It is widely accepted that conditions exist for the evolution of a new strain of influenza virus with the potential to cause a human pandemic 1. The biggest challenge in planning for an influenza pandemic is the range of unknown factors; its nature and impact cannot be fully predicted until the pandemic virus actually emerges. Those planning for a pandemic must, therefore, work from a number of assumptions based on knowledge gained from previous pandemics and scientific modelling of a range of potential scenarios. The UK Pandemic Influenza Plan 2 sets out a range of possible scenarios for clinical attack rates and case fatality rates during a pandemic, including the potential for more than one wave. The base scenario assumes a clinical attack rate of 25% and a case fatality rate of 0.37%, giving rise to 53,700 excess deaths in the UK. A reasonable worst case scenario involves a cumulative clinical attack rate of 50% with 2.5% case fatality, causing 709,300 excess deaths. Similarly, the US Department of Health and Human Services predicts that in a "moderate" scenario based on a virus with 1968-like pathogenicity, 865,000 will require hospitalisation and 65,000 (7.5%) will require ventilation. They also outline a "severe" 1918-like scenario with 9.9 million hospitalisations and 743,000 patients requiring ventilation 3.

An influenza pandemic will undoubtedly create a major increase in demand for critical care services. The majority of UK hospital intensive care units (ICUs) are already operating at > 98% bed occupancy. Integral to the success of any emergency planning strategy is 'surge capability', incorporating the ability to scale up the delivery of appropriate specialist care to those that require it 4. Modelling of the impact of an influenza pandemic on UK critical care services has been carried out using the FluSurge 1.0 programme developed at the US Centers for Disease Control 5. With simulation of an 8-week epidemic and 25% attack rate the demand for critical care beds from patients with influenza would represent 208% of current combined level 2 (high-dependency unit) and level 3 (ICU) bed capacity, and 231% of current level 3 capacity 6. Even allowing for optimistic estimates of other modulating factors (50% reduction in ICU demand with use of neuraminidase inhibitors and 50% upgrade of level 2 to level 3 beds), level 3 bed occupancy due to the pandemic would remain at 75%. Furthermore, occupancy of level 3 beds by 'flu patients' was unsustainable at approximately 50% in terms of care for other patients even in the most optimistic conditions.

Although some research and modelling exists regarding hospital surge capacity for major incidents, this generally relates to 'big bang' single incidents rather than 'rising tide' prolonged problems 7891011. The closest objective evidence for efficacy of critical care in the event of a flu pandemic is extrapolated from H5N1 influenza and the recent SARS outbreak in Toronto. Of the H5N1 admissions to hospital in Thailand, 75% developed respiratory failure. Hospital mortality in these cases was 75% 12. During the Toronto SARS outbreak, up to 32% of cases were admitted to ICU, 25% were mechanically ventilated and 28 day mortality for ventilated patients was 45% 13. In Singaporean SARS patients admitted to ICU, 98% developed ARDS 13.

Properly constructed plans for the delivery of critical care during an influenza pandemic must include the ability to deal with excessive demand, high and possibly extreme mortality, and the risk to the health of critical care staff.

The consequences of a pandemic, both in terms of numbers of patients and the effect on the healthcare system, are likely to precipitate a 'major incident' where special arrangements are needed to manage the system while it is under extreme pressure. It is anticipated that there will be an overwhelming demand for critical care services, not only for respiratory support through mechanical ventilation but also for a full range of care to manage multiorgan failure. Assuming that the next pandemic derives from the H5N1 strain, the epidemiological evidence to date suggests extremely high mortality and, although not precisely quantifiable, a significant risk to health care workers. Both of these will undermine the ability to deliver critical care to influenza patients even before consideration is given to the duty of care to other critically ill patients.

