The Swine Flu Triage (SWIFT) study: Development and ongoing refinement of a triage tool to provide regular information to guide immediate policy and practice for the use of critical care services during the H1N1 swine influenza pandemic

Study infrastructure

On 7 July 2009, prior to provisional funding approval, five co-investigators (KMR, DAH, DFMcA, GDP and DKM) met to discuss and agree the proposed data set for primary collection of pandemic-related data. Using the ICNARC’s existing international links, data collection forms from Mexico, Canada and Australasia were used to inform the SwiFT data set and to enable compatibility between common variables for international comparison. 2 From this, the final data set specification was developed by the two ICNARC co-investigators (DAH and KMR). Three members of the ICNARC information technology staff immediately commenced development of a secure, web-based, data entry system for SwiFT. Following comprehensive testing by ICNARC staff members, the SwiFT secure, web-based, data entry system went live on 17 September 2009.

In parallel with internal study infrastructure work, central and local research governance approvals were required for approximately 220 organisations (Trusts, Health Boards, etc.) in five countries [England, Wales, Northern Ireland, Scotland and the Republic of Ireland (ROI)]. Five senior members of ICNARC staff, supported by the entire staff, were allocated responsibility for research governance in each of the five countries.

The ICNARC used every opportunity to promote and outline the objectives of SwiFT, either directly as presentations on SwiFT or indirectly by mentioning SwiFT at other ICNARC audit/research study meetings. SwiFT was presented at: the Scottish Intensive Care Society meeting on 4 September 2009; a national critical care meeting on the H1N1 pandemic on 10 September 2009; the twelfth Current Controversies in Anaesthesia & Peri-Operative Medicine meeting (Dingle, Ireland) on 15 October 2009; and a national meeting on infection in critical care on 4 November 2009. SwiFT was also highlighted at data collection training courses for the national clinical audit of adult critical care co-ordinated by the ICNARC, the Case Mix Programme (CMP), and for ongoing ICNARC research studies, Fungal Infection Risk Evaluation (FIRE, 07/29/01) and Risk Adjustment In Neurocritical care (RAIN, 07/37/29) at five courses in September 2009, one in October 2009 and one in November 2009.

Research governance

Participation

Local R&D approval was gained at the organisational level (e.g. Trust, Health Board, etc.). Of the 221 organisations identified, submission for local R&D approval was achieved for 192 (87%; see Figure 2). The main reason for not submitting for local R&D approval was failure to identify a local collaborator to lead SwiFT. Only in one Trust (Stockport NHS Foundation Trust – Stepping Hill Hospital) did an identified local collaborator (the clinical lead for the critical care unit) actually refuse to participate. Local R&D approval for participation in SwiFT was received for 81% of organisations representing 85% of acute hospitals (Figure 2). Local R&D approval converted into actual commencement of data collection in only 64% of the 301 identified acute hospitals.

FIGURE 2.

Research governance and participation in SwiFT. NI, Northern Ireland.

Local R&D approval was quickest in the ROI, but for only a small number (n = 9, 24%) of acute hospitals. England achieved both quick (Figure 3) and timelier (Figure 4) local R&D approval for a much larger number of NHS Trusts (n = 150, 96%) – with 35 English NHS Trusts providing local R&D approval pending central R&D approval (REC approval) and a further 56 providing local R&D approval within 1 day of central R&D approval. Northern Ireland was similarly quick to gain local R&D approval, but for a much smaller number of only five HSC Trusts. Scotland appeared slower for their 11 Health Boards but, as the decision on REC approval for Scotland (i.e. it was ultimately deemed not to be required) did not occur until day 25, local R&D approval was delayed by this fact. The five NHS Trusts in Wales took much longer to gain local R&D approval.

FIGURE 3.

Mean time (days) to obtain local R&D approval for organisations for SwiFT.

FIGURE 4.

Timeliness of local R&D approval for SwiFT – day 0 indicates central R&D approval for England, Wales and Northern Ireland.

SwiFT commenced in 192 (64%) acute hospitals in 158 organisations (Figure 5).

FIGURE 5.

Participation of acute hospitals in SwiFT.

Facilitators

• Accelerated research governance The accelerated processes, developed by/for the NIHR CRN, CSPU/CSP, NIGB and NRES/REC, which were activated in England appeared to deliver for SwiFT. 4

• ICNARC The ICNARC had the infrastructure (staff skilled in research design, data set development, web-based data entry development/testing, data validation, analysis and reporting), knowledge (in conducting research, in navigating the research governance system and in critical care), contacts, relationships, trust and track record to conceive, design and deliver SwiFT. In addition, the ‘permission’ from the NETSCC to divert required ICNARC staff members, without worry of the knock-on effect to other ongoing research studies, allowed for rapid initiation of SwiFT. Despite the existing heavy workload at the ICNARC, ICNARC staff were motivated by the challenge and topical nature of SwiFT and every individual, at every level in the organisation, played a role in supporting its set-up and co-ordination.

• External support External support came from four main sources: first, through direct access to individuals with high levels of authority to make decisions in the situation where any potential barriers in the research governance system were identified; second, from the professional organisations, the UK and national Societies, who supported SwiFT; third, from the members of the UK CRN Critical Care Specialty Group and from the directors and leads for Research Management and Governance in the CLRNs (or their direct/indirect equivalents in Wales, Northern Ireland, Scotland and the ROI) who supported acute hospitals in identifying and accessing local resources; and, fourth, from the critical care staff, who rose to the challenge that SwiFT posed.

Barriers

• Failure to identify local collaborators With respect to national clinical audit, the ICNARC has a remit solely for adult critical care in England, Wales and Northern Ireland. While the Scottish co-investigator (TSW) helped to facilitate identification of local collaborators in Scotland, the absence of an existing network of contacts in the ROI (and no ROI co-investigator) potentially hindered participation in SwiFT there. Paediatric critical care colleagues were also more reticent about participating owing to insufficient time for the degree of involvement/negotiation being requested by the professional society, the Paediatric ICS, for them to fully endorse and advertise SwiFT.

