of these events are potentially predictable and preventable by a timely recognition of the patient derangement and a proper management by a skilled team.1,4–6 The failure to identify and monitor the warning signs, and to escalate the care for the patients at risk for clinical deterioration, has led to the development of rapid response systems.3,6–8
These systems usually have a “track and trigger” system that allows the identification of the patients with physiological abnormalities, allowing a fast and effective response.3,6,9 These “track and trigger” systems are frequently known as early warning score (EWS) systems,1 and score the degree of vital sign derangement.3,6,10 This score is used to define an individual care pathway for these patients, such as more frequent monitoring of vital signs or the involvement of more experience professionals, in particular the rapid response team.10,11 The use of EWS systems is recommended in the United Kingdom for the early recognition and response to patients’ clinical deterioration.11–14 There are more than 100 EWS systems available,1 such as Modified Early Warning Score,5 Vital PacTM
∗ Corresponding author. E-mail address: [email protected] (C. Nunes). https://doi.org/10.1016/j.aucc.2017.11.004 1036-7314/© 2017 Australian College of Critical Care Nurses Ltd. Published by Elsevier Ltd. All rights reserved.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
2
Early Warning Score (ViEWS)15 or National Early Warning Score (NEWS).14 ViEWS15 and NEWS,14 in their development studies, had presented better discriminating capability than the other 33 EWS systems in the United Kingdom for the 24-h period prior to a death versus survival outcome.11,15 A study from Luís16 has validated these two systems for Portugal. Therein, both systems showed an excellent discriminating capability to the outcomes survival vs unanticipated intensive care unit admission (UICUA) and survival vs death.16 Some studies suggest that the collection of all the necessary data to complete EWS scores in short periods may be time consuming for health professionals.17–20 In addition, the absence of missing values or errors in these vital signs records are essential and may have consequences in the EWS use.16–18,21,22 The number of variables used in the EWS systems may influence the data collection workload for its use and add complexity to the healthcare professionals in their work, increasing the risk of errors. A simpler EWS with less variables can help to reduce the risk of error and unnecessary effort and time, leading to more reliable scores for an identical system performance.21,22 Therefore, recognising the specificities of the clinical practice and the workload imposed by a systematic collection of several datasets of vital parameters, we assume the need to develop a modified NEWS system with fewer parameters that can contribute to better adherence and fewer errors in its use. The main goal of this study is to develop a simpler modified NEWS with the vital parameters that most significantly anticipate a negative outcome (specifically death or unanticipated intensive care unit admission) for hospital inpatients.
2. Methods We evaluated the individual parameters of the NEWS model for an appropriate detection of the outcomes, building an aggregate system properly adapted to the suited sample. We intended to make the NEWS model simpler and more user-friendly, maintaining a high discriminating capability. 2.1. Study design A prospective cohort study, with retrospective analysis, was conducted in 2 medical wards, 3 surgical wards and 1 Haematology ward in Lisbon Central Hospital Centre (LCHC). 2.2. Sample The current study was based on data from another study,16 authorised by the LCHC board and ethics committee (LCHC Ethics Committee approval: 107/2012). It used a convenience sampling based on nurses’ records for the patients admitted to the study wards between December 1st and 31st, 2012. Patients were excluded if they were less than 18 years old, were pregnant, were in a preoperative phase for elective surgery, or in palliative care. Incomplete records (with missing values in at least one NEWS variables) were also excluded from the sample. For survival outcome, the highest NEWS score in patients with all the parameters filled during hospitalization episodes in the LCHC wards was used. For the other outcome, including a combination of both death and unanticipated intensive care unit admission (UICUA), we considered the NEWS score obtained 24 h before the outcome. This option was made to allow a better analysis of the performance of the EWS score in discriminating the outcomes,
allowing to compare the maximum derangement of patients who survived with those who died or had an UICUA. To reduce the frequency of absent values, nurses were asked to register all the NEWS parameters. All data were collected daily and registered in a digital document, which computed a NEWS score automatically. The data were anonymised for analysis purposes. 2.3. Variables The variables of interest in the study were NEWS score (Numerical Scale between 0 and 20) and the individual scores of each NEWS parameter: respiratory rate, oxygen saturations, body temperature, systolic blood pressure (SBP), heart rate, level of consciousness (numerical scale between 0 and 3) and any supplemental oxygen (numerical scale between 0 and 2). The study outcomes were survival or death/UICUA. Survival meant that the patient had been discharged from the hospital, with an improvement of his or her health status. Death or UICUA represented a deterioration of the clinical situation of the patients, considering only “intended to treat situations” in UICUA situations as suggested by National Institute for Health and clinical Excellence (NICE) in 2007.12 Therefore, these last two results (death and UICUA) were combined into a single outcome. 2.4. Data analysis We conducted a descriptive analysis of survival or death/UICUA distribution by gender (Male/Female), admission diagnosis (Medical/Surgical/Haematological) and age group (18–44 years, 45–64 years, 65–84 years, above 85 years). The Area Under the Curve of the Receiver Operating Characteristic curve (AUROC) was calculated for the outcomes death/UICUA versus survival, using the single NEWS categories defined in its development.14 This allowed us to evaluate the discriminant capability of the system (total score) and single parameters (independently) to detect a critical case — associated with a high probability to have a negative event (death/UICUA) in the next 24 h. We assumed an AUROC value between 0.5 (no better than chance at predicting the outcome) and 1 (perfect prediction of the outcome). An AUROC between 0.700–0.800 indicated a reasonable discrimination capability and values exceeding 0.800 a good discrimination capability.11 We compared the highest NEWS score reached during the hospital ward stay for each patient (survival group) with the NEWS score from the death/UICUA group in the 24 h prior to the outcome. We began our study by assessing the individual contribution of each NEWS parameter to predict the outcome death/UICUA using binary logistic regression methods. This test helped us to decide which variables were important for the NEWS predictive capability performance, allowing us to evaluate alternative NEWS models. Models’ calibration was assessed using the Hosmer–Lemeshow goodness of fit test. Furthermore, we estimated Odds Ratios and respective 95% confidence intervals. In the second phase of the study, we developed two new aggregated scales, considering both AUROC values and statistical significance in regression models. Then, the discriminant capability of these two scales was analysed using the AUROC.23,24 The new scales were compared with the original NEWS scale. The discriminant capability evaluation of each EWS was complemented by calculating its predictive power. We analysed the sensitivity and specificity in the optimal threshold for the scales using the Youden index (J).25 All calculations were performed using the IBM software Statistical Package for Social Sciences, v.22 (Armonk, NY: IBM Corp).
