Objectives Prepandemic projections anticipated huge excess attendances and mortality in an influenza pandemic. A number of tools had been suggested for triaging patients with influenza for inpatient and critical care admission, but none had been validated in these patients. The authors aimed to evaluate three potential triage tools—CURB-65, PMEWS and the Department of Health community assessment tool (CAT)—in patients in the first waves of the 2009 H1N1 pandemic.
Setting Prospective cohort study in three urban emergency departments (one adult, one paediatric, one mixed) in two cities.
Participants All patients presenting to the three emergency departments fulfilling the national definition of suspected pandemic influenza.
Outcome measures 30-day follow-up identified patients who had died or had required advanced respiratory, cardiovascular or renal support.
Results The pandemic was much less severe than expected. A total of 481 patients (347 children) were recruited, of which only five adults fulfilled the outcome criteria for severe illness. The c-statistics for CURB-65, PMEWS and CAT in adults in terms of discriminating between those admitted and discharged were 0.65 (95% CI 0.54 to 0.76), 0.76 (95% CI 0.66 to 0.86) and 0.62 (95% CI 0.51 to 0.72), respectively. In detecting adverse outcome, sensitivities were 20% (95% CI 4% to 62%), 80% (95% CI 38% to 96%) and 60% (95% CI 23% to 88%), and specificities were 94% (95% CI 88% to 97%), 40% (95% CI 32% to 49%) and 81% (95% CI 73% to 87%) for CURB-65, PMEWS and CAT, respectively.
Conclusions Although limited by a paucity of cases, this research shows that current triage methods for suspected pandemic influenza did not reliably discriminate between patients with good and poor outcomes.
- Cardiac care
- care systems
- infectious diseases
- major incidents
Statistics from Altmetric.com
An influenza pandemic could potentially place a huge strain on health and emergency care services, which may be exacerbated by staff sickness. The 2007 UK influenza pandemic contingency plan predicted 750 000 excess emergency department (ED) attendances and 82 500 excess hospitalisations.1 Under these circumstances, it would be impractical for all patients to be fully assessed by a senior clinician. We therefore need methods of triage and resource allocation that are fair, robust and reproducible.2 ED triage methods need to predict accurately the individual patient's risk of death or severe illness; low-risk patients may be discharged home, high-risk patients admitted to hospital, and those at very high risk referred for high dependency or intensive care.
In April 2009, a new strain of A/H1N1 influenza (swine flu) was detected in Mexico and started to spread globally, being declared a pandemic in June. Before this, Health Protection Agency guidance recommended the use of the CURB-65 pneumonia score3 (appendix 1) for risk stratification of influenza-related pneumonia. Department of Health guidelines on surge capacity recommended use of CURB-65 ‘when assessing patients with influenza-like illness during a pandemic’, and also considered use of a physiological-social score (PMEWS)4 (appendix 2), which includes physiological variables, age, social factors, chronic disease and performance status. National guidance produced in early 2009 included a new swine flu hospital pathway, based on a community assessment tool (CAT) with seven criteria, any one of which predicts increased risk and the need for hospital assessment5 (appendices 3 and 4). This was flagged as being for use only where demand placed severe strain on surge capacity.
CURB-65 performs reasonably as a mortality predictor in adults with community-acquired pneumonia (area under the receiver operating characteristic curve (AUROC) 0.76),6 but less well in predicting the need for critical care (AUROC 0.697 and 0.648). PMEWS is not a particularly good predictor of death in community-acquired pneumonia (AUROC 0.66), but performs better at predicting a requirement for critical care (AUROC 0.83)8 and has shown promise in prehospital care in determining the need for ED attendance (AUROC 0.719 and 0.810). The CAT appears to have been developed by expert consensus without validation in the appropriate patient populations. To our knowledge there have been no studies evaluating any of these triage methods in patients with suspected pandemic influenza. We therefore aimed to use the initial waves of the H1N1 pandemic to evaluate ED triage methods for predicting severe illness or death in patients with suspected pandemic influenza.
