Clinical review: Allocating ventilators during large-scale disasters – problems, planning, and process

Catastrophic disasters, particularly a pandemic of influenza, may force difficult allocation decisions when demand for mechanical ventilation greatly exceeds available resources. These situations demand integrated incident management responses on the part of the health care facility and community, including resource management, provider liability protection, community education and information, and health care facility decision-making processes designed to allocate resources as justly as possible. If inadequate resources are available despite optimal incident management, a process that is evidence-based and as objective as possible should be used to allocate ventilators. The process and decision tools should be codified pre-event by the local and regional healthcare entities, public health agencies, and the community. A proposed decision tool uses predictive scoring systems, disease-specific prognostic factors, response to current mechanical ventilation, duration of current and expected therapies, and underlying disease states to guide decisions about which patients will receive mechanical ventilation. Although research in the specifics of the decision tools remains nascent, critical care physicians are urged to work with their health care facilities, public health agencies, and communities to ensure that a just and clinically sound systematic approach to these situations is in place prior to their occurrence.


Introduction
Current health care emergency preparedness planning recognizes that excess critical care capacity for a disaster is extremely limited. Most critical care units routinely function at or near capacity [1][2][3], significantly decreasing available critical care response capabilities for disasters [4][5][6]. These limitations are not merely a consequence of space and staffing issues; countless medical equipment and supplies are maintained in quantities sufficient only for daily operations. Unfortunately, usual medical material distribution chains are vulnerable to disruption during disasters, so additional supplies and equipment may not be readily obtainable. For some equipment and medications, traditional allocation units can be reduced and substitutions are possible (for example, oral instead of intravenous medications) to allow all in need to have adequate access to resources rather than fewer to have unlimited access.
For a number of complex, durable medical devices such as mechanical ventilators, there are few acceptable alternatives [7,8]. In some circumstances, temporizing measures such as manually ventilating patients may be adequate. Such a strategy is likely to be inadequate for disasters requiring days of ventilatory support, especially for people with severe respiratory failure. Even if enough mechanical ventilators were available to meet the need of hundreds or thousands of additional critically ill patients in a community, critical care personnel may be in too short of supply to withstand the demands of pervasive disasters, like a severe influenza pandemic. The critical care response to a severe influenza pandemic will be further challenged by insufficient external relief and augmentation assets; the critical care needs of all communities may not be met [9][10][11]. Under these circumstances, triage of patients who are newly requiring critical care interventions and of those who have already received some degree of definitive medical care may be required in order to justly allocate the limited functional capability of mechanical ventilation at a health care facility. To optimize these decisions, there are several key constructs that hospitals must discuss and predefine: 1. A well-practiced incident management system (for example, Hospital Incident Command System) [ A few recent publications have examined both the process [10] of decision-making and possible decision tools for burns [16] and mechanical ventilation [9,11]. Prognostic data can be used to triage patients after radiation exposures [17,18]. Research in trauma mass casualty triage is scant [19,20] and is often based on military experience [21][22][23]. Pandemic influenza triage has been examined in the context of prediction of the need for hospital admission and mortality, but not in terms of resource triage [24].
Given the heterogeneity of possible events and agents that may result in large numbers of patients requiring critical care, no single tool or scale can be expected to provide adequate decision-making power. There are also many injury and disease states (such as pandemic influenza) for which decision tools will have to be developed and validated during the event as evidence of disease epidemiology and patient response to therapy accrues. The availability (or lack) of accurate and rapid testing for certain diseases may also have significant impact on the ability to predict outcome or make definitive diagnoses [25,26].
Thus, the focus at the hospital level should be on establishing the process that will be followed at the health care facility and within the geographic region in a resource-scarce situation, knowing that the specific decision support tool may be eventdriven. This is crucial because regardless of the origin of the decision tool (national government, local/regional government, specialty medical society, and so on) the implementation of the tool occurs at the hospital level. This article will discuss a sample process for resource-scarce situations and a prototype decision matrix for allocating ventilators.