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
3
Table 1 Sample sociodemographic characterization. Outcome
Global
Variables
Absolute frequency (Episodes)
Relative frequency (%)
Survival Absolute frequency (Episodes)
Relative frequency (%)
Absolute frequency (Episodes)
Death/UICUA Relative frequency (%)
Gender
Female Male
171 159
51.8 48.2
155 141
90.6 88.7
16 18
9.4 11.3
Admission diagnosis
80 Surgical Medical 145 Haematological 105
24.2 43.9 31.8
70 130 96
87.5 89.7 91.4
10 15 9
12.5 10.3 8.6
Age group
18–44 years 45–64 years 65–84 years ≥85 years
38 96 156 40
11.5 29.1 47.3 12.1
36 89 136 35
94.7 92.7 87.2 87.5
2 7 20 5
5.3 7.3 12.8 12.5
330
100
296
89.7
34
10.3
Total
Table 2 AUROC for NEWS global model and Single parameters. Variables
NEWS Heart rate Respiratory rate Temperature Systolic blood pressure Oxygen saturations Any supplemental oxygen Level of consciousness
AUROC
0.944 0.766 0.618 0.51 0.711 0.729 0.827 0.783
Standard error
0.016 0.05 0.056 0.051 0.048 0.053 0.03 0.05
p-value
<0.001 <0.001 0.025 0.846 <0.001 <0.001 <0.001 <0.001
Confidence interval (CI) 95% IL
UL
0.912 0.668 0.509 0.411 0.616 0.626 0.768 0.684
0.975 0.864 0.727 0.609 0.806 0.833 0.886 0.882
Table 1 shows the sociodemographic characteristics of our sample (n = 330). The sample contained slightly more females (51.8%) than males. Medical inpatient diagnosis was predominant in the hospitalization episodes (43.9%), compared to haematological diagnoses (31.8%) and surgical diagnoses (24.2%). There was a predominance of hospitalisation episodes of older people aged between 65 and 84 years (47.3%). Concerning the hospitalisation outcome, there were more cases of death or UICUA in males (Male = 11.3%; Female = 9.4%), surgical admissions (Surgical = 12.5%; Medical = 10.3%; Haematological = 8.6%) and in ages equal or above 65 years old (65–84 years = 12.8%; ≥85 years = 12.5%). 3.2. Discriminant power analysis of NEWS vital parameters
Fig. 1. ROC curves for NEWS, Model 1 and Model 2 legend. NEWS. Model 1 — Short NEWS. Model 2. Reference Line.
3. Results
The discriminating capability analysis of NEWS and their individual parameters are presented by the AUROC values in Table 2. NEWS had an excellent discriminating capability, with an AUROC value of 0.944 (p-value <0.001; 95% CI: 0.912–0.975). NEWS model single vital parameters had AUROC values between 0.618 (respiratory rate) and 0.827 (Oxygen Saturation). Temperature was the only parameter that had no discriminant capability, with an AUROC value of 0.51. This value was not statistically significant (p-value >0.05). This suggested the need to evaluate the individual contributions of the parameters to the NEWS model. 3.3. NEWS model individual vital parameters contribution analysis
3.1. Sample description From an initial sample of 439 patients admitted to the wards, 109 patients were excluded due to missing values (24.8%). The remaining 330 patients were used for the purpose of this analysis.
The analysis of single parameters’ contribution to the NEWS model was performed with binary logistic regressions, using Death/UICUA as the event being modelled. Likelihood ratio tests allowed us to verify that at least one of the parameters was a sig-
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
4 Table 3 Variables included in the model. Variables
ˇ
Standard error
Wald’s test
p-value
Odds ratio
Heart rate Respiratory rate Temperature Systolic blood pressure Oxygen saturations Supplemental oxygen Level of consciousness Constant
1.759 0.884 0.407 −0.024 0.806 1.852 1.288 −9.79
0.403 0.287 0.375 0.249 0.301 0.516 0.27 1.743
19.09 9.497 1.177 0.009 7.17 12.876 22.723 31.528
<0.001 0.002 0.278 0.922 0.007 <0.001 <0.001 <0.001
5.805 2.421 1.502 0.976 2.238 6.373 3.627 0
Odds ratio interval (CI:95%) Inferior limit
Upper limit
2.637 1.38 0.72 0.599 1.241 2.317 2.136
12.778 4.249 3.135 1.591 4.038 17.527 6.161
Table 4 Short NEWS. Physiological parameters Respiratory rate Oxygen saturations Any supplemental oxygen Systolic blood pressure Heart rate Level of consciousness
3 ≤8 ≤91 ≤90 ≤40
2 92–93 Sim 91–100
1 9–11 94–95 101–110 41–50
nificant predictor to the outcome (chi-square = 141.724; p-value <0.01). Model fit was evaluated by performing the Hosmer–Lemeshow test, which revealed that the variables fit the model (chisquare = 2.522; p-value = 0.961). Each variable contributing to the model is shown in the next table (Table 3). Temperature and SBP were found not to be significant predictive variables in the model, showing a p-value for the Wald test higher than 0.05. The parameters that contributed the most to the model were heart rate (OR = 5.805; p-value <0.001) and the presence of supplemental oxygen (OR = 6.373; p-value <0.001), with the higher odds ratio. Considering these results and the clinical value of each parameter for derangement prediction, we developed two test models, which were assessed for their discriminatory capacity for the outcomes survival versus death/UICUA. In one model, called Model 1, we excluded temperature. In the second model, Model 2, we excluded temperature and SBP.