• Method of identifying local collaborators The usual mode of engaging clinicians in a specific research study (i.e. identifying them, engaging their interest and jointly submitting for approval, etc.) was circumnavigated in the interest of speed for SwiFT in England. SSI forms were populated with individuals, probably busy owing to the impending pandemic, whose interest in SwiFT had not previously been identified. While most engaged with the pandemic situation, the speed of the process and the failure to engage up-front did put a few ‘noses out of joint’. However, the process followed was not unreasonable and was the only practicable way to get the R&D process ‘up and running’, in such a large number of acute hospitals, over such a short time frame.

• Lack of local resources Accessing local resources to support SwiFT appeared to be much harder in countries outside England, leading to much lower participation in these countries. The exception was Scotland, where participation was higher owing to two apparent reasons – first, the efforts of the Scottish co-investigator (TSW) and second, the existence of a genetic H1N1 study in Scottish critical care units to which SwiFT linked to supply the phenotype data for patients, thus avoiding duplicate data requirements.

• The existence of the CLRN system enhanced participation in England; however, following local R&D approval, the main reason cited for not participating appeared to be the absence of local resources. This appeared to be as much down to a lack of knowledge of acute hospital staff in how to access these resources, possibly due to the newness of the system and the staff’s newness to research, than to a refusal to provide the resources.

• Lack of an integrated, centralised system for research governance The relative effort required to gain approvals for acute hospitals outside England was immense. While Scotland had a centralised system for R&D approvals, gaining information governance clearance for each NHS Health Board from each Caldicott Guardian proved very time-consuming; one individual could hold up a number of acute hospitals ready to start for some considerable time.

• Other competing activities The European Society of Intensive Care Medicine initiated a European H1N1 registry which was promoted to the UK critical care community. The UK Extracorporeal Membrane Oxygenation (ECMO) Consortium had their own H1N1 initiative and acute hospitals providing ECMO tended not to, or only partially, participate in SwiFT. The Paediatric Intensive Care Audit Network was pooling data on paediatric critical care unit admissions with H1N1.

The next chapters present the results of SwiFT.

Selection of data

The CMP is the national, comparative audit of patient outcomes from adult, critical care units in England, Wales and Northern Ireland, co-ordinated by the ICNARC. The CMP is a voluntary performance assessment programme using high-quality clinical data to facilitate local quality improvement through routine feedback of comparative outcomes and key quality indicators to clinicians/managers in adult critical care units. The CMP recruits predominantly adult, general critical care units, either standalone intensive care units or combined intensive care/high-dependency units. Currently, approximately 90% of adult, general critical care units in England, Wales and Northern Ireland are participating in the CMP.

The CMP-specified data are recorded prospectively and abstracted retrospectively by trained data collectors according to precise rules and definitions, set out in the ICNARC CMP Dataset Specification. Data collectors from each unit are trained prior to commencing data collection, with retraining of existing staff or training of new staff also available. CMP training courses are held at least four times per year.

The CMP-specified data are collected on consecutive admissions to each participating critical care unit and are submitted to the ICNARC quarterly. Data are validated locally, on data entry, and then undergo extensive central validation for completeness, illogicalities and inconsistencies, with data validation reports returned to units for correction and/or confirmation. The validation process is repeated until all queries have been resolved and then the data are incorporated into the CMP Database (CMPD).

The CMPD has been evaluated according to the quality criteria of the Directory of Clinical Databases (www.icapp.nhs.uk/docdat/) and scored highly. 18

All admissions in the CMPD to adult, general critical care units in England, Wales and Northern Ireland from 1 January 2007 to 31 March 2009, collected to Version 3.0 of the ICNARC CMP Dataset Specification, incorporating the Department of Health Critical Care Minimum Data Set (CCMDS),19 were extracted.

Impact of cancelled/postponed elective and scheduled surgery

Critical care unit admissions that could potentially be avoided by the cancellation/postponement of elective and scheduled surgery were identified by the following criteria:

1. admissions direct from theatre with a classification of surgery as ‘elective’ or ‘scheduled’

2. all subsequent admissions during the same hospital stay of patients identified from criterion (1)

3. admissions for pre-surgical preparation.

The impact of cancellation/postponement of these admissions was estimated by:

• the percentage of admissions avoided/postponed

• the percentage of calendar days of critical care saved

• the percentage of calendar days at Level 3 saved

• the percentage of days of advanced respiratory support saved,

both overall and across units.

Exclusions

For the purpose of developing the triage models, the following admissions were excluded:

• admissions associated with elective/scheduled surgery (as defined above)

• admissions missing all basic vital signs (temperature, systolic blood pressure, heart rate and respiratory rate)

• admissions missing status at ultimate discharge from acute hospital

• readmissions of the same patient within the same acute hospital stay.

Patient cohorts

Models were developed using both the following two cohorts:

1. (a) all admissions, except those excluded based on the criteria listed above

2. (b) admissions for acute exacerbations of respiratory illness, defined by the presence of any of the following conditions as the primary or secondary reason for admission to the critical care unit:

   1. – acute respiratory distress syndrome (ARDS) [‘non-cardiogenic pulmonary oedema (ARDS)’]

   2. – pneumonia (‘bacterial pneumonia’; ‘viral pneumonia’; ‘pneumonia, no organism isolated’)

   3. – acute exacerbations of chronic airways disease [‘chronic obstructive pulmonary/airways disease (COPD/COAD)’; ‘chronic obstructive pulmonary disease with acute lower respiratory infection’; ‘chronic obstructive pulmonary disease with acute exacerbation, unspecified’; ‘emphysema’; ‘asthma attack in new or known asthmatic’].

