Simulation of Ventilator Allocation Strategies During the COVID-19 Pandemic

As the number of SARS-CoV-2 infections mount globally, hospital systems face unprecedented demands on limited resources, including ventilators. Existing ventilator allocation strategies have not undergone rigorous analyses to evaluate their real-world implications. To address this critical gap in knowledge with profound societal implications, we present a simulation model to evaluate the proposed allocation strategies and report the trade-offs between these strategies that merit careful consideration. effects models dynamic power simulation modeling traditional ethical reasoning difficult choices ventilator allocation.


Introduction, Results And Discussion
The COVID-19 pandemic is overwhelming healthcare systems around the world. The number of available ventilators currently falls far short of the anticipated need by tens or even hundreds of thousands in the United States and elsewhere. 1 Given this dire reality, there is consensus that institutions should establish algorithmic patient ranking systems to allocate ventilators rather than compel subjective rationing decisions to be made at the bedside. 1,2 Proposed ventilator allocations systems have uniformly abandoned the "first-come, first serve" and "sickest-first" principles and moved to utilitarian considerations of maximization of benefits for the population. 1,2 There is no consensus, however, on whether benefit maximization should focus on lives saved or life-years saved and to what extent benefit maximization should be constrained by the ethical value of treating people equally. 1,2 Some advocate that younger patients should be prioritized to ensure as many achieve a "complete life" as possible, 3 while others argue that age should not be used as an explicit factor in allocation at all. 1 As a result, there is considerable variation in the translation of these ethical principles into concrete triage rules. For example, early reports and guidelines from Italy suggest physicians are employing age cutoffs. 4,5 New York state guidelines categorize patients needing mechanical ventilation into three priority tiers based on the Sequential Organ Failure Assessment (SOFA) score and use a lottery to break ties within tiers; age is explicitly ignored. 6 Another proposed priority score from researchers at the University of Pittsburg would linearly assign the patient 1-4 points from three different 3 categories: SOFA, medical comorbidities, and age. 2 Maryland's state proposal would give patients a score ranging from 1-8 based on SOFA and comorbidities, using age as only a tiebreaker. 7 Each of these varying formulas represents the mathematical summarization of a nuanced ethical framework.
The ethical consequences of applying specific triage rules to a complex system like ventilator allocation can be better appreciated with simulation modeling.
We developed a simulation model to illustrate the downstream consequences of different triage rules under varying conditions of patient volume, patient severity of illness, and ventilator availability. The model draws a patient sample from the age distribution of COVID-19-infected patients as reported by the Centers for Disease Control 8 and the distribution of SOFA score and comorbidities at the time of intubation derived from a large tertiary care hospital system. 9 As recommended by the corresponding priority scores, 2,10 chronic disease categorization (none, major, severe) was assigned based on the presence of qualifying diagnoses and the value of the AHRQ Elixhauser Co-morbidity Index Score for in-hospital mortality. 11 The model then simulates patient survival with ventilation based on the calibration of the SOFA score in a multi-center cohort of over 180,000 critically ill patients with infections. 12 The life-years saved by each model was calculated by subtracting the patient's age from one hundred (maximum lifespan). Life-years saved were discounted 50% if a major chronic disease was present and reduced to one if a severe chronic condition was present. We applied eight different ventilator allocation rules to this population as described in detail in Table 1.
Our simulation model (Figure 1) demonstrates that: 1) Any triage system that incorporates age, SOFA or comorbidities would result in more lives and life-years saved than a simple lottery allocation system; 2) Youngest first system results in more life-years saved but less total lives saved compared to methods incorporating SOFA and comorbidities (Maryland and Pittsburgh); 3) Multi-principle systems incorporating age, SOFA and comorbidities have better performance in both lives and lifeyears saved than New York's SOFA tier system. Importantly, there is much wider variation between the allocation systems in terms of life-years saved compared to lives saved.
Given the ethical complexity of these decisions, no ventilator allocation system will be universally 4 satisfactory. A difficult balance between competing ethical principles must be struck. Simulation For this study, we set the conditions to 1,000 patients with 500 available ventilators to simulate severe scarcity across a regional area (i.e., multiple hospitals across a city). The model created a simulated dataset of 1,000 patients and their characteristics (age, comorbidity burden and SOFA score). The age of these patients were sampled from the age distribution of COVID-19-infected patients as reported by the Centers for Disease Control. 8 The relationships between age, comorbidities and SOFA score for these simulated patients were derived from a large tertiary care hospital system. 9 Table 1. The comorbidity burden for these patients was categorized as none, major or severe based on a combination of currently utilized priority scores 2,10 and the value of the AHRQ Elixhauser Co-morbidity Index Score. 11 A multinomial logistic regression was performed on age to comorbidity burden, and then linear regression was done to regress the SOFA score on the age and comorbidity burden of the patients. The coefficients from these regressions are presented in Methods Table 2. These relationships between age, comorbidity burden and SOFA were used to simulate patient data for the simulation model.
We applied eight different ventilator allocation rules to the simulated population, and these allocation rules are described in further detail in Table 1 in the manuscript. If the allocation rule assigned a patient to a ventilator, the model then determined patient survival based on the calibration of the SOFA score to survival from a multi-center cohort of over 180,000 critically ill patients with infections. 12 If the patient survived, the life-years saved was determined by subtracting the patient's age from one hundred (maximum lifespan). Life-years saved was discounted by 50% if a major 7 chronic disease is present. Since the definition of severe chronic condition in allocation scores is an expected lifespan of one year, the life-years saved was reduced to one if such a condition was present. If the allocation rule did not assign a patient to a ventilator, the patient did not survive. The simulation was repeated for 10,000 runs. Lives saved and life-years saved were averaged over the 10,000 runs and compared across the different allocation strategies.
The simulated data used age, comorbidity and SOFA score relationships based on clinical data from a single medical center. In order to make the simulation more generalizable and widely available, we

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download. 2521310addinfo0q8w18d.txt Appendix.docx