Coherent incident response requires a robust command and control structure, with the ability to make rapid informed decisions across an organisation and also across a health economy. In the UK, health incident management is based on a 'medallion' structure, with gold, silver and bronze corresponding to strategic, tactical and operational command levels 14. North American and Asian health institutions tend to use the Hospital Emergency Incident Command System 15. The common theme in both systems is a clear command and control structure with which healthcare staff should be familiar 41416171819. Their generic hierarchical structure allows application to a wide range of incidents whilst retaining familiarity gained from training and exercises. The importance of familiarity with the command and control structure was highlighted in a recent Delphi study 20 and European survey 21.

Critical care contingency planning guidance from the UK Department of Health places an expectation on providers to expand their level 3 bed capacity by a factor of 3 but no more. Provision of full multiorgan level 3 support is recognised to be unrealistic, but principally respiratory support is felt to be achievable. Cancellation of elective surgery to minimise alternative sources of demand for critical care, upgrading level 2 to level 3 facilities and recruitment of theatre recovery areas and even operating theatres may allow expansion of ICU-like care capacity. Staff in these areas already have the competencies to manage sedated patients and those receiving respiratory support. Escalating their clinical role should require relatively limited focussed training 22. Other staff may need to be redeployed and receive training in the management of critical care patients to support fully trained staff, permitting a dilution of the standard critical care nurse to patient ratio 23. Flexibility around dependency level and staff experience will be required 24. The expansion of ICU capacity to provide critical care in other areas will require the pre-emptive identification, tracing and maintenance of all usable equipment and potentially the stockpiling of key items to allow for rapid up-scaling of activity in response to demand.

It is likely that there will be some variability in the prevalence of influenza across the country during a pandemic wave, with peaks in demand staggered across geographical areas. It may be possible to disperse some of the patient load by inter-facility transfer if this occurs to any significant extent.

The expansion of ICU facilities during the SARS epidemic in Hong Kong and Singapore was recently described 25. Infection control is recognised as an overriding priority for the delivery of critical care, including the ability, in the early stages, to cohort cases. This should ideally include the use of separate entrances and exits, isolation rooms with negative pressure ventilation and dedicated separate healthcare staff. The Toronto experience identified 21 secondary cases of nosocomial transmission of SARS in ICU from an initial index case before infection control measures were introduced. Even following the introduction of extensive protective equipment, nine healthcare workers developed SARS as a result of being present in the room during the intubation of a single patient. In terms of personal protection, planning and practice in the donning of protective equipment (PPE) and prior fit testing is essential 26. The practicalities of being able to manage patients when fully attired must be understood and consideration given to the fact that any procedure or task will take longer. This will impact on care efficiency and the staff to patient ratio.

While beds can be scaled up and extra areas recruited to provide critical care, without trained staff the planning will be ineffective. Staff illness rates and the risk to staff must be factored into the planning process. In the UK, staff illness has been estimated at 30% with work absences of up to 8 days 2. Normal working patterns may need to be revised and facilities provided for staff to stay on site rather than go home to their families. Staff absence tends to be greater the longer special circumstances apply and the greater the impact on the lives of the staff 27. The preventive effectiveness of neuraminidase inhibitors may make focussed chemoprophylaxis a strategy for reducing staff illness in critical care areas 28.

The evolution of a new pandemic strain of influenza will inevitably result in a major increase in demand for critical care services. It is likely that these services will rapidly reach their capacity and even their contingency arrangements for extended facilities will be overwhelmed. Excessive demand where resources are finite creates an ethical dilemma and many emergency plans apply a utilitarian approach of 'best care for the greatest number' 29. There is a legitimate debate about how limited capacity can best be utilised, but a number of themes are recurrent. There needs to be a legal and ethical framework for the process decided in advance, the rationale for triage should be fair and transparent and it should meet the principles of distributive justice 303132. Triage can conflict with human rights legislation and even humanitarian laws but 'accountability for reasonableness' can temper the disagreements about priority setting 33.