We undertook a prospective cohort study of patients presenting to the ED between September 2009 and February 2010 with suspected pandemic influenza. Patients were eligible for inclusion if they met the criteria of (1) fever (fever ≥38°C) or a history of fever and (2) influenza-like illness (two or more of cough, sore throat, rhinorrhoea, limb or joint pain, headache, vomiting or diarrhoea) or severe and/or life-threatening illness suggestive of an infectious process. ED staff identified eligible patients and then completed a clinical assessment form that doubled as clinical notes. It included the elements of the CURB-65 score, PMEWS, the CAT and other measures routinely recorded in the ED (co-morbidities, physiological observations, blood tests, ECG and chest x-ray), pre-presentation antivirals, antibiotics and immunisation status. Patients were managed according to normal ED practices. We did not aim to collect data on PCR confirmation of pandemic H1N1, as at the time of the study this was not recommended for all patients with influenza-like illness.
Research staff followed patients up until 30 days after attendance by hospital record review and, if appropriate, general practitioner contact to identify patient outcomes. Anonymised data were kept on a secure online database at the University of Sheffield.
A poor outcome was defined as death or requirement for advanced respiratory, cardiovascular or renal support during the 30-day follow-up. Respiratory support included any intervention to protect the patient's airway or assist their ventilation, including non-invasive ventilation or acute administration of continuous positive airway pressure. It did not include supplemental oxygen alone or nebulised bronchodilators. Cardiovascular support included any intervention to maintain organ perfusion (inotropic drugs) or invasively monitor cardiovascular status (central venous, pulmonary artery or arterial blood pressure monitoring). It did not include peripheral intravenous cannulation and/or fluid administration. Renal support included any intervention to assist renal function (haemoperfusion, haemodialysis, peritoneal dialysis). It did not include intravenous fluid administration. Advanced support was specifically selected as an outcome measure to identify those patients at highest risk who were most likely still to qualify for critical care even in the event of overwhelming patient numbers. We note that the severe viral pneumonia observed in H1N1 patients was often rapidly progressive, and that early identification of those patients likely to progress to critical illness was therefore important.11
Outcome assessment was based primarily on researcher review of hospital computer records and case notes. If there was no evidence of a poor outcome in these, the patient was recorded as having a good outcome. If the outcome was uncertain (eg, if the patient was transferred to another hospital or left hospital against medical advice), the researcher contacted the patient's general practitioner for clarification.
We assessed CURB-65, PMEWS and the swine flu clinical pathway in adults and the swine flu clinical pathway in children by calculating the AUROC curve (c-statistic) for discrimination. A AUROC curve plots sensitivity against 1−specificity at various cut-off points of a test as an assessment of discriminatory power. An area under the curve of 0.5 is considered no better than chance, while an area of 1.0 would constitute a perfect test. The AUROC curve equates to the probability that a randomly selected patient with a poor outcome has a worse score than a randomly selected patient with a good outcome. For each score, we assumed a score of zero or a negative categorisation for any variable or criterion that was missing. The sample size depended on the size and severity of the pandemic. We planned to collect data during the pandemic at four hospitals in Sheffield and Manchester covering a population of over one million. Before the pandemic the Department of Health estimated that a 25% clinical attack rate with illustrative case hospitalisation and case death rates of 0.55% and 0.37%, respectively,1 suggested that a pandemic could lead to 12 500 ED attendances, 1400 hospitalisations and 900 excess deaths in our population. Accepting that the difficulties of undertaking research in a pandemic would reduce case identification, we estimated that a sample including 283 positive cases with a poor outcome would ensure a power of 80% to compare an AUROC curve of 0.85 versus 0.90 at 5% significance, assuming a correlation of 0.6 between scores.12
Ethics and governance arrangements
We received approval from North West Research Ethics Committee and the National Information Governance Board. We did not seek patient consent to participate, as the study was limited to collection of routinely available data and any delays in patient assessment could have risked compromising patient care. However, patients were informed of the study via posters displayed in the ED and leaflets distributed from reception and the pandemic flu assessment area and were given the opportunity to withdraw their data from the study.
We collected data at three hospitals in the second wave of the pandemic 24 September 2009 and 7 February 2010. Research governance delays prevented data collection at the fourth hospital.13 We identified 492 cases, data for 11 of which were asked to be withdrawn, leaving 481 for analysis. Ages ranged from under 1 to 96 years. Most were children, with 347/481 (72%) aged 16 years or less. The modal age group was 1–2 years, accounting for 69/481 (14%). There were 237 female patients (49%) and 244 male patients (51%). Most patients self-referred (399/481, 83%), while 41 (8%) were referred by their general practitioner, and 15 (3%) were referred via NHS Direct.