• In a disaster situation, the focus of medical care shifts from the needs of the individual (autonomy) to the needs of the community as a whole (distributive justice) so that the 'greatest good for the greatest number' is the goal. Actual application of this ethical principle is complex and is the subject of current debate and interpretation. It is customary for the critical care physician to heavily weigh patient and family wishes and subjective considerations in determining 'futile care'. This calculus is reversed during a disaster so that the weight is on objective prognostic criteria and less on subjective and individual patient factors (Table 1). This shift in priorities will require significant pre-event education and training for critical care staff. • An additional overall goal-which has received inadequate attention-should be to provide patients as much comfort and dignity as the situation allows regardless of other interventions available. • If a specific scarce resource that is life-saving or potentially life-saving is not available in sufficient quantity to meet patient demand despite all efforts to obtain adequate resources from other local, regional, and national partners in a timely manner and -No temporizing measures are available (for example, when manual ventilation is not an option) and -Resource cannot be 'titrated' (for example, drugs or oxygen) or substituted (for example, oral instead of intravenous antibiotics) and -All available resources and resource surrogates (for example, bi-level positive airway pressure and anesthesia machines) have been repurposed to manage respiratory failure, but these efforts are inadequate to meet the demand, then -The overarching goal is to allocate facility resources to those likeliest to benefit, taking into consideration [39]: • Medical prognosis • Underlying disease • Expected duration of resource need • Duration of benefit • Quality of life after intervention (unfortunately a criteria subject to significant bias and interpretation). -Any other considerations involving weighing subjective measures (for example, the role of the patients, including health care workers, in the community) must be determined by public discussion, and a means for a lay panel/team to assess these factors must be available if the community determines that this is important. Medical personnel should not apply subjective criteria in their decision-making. This is consistent with the American Medical Association position on scarce-resource allocation [39]. • Due to government control of practitioner licensure and liability, any such system of resource allocation should be part of a planned state/province or national government response to an overwhelming emergency. As part of this planned response, medical providers must be protected legally for making these difficult decisions. Under no circumstances should a health care facility be in a position of having to make systematic triage decisions without activation of state or national emergency health powers [10,40] to enable legal protection for providers who in good faith are complying with pre-existing response plans or event-specific state and jurisdictional directives.

Allocation decision-making
The first goal of health care emergency preparedness planning is to augment the actual capacity and capabilities of the system [5,10,14,[41][42][43][44]. Should this capacity prove inadequate to meet the demands of a catastrophic disaster, the secondary goal is to make the system 'fault-tolerant' or to 'bend' rather than 'break'-making the minimum adaptations necessary to cope with resource shortages ( Table 2).
Many incidents with multiple victims may require some degree of adaptation or resource triage. (For example, in a car crash with multiple victims, who goes to computed tomography or the operating room first?) However, these situations are generally mitigated rapidly and do not usually require a systematic, ongoing, and institution-wide process to determine what medical care can be provided based on the demand and the resources available. Resource stockpiling, staff cross-training, and altering staff responsibilities and documentation requirements can help increase the capacity for these incidents and thus prevent a resource-scarce situation, as can timely resource requests based on anticipation of deficits prior to their actual occurrence [4,5,10,14,[41][42][43][44].
When an event is ongoing and when it is likely that resources will continue to be inadequate to meet immediate or future needs, changes to the usual process of patient care become necessary to provide sufficient medical care relative to the resources available and the patient demand. The goals are to make the minimum adaptive changes necessary, with an emphasis on administrative changes (for example, limiting documentation and having non-clinical staff serve meals) rather than clinical changes (for example, triaging persons away from the emergency department and allocating limited ventilators), and to reverse the adaptations as soon as adequate resources become available [10].
As part of this response, mobilization of available health care and non-health care personnel to provide patient care should occur. Internally, staff should be reassigned and roles redefined to maximize patient care time. External workforce augmentation via agreements with the Medical Reserve Corps and other sources of personnel augmentation (such as Federal Disaster Medical Assistance Teams) should be preplanned with relevant agencies. Also, this planning should include appropriate just-in-time training for all personnel taking on unfamiliar tasks and should credential, privilege, and assign mentors to outside personnel assisting in the institution. A clinical care committee or similar group (Table 3) must review the facility resources and community needs and determine what services will be offered, how and where these services will be provided, and by what process triage decisions will be made should that become necessary [10].
The process for institutional response (a sample of which is detailed in Table 4) will vary depending on the size and mission of the facility. For this to be successful, timely and accurate unit and individual patient data must be available to the committee. Note that this represents an ideal situation. In times of disaster, the incident commander must be able to account for the actions required to fairly institute such protocols even if the full clinical care committee is unavailable. However, the full committee as determined by the institution is critical for effective planning, training, and Available online http://ccforum.com/content/11/3/217 drilling of the scarce-resource response to ensure that the institutional plan is realistic and sound.