Note: complete identification of such cases was dependent on the recording of one of the above conditions as either the primary (mandated) or secondary (optional) reason for admission. Primary and secondary reasons for admission to the critical care unit are coded using the ICNARC Coding Method,20 which was developed specifically for the CMP. It is a five-tiered, hierarchical method for coding reasons for admission or underlying conditions in critical care consisting of the type of condition (surgical/non-surgical), body system, anatomical site, pathological/physiological process and specific condition.

The case mix of admissions in the two cohorts was described by their age, sex, source of admission to the critical care unit, Acute Physiology And Chronic Health Evaluation (APACHE) II Acute Physiology Score and APACHE II score,21 ICNARC Physiology Score,22 and, for cohort B, CURB-65 (new onset confusion; urea > 7 mmol l^-1; respiratory rate ≥ 30 breaths per minute; systolic blood pressure < 90 mmHg or diastolic blood pressure ≤ 60 mmHg; age ≥ 65 years). 23

Triage modelling

The primary outcome for the triage models was an ordinal outcome on the following scale:

1. Potentially avoidable admission: admission did not receive advanced respiratory support, advanced cardiovascular support, renal support, neurological support or liver support (as defined by the CCMDS)19 at any time during the stay in the critical care unit and survived to leave acute hospital.

2. Critical care required: admission survived to leave acute hospital but was not a ‘potentially avoidable admission’ (i.e. received advanced respiratory support, advanced cardiovascular support, renal support, neurological support and/or liver support at any time during the stay in the critical care unit).

3. Death: admission did not survive to leave acute hospital.

The following routine physiological variables, termed ‘core variables’, measured and recorded during the first 24 hours following admission to the critical care unit, were included in the modelling:

• highest temperature (central or, if no central available, non-central)

• lowest systolic blood pressure

• highest heart rate

• highest respiratory rate

• neurological status, using Glasgow Coma Score (GCS) to approximate AVPU (Alert, Voice, Pain, Unresponsive) categories of Alert (GCS 15), Voice (GCS 10–14), Pain (GCS 7–9) and Unresponsive (GCS 3–6)24 and a separate category for patients for whom GCS could not be assessed owing to sedation.

In addition, the effect of adding the following variables, termed ‘additional variables’, to the model (alone or in combination) was explored:

• lowest partial pressure of oxygen (PaO₂), associated fraction of inspired oxygen (FiO₂) or PaO₂ : FiO₂ ratio

• base excess (calculated from the arterial blood gas with the lowest pH and from the lowest haemoglobin)

• highest blood lactate

• highest serum urea.

The physiological variables were divided into categories using small intervals at round values (e.g. multiples of 5 or 10, dependent on the scale of the variable).

Finally, the effect of incorporating severe comorbidity and/or age into the triage models was explored by selecting the preferred model based on physiological variables only and adding severe comorbidity (either in five individual organ systems or overall) and age (as a linear term).

Models were fitted using ordered logistic regression, with the primary performance measure of interest being the ability of the model to discriminate between the three outcome categories. Discrimination was assessed by Harrell’s concordance statistic, a natural generalisation of the area under the receiver operating characteristic curve for binary logistic regression. 25 A concordance of 1 corresponds to perfect discrimination – that is, for any pair of admissions with different outcomes, the admission with the worse outcome will have the higher score. A concordance of 0.5 corresponds to discrimination that is no better than chance. Concordance values > 0.7 are generally considered to be ‘satisfactory’, > 0.8 ‘good’ and > 0.9 ‘excellent’. 26 The standard error of the concordance statistic was calculated using a jack-knife procedure adjusted for clustering of the outcome at the unit level. As the models were developed and validated on the same data set, performance measures may be subject to ‘shrinkage’. Efron’s optimism bootstrap (with 200 bootstrap samples) was used to estimate the optimism in the concordance due to overfitting. 27

The effect of using a model to triage patients with low or high scores was explored by modelling potential outcomes for triaged patients. For patients with low scores triaged to temporary critical care areas, it was assumed that those with outcomes classified as ‘potentially avoidable admissions’ would be managed safely in the temporary critical care area and discharged alive, but that those classified as ‘critical care required’ or ‘death’ would subsequently be transferred to the critical care unit. The number of bed days saved by this approach was therefore the bed days of critical care occupied by the diverted admissions that were ‘potentially avoidable’. For patients with high scores triaged to no critical care, it was assumed that those with outcomes classified as ‘potentially avoidable admissions’ would survive without critical care, but that those classified as ‘critical care required’ or ‘death’ would die, with the ‘critical care required’ group representing deaths that were potentially avoidable if more critical care capacity had been available.

All analyses were performed using stata/version 10.1 (StataCorp LP, College Station, Texas, USA). The concordance statistic was calculated using the somersd package. 28

Selection of data

Data were extracted from the CMPD for 105,397 admissions to 148 adult, general critical care units in England, Wales and Northern Ireland from 1 January 2007 to 31 March 2009. Excluding admissions whose last known status was still in the critical care unit (n = 7), admissions missing the date of discharge from, or death in, the critical care unit (n = 2), and admissions with inconsistencies in CCMDS data (n = 8) left 105,380 admissions to 148 units (99.98%) included in the analysis.

Impact of cancelled/postponed elective and scheduled surgery

Overall, 25,828 (24.5%) admissions were associated with elective/scheduled surgery. These consisted of 23,548 admissions from theatre, 1240 subsequent readmissions, and 1040 admissions for pre-surgical preparation. Cancellation/postponement of these admissions would result in saving 17.0% of calendar days of critical care, 10.9% of calendar days of Level 3 care and 9.9% of calendar days of advanced respiratory support.