Mean symptom duration (N=379) was 3.1 days, median was 2 days, and most patients (213/379, 56%) had 1–2 days of symptoms. Before their index hospital attendance, 30 patients (6%) had attended hospital with the same complaint, eight (2%) had received vaccination against H1N1, 39 (8%) had been given oseltamivir, and 46 (10%) had been given antibiotics, although not always specifically for their presenting complaint.
Influenza was thought by the physician to be the most likely diagnosis in 214/368 cases (58%). The most common alternatives were upper respiratory tract infection (79 cases) and tonsillitis (23 cases).
Presenting physiological features were not recorded in all cases. Of note was the absence of a recorded respiratory rate in 19% of cases. Mean temperature was 37.8 (N=425, SD 1.1, range 35.0–40.7), and mean arterial oxygen saturation (Sao2) was 97% (N=369, SD 6, range 79–100). In 19/369 (5%) cases, Sao2 was below 94%. Pulse rate, respiratory rate and blood pressure are given in table 1, categorised by age group. Tachycardia and tachypnoea were relatively common, while blood pressure was generally normal.
Blood tests were requested in 55/481 cases (11%). Chest x-rays were abnormal in 12 cases, normal in 19, not performed in 284, and not recorded in 166. They were performed at clinician discretion and so non-performance should not be taken to imply lack of clinical chest signs. An ECG was abnormal in 10 cases, normal in 24, not performed in 67, and not recorded in 380.
Oseltamivir was prescribed for 58 cases and antibiotics for 56 (amoxicillin, 22; augmentin, 9; cefotaxime, 1; ceftriaxone, 2; clarithromycin, 3; gentamicin, 1; penicillin, 18). The patient in 83/481 cases (17%) was admitted.
Tables 2 and 3 show the CURB-65 scores and PMEWS scores (adults only), respectively. Using the recommended CURB-65 threshold for admission of ≥2,3 9/134 (7%) of patients should have been admitted. Applying a similar threshold for PMEWS would have resulted in 81/134 (60%) being admitted.
Patients with a higher CURB-65 score were more likely to be admitted (table 2: p=0.001, χ2 for trend). Admitted patients had a higher mean PMEWS score (table 3: 4.6 vs 2.0, p<0.001, t test). The c-statistics for CURB-65, PMEWS and the CAT in adults in terms of discriminating between those admitted and discharged were 0.65 (95% CI 0.54 to 0.76), 0.76 (95% CI 0.66 to 0.86) and 0.62 (95% CI 0.51 to 0.72), respectively.
Only 5/481 (1%) patients had a poor outcome according to our definition. Their details are as follows:
|Patient details||Physiological variables at presentation||Blood results at presentation||Score according to assessment tools||Outcome|
|Died 5 days after admission.|
|Required non-invasive ventilation.|
|Required non-invasive ventilation and haemodialysis.|
|Required positive pressure ventilation and died.|
|Required non-invasive and positive pressure ventilation.|
RR, respiratory rate; SaO2, transcutaneous oxygen saturation; GCS, Glasgow Coma Score; Na, sodium; K, potassium.
All five patients were admitted to hospital at the initial attendance. The c-statistic for discriminating between those with good and poor outcomes for each method was CURB-65 0.78 (95% CI 0.58 to 0.99), PMEWS 0.77 (95% CI 0.55 to 0.99), and the swine flu hospital pathway 0.70 (95% CI 0.45 to 0.96). Table 6 shows sensitivity and specificity for CURB-65 and PMEWS with a threshold of >1 and the swine flu hospital pathway with any criterion positive.