3.4. Discriminant capacity of the new models Considering that temperature and SBP did not have a significant contribution to the NEWS model predictive power, we developed two alternative models. These two models maintained the vital parameters: heart rate, respiratory rate, oxygen saturations, presence of supplemental oxygen and level of consciousness. They were tested for their discriminant power against the original model. The results are presented in Fig. 1. The new models had excellent discriminatory capability, with similar values as NEWS. However, Model 1 AUROC (removing temperature) had a higher value than the other models (AUROC = 0.965; p-value <0.001; 95% CI: 0.942–0.988). Removing the SBP parameter as well as temperature decreased the discriminant capability of the aggregated model. This second model had worse results than the original NEWS and model 1, with an AUROC = 0.903 (p-value <0.001; 95% CI: 0.855–0.950). The predictive capability evaluation of the models showed that NEWS had a sensitivity of 91.2% and a specificity of 78.4% at an optimum threshold of 6.5 (Youden Index (J) = 0.696). Model 1 correctly identified 80.7% of survivals when the score is below 5.5, and 97.2%
0 12–20 ≥96 Não 111–219 51–90 Alert (A)
1
2 21–24
91–110
111–130
3 ≥25
≥220 ≥131 Verbal (V) Pain (P) Unresponsive (U)
of Death/UICUA when the score is over that value (J = 0.778). Model 2 had a sensitivity of 82.4% and a specificity of 82.8% at an optimum threshold of 4,5 (J = 0.652). The models’ comparison allowed us to verify that Model 1 had a better overall performance predicting the outcomes for the optimum threshold. However, Model 2 had a slightly higher specificity than Model 1 in the optimum threshold. According to Hosmer–Lemeshow test, all the models had correctly fitted the data (p-value >0,05). Model 1, referred to here as Short NEWS, is presented in Table 4.
4. Discussion The current study detailed the NEWS model and, through the clarification of each parameter’s importance, offers a simpler EWS for broader use and dissemination in the clinical field. The study results have shown excellent discriminating power of NEWS for the outcomes of Survival and Death/UICUA (AUROC = 0.944; 95% CI: 0.912–0.975). The results obtained by Smith et al.11 demonstrated good discriminant power (AUROC = 0.865; 95% CI: 0.858–0.872) of NEWS for an outcome that grouped Death, UICUA and cardiorespiratory arrest. The actual study results were even better than the Smith et al.11 study, but the sample was smaller and therefore, does not allow us to generalise the results. This study aimed to test if any NEWS vital parameter had low or no influence on the discriminant and predictive capability of the NEWS aggregated model. We wanted to optimise the NEWS model and present a simpler model, with fewer parameters and the best clinical applicability. It was possible to verify that temperature and SBP had a low contribution for the discriminant and predictive capability of NEWS. Without these parameters, the aggregated model with the other parameters had similar results. Scarce evidence exists about temperature’s impact in acute clinical deterioration predictive models. However, since temperature is considered a physiological disorder and an important disease severity signal, it is used in the NEWS model.14 In the case of SBP, the Royal College of Physicians14 refers to it as a central element in the definition of clinical deterioration of the patient. NICE12 refers to the importance of temperature as a central element to the EWS. This is explained by referring to studies where this parameter was included. However, there is no explanation or empirical support
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
of the capability of this parameter to predict death. Despite this consideration, changes in temperature are in many situations associated with infections that do not result in an immediate clinical deterioration of patients. These changes may be present in isolation for extended periods. Considering the NEWS model in the analysed period, we verified that temperature had no effect as predictor of death or UICUA, so its exclusion led to similar results to those obtained with the original model. Temperature is an important vital sign, but its use in a EWS system may pose some difficulties, especially when there is a clinical escalation protocol that increases its evaluation frequency. The study by Clifton et al.18 showed in a review of 16 795 EWS vital signs observation sets that temperature was the most commonly missed vital sign, being absent in 11.4% of total EWS. This may be a problem when the parameter is essential to the model. Observing Smith et al.26 study about single parameters track and trigger systems, temperature was not presented as a central element of these systems. In this same study, SBP was always present and was considered relevant for the detection of patient’s clinical deterioration. The same authors have shown the same assumptions in another study about aggregated track and trigger systems.10 Since temperature was the only excluded variable in the comparisons made in these studies, it seems that there is a similarity of the discriminant capability between the models with and without this variable. This may suggest a weak contribution from this parameter to model discrimination, which concurs with our results. Some studies revealed that a simpler EWS had better data collection with fewer errors and a significant improvement in the quality of the results.12,21,22 Considering the study results, we have performed a discriminant analysis for the NEWS derived models, one without temperature (Model 1 — named here as Short NEWS) and the other without SBP and temperature (Model 2). Short NEWS showed slightly better results than the original NEWS. The option of maintaining SBP was due to the better results of the model, because it was a routinely and frequently collected parameter and for the importance given to this parameter in the literature observed.14 We verified that the use of a simpler model may allow for better agreement in its application in clinical practice.21,22 This may increase the health professional’s compliance with the data collection required for the EWS. The comparison of the discriminant and predictive capacity of Short NEWS against NEWS and Model 2 have shown similarity in the sensitivity and specificity of the scale. However, Short NEWS had better results for the optimum threshold than the other models. This may support the use of Short NEWS as a good alternative to NEWS in Portugal. 4.1. Limitations The small sample of patients with the outcome of UICUA or death was a limitation to any strong conclusion in this study. The short timeframe was another limitation, because seasonal effects may influence the results. Restricting the study to only a hospital limited the generalisability of the results, but allowed a demonstration of the model’s potential. 5. Conclusions This study presented a different approach to the NEWS model, with a detailed analysis of the NEWS parameters and their contribution to its goal of detecting the clinical deterioration of in-hospital patients. The weak contribution of the SBP and temperature parameters to NEWS performance allowed the opportunity to develop two
5