There was considerable variation in admissions associated with elective/scheduled surgery across the 148 critical care units (Figure 6). The median number of admissions in the available data from these units for elective/scheduled surgery was 611 [interquartile range (IQR) 368 to 973; and range 44 to 1966, respectively] and the median number of calendar days of critical care delivered was 3490 (IQR 2280 to 5526; and range 326 to 12,361, respectively).

FIGURE 6.

Variation across 148 adult, general critical care units in (a) the percentage of admissions prevented, and (b) calendar days of critical care, (c) Level 3 care and (d) advanced respiratory support saved by cancellation/postponement of elective/scheduled surgery.

Exclusions

After exclusion of admissions associated with elective/scheduled surgery, a total of 79,552 admissions remained. For triage modelling, the following additional admissions were excluded: admissions missing status at ultimate discharge from acute hospital (n = 867, 1.1%); readmissions to the critical care unit within the same acute hospital stay (n = 3475, 4.4%); admissions missing all basic vital signs (n = 700, 0.9%). A total of 74,510 admissions to 148 adult, general critical care units remained for analysis.

Patient cohorts

Of 74,510 admissions included in the analysis, 15,996 (21.5%) admissions were identified as admissions for acute exacerbations of respiratory illness. A brief summary of the case mix of all admissions and of admissions for acute exacerbations of respiratory illness is presented in Table 1.

Triage modelling

Of 74,510 admissions, 19,557 (26.2%) were classified as ‘potentially avoidable’, 31,074 (41.7%) as ‘critical care required’, and the remaining 23,879 (32.0%) died before discharge from acute hospital. Of 15,996 admissions with acute exacerbations of respiratory illness, 4098 (25.6%) were classified as ‘potentially avoidable’, 5800 (36.3%) as ‘critical care required’ and 6098 (38.1%) died before discharge from acute hospital.

The concordance statistics for triage models based on the core variables only, and core variables plus one or more additional variables, are shown in Table 2. Data on blood lactate were only available for 57,551 admissions to 118 units (12,144 with acute exacerbations of respiratory illness), therefore models incorporating blood lactate were restricted to these data. The model based on core variables alone produced a concordance of 0.75 (which may be considered ‘satisfactory’). Incorporating all additional variables raised this to a maximum of 0.79. The discrimination of the models among admissions with acute exacerbations of respiratory illness was generally worse, with concordance statistics ranging from 0.71 to 0.75. Among all admissions, the single additional variables that added most discriminatory ability to the core variables were FiO₂ and urea (each raising the concordance to 0.77). Among admissions with acute exacerbations of respiratory illness, the addition of urea marginally outperformed the addition of FiO₂ (concordance 0.74 vs 0.73), supporting the presence of urea within the CURB-65 score for community-acquired pneumonia. However, the difficulty in obtaining a urea value quickly enough to use as the basis for a triage decision has previously been identified as a limitation of using CURB-65 for triage. 15 It was proposed, therefore, that the most practical physiology-based triage model should be based on the core variables plus FiO₂. Applying Efron’s optimism bootstrap to this model indicated that the anticipated degree of shrinkage when applying this model in future populations was small. The estimated optimism (95% CI) in the concordance was 0.00054 (0.00036 to 0.00073) for all admissions and 0.00099 (0.00055 to 0.00144) for admissions with acute exacerbations of respiratory illness. Adjusted estimates of the concordance for this model were therefore 0.769 and 0.727, respectively.

Incorporating severe comorbidity into the triage model based on core variables and FiO₂ resulted in a negligible improvement in concordance (Table 3). Incorporating age into the triage model, in addition to comorbidity, produced a small improvement in concordance to 0.788 for all admissions and 0.761 for admissions with acute exacerbations of respiratory illness.

A potential simple, physiology-based, triage model is described in Table 4. This model produced a score with a range from 0 to 12. The distribution of the score and associated outcomes are illustrated in Figure 7. As a result of combining adjacent categories from the original fine categorisation to produce this score, there was some loss of discrimination, resulting in a concordance statistic for the score of 0.754 (95% CI 0.747 to 0.760) among all admissions and 0.713 (95% CI 0.705 to 0.721) among admissions with acute exacerbations of respiratory illness. By comparison, the concordance statistic for CURB-65 among admissions with acute exacerbations of respiratory illness was 0.680 (95% CI 0.671 to 0.688).

FIGURE 7.

Distribution of the score (bars) and relationship with outcomes (line: predicted by model; points: observed) for a 12-point triage score based on core variables plus FiO₂.

The effect of using this model to triage patients with low and high scores is shown in Tables 5 and 6, respectively. Although the strategy of diverting patients with a low score to temporary critical care areas may seem appealing in terms of initially diverting a relatively large proportion of admissions from the critical care unit, the effect on bed occupancy is much less as the ‘potentially avoidable admissions’ generally had short stays in the critical care unit. The upper limit of any such strategy would be to save the 12.5% of bed days occupied by ‘potentially avoidable admissions’ overall. By comparison, although triaging patients with high scores diverts a much smaller proportion of admissions, the effect in critical care unit bed days saved is greater as these patients had longer stays in the critical care unit. However, the savings in terms of bed days would be small and the cost of these savings would potentially be a substantial number of avoidable deaths.

Coverage

All acute hospitals in England, Scotland, Wales, Northern Ireland and the ROI were encouraged to participate in SwiFT. As soon as each site completed governance checks, data were collected on consecutive patients meeting the inclusion criteria until SwiFT closed to recruitment on 31 January 2010 (in consultation with the Department of Health). At the end and following completion of recruitment, the local collaborator at each site signed a final declaration, confirming the period of recruitment and indicating that all consecutive, eligible patients had been recruited, or reporting any exceptions. The overall coverage of SwiFT across sites in England was assessed by comparing data from SwiFT with weekly prevalence figures for critical care compiled and published by the Health Protection Agency (HPA) from daily situation report data submitted to the Department of Health from individual NHS Trusts in England. The coverage of SwiFT, relative to the pandemic as a whole, was estimated by comparing the numbers of confirmed H1N1 cases recruited in SwiFT with the total numbers of intensive care unit and high-dependency unit admissions reported by the Department of Health and the devolved administrations. 1

Inclusion criteria

All patients (adult or paediatric) were included in SwiFT if they were either:

• confirmed or suspected pandemic H1N1 patients referred and assessed as requiring critical care; or

• non-H1N1 patients referred and assessed as requiring critical care (under usual/non-pandemic circumstances), but not admitted to a critical care unit in the hospital where referred and assessed.