The decision making process needs to be valid and reproducible. Although there are a number of triage systems available for mass casualty incidents, there has been little validation of any of them in the field 34, and what there has been relates to 'big bang' single incidents and the apparent unreliability of triage 3536. While it does not need to be explicit ahead of time, the decision thresholds should be based on both the cumulative evidence about the disease process and prognosis, and the number of patients and severity of illness making the demands on the service 31. In effect, triage may result in a gradual degradation of care with the increasing scale of the incident and become a 'societally mandated Do Not Resuscitate order'. On these grounds the process needs to be carefully considered at an appropriately senior level and applied consistently 32.

Allowing for the utilitarian approach, it is recognised that in mass casualty incidents, the standard of care for all patients, including those not immediately related to the incident, may need to be adjusted and reduced. While this may infringe individual rights, the higher ethical principle of 'wellness of society as a whole' allows for the direction of resources to those where it is felt most effective. It may also allow for an expansion in the scope of practice of non-physicians 37.

It may be unrealistic and impractical to expect that senior medical intensive care staff will make all decisions regarding instituting critical care and there will be a need to empower more referring general clinicians to do so. This is at odds with the need for decision making by the most senior person 32 and will require a change in practice for many clinicians; it is not current practice in the UK. The use of track and triage protocols will be essential to direct this decision making and ensure its consistency. Ardagh 38 has developed a set of pragmatic questions for the clinician facing acute problems of resource allocation; the only point lacking in his assessment process is a tool for the 'ranking' of patients in terms of likelihood of benefit from the limited resources.

We believe that the basic criteria for a system for triage to critical care in a pandemic are fourfold; it should identify patients sick enough to require higher level care at some stage in their illness, it should be able to recognise those patients who are too acutely or chronically unwell to benefit from critical care, it should be consistently applicable by healthcare professionals and support workers from a variety of backgrounds within the constraints of the pandemic and should ideally also be scalable to reflect any mismatch between need and capacity. In order to fairly allocate resources across both flu and non-flu patients it should also be disease non-specific and allow prognostic comparisons across disease categories.

A number of scoring systems have been advocated for use in a pandemic. The UK Department of Health currently recommends a six-point pneumonia severity score 2. Although US guidelines emphasise the importance of triage in primary influenza, specific tools are only recommended for assessment of post-influenza bacterial pneumonia 39.

The majority of available potential scores were developed as mortality indicators and perform less well for predicting critical care usage. Amongst ICU admissions with community-acquired pneumonia in Massachusetts in 1996 to 1997, 10/32 scored CURB-65 1 or 2 (that is, low risk) and 5/32 were classified as PSI (Pneumonia Severity Index) class III (intermediate risk) 40. Even amongst patients with pneumonia included in the PROWESS study, only 90.5% were PSI class IV or V, and only 70.3% had a CURB-65 score of 3 or above 41.

There is no guarantee that pandemic influenza will be primarily pneumonic in its presentation; case reports have documented H5N1 influenza presenting with diarrhoea 4243 and coma 43 and a World Health Organisation summary has described absence of respiratory symptoms in a number of cases 44. The utility of disease-specific pneumonia scores may also be limited by mortality from comorbidities such as cardiovascular disease.

A number of intensive care scoring systems have demonstrated their power in using physiological derangement to predict mortality or higher resource requirements, whatever the presenting diagnosis 4546474849. Physiological scores have also been demonstrated to be good predictors of requirement for higher level care on hospital wards 50, in medical assessment units 5152 and in the Emergency Department 53. We have demonstrated that a purely clinical score incorporating acute physiological derangement and chronic health and performance status can reliably predict requirement for critical care 54.

It is inevitable that if an influenza pandemic reaches the scale of some predictions, some patients who, in normal circumstances, would benefit from critical care will not be offered it. Critical care triage will need to evolve from a process of identifying cases who need high level care to one that determines those patients most likely to benefit from the limited resources available and distinguishes them from those where care is likely to be futile.