The importance of incident command awareness of and facility with these situations and the adaptive structures and mechanisms that exist at the institution cannot be underemphasized. Furthermore, critical care staff should be aware of the process for decision-making when demand exceeds resources so that they are prepared to act, rather than spend valuable time reacting to a completely new health care paradigm.

Decision support tools
Determining which patients shall receive mechanical ventilation when triage is required will be a difficult process from both a clinical and a psychological perspective. There are currently no models that provide a framework for considering what we believe are three key areas that should be evaluated when considering a patient for critical care services: • Organ system function (and severity of impairment) • Duration of ventilator use and duration of benefit related to: -Disease-specific predictors (for example, pandemic influenza) -Underlying disease (for example, severe chronic obstructive pulmonary disease) • Response to a trial of mechanical ventilation (if received based on above factors) An ideal decision tool for allocating mechanical ventilation would likely be graphically clean, easy to apply, objective, accurate, reproducible, and predictive of resource use and outcome and would not discriminate against vulnerable populations. No such tool is available, but a matrix (developed as a discussion draft for the Minnesota Department of Health) that captures the core elements above is presented in Figure 1. The matrix provides a supportive framework for clinical decision-making but allows the 'best method/ evidence available' to be used within the matrix boxes and adjusted on an ongoing basis.
A standardized method of assessing multi-organ function (and failure) is a key step in determining prognosis, and several scoring systems have significant ability to predict mortality. Prior authors have suggested using the Sequential Organ Failure Assessment (SOFA) [45] in a strategy considering underlying disease and organ failure [9,11].
The SOFA score has wide validity across a range of patient pathologies and is easy to calculate, using few laboratory and no invasive hemodynamic variables. Changes in SOFA scores over time may be of value in determining prognosis as well [46][47][48][49][50][51][52][53][54][55]. Notably, SOFA and most scoring systems have not been validated in pediatric populations. The SOFA score provides an objective result relating to patient prognosis and is a numeric that may be compared between institutions and Table 2 Fault-tolerant systems 'Fault-tolerance or graceful degradation is the property that enables a system to continue operating properly in the event of the failure of some of its components. If its operating quality decreases at all, the decrease is proportional to the severity of the failure, as compared to a naively designed system in which even a small failure can cause total breakdown. Fault-tolerance is particularly sought after in high-availability or life-critical systems' [61].
Many systems must be engineered to be fault-tolerant. The same principles must be applied to critical services provided by hospitals during a disaster. The following strategies can be applied to manage demand that would otherwise prompt system failure: • Engineered system failure -Similar to a circuit breaker, this allows system components to fail in order to prevent catastrophic damage to the system as a whole. An example might be a hospital switchboard that gives preference to internal hospital calls (rather than to calls from external sources) to preserve internal communications during an emergency. • Redundancy -Having adequate duplicate supplies or services available in case of failure (for example, extra intravenous pumps or ventilators). • Diversity -Having many ways of providing the same service, but via different techniques (for example, triaging patients in multiple areas of the hospital: emergency department, lobby areas, and so on).  [56]. This inaccuracy compounds the difficulty in deciding whom to offer ventilation as there would need to be a substantial difference (for example, 25%) in survival advantage predicted by the scoring system to justify removing one patient from a ventilator and giving the resource to another. Furthermore, this use repurposes these prognostic systems to tasks for Table 4 Sample process for healthcare facility response during resource-poor situation • Incident commander recognizes that systematic changes are or will be required to allocate scarce facility resources and that no regional resources are available to offset demand.
• Planning chief gathers any guidelines, epidemiologic information, resource information, and regional hospital information.
• Clinical care committee reviews facility/regional situation and examines the following: -Alternate care sites -Can additional areas of the building or external sites be used for patient care? (This should be planned in advance.) -Medical care adaptations (for example, use of non-invasive ventilation techniques, changes in medication administration techniques, and use of oral medications and fluids instead of intravenous).
-Changes in staff responsibilities to allow specialized staff to redistribute workload (for example, floor nurses provide basic patient care in the intensive care unit while critical care nurses 'float' and troubleshoot) [5] and/or incorporate other health care providers, lay providers, or family members where practical.
-Triage plan describing how the use of scarce resources at the facility (emergency department [ED] resources, beds, operating rooms, and ventilators) will be allocated. (What level of severity will receive care? What tool or process will be used to make decisions when there are competing demands for the same resource?) -Community/regional strategies to cope with the situation and how the institutional response contributes to those efforts.