alternative models, without losing quality in detecting death or unanticipated intensive care unit admission. Short NEWS (not including the temperature parameter) has shown discriminant and predictive power superior to the others, using fewer parameters. Considering this and the scientific evidence that holds that SBP is an important predictor of mortality, Short NEWS was suggested for use in clinical practice. The study proposed a tool to support clinical practice decisions in the Portuguese reality, based on less data, a more rational use of worktime for health professionals and without losing its discriminant capacity. This methodology is relatively easy to be used in other countries, potentially bringing similar results or the opportunity to parameterize a tailored local NEWS. Authors’ contributions Conception, design and performing the research: LL; Analysis and interpretation of data: LL, CN; Drafting the manuscript: LL, CN; Final approval of the version to be submitted: LL, CN; Agreement to be accountable for all aspects of the work: LL, CN. Funding None. References 1. Churpek MM, Yuen TC, Edelson DP. Risk stratification of hospitalized patients on the wards. Chest 2013;143(6):1758–65. 2. Jansen JO, Cuthbertson BH. Detecting critical illness outside the ICU: the role of track and trigger systems. Curr Opin Crit Care 2010;16:184––190. 3. Smith GB, Prytherch DR, Schmidt P, Featherstone PI, Knight D, Clements G, et al. Hospital-wide physiological surveillance — a new approach to the early identification and management of the sick patient. Resuscitation 2006;71:19–28. 4. Goldhill DR, Mcnarry AF, Mandersloot G, Mcginley A. A physiologically-based early warning score for ward patients: the association between score and outcome. Anaesthesia 2005;60:547–53. 5. Subbe CP, Kruger M, Rutherford P, Gemmel L. Validation of a modified Early Warning Score in medical admissions. QJM 2001;94(10):521–6. 6. Jones DA, DeVita MA, Bellomo R. Rapid-response teams. N Engl J Med 2011;365:139–46. 7. Maharaj R, Raffaele I, Wendon J. Rapid response systems: a sysreview and meta-analysis. Crit Care 2015;19(1):254, tematic http://dx.doi.org/10.1186/s13054-015-0973-y. 8. Stafseth SK, Grønbeck S, Lien T, Randen I, Lerdal A. The experiences of nurses implementing the Modified Early Warning Score and a 24-hour oncall Mobile Intensive Care Nurse: an exploratory study. Intensive Crit Care Nurs 2016;34:33–41. Elsevier Ltd. 9. DeVita MA, Smith GB, Adam SK, Adams-pizarro I, Buist M, Bellomo R, et al. “Identifying the hospitalised patient in crisis” — a consensus conference on the afferent limb of Rapid Response Systems. Resuscitation 2010;81(4):375–82. Elsevier Ireland Ltd. 10. Smith GB, Prytherch DR, Schmidt PE, Featherstone PI. Review and performance evaluation of aggregate weighted ‘track and trigger’ systems. Resuscitation 2008;77:170–9. 11. Smith GB, Prytherch DR, Meredith P, Schmidt PE, Featherstone PI. The ability of the National Early Warning Score (NEWS) to discriminate patients at risk of early cardiac arrest, unanticipated intensive care unit admission, and death. Resuscitation 2013;84(4):465–70. 12. National Institute for Health and Clinical Excellence. Acutely ill patients in hospital. London: NICE; 2007. 13. National Confidential Enquiry into Patient Outcome and Death. Time to Intervene? A review of patients who underwent cardiopulmonary resuscitation as a result of an in-hospital cardiorespiratory arrest. London: NCEPOD; 2012. 14. Royal College of Physicians. National Early Warning Score (NEWS) — Standardising the assessment of acute-illness severity in the NHS. Report of a working party. RCP; 2012, 47 p. 15. Prytherch DR, Smith GB, Schmidt PE, Featherstone PI. ViEWS — towards a national early warning score for detecting adult inpatient deterioration. Resuscitation 2010;81(8):932––937. Elsevier Ireland Ltd. 16. Luís L. Traduc¸ão, Validac¸ão e Aplicac¸ão dos Sistemas de Pontuac¸ão de Alerta Precoce “ViEWS” e “NEWS” em Portugal. Escola Superior de Tecnologias da Saúde de Lisboa — Instituto Politécnico de Lisboa, Escola Superior de Saúde da Universidade do Algarve; 2014. 17. Prytherch DR, Smith GB, Schmidt P, Peter I, Stewart K, Knight D, et al. Calculating early warning scores — a classroom comparison of pen and paper and hand-held computer methods. Resuscitation 2006;70:173–8.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
18. Clifton DA, Clifton L, Sandu D, Smith GB, Tarassenko L, Vollam SA, et al. ‘Errors’ and omissions in paper-based early warning scores: the association with changes in vital signs — a database analysis. BMJ Open 2015;5, http://dx.doi.org/10.1136/bmjopen-2014-007376. 19. Hands C, Reid E, Meredith P, Smith GB, Prytherch DR, Schmidt PE, et al. Patterns in the recording of vital signs and early warning scores: compliance with a clinical escalation protocol. BMJ Qual Saf 2013;(April):1–8. 20. Kolic I, Crane S, McCartney S, Perkins Z, Taylor A. Factors affecting response to National Early Warning Score (NEWS). Resuscitation 2015;90:85–90. 21. Subbe CP, Harrison DA. Reproducibility of physiological track-and-trigger warning systems for identifying at-risk patients on the ward. Intensive Care Med 2007;33:619–24.
22. Jarvis S, Kovacs C, Briggs J, Meredith P, Schmidt PE, Featherstone PI, et al. Can binary early warning scores perform as well as standard early warning scores for discriminating a patient’s risk of cardiac arrest, death or unanticipated intensive care unit admission? Resuscitation 2015;93:46––52. 23. Pepe MS. The statistical evaluation of medical tests for classification and prediction. New York: Oxford University Press; 2009. 24. Zou H, Liu A, Bandos AI, Ohno-Machado L, Rockette E. Statistical evaluation of diagnostic performance — topics in ROC analysis. Boca Raton, Florida: CRC Press; 2012. 25. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3(1):32–5. 26. Smith GB, Prytherch DR, Schmidt PE, Featherstone PI, Higgins B. A review, and performance evaluation, of single-parameter “track and trigger” systems. Resuscitation 2008;79(1):11–21.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
Australian Critical Care xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Australian Critical Care journal homepage: www.elsevier.com/locate/aucc
Research paper
Short National Early Warning Score — Developing a Modified Early Warning Score Leandro Luís a Carla Nunes b,∗ a
Centro Hospitalar Lisboa Central, Escola de Nacional de Saúde Pública — Universidade Nova de Lisboa, Portugal Centro de Investigac¸ão em Saúde Pública, Escola de Nacional de Saúde Pública — Universidade Nova de Lisboa, Avenida Padre Cruz, 1600 560 Lisboa, Portugal b
article information Article history: Received 16 May 2017 Received in revised form 7 November 2017 Accepted 12 November 2017 Keywords: NEWS Hospital Clinical derangement Vital signs Patient safety Statistic models
a b s t r a c t Introduction: Early Warning Score (EWS) systems have been developed for detecting hospital patients clinical deterioration. Many studies show that a National Early Warning Score (NEWS) performs well in discriminating survival from death in acute medical and surgical hospital wards. NEWS is validated for Portugal and is available for use. A simpler EWS system may help to reduce the risk of error, as well as increase clinician compliance with the tool. Objectives: The aim of the study was to evaluate whether a simplified NEWS model will improve use and data collection. Methods: We evaluated the ability of single and aggregated parameters from the NEWS model to detect patients’ clinical deterioration in the 24 h prior to an outcome. There were 2 possible outcomes: Survival vs Unanticipated intensive care unit admission or death. We used binary logistic regression models and Receiver Operating Characteristic Curves (ROC) to evaluate the parameters’ performance in discriminating among the outcomes for a sample of patients from 6 Portuguese hospital wards. Results: NEWS presented an excellent discriminating capability (Area under the Curve of ROC (AUCROC) = 0.944). Temperature and systolic blood pressure (SBP) parameters did not contribute significantly to the model. We developed two different models, one without temperature, and the other by removing temperature and SBP (M2). Both models had an excellent discriminating capability (AUCROC: 0.965; 0.903, respectively) and a good predictive power in the optimum threshold of the ROC curve. Conclusions: The 3 models revealed similar discriminant capabilities. Although the use of SBP is not clearly evident in the identification of clinical deterioration, it is recognized as an important vital sign. We recommend the use of the first new model, as its simplicity may help to improve adherence and use by health care workers. © 2017 Australian College of Critical Care Nurses Ltd. Published by Elsevier Ltd. All rights reserved.