This is recognised by the Emergency Medicine community and the US administration in terms of disaster triage 3755. The American Thoracic Society adopted the utilitarian principle a decade ago, stating that "the duty of health providers to benefit an individual patient has limits when doing so unfairly compromises the availability of resources needed by others" 56. The problem now facing policymakers and clinicians is defining a process for resource allocation that meets the requirements of distributive justice and accountability for reasonableness 33. As the Working Group on Emergency Mass Critical Care of the Society for Critical Care Medicine recognised, "an ideal triage system is based on data collected at hospital admission, requires little or no laboratory testing, and has been proven to predict hospital survival" 57.

The Ontario Ministry of Health Long-term Care working group have courageously taken the first steps in defining a triage protocol for critical care 58 and their use of serial Sequential Organ Failure Assessment (SOFA) scores to place a ceiling on care provided to non-responding patients is to be supported. However, it is unlikely to be feasible for all patients to have a trial of inotropes and/or ventilation and some way of screening out the sicker patients at ward/floor level will be required.

We are not aware of the use of objective prognostic scores to allocate or refuse critical care resources at present and indeed most research demonstrates the ad hoc nature of admission decision-making 59. However, if, as is likely, review by experienced critical care physicians is impractical, decision support will be required for the non-critical care specialist. Emergency physicians, for example, had a positive predictive value (PPV) of only 73% in identifying those with a low chance of survival, as opposed to critical care fellows (PPV 83%) and the Mortality Probability Model (MPM₀; PPV 86%) 60.

SOFA scoring has previously been demonstrated on a multinational basis to predict high risk of mortality (a SOFA score of over 15 was 98.9% specific for mortality) 61. Other critical care scoring systems show comparable performance in mortality prediction; discrimination as measured by area under Receiver Operator Characteristic (ROC) curve was 0.825 to 0.901 for Acute Physiology and Chronic Health Evaluation III (APACHE III) 62636465, 0.79 to 0.846 for Simplified Acute Physiology Score II (SAPS II) 626466, and 0.928 for the Multiple Organ Dysfunction Score 67. However, calibration of these scores to give absolute risks of mortality has not always been reliable 65 and has required customisation for international use 6869.

Concentrated work is clearly required to amend and validate existing scoring systems so that they are suitable for use as triage tools. We suggest that this should be done on two levels. While disease specific scoring systems are valuable and should continue to be refined, there is a need to develop an appropriately generalisable scoring system for as unselected a group of patients as possible. To have the discriminating power, it will need to take place on a multi-centre or, preferably, on a multi-national basis.

It is a general principle of major incident planning that procedures should not be changed at precisely the moment when the system or institution is under its greatest stress, so planning for pandemic flu needs to make use as much as possible of systems and procedures already in place. Development of a triage system and tool needs to be accompanied by planning for hospital command and control (to dictate scalability as related to available resources) and by training for staff whose roles may change.

Researchers, clinicians and policymakers in the field need to analyse systems and scores already in existence and improve and validate them as triage tools (though this may not be the purpose for which they were originally developed). At the same time ethical principles require transparency and consistency in the decision-making process, and involvement of public in its development.

In reality, perhaps the question we need to address is the action required when critical care services are overwhelmed. The scalability of triage tools may aid in decision making by objectively altering the threshold for admission to critical care. However, the time may come when we need realistically to evaluate the effectiveness of critical care in influenza. If survival with the benefit of critical care is marginal (for example, <10%) and there is a significant cross-infection risk, perhaps critical care should then close and concentrate its efforts on outreach to other areas, including wards. Direction and support from professional bodies and health departments will be required to support the medical staff with such difficult decisions possibly against a ground swell of media-driven public opinion.

ICU = intensive care unit.

DW is a member of the UK Department of Health Critical Care Contingency Planning Working Group. The other authors declare that they have no competing interests.

This article is part of a thematic series on Disaster management edited by J Christopher Farmer.

Other articles in this series can be found online at http://ccforum.com/articles/theme-series.asp?series=CC_Disaster