-Committee summarizes recommendations for next operational period and determines meeting and review cycles for subsequent periods (may involve conference calls or similar means to avoid face-to-face meetings during a pandemic).
• Incident commander approves committee recommendations as part of incident action plan. Plan is operationalized. Public information officer communicates updates to staff, patients, families, and the public.
• Current inpatients, patients presenting to the hospital, and their family members are given verbal and printed information (ideally by the triage nurse in the ED or, for inpatients, by their primary nurse or physician) explaining the situation and that resources may have to be reallocated, even once assigned, in order to provide care to those who will most benefit. A mechanism for responding to patient/family questions and concerns should also be detailed.
• Security and behavioral health response plans should be implemented.
• Triage plan (which may affect all units equally or some more than others) implemented: -ED/outpatient screening of patients (and denial of service to patients either too sick or too well to benefit from evaluation/admission) based on guidance disseminated by the clinical care team.
-Tertiary triage team (ideally NOT the physicians directly providing the patients' care and ideally two physicians of equal 'rank' in the institution) considers situations in which there are competing patient demands for a scarce resource. The resource should be assigned as follows: When two patients have essentially equal claim to the resource, a 'first-come, first-served' policy should be used.
When, according to guidelines or the triage team's clinical experience, the claim to the resource is clearly not equal, the patient with a more favorable prognosis/prediction shall receive the resource.
The triage team should ask for and receive whatever patient information is necessary to make a decision but should NOT consider subjective assessments of the quality of the patients' life or value to society and, in fact, should ideally be blinded to such information when possible.
• A 'bed czar' (under the Hospital Incident Command System, this is the inpatient unit leader) should be appointed to make final decisions on bed assignments. This individual should have access to real-time inpatient and outpatient system status and, when needed, patient clinical information.
• Whenever a decision is made to reallocate a ventilator or similar critical resource, the treating physician and family should be provided with the grounds for the decision (which should be documented for the record at the facility) and a rapid appeals process if there is additional or new information that the family or a treating physician feels would affect the decision.
which they were not designed and which were not considered during the original mortality 'cut-score' determinations. Thus, using a numeric score based on organ function is not likely to be a satisfactory sole discriminator (unless the difference between patient scores is large) since it is not enough to 'rank' patients in order to place them on available ventilators as there are likely none available.
Prospective validation of the use of the SOFA score and other scoring systems for these purposes should be a focus of research efforts. Awaiting better prognostic tools, we would recommend the SOFA score due to its ease of calculation and its reliance on a minimum of laboratory testing.
Duration of benefit and duration of resource need are likely to involve more subjective interpretation of data and are thus more fraught with ethical peril. However, we feel that these considerations are important when determining ventilator allocation.
Assessment of likely duration of benefit and duration of need for ventilation must account for both current disease-specific prognostic factors and underlying disease states (particularly ones that severely shorten life span or impact quality of life). Prognosis is easier to estimate for some pathologies (pulmonary contusion from trauma and severe burns) but less so for disease states such as pandemic influenza, in which the age groups affected, response to treatment, and other factors are unknown. Assessment of expected duration of mechanical ventilation aims to successfully manage more patients with the same resource over a prolonged event, thus offering greater community benefit, but may be invoked only when there is a clear-cut difference between patients (for Decision matrix for ventilator allocation during resource-poor situation. Compared to other patient(s) requiring and awaiting mechanical ventilation, does this patient have significant differences in prognosis or resource use in one or more categories above that would justify reallocation of the ventilator? a The SOFA (Sequential Organ Failure Assessment) score is a currently preferred scoring system based on type of data required and ease of calculation. b Examples of underlying diseases that predict poor short-term survival include (but are not limited to) the following: congestive heart failure with an ejection fraction of less than 25% (or persistent ischemia unresponsive to therapy or ischemia with pulmonary edema); acute renal failure requiring hemodialysis (related to illness); severe chronic lung disease, including pulmonary fibrosis, cystic fibrosis, or obstructive or restrictive diseases requiring continuous home oxygen use prior to onset of acute illness; immunodeficiency syndromes with evidence of opportunistic pathogen infection; central nervous system, solid organ, or hematopoietic malignancy with poor prognosis for recovery; cirrhosis with ascites, history of variceal bleeding, fixed coagulopathy, or encephalopathy; acute hepatic failure with hyperammonemia; acute and chronic and irreversible neurologic impairment that makes patient dependent for all personal care (for example, severe stroke, congenital syndrome, persistent vegetative state, and severe dementia). c Changes in oxygenation index (OI) over time may provide comparative data, though of uncertain prognostic significance. OI = MAWP × FiO 2 /PaO 2 , where MAWP is mean airway pressure, FiO 2 is inspired oxygen concentration, and PaO 2 is arterial oxygen pressure. PaO 2 may be estimated from peripheral oxygen saturation by using the oxygen dissociation curve if blood gas measurements are unavailable.
example, pulmonary edema secondary to missed hemodialysis versus acute respiratory distress syndrome [ARDS]). Though important, these predictions are imprecise and often may be subject to bias.