1. Introduction Patients who suffer from an adverse event on the ward, such as cardiac arrest or death, often have a physiological deterioration several hours before this event.1–3 Many of these events are potentially predictable and preventable by a timely recognition of the patient derangement and a proper management by a skilled team.1,4–6 The failure to identify and monitor the warning signs, and to escalate the care for the patients at risk for clinical deterioration, has led to the development of rapid response systems.3,6–8
These systems usually have a “track and trigger” system that allows the identification of the patients with physiological abnormalities, allowing a fast and effective response.3,6,9 These “track and trigger” systems are frequently known as early warning score (EWS) systems,1 and score the degree of vital sign derangement.3,6,10 This score is used to define an individual care pathway for these patients, such as more frequent monitoring of vital signs or the involvement of more experience professionals, in particular the rapid response team.10,11 The use of EWS systems is recommended in the United Kingdom for the early recognition and response to patients’ clinical deterioration.11–14 There are more than 100 EWS systems available,1 such as Modified Early Warning Score,5 Vital PacTM
∗ Corresponding author. E-mail address: [email protected] (C. Nunes). https://doi.org/10.1016/j.aucc.2017.11.004 1036-7314/© 2017 Australian College of Critical Care Nurses Ltd. Published by Elsevier Ltd. All rights reserved.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
2
Early Warning Score (ViEWS)15 or National Early Warning Score (NEWS).14 ViEWS15 and NEWS,14 in their development studies, had presented better discriminating capability than the other 33 EWS systems in the United Kingdom for the 24-h period prior to a death versus survival outcome.11,15 A study from Luís16 has validated these two systems for Portugal. Therein, both systems showed an excellent discriminating capability to the outcomes survival vs unanticipated intensive care unit admission (UICUA) and survival vs death.16 Some studies suggest that the collection of all the necessary data to complete EWS scores in short periods may be time consuming for health professionals.17–20 In addition, the absence of missing values or errors in these vital signs records are essential and may have consequences in the EWS use.16–18,21,22 The number of variables used in the EWS systems may influence the data collection workload for its use and add complexity to the healthcare professionals in their work, increasing the risk of errors. A simpler EWS with less variables can help to reduce the risk of error and unnecessary effort and time, leading to more reliable scores for an identical system performance.21,22 Therefore, recognising the specificities of the clinical practice and the workload imposed by a systematic collection of several datasets of vital parameters, we assume the need to develop a modified NEWS system with fewer parameters that can contribute to better adherence and fewer errors in its use. The main goal of this study is to develop a simpler modified NEWS with the vital parameters that most significantly anticipate a negative outcome (specifically death or unanticipated intensive care unit admission) for hospital inpatients.
2. Methods We evaluated the individual parameters of the NEWS model for an appropriate detection of the outcomes, building an aggregate system properly adapted to the suited sample. We intended to make the NEWS model simpler and more user-friendly, maintaining a high discriminating capability. 2.1. Study design A prospective cohort study, with retrospective analysis, was conducted in 2 medical wards, 3 surgical wards and 1 Haematology ward in Lisbon Central Hospital Centre (LCHC). 2.2. Sample The current study was based on data from another study,16 authorised by the LCHC board and ethics committee (LCHC Ethics Committee approval: 107/2012). It used a convenience sampling based on nurses’ records for the patients admitted to the study wards between December 1st and 31st, 2012. Patients were excluded if they were less than 18 years old, were pregnant, were in a preoperative phase for elective surgery, or in palliative care. Incomplete records (with missing values in at least one NEWS variables) were also excluded from the sample. For survival outcome, the highest NEWS score in patients with all the parameters filled during hospitalization episodes in the LCHC wards was used. For the other outcome, including a combination of both death and unanticipated intensive care unit admission (UICUA), we considered the NEWS score obtained 24 h before the outcome. This option was made to allow a better analysis of the performance of the EWS score in discriminating the outcomes,
allowing to compare the maximum derangement of patients who survived with those who died or had an UICUA. To reduce the frequency of absent values, nurses were asked to register all the NEWS parameters. All data were collected daily and registered in a digital document, which computed a NEWS score automatically. The data were anonymised for analysis purposes. 2.3. Variables The variables of interest in the study were NEWS score (Numerical Scale between 0 and 20) and the individual scores of each NEWS parameter: respiratory rate, oxygen saturations, body temperature, systolic blood pressure (SBP), heart rate, level of consciousness (numerical scale between 0 and 3) and any supplemental oxygen (numerical scale between 0 and 2). The study outcomes were survival or death/UICUA. Survival meant that the patient had been discharged from the hospital, with an improvement of his or her health status. Death or UICUA represented a deterioration of the clinical situation of the patients, considering only “intended to treat situations” in UICUA situations as suggested by National Institute for Health and clinical Excellence (NICE) in 2007.12 Therefore, these last two results (death and UICUA) were combined into a single outcome. 2.4. Data analysis We conducted a descriptive analysis of survival or death/UICUA distribution by gender (Male/Female), admission diagnosis (Medical/Surgical/Haematological) and age group (18–44 years, 45–64 years, 65–84 years, above 85 years). The Area Under the Curve of the Receiver Operating Characteristic curve (AUROC) was calculated for the outcomes death/UICUA versus survival, using the single NEWS categories defined in its development.14 This allowed us to evaluate the discriminant capability of the system (total score) and single parameters (independently) to detect a critical case — associated with a high probability to have a negative event (death/UICUA) in the next 24 h. We assumed an AUROC value between 0.5 (no better than chance at predicting the outcome) and 1 (perfect prediction of the outcome). An AUROC between 0.700–0.800 indicated a reasonable discrimination capability and values exceeding 0.800 a good discrimination capability.11 We compared the highest NEWS score reached during the hospital ward stay for each patient (survival group) with the NEWS score from the death/UICUA group in the 24 h prior to the outcome. We began our study by assessing the individual contribution of each NEWS parameter to predict the outcome death/UICUA using binary logistic regression methods. This test helped us to decide which variables were important for the NEWS predictive capability performance, allowing us to evaluate alternative NEWS models. Models’ calibration was assessed using the Hosmer–Lemeshow goodness of fit test. Furthermore, we estimated Odds Ratios and respective 95% confidence intervals. In the second phase of the study, we developed two new aggregated scales, considering both AUROC values and statistical significance in regression models. Then, the discriminant capability of these two scales was analysed using the AUROC.23,24 The new scales were compared with the original NEWS scale. The discriminant capability evaluation of each EWS was complemented by calculating its predictive power. We analysed the sensitivity and specificity in the optimal threshold for the scales using the Youden index (J).25 All calculations were performed using the IBM software Statistical Package for Social Sciences, v.22 (Armonk, NY: IBM Corp).