The pandemic was much less severe than predicted. As of 5 January 2010, there had been 28 456 laboratory-confirmed cases of H1N1 in the UK, with 4930 reported as being hospitalised and 355 deaths.14 Although limited by the small number of cases, our findings suggest that currently recommended triage methods do not reliably discriminate between patients with influenza-like illness who have a good outcome and those at risk of a poor outcome. A CURB-65 score of ≥2 is recommended to trigger admission.4 CURB-65 was ≥2 in 7% of our adult patients and 1/5 cases with a poor outcome. The swine flu hospital pathway was positive for 21% of adult patients and 3/5 cases with a poor outcome. PMEWS does not have a recommended threshold, but a score of ≥2 appears to mirror non-pandemic routine practice (K Challen, personal communication), rendering PMEWS positive in 60% of adult patients and 4/5 cases with a poor outcome. We did not use hospital admission as an outcome because we thought that this would be heavily influenced by the triage method in use. However, we note that CURB-65 and the swine flu hospital pathway appeared to discriminate poorly between those admitted and those discharged, despite being recommended for decision-making. It appears that participating clinicians were basing their decisions on other criteria. We did not formally assess clinicians' decision-making, but we noted that, where patients meeting the recommended threshold for admission were discharged, they tended to be young adults with a high fever.
We accept that our outcome measures (other than death) are all subject to local variability in practice. However, we present them as proxies for illness severity, as all mandate assessment by a senior clinician before initiation. We may have missed cases of severe illness that responded to simple interventions, such as oxygen therapy or intravenous fluids, but including these interventions in our outcome measure would have carried a much greater risk of misclassifying patients with mild illness as being severely ill if they were given simple interventions without a strong indication.
Our findings are limited by the small sample and paucity of cases with poor outcomes. These five cases may not have been typical of critically ill patients across the pandemic, so conclusions regarding the value of the triage tools must be cautious, but we raise valid concerns about their discriminant value. We did not test the application of the methods in practice, but calculated or inferred their performance from clinical data. It is possible that some criteria, such as the swine flu hospital pathway criterion G (other clinical concern), may have identified some of the positive cases in practice. However, we note that other investigators have identified significant discrepancies between the application of swine-flu-specific algorithms and other evidence-based guidelines.15
As the 2009–2010 H1N1 pandemic in the UK has not produced adequate numbers of severely ill patients from which to draw robust conclusions, health service planners must revert to the pre-existing evidence base: non-flu risk stratification tools, SARS and H5N1, and international experience of H1N1. Prepandemic advice was to use pneumonia severity scores to risk stratify patients with flu, with3 or without4 secondary bacterial pneumonia. Some evidence exists to support their use in identifying patients with bacterial pneumonia likely to require critical care facilities.16–19 Other tools designed specifically to predict requirement for critical care in patients with pneumonia exist,20 21 but have yet to be fully validated. However, in extrapolating from pneumonia-specific severity scores, it should be remembered that, in the FLU-CIN study of patients admitted to hospital with confirmed H1N1 disease, only 29% had radiological evidence of pneumonia.22 A significant minority of both paediatric and adult patients eventually diagnosed with H1N1 did not fulfil Health Protection Agency screening criteria, notably for fever.23 24 Little literature exists on risk assessment of undifferentiated emergency patients, and what there is concentrates on mortality risk.25–27 It appears from the international experience that obesity,28 pre-existing comorbidity,29 and pregnancy28 30 convey a worse prognosis during pandemic influenza infection.
In conclusion, current triage methods for suspected pandemic influenza did not reliably discriminate between patients with good and poor outcomes and did not appear to guide admission decisions in the second wave of the 2009 H1N1 pandemic. EDs should remain prepared to deal with patients with diffuse non-specific symptomatology from flu, and retain the capability to cohort these potentially infectious patients in the ED and hospital. Risk assessment will still take place in the absence of evidence, but should be guided by information from the international experience.
Appendix 1 The CURB-65 score
One point each for:
Urea >7 mmol/l
Respiratory rate ≥30 min
low systolic (<90 mm Hg) or diastolic (≤60 mm Hg) Blood pressure
age ≥65 years
Appendix 2 The Pandemic Modified Early Warning Score PMEWS score
Appendix 3 Community Assessment Tool for Adults
Appendix 4 Community Assessment Tool for Children
Department of Health Disclaimer: the views and opinions expressed herein are those of the authors and do not necessarily reflect those of the Department of Health.
Funding This project was funded by the NIHR Health Technology Assessment programme (project number 09/84/66) and was published in full in Health Technology Assessment Vol 14, No 46, pp 173–236. See the HTA programme website for further project information.
Competing interests KC, AB and DW developed the PMEWS score, evaluated in this study. SG, MC, RW and CF have no conflicts of interest.
Ethics approval This study was conducted with the approval of the North West MREC and NIGB.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.