Data collection

Selected clinical data were collected from both the point of referral and assessment for critical care and daily (calendar day 00:00–23:59) while receiving critical care. Daily data collection continued until either tested negative for suspected H1N1 cases or critical care ended for confirmed H1N1 cases (Figure 8). Confirmed H1N1 was defined as tested positive for H1N1 by real-time polymerase chain reaction (PCR) or viral culture from upper respiratory swabs or tracheobronchial aspirate. Suspected H1N1 was defined as any patient believed to have H1N1 and managed as such. Tested negative for H1N1 was defined as at least one negative real-time PCR from both upper respiratory swabs and tracheobronchial aspirate. Date, time and status when critical care ended were collected for all suspected H1N1, confirmed H1N1 and non-H1N1 cases.

FIGURE 8.

Overview of data collection for SwiFT.

Assessment data included event (date, time and location of assessment); sociodemographic (age, sex and ethnicity); body composition [ranges of body mass index (BMI)]; pregnancy status; H1N1-related (status, vaccine status, antivirals and presentation); chronic organ dysfunction (mild/severe); responsiveness (Alert/Voice/Pain/Unresponsive, confusion); vital signs (temperature, blood pressure, heart and respiratory rate); oxygenation (oxygen saturation and fraction of inspired oxygen); and selected results (base excess, blood lactate, serum urea and creatine kinase). All elements to calculate the CURB-65 assessment of severity of pneumonia score23 were included.

Daily data included H1N1-related (status and antivirals); organ support (by body system including type of ventilatory support); selected drugs; responsiveness (lowest GCS); vital signs (lowest systolic blood pressure and highest heart rate); oxygenation (lowest PaO₂ with associated FiO₂); selected results (highest bilirubin, lowest platelet count, highest blood lactate and highest creatinine) and fluids (total urine output and overall balance). All elements to calculate the SOFA score32 were included.

SwiFT data were entered onto a secure web-based data entry system developed and hosted by the ICNARC. Data collection manuals and forms (see Appendix 2), definitions (either as help text or as answers to frequently asked questions) and error checking were either available for download or built into the design of the web portal.

Reporting during the pandemic

Once a suitable sample size of cases had been accrued in the SwiFT database, weekly reports were circulated to policy-makers and to critical care clinicians. Key policy-makers receiving the reports included the Department of Health Pandemic Flu Team, Chief Medical Officer, Scientific Advisory Group for Emergencies, Pandemic Influenza Clinical and Operational Group, Director General responsible for Pandemic Influenza, Director of NHS Flu Resilience, Scientific Pandemic Influenza sub-group on Modelling Operational group, Influenza Clinical Information Network, and Swine Flu Critical Care Clinical Group. To maximise access to all clinicians, the weekly reports were uploaded onto the SwiFT web portal to provide regular updates on the evolving pandemic.

The content of the reports was developed initially by two SwiFT co-investigators (KMR, DAH) and circulated to key policy-makers and clinicians for comment and input.

System capacity

The impact of the H1N1 pandemic on critical care system capacity was assessed by: the numbers of patients refused critical care (owing to either perceived futility or the lack of available staff and/or beds); the numbers of patients managed in extended critical care areas or in non-critical care areas; and the numbers transferred to receive critical care in another acute hospital.

The wider impact of the H1N1 pandemic was assessed by evaluating data from the CMPD for the year April 2009 to March 2010 compared with the previous 2 years. Data were extracted for all adult, general critical care units with complete data for the entire 3-year period, and the following were plotted for each month: total number of admissions; number of admissions direct from theatre following elective or scheduled surgery; number of admissions from recovery only (representing use of recovery as an extended critical care area); number of admissions transferred to a critical care unit in another hospital; and the number of these transfers that were for non-clinical reasons (i.e. not for more specialised critical care or repatriation).

Description of cases

Suspected/confirmed H1N1 cases were classified, for presentation, into the following three categories:

• Confirmed: H1N1 confirmed either on initial assessment or at any time during critical care.

• Suspected: H1N1 suspected on initial assessment, but never confirmed nor tested negative.

• Tested negative: H1N1 suspected on initial assessment, but never confirmed and subsequently tested negative.

The patients identified in these three categories were described and compared by the following factors.

Patient demographics were described by age on initial assessment, ethnicity, sex, pregnancy status and body composition. Ethnicity was reported by collapsing the ethnic categories, as used in the 2001 UK census, into the following five groups: white (A, B, C); mixed (D, E, F, G); Asian or Asian British (H, J, K, L); black or black British (M, N, P); other ethnic groups (R, S); or not stated (Z). Pregnancy status was recorded as currently pregnant, recently pregnant (within the past 42 days), or neither. Body composition was assessed, either objectively by calculation of BMI or subjectively, and categorised as: very thin (BMI < 16 kg m^-2); thin (BMI 16–18.5 kg m^-2); acceptable weight (BMI 18.6–24.9 kg m^-2); overweight (BMI 25–29.9 kg m^-2); obese (BMI 30–39.9 kg m^-2); or morbidly obese (BMI ≥ 40 kg m^-2).