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
3
Table 1 Sample sociodemographic characterization. Outcome
Global
Variables
Absolute frequency (Episodes)
Relative frequency (%)
Survival Absolute frequency (Episodes)
Relative frequency (%)
Absolute frequency (Episodes)
Death/UICUA Relative frequency (%)
Gender
Female Male
171 159
51.8 48.2
155 141
90.6 88.7
16 18
9.4 11.3
Admission diagnosis
80 Surgical Medical 145 Haematological 105
24.2 43.9 31.8
70 130 96
87.5 89.7 91.4
10 15 9
12.5 10.3 8.6
Age group
18–44 years 45–64 years 65–84 years ≥85 years
38 96 156 40
11.5 29.1 47.3 12.1
36 89 136 35
94.7 92.7 87.2 87.5
2 7 20 5
5.3 7.3 12.8 12.5
330
100
296
89.7
34
10.3
Total
Table 2 AUROC for NEWS global model and Single parameters. Variables
NEWS Heart rate Respiratory rate Temperature Systolic blood pressure Oxygen saturations Any supplemental oxygen Level of consciousness
AUROC
0.944 0.766 0.618 0.51 0.711 0.729 0.827 0.783
Standard error
0.016 0.05 0.056 0.051 0.048 0.053 0.03 0.05
p-value
<0.001 <0.001 0.025 0.846 <0.001 <0.001 <0.001 <0.001
Confidence interval (CI) 95% IL
UL
0.912 0.668 0.509 0.411 0.616 0.626 0.768 0.684
0.975 0.864 0.727 0.609 0.806 0.833 0.886 0.882
Table 1 shows the sociodemographic characteristics of our sample (n = 330). The sample contained slightly more females (51.8%) than males. Medical inpatient diagnosis was predominant in the hospitalization episodes (43.9%), compared to haematological diagnoses (31.8%) and surgical diagnoses (24.2%). There was a predominance of hospitalisation episodes of older people aged between 65 and 84 years (47.3%). Concerning the hospitalisation outcome, there were more cases of death or UICUA in males (Male = 11.3%; Female = 9.4%), surgical admissions (Surgical = 12.5%; Medical = 10.3%; Haematological = 8.6%) and in ages equal or above 65 years old (65–84 years = 12.8%; ≥85 years = 12.5%). 3.2. Discriminant power analysis of NEWS vital parameters
Fig. 1. ROC curves for NEWS, Model 1 and Model 2 legend. NEWS. Model 1 — Short NEWS. Model 2. Reference Line.
3. Results
The discriminating capability analysis of NEWS and their individual parameters are presented by the AUROC values in Table 2. NEWS had an excellent discriminating capability, with an AUROC value of 0.944 (p-value <0.001; 95% CI: 0.912–0.975). NEWS model single vital parameters had AUROC values between 0.618 (respiratory rate) and 0.827 (Oxygen Saturation). Temperature was the only parameter that had no discriminant capability, with an AUROC value of 0.51. This value was not statistically significant (p-value >0.05). This suggested the need to evaluate the individual contributions of the parameters to the NEWS model. 3.3. NEWS model individual vital parameters contribution analysis
3.1. Sample description From an initial sample of 439 patients admitted to the wards, 109 patients were excluded due to missing values (24.8%). The remaining 330 patients were used for the purpose of this analysis.
The analysis of single parameters’ contribution to the NEWS model was performed with binary logistic regressions, using Death/UICUA as the event being modelled. Likelihood ratio tests allowed us to verify that at least one of the parameters was a sig-
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model
ARTICLE IN PRESS
AUCC-405; No. of Pages 6
L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
4 Table 3 Variables included in the model. Variables
ˇ
Standard error
Wald’s test
p-value
Odds ratio
Heart rate Respiratory rate Temperature Systolic blood pressure Oxygen saturations Supplemental oxygen Level of consciousness Constant
1.759 0.884 0.407 −0.024 0.806 1.852 1.288 −9.79
0.403 0.287 0.375 0.249 0.301 0.516 0.27 1.743
19.09 9.497 1.177 0.009 7.17 12.876 22.723 31.528
<0.001 0.002 0.278 0.922 0.007 <0.001 <0.001 <0.001
5.805 2.421 1.502 0.976 2.238 6.373 3.627 0
Odds ratio interval (CI:95%) Inferior limit
Upper limit
2.637 1.38 0.72 0.599 1.241 2.317 2.136
12.778 4.249 3.135 1.591 4.038 17.527 6.161
Table 4 Short NEWS. Physiological parameters Respiratory rate Oxygen saturations Any supplemental oxygen Systolic blood pressure Heart rate Level of consciousness
3 ≤8 ≤91 ≤90 ≤40
2 92–93 Sim 91–100
1 9–11 94–95 101–110 41–50
nificant predictor to the outcome (chi-square = 141.724; p-value <0.01). Model fit was evaluated by performing the Hosmer–Lemeshow test, which revealed that the variables fit the model (chisquare = 2.522; p-value = 0.961). Each variable contributing to the model is shown in the next table (Table 3). Temperature and SBP were found not to be significant predictive variables in the model, showing a p-value for the Wald test higher than 0.05. The parameters that contributed the most to the model were heart rate (OR = 5.805; p-value <0.001) and the presence of supplemental oxygen (OR = 6.373; p-value <0.001), with the higher odds ratio. Considering these results and the clinical value of each parameter for derangement prediction, we developed two test models, which were assessed for their discriminatory capacity for the outcomes survival versus death/UICUA. In one model, called Model 1, we excluded temperature. In the second model, Model 2, we excluded temperature and SBP.