Consideration of significant underlying disease states and their effect on patient life span, quality of life, duration of resource benefit, and resource demands is important for determining whether there is a substantial difference between the patients requiring the therapy, but is most subject to interpretation bias and is least specific. Triage based on underlying disease states aims to ventilate patients who do not have diseases that would predict a protracted or poor response to treatment or limited duration and scope of benefit following recovery. Use of hospice and other welldefined 'quality of life' standards and longevity predictors may be helpful strategies, and this is being explored currently by a task force in King County, WA, USA (L Rubinson).
Patients already on a ventilator or those who are offered a trial of mechanical ventilation (and it should be viewed as a therapeutic trial rather than a resource 'assignment' in a resource-scarce situation) should be reassessed on an ongoing basis. If their condition (and prognosis) continues to worsen (or simply fails to improve after an appropriate duration of time given their disease state), this should be weighed when considering whether to continue ventilatory support. An objective tool to assess response to mechanical ventilation is needed and should be developed. The oxygenation index has been correlated with mortality in certain groups and is one of the few calculations available that may have utility as an objective marker as it changes over time. Although its applicability to triage is unknown, a worsening oxygenation index over time may provide helpful information to the critical care physician and triage team about the benefit of mechanical ventilation [57][58][59][60]. Teams may find it helpful to track patient data and history factors on the matrix tool so that these can be maintained for ongoing comparisons/updates and changes tracked easily over time. Note that as clinical conditions change, patients C or D may become the most marginal recipient of therapy and thus would be the basis for comparison rather than patient A. It is extremely important that the triage team be aware of evolving changes in patients' clinical conditions and be able to maintain a relative 'rank list' of intensive care unit patients on ventilators relative to their clinical condition and other variables.
The science of triage (in particular, tertiary triage) is nascent, and we believe that our suggestions above will soon be replaced by more robust and researched strategies. Predictive models that are more specific may be introduced into the 'disease-specific prognosis' category (for example, neurotrauma, burns, and ARDS). The matrix allows better predictive systems to be incorporated without changing the basic framework or factors considered. This matrix may also be used to consider whether to offer other therapies to patients (such as intensive care admission), but there may be other simpler tools that rely on clinical assessment and that could be used in an emergency department setting to accomplish this type of secondary triage (after initial stabilization).
Additionally, although the matrix attempts to incorporate relevant variables for triage of mechanical ventilation, it has not been tested for reproducibility between practitioners and patient groups, which limits its inherent validity. However, we feel that this tool represents a significant step forward from prior empiric efforts and reflects realistic considerations that must be balanced when making these difficult decisions. It is also a concrete reference point for discussions about limiting care in resource-scarce situations, as theoretical discussions about clinical care committees and the ethical basis of withdrawing and withholding medical care may miss practical issues and problems that the institution would face in such a circumstance.

Conclusion
Critical care providers are in a position to help inform and lead hospital discussions about resource allocation decisions, critical care surge capacity, and the exercising of these plans at their facility. Indeed, their absence from the process predicts failure of the systems in crisis. Additionally, critical care physicians must understand that the patients being hospitalized during such an event will be far sicker than the usual caseload and must determine in advance as part of their departmental plan how their limited nursing and physician staff would be extended (consultative role, documentation, and external consultations for hospitals not used to providing long-term critical care) [5].
Finally, critical care physicians must exert leadership to ensure that their institution is prepared for a situation in which critical care resources, including mechanical ventilation, would be inadequate to meet the community need.
Determining 'futile care' in the context of daily patient care is far different from a resource-poor situation (Table 1), and critical care physicians can assist their hospital ethics committee and administration in developing plans and providing education that will facilitate such a paradigm shift should that become necessary. We urge that these conversations begin now, so that should such a tragic situation develop, providers and the public alike understand our limitations, our goals, and the systems that we have designed to make the best of our limitations.