Chronic health status was described by reported chronic organ dysfunction of respiratory, cardiovascular, renal, hepatic and neurological systems (classified as either moderate or severe) and by the presence of immunocompromise (due to disease or therapy). For respiratory and cardiovascular organ dysfunction, moderate dysfunction was defined as outpatient management and severe dysfunction as severe impairment in activities of daily living. For renal organ dysfunction, moderate dysfunction was defined as outpatient management and severe dysfunction as requiring chronic renal replacement therapy. For hepatic organ dysfunction, moderate dysfunction was defined as compensated liver disease and severe dysfunction as decompensated liver disease or awaiting transplantation. For neurological organ dysfunction, moderate dysfunction was defined as some impairment in activities of daily living and severe dysfunction as severe impairment in activities of daily living.

Reported presentation was classified as: viral pneumonitis/ARDS; secondary bacterial pneumonia; exacerbation of airflow limitation, e.g. COPD or asthma; or another intercurrent illness with H1N1 (coded with the ICNARC Coding Method20).

Acute severity on initial assessment was described by the CURB-65 score,23 which is a severity assessment score for community-acquired pneumonia. It consists of five points, one point each for the presence of new onset confusion, urea > 7 mmol l^-1, respiratory rate of ≥ 30 breaths minute^-1, low systolic (< 90 mmHg) or diastolic (≤ 60 mmHg) blood pressure, and age ≥ 65 years. Acute severity on the first calendar day of critical care was summarised by the SOFA score. 32 The SOFA score is a scoring system summarising the degree of organ dysfunction with 0–4 points assigned to each of: respiratory (PaO₂ : FiO₂ and mechanical ventilation); cardiovascular (mean arterial pressure and administration of vasopressors); renal (creatinine and urine output); hepatic (bilirubin); neurological (GCS); and coagulation (platelets), giving a total score ranging from 0 to 24.

For patients receiving critical care, the duration of critical care was calculated from the date and time critical care commenced to the date and time critical care ended. The duration of critical care was displayed graphically by cumulative frequency plots and summarised by the median and mean. The survival status for the three groups was summarised at the end of critical care within the original hospital.

Performance of triage models

The performance of the simple physiology-based triage model developed was assessed among patients with confirmed or suspected H1N1 at initial assessment for critical care and compared with that of CURB-65. The outcome variable differed slightly from that used to derive the model, as follow-up was available only to the end of critical care and not to ultimate discharge from acute hospital. This may therefore be considered to represent the performance of the model under the assumption that all those patients surviving the critical care episode would go on to leave acute hospital alive. [Note: data from all adult, general critical care admissions in the CMP indicate that, of those who survive to leave the critical care unit, a further 8% die before discharge from acute hospital.]

Risk factors for death and duration of critical care

For patients receiving critical care within the original hospital, risk factors for death, while receiving critical care within the original hospital, were assessed with a Cox proportional hazards regression model. The outcome for the model was the time to death while receiving critical care. Patients ending critical care or lost to follow-up (owing to incomplete daily data) were treated as censored. The potential risk factors included in the model were: age (non-linear relationship fitted using restricted cubic splines); sex; pregnancy status; ethnicity (white vs non-white); H1N1 status (confirmed vs suspected) on initial assessment; reported presentation; location of assessment (emergency department vs other); chronic organ dysfunction (any severe vs any moderate vs none); being immunocompromised; and SOFA score on first calendar day of critical care.

For patients surviving to the end of critical care within the original hospital, risk factors for longer duration of critical care were assessed with a Cox proportional hazards model. The outcome for the model was time to end of critical care. Deaths were excluded and patients lost to follow-up were treated as censored. The same potential risk factors were explored as listed for time to death while receiving critical care (above).

Coverage

SwiFT received central research governance approval on 3 September 2009 and the secure, web-based, data entry system went live on 17 September 2009. A total of 192 acute hospitals participated in SwiFT, including 154 of 214 acute hospitals in England (72%), 6 of 16 in Wales (38%), 4 of 9 in Northern Ireland (44%), 19 of 25 in Scotland (76%) and 7 of 37 in ROI (19%). Owing to the very variable time taken to complete local governance checks, participation increased over time.

Final declaration forms were received from the local collaborators representing 188 (98%) of the 192 acute hospitals that commenced recruitment. Of the returned declarations, 64 (34%) indicated that not all paediatric cases were included, owing to a misunderstanding of the inclusion criteria and only recording data for patients admitted to the adult critical care unit (n = 59), owing to lack of resources (n = 3) or owing to cases being missed (n = 2). Capture of all confirmed and suspected H1N1 cases was reported as complete for 178 (95%) acute hospitals, with 10 reporting potentially incomplete capture owing to either lack of resources (n = 3) or cases being missed (n = 7). Capture of non-H1N1 cases affected by the pandemic was reported as complete for 170 (90%), with 18 reporting potentially incomplete capture owing to a misunderstanding of the inclusion criteria (n = 11), to lack of resources (n = 5), or to cases potentially being missed (n = 2).

The number of acute hospitals participating by week, of the 174 acute hospitals admitting adult patients and returning completed final declaration forms confirming periods of continuous screening for eligible cases, is shown in Figure 9 (four paediatric hospitals that actively participated in SwiFT have been excluded from this figure owing to the lack of complete capture of paediatric cases in other hospitals). Extrapolating from these figures indicated overall coverage of 39% (46% in England) of acute hospitals with adult critical care facilities for the period 3 September 2009 to 31 January 2010. Figure 10 shows the number of new adult cases by week compared with the number of participating hospitals.

FIGURE 9.

Number of acute hospitals admitting adult patients participating in SwiFT by week (week 36 = week commencing 31 August 2009).

FIGURE 10.

Number of new confirmed/suspected H1N1 cases (aged ≥ 16 years) in SwiFT, by week, compared with number of participating acute hospitals (week 36 = week commencing 31 August 2009).