3.4. Discriminant capacity of the new models Considering that temperature and SBP did not have a significant contribution to the NEWS model predictive power, we developed two alternative models. These two models maintained the vital parameters: heart rate, respiratory rate, oxygen saturations, presence of supplemental oxygen and level of consciousness. They were tested for their discriminant power against the original model. The results are presented in Fig. 1. The new models had excellent discriminatory capability, with similar values as NEWS. However, Model 1 AUROC (removing temperature) had a higher value than the other models (AUROC = 0.965; p-value <0.001; 95% CI: 0.942–0.988). Removing the SBP parameter as well as temperature decreased the discriminant capability of the aggregated model. This second model had worse results than the original NEWS and model 1, with an AUROC = 0.903 (p-value <0.001; 95% CI: 0.855–0.950). The predictive capability evaluation of the models showed that NEWS had a sensitivity of 91.2% and a specificity of 78.4% at an optimum threshold of 6.5 (Youden Index (J) = 0.696). Model 1 correctly identified 80.7% of survivals when the score is below 5.5, and 97.2%
0 12–20 ≥96 Não 111–219 51–90 Alert (A)
1
2 21–24
91–110
111–130
3 ≥25
≥220 ≥131 Verbal (V) Pain (P) Unresponsive (U)
of Death/UICUA when the score is over that value (J = 0.778). Model 2 had a sensitivity of 82.4% and a specificity of 82.8% at an optimum threshold of 4,5 (J = 0.652). The models’ comparison allowed us to verify that Model 1 had a better overall performance predicting the outcomes for the optimum threshold. However, Model 2 had a slightly higher specificity than Model 1 in the optimum threshold. According to Hosmer–Lemeshow test, all the models had correctly fitted the data (p-value >0,05). Model 1, referred to here as Short NEWS, is presented in Table 4.
4. Discussion The current study detailed the NEWS model and, through the clarification of each parameter’s importance, offers a simpler EWS for broader use and dissemination in the clinical field. The study results have shown excellent discriminating power of NEWS for the outcomes of Survival and Death/UICUA (AUROC = 0.944; 95% CI: 0.912–0.975). The results obtained by Smith et al.11 demonstrated good discriminant power (AUROC = 0.865; 95% CI: 0.858–0.872) of NEWS for an outcome that grouped Death, UICUA and cardiorespiratory arrest. The actual study results were even better than the Smith et al.11 study, but the sample was smaller and therefore, does not allow us to generalise the results. This study aimed to test if any NEWS vital parameter had low or no influence on the discriminant and predictive capability of the NEWS aggregated model. We wanted to optimise the NEWS model and present a simpler model, with fewer parameters and the best clinical applicability. It was possible to verify that temperature and SBP had a low contribution for the discriminant and predictive capability of NEWS. Without these parameters, the aggregated model with the other parameters had similar results. Scarce evidence exists about temperature’s impact in acute clinical deterioration predictive models. However, since temperature is considered a physiological disorder and an important disease severity signal, it is used in the NEWS model.14 In the case of SBP, the Royal College of Physicians14 refers to it as a central element in the definition of clinical deterioration of the patient. NICE12 refers to the importance of temperature as a central element to the EWS. This is explained by referring to studies where this parameter was included. However, there is no explanation or empirical support
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
of the capability of this parameter to predict death. Despite this consideration, changes in temperature are in many situations associated with infections that do not result in an immediate clinical deterioration of patients. These changes may be present in isolation for extended periods. Considering the NEWS model in the analysed period, we verified that temperature had no effect as predictor of death or UICUA, so its exclusion led to similar results to those obtained with the original model. Temperature is an important vital sign, but its use in a EWS system may pose some difficulties, especially when there is a clinical escalation protocol that increases its evaluation frequency. The study by Clifton et al.18 showed in a review of 16 795 EWS vital signs observation sets that temperature was the most commonly missed vital sign, being absent in 11.4% of total EWS. This may be a problem when the parameter is essential to the model. Observing Smith et al.26 study about single parameters track and trigger systems, temperature was not presented as a central element of these systems. In this same study, SBP was always present and was considered relevant for the detection of patient’s clinical deterioration. The same authors have shown the same assumptions in another study about aggregated track and trigger systems.10 Since temperature was the only excluded variable in the comparisons made in these studies, it seems that there is a similarity of the discriminant capability between the models with and without this variable. This may suggest a weak contribution from this parameter to model discrimination, which concurs with our results. Some studies revealed that a simpler EWS had better data collection with fewer errors and a significant improvement in the quality of the results.12,21,22 Considering the study results, we have performed a discriminant analysis for the NEWS derived models, one without temperature (Model 1 — named here as Short NEWS) and the other without SBP and temperature (Model 2). Short NEWS showed slightly better results than the original NEWS. The option of maintaining SBP was due to the better results of the model, because it was a routinely and frequently collected parameter and for the importance given to this parameter in the literature observed.14 We verified that the use of a simpler model may allow for better agreement in its application in clinical practice.21,22 This may increase the health professional’s compliance with the data collection required for the EWS. The comparison of the discriminant and predictive capacity of Short NEWS against NEWS and Model 2 have shown similarity in the sensitivity and specificity of the scale. However, Short NEWS had better results for the optimum threshold than the other models. This may support the use of Short NEWS as a good alternative to NEWS in Portugal. 4.1. Limitations The small sample of patients with the outcome of UICUA or death was a limitation to any strong conclusion in this study. The short timeframe was another limitation, because seasonal effects may influence the results. Restricting the study to only a hospital limited the generalisability of the results, but allowed a demonstration of the model’s potential. 5. Conclusions This study presented a different approach to the NEWS model, with a detailed analysis of the NEWS parameters and their contribution to its goal of detecting the clinical deterioration of in-hospital patients. The weak contribution of the SBP and temperature parameters to NEWS performance allowed the opportunity to develop two