Initial coverage of SwiFT in England was low compared with the weekly prevalence figures published by the HPA. However, by the end of SwiFT, a greater number of prevalent cases were identified from SwiFT than were reported by the HPA (Figure 11).

FIGURE 11.

Weekly reported prevalent cases of confirmed or suspected H1N1 receiving critical care in England from SwiFT compared with HPA figures (week 38  =  week commencing 14 September 2009).

All cases, including those added to the portal after the close of recruitment, were included in this report. Overall, 1,725 confirmed or suspected H1N1 cases and three non-H1N1 cases were reported in SwiFT (Figure 12). Of these, 268 (16%) cases and 2,098 (19%) daily assessments (of a final total of 11,322) were added to the web portal after 31 January 2010, when SwiFT closed to new cases.

FIGURE 12.

SwiFT case flow.

Of the 1725 H1N1 cases, 562 (33%) were confirmed to have H1N1, either on initial assessment or at any time during critical care, 899 (52%) tested negative for H1N1 having initially been suspected, and 264 (15%) suspected cases were neither confirmed nor tested negative. Of 1679 cases assessed with information on location available, 603 (36%) assessments took place on the ward, 506 (30%) took place in the emergency department, 521 (31%) took place in a critical care or extended critical care area, and 49 (3%) took place in other locations.

Relative to the total number of admissions with H1N1 to intensive care units and high-dependency units throughout the entire pandemic (waves 1 and 2 combined), as reported by the Department of Health and devolved administrations, 474/2326 (20%), 6/64 (9%), 6/50 (12%) and 60/187 (32%) were recorded in SwiFT from England, Wales, Northern Ireland and Scotland, respectively. Total figures for the ROI were not available.

Reporting during the pandemic

The first weekly report from SwiFT was submitted to the Department of Health on 13 November 2009 and was uploaded to the SwiFT web portal on 16 November 2009. Reports continued weekly until the final weekly report on 5 February 2010. An example of a weekly report is included in Appendix 3. Weekly dialogue was maintained with the Deputy Director of the Department of Health Pandemic Flu Team and additional ad hoc analyses were conducted on specific issues (e.g. on pregnancy, on vaccination, on obesity, etc.), as required. In addition, the SwiFT Chief Investigator presented an update on the impact of the pandemic on critical care services to the Swine Flu Critical Care Clinical Group on 11 December 2009.

A comparison of results using those data available (i.e. entered) by end of week 47 with all data up to week 47 (i.e. including those data subsequently entered) indicated no difference in the substantive results provided in the weekly reports.

System capacity

Of the three non-H1N1 cases reported in SwiFT, one patient was reported to have been refused critical care owing to lack of available staff and beds, and the remaining two patients received critical care in an extended critical care area.

Of the suspected and confirmed H1N1 cases reported in SwiFT, one patient was reported to have been refused critical care owing to perceived futility, one patient was reported refused critical care owing to lack of available staff and beds, two patients died while under assessment before transfer to a critical care unit could be arranged, 42 patients received critical care in an extended critical care area, two patients received critical care in a non-critical care area, and 11 patients were transferred to receive critical care in another acute hospital.

Complete data for the period April 2007 to March 2010 were available for 125 adult, general critical care units in the CMPD. There was no clear pattern to indicate either higher absolute numbers of admissions during the pandemic (Figure 13) or lower numbers of admissions following elective or scheduled surgery (Figure 14). Admissions from recovery only (Figure 15), representing use of recovery as an extended critical care area, were more frequent in winter, but no more common in 2009–10 than in previous years. Similarly, there was no evidence that transfers between critical care units increased during the pandemic, either overall (Figure 16) or for non-clinical reasons (Figure 17). It should be noted, however, that none of these figures would identify cases admitted directly to an extended critical care area and managed there without ever being admitted to a critical care unit, as these cases would be outside the scope of the CMP.

FIGURE 13.

Total number of admissions by month to 125 adult, general critical care units in England, Wales and Northern Ireland with complete data from April 2007 to March 2010.

FIGURE 14.

Number of admissions following elective/scheduled surgery by month to 125 adult, general critical care units in England, Wales and Northern Ireland with complete data from April 2007 to March 2010.

FIGURE 15.

Number of admissions from recovery only (representing use of recovery as an extended critical care area) by month to 125 adult, general critical care units in England, Wales and Northern Ireland with complete data from April 2007 to March 2010.

FIGURE 16.

Number of admissions transferred to a critical care unit in another acute hospital by month from 125 adult, general critical care units in England, Wales and Northern Ireland with complete data from April 2007 to March 2010.

FIGURE 17.

Number of admissions transferred to a critical care unit in another acute hospital for non-clinical reasons by month from 125 adult, general critical care units in England, Wales and Northern Ireland with complete data from April 2007 to March 2010.

Description of cases

Demographics of the cases on initial assessment for critical care are shown in Table 7. Confirmed H1N1 cases were younger than those suspected or tested negative, with 92% aged < 65 years. Confirmed cases were more likely to be pregnant (13% of female patients vs 2%–3%) and more likely to be obese/morbidly obese (25% vs 20% for suspected and 13% for tested negative).

Chronic organ dysfunctions, presentation, physiology and acute severity of illness on initial assessment for critical care are reported in Table 8. Overall, one-third of confirmed H1N1 cases had one or more severe chronic organ dysfunctions (including being immunocompromised). This was similar for those suspected or tested negative. Unsurprisingly, the most common severe chronic organ dysfunction was of the respiratory system (26% of confirmed H1N1 cases). Vital signs and oxygenation data were available for between 84% (temperature) and 92% (heart rate) of cases. Further test results were available for between 26% (creatine kinase) and 73% (base excess) of cases. Acute severity of illness on initial referral and assessment for critical care, as assessed by the CURB-65 score, was low with 61% of confirmed H1N1 cases scoring 0 or 1 point.