5
alternative models, without losing quality in detecting death or unanticipated intensive care unit admission. Short NEWS (not including the temperature parameter) has shown discriminant and predictive power superior to the others, using fewer parameters. Considering this and the scientific evidence that holds that SBP is an important predictor of mortality, Short NEWS was suggested for use in clinical practice. The study proposed a tool to support clinical practice decisions in the Portuguese reality, based on less data, a more rational use of worktime for health professionals and without losing its discriminant capacity. This methodology is relatively easy to be used in other countries, potentially bringing similar results or the opportunity to parameterize a tailored local NEWS. Authors’ contributions Conception, design and performing the research: LL; Analysis and interpretation of data: LL, CN; Drafting the manuscript: LL, CN; Final approval of the version to be submitted: LL, CN; Agreement to be accountable for all aspects of the work: LL, CN. Funding None. References 1. Churpek MM, Yuen TC, Edelson DP. Risk stratification of hospitalized patients on the wards. Chest 2013;143(6):1758–65. 2. Jansen JO, Cuthbertson BH. Detecting critical illness outside the ICU: the role of track and trigger systems. Curr Opin Crit Care 2010;16:184––190. 3. Smith GB, Prytherch DR, Schmidt P, Featherstone PI, Knight D, Clements G, et al. Hospital-wide physiological surveillance — a new approach to the early identification and management of the sick patient. Resuscitation 2006;71:19–28. 4. Goldhill DR, Mcnarry AF, Mandersloot G, Mcginley A. A physiologically-based early warning score for ward patients: the association between score and outcome. Anaesthesia 2005;60:547–53. 5. Subbe CP, Kruger M, Rutherford P, Gemmel L. Validation of a modified Early Warning Score in medical admissions. QJM 2001;94(10):521–6. 6. Jones DA, DeVita MA, Bellomo R. Rapid-response teams. N Engl J Med 2011;365:139–46. 7. Maharaj R, Raffaele I, Wendon J. Rapid response systems: a sysreview and meta-analysis. Crit Care 2015;19(1):254, tematic http://dx.doi.org/10.1186/s13054-015-0973-y. 8. Stafseth SK, Grønbeck S, Lien T, Randen I, Lerdal A. The experiences of nurses implementing the Modified Early Warning Score and a 24-hour oncall Mobile Intensive Care Nurse: an exploratory study. Intensive Crit Care Nurs 2016;34:33–41. Elsevier Ltd. 9. DeVita MA, Smith GB, Adam SK, Adams-pizarro I, Buist M, Bellomo R, et al. “Identifying the hospitalised patient in crisis” — a consensus conference on the afferent limb of Rapid Response Systems. Resuscitation 2010;81(4):375–82. Elsevier Ireland Ltd. 10. Smith GB, Prytherch DR, Schmidt PE, Featherstone PI. Review and performance evaluation of aggregate weighted ‘track and trigger’ systems. Resuscitation 2008;77:170–9. 11. Smith GB, Prytherch DR, Meredith P, Schmidt PE, Featherstone PI. The ability of the National Early Warning Score (NEWS) to discriminate patients at risk of early cardiac arrest, unanticipated intensive care unit admission, and death. Resuscitation 2013;84(4):465–70. 12. National Institute for Health and Clinical Excellence. Acutely ill patients in hospital. London: NICE; 2007. 13. National Confidential Enquiry into Patient Outcome and Death. Time to Intervene? A review of patients who underwent cardiopulmonary resuscitation as a result of an in-hospital cardiorespiratory arrest. London: NCEPOD; 2012. 14. Royal College of Physicians. National Early Warning Score (NEWS) — Standardising the assessment of acute-illness severity in the NHS. Report of a working party. RCP; 2012, 47 p. 15. Prytherch DR, Smith GB, Schmidt PE, Featherstone PI. ViEWS — towards a national early warning score for detecting adult inpatient deterioration. Resuscitation 2010;81(8):932––937. Elsevier Ireland Ltd. 16. Luís L. Traduc¸ão, Validac¸ão e Aplicac¸ão dos Sistemas de Pontuac¸ão de Alerta Precoce “ViEWS” e “NEWS” em Portugal. Escola Superior de Tecnologias da Saúde de Lisboa — Instituto Politécnico de Lisboa, Escola Superior de Saúde da Universidade do Algarve; 2014. 17. Prytherch DR, Smith GB, Schmidt P, Peter I, Stewart K, Knight D, et al. Calculating early warning scores — a classroom comparison of pen and paper and hand-held computer methods. Resuscitation 2006;70:173–8.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004
G Model AUCC-405; No. of Pages 6 6
ARTICLE IN PRESS L. Luís, C. Nunes / Australian Critical Care xxx (2017) xxx–xxx
18. Clifton DA, Clifton L, Sandu D, Smith GB, Tarassenko L, Vollam SA, et al. ‘Errors’ and omissions in paper-based early warning scores: the association with changes in vital signs — a database analysis. BMJ Open 2015;5, http://dx.doi.org/10.1136/bmjopen-2014-007376. 19. Hands C, Reid E, Meredith P, Smith GB, Prytherch DR, Schmidt PE, et al. Patterns in the recording of vital signs and early warning scores: compliance with a clinical escalation protocol. BMJ Qual Saf 2013;(April):1–8. 20. Kolic I, Crane S, McCartney S, Perkins Z, Taylor A. Factors affecting response to National Early Warning Score (NEWS). Resuscitation 2015;90:85–90. 21. Subbe CP, Harrison DA. Reproducibility of physiological track-and-trigger warning systems for identifying at-risk patients on the ward. Intensive Care Med 2007;33:619–24.
22. Jarvis S, Kovacs C, Briggs J, Meredith P, Schmidt PE, Featherstone PI, et al. Can binary early warning scores perform as well as standard early warning scores for discriminating a patient’s risk of cardiac arrest, death or unanticipated intensive care unit admission? Resuscitation 2015;93:46––52. 23. Pepe MS. The statistical evaluation of medical tests for classification and prediction. New York: Oxford University Press; 2009. 24. Zou H, Liu A, Bandos AI, Ohno-Machado L, Rockette E. Statistical evaluation of diagnostic performance — topics in ROC analysis. Boca Raton, Florida: CRC Press; 2012. 25. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3(1):32–5. 26. Smith GB, Prytherch DR, Schmidt PE, Featherstone PI, Higgins B. A review, and performance evaluation, of single-parameter “track and trigger” systems. Resuscitation 2008;79(1):11–21.
Please cite this article in press as: Luís L, Nunes C. Short National Early Warning Score — Developing a Modified Early Warning Score. Aust Crit Care (2017), https://doi.org/10.1016/j.aucc.2017.11.004