The duration of critical care is presented in Figure 18, with patients whose last known status was receiving critical care (n = 18) treated as being censored at this date. The median duration of critical care was 8.5 days for confirmed H1N1, 1.3 days for suspected H1N1 and 5.4 days for those patients tested negative. The mean duration of critical care was 13.5 days, 2.1 days and 10.4 days, respectively. Of those receiving critical care in the original hospital, data on survival status were available for 537 of 556 (97%) confirmed H1N1, 223 of 255 (87%) suspected H1N1 and 891 of 899 (99%) of those tested negative. Of those with complete data, 423 of 537 (79%), 154 of 223 (69%) and 754 of 891 (85%), respectively, survived to the end of critical care.

FIGURE 18.

Duration of critical care for confirmed, suspected and tested negative H1N1 cases.

Performance of triage models

Of 1651 patients with confirmed or suspected H1N1 on initial assessment: 320 (19.4%) died while receiving critical care; 850 (51.5%) survived to the end of critical care and were reported to have received advanced respiratory support, advanced cardiovascular support, renal support, hepatic support and/or neurological support; and 481 (29.1%) survived to the end of critical care and were not reported to have received support for any of these organ systems. The concordance (95% CI) of the simple physiology-based triage score and CURB-65 for discriminating among these groups was 0.613 (95% CI 0.590 to 0.637) and 0.615 (95% CI 0.595 to 0.634), respectively.

Risk factors for death and duration of critical care

The results of the Cox proportional hazards regression model for death while receiving critical care are displayed in Table 9 and Figure 19. A hazard ratio > 1 represents a higher risk of death and a hazard ratio < 1 represents a lower risk of death. Increasing risk of death was associated with increasing age > 30 years, increasing SOFA score, severe chronic organ dysfunction and being immunocompromised. Pregnancy was associated with a lower risk of death.

FIGURE 19.

Hazard ratio and 95% CI for death while receiving critical care by age relative to age 45 years.

The results of the Cox proportional hazards regression model for duration of critical care among survivors are displayed in Table 10 and Figure 20. A hazard ratio < 1 represents a longer duration of critical care and a hazard ratio > 1 represents a shorter duration of critical care. Increasing duration of critical care was associated with increasing age up to 50 years (with a possible decrease at the oldest ages), increasing SOFA score, overweight or obesity, pregnancy, confirmed H1N1 on initial assessment, severe chronic organ dysfunction, and presentation with viral pneumonitis/ARDS or secondary bacterial pneumonia.

FIGURE 20.

Hazard ratio and 95% CI for duration of critical care among survivors by age relative to age 45 years.

Identification of comparator data – UK wave 1 H1N1 pandemic and pre-pandemic cohorts

Data for critically ill, UK patients, from the first wave of the H1N1 pandemic and for pre-pandemic comparator cohorts, were identified from the CMP (see Chapter 2 for a description of data collection and validation).

Early in the pandemic, all critical care units participating in the CMP were contacted with a request to submit individual data files for patients with confirmed or suspected H1N1 rather than waiting to submit these cases in their routine quarterly data submission for the CMP. Cases were identified in the free text field of the data set with the words ‘Confirmed Swine Flu Case’ or ‘Suspected Swine Flu Case’. The free text of all routinely submitted data was also searched for relevant terms (e.g. ‘H1N1’, ‘pandemic influenza’) to identify any additional cases that were not reported in this way. For the purpose of this comparison, the wave 1 H1N1 cohort has been defined to be all confirmed H1N1 cases, identified in this way, that were admitted to critical care units participating in the CMP between 1 June 2009 and 31 August 2009.

Pre-pandemic cohorts were selected using data from 1 December 2007 to 31 March 2008 and 1 December 2008 to 31 March 2009 to represent the previous two influenza seasons. Two, pre-pandemic cohorts were identified based on the primary, secondary and ultimate primary reason for admission to the critical care unit, coded using the ICNARC Coding Method. 20 The primary (mandatory) and secondary (optional) reasons for admission are coded based on information available up to and including the first 24 hours in the critical care unit. The primary reason for admission may be updated to an ultimate primary reason for admission based on information that became available later. The two cohorts were defined as follows:

• Viral pneumonia: admissions with ‘Viral pneumonia’ coded as the primary, secondary or ultimate primary reason for admission.

• Bacterial pneumonia: admissions with ‘Bacterial pneumonia’ or ‘Pneumonia, no organism isolated’ coded as the primary, secondary or ultimate primary reason for admission and without ‘Viral pneumonia’ in any of these fields.

Identification of comparator data – international cohorts

International comparator data were extracted from published cohorts of critically ill patients with H1N1. Four appropriate cohorts were identified:

• Seven hundred and twenty-two patients from wave 1 of the pandemic in Australia and New Zealand (June–August 2009) admitted to an adult or paediatric intensive care unit with confirmed H1N1 according to WHO definitions (see Appendix 4). 3

• One hundred and sixty-eight patients from wave 1 of the pandemic in Canada (April–August 2009) identified as critically ill (defined as admitted to an adult or paediatric intensive care unit, or mechanically ventilated, or FiO₂ ≥ 60%, or receiving intravenous inotrope or vasopressor) with confirmed or probable H1N1 according to WHO definitions. 40

• Fifty-eight patients from wave 1 of the pandemic in Mexico (March–June 2009) identified as critically ill (defined as for Canada) with confirmed, probable or suspected H1N1 according to WHO definitions. 41

• Thirty-two patients from wave 1 of the pandemic in Spain (June–July 2009) admitted to an adult intensive care unit (age ≥ 15 years) with confirmed H1N1 according to WHO definitions. 42

Descriptive analysis

Confirmed H1N1 cases from SwiFT were compared with the wave 1 H1N1 pandemic and pre-pandemic cohorts based on information that was available and similarly defined in SwiFT, and in at least one comparator data set.