Comparison of 2 Triage Scoring Guidelines for Allocation of Mechanical Ventilators

Key Points Question What are the characteristics of intensive care unit admissions identified by 2 proposed pandemic ventilator allocation triage guidelines using Sequential Organ Failure Assessment scores when applied retrospectively to critically ill US patients who received mechanical ventilation? Findings In this cohort study of 40 439 admissions to intensive care units that received mechanical ventilation, the New York State guideline identified 9% who would likely meet criteria for the lowest priority for ventilator allocation compared with 4% from the original White and Lo guideline. Only 655 admissions (1.6%) were in the lowest priority category for both guidelines, with 39% survival to hospital discharge for admissions identified as lowest priority using the New York State guideline compared with 56% for admissions identified using White and Lo. Meaning Two distinct approaches to triage for mechanical ventilation showed little agreement, suggesting that further clinical assessment of different potential criteria for triage decisions is important to ensure equitable allocation of resources.


Introduction
The potential need to ration mechanical ventilators has been a worldwide focus during the coronavirus disease 2019 (COVID- 19) pandemic. 1,2 Many regions and countries have been forced to address the possible need for ventilator allocation, which refers to identification of critically ill patients for whom mechanical ventilation should be withheld or withdrawn if it has already been initiated. 3 Different variations of triage criteria for this rationing have been proposed. In the US, 2 of the most prominent triage criteria are the New York State protocol 4 and one initially proposed by White and Lo. 5 Both protocols require assessment of a combination of burden of comorbidities and severity of illness on admission.
Despite extensive literature on resource allocation in pandemics, these criteria have not been tested on a large scale in high-income countries. It is unclear how many individuals would meet different triage criteria, which of the specific criteria would most commonly be met, and how many ventilators would be made available if these criteria were applied to deny care to patients deemed less likely to benefit. Therefore, we sought to apply these criteria to a heterogeneous cohort of patients across US intensive care units (ICUs) before the current COVID-19 pandemic as a first step to understanding the impact of applying these criteria to a critically ill population. Specifically, we assessed all admissions to ICUs for patients who received mechanical ventilation to determine how many would meet initial criteria for withholding of mechanical ventilation depending on the level of triage activated and then how many would meet criteria at a later time for withdrawal of mechanical ventilation.

Methods Overview
We assessed triage criteria from 2 documents. The first is the New York State Ventilator Allocation Guidelines. 4 The second is the original guideline proposed by White and Lo (referred to herein as the White and Lo guideline). 5 This second guideline was subsequently updated on April 22, 2020, owing to concerns raised that it violated US antidiscrimination laws. However, the new version cannot be easily operationalized, and other guidelines based on it (eg, Colorado 6 ) have fallen back on use of specific diagnoses to make it feasible to apply it to individual patients. Therefore, for this study, we have chosen to assess the original guideline. For each, we extracted explicit criteria used to determine patient assignment to each level of priority for ventilator allocation. We then applied these criteria to a cohort of ICU admissions from a non-COVID-19 period. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. This study used deidentified and publicly available data. 7 Ethics approval was not applicable because the Philips eICU Collaborative Research Database 7 is released under the Health Insurance Portability and Accountability Act safe harbor provision. The requirement to obtain informed patient consent was not applicable because the re-identification risk was certified as meeting safe harbor standards by Privacert (in Cambridge, Massachusetts).

Cohort
This was a retrospective cohort study using the Philips eICU database, 7 a large database of a random sample of ICU admissions to 291 ICUs in 208 US hospitals participating in Philips' telehealth program from 2014 to 2015. We identified adult admissions (aged Ն18 years) for patients who received mechanical ventilation at any time during their ICU stay. We excluded admissions of patients who were never ventilated, those admitted after elective surgery, and anyone who had inaccurate data for included variables (eFigure 1 in the Supplement). Because patients could be admitted to the ICU more than once during a hospitalization and would on each admission be eligible for assessment regarding triage level, we included all admissions to the ICU.

Outcomes
The outcomes assessed included (1) proportion of all admissions mechanically ventilated and patients who met each level of priority, (2) distribution of specific criteria that patients met for each level of priority, (3) proportion of all ventilator hours, (4) ICU mortality, and (5) hospital mortality.
Among those who were not the lowest priority level on ICU admission, we also assessed how many additional patients met each level of priority 48 hours and 120 hours after admission. For outcomes and descriptors that could only occur once for a patient during a hospitalization (ie, hospital length of stay, hospital mortality, and discharge destination after the hospitalization), we restricted the analysis to the first ICU admission for each patient.

New York State Guidelines
The New York State guidelines include 2 steps that are applied to patients on admission. The first is a list of exclusion criteria for adult patients with "medical conditions that result in immediate or nearimmediate mortality even with aggressive therapy" 4(p57) (eTable 1 in the Supplement). To operationalize the exclusion criteria, we mapped them to available variables in the Philips eICU database (eTable 2 in the Supplement). The data available did not contain all of the granular criteria listed in the triage documents, and some criteria were not quantitative and required clinical interpretation to operationalize them using retrospective data (eTable 2 in the Supplement). The second step in the New York State guidelines is the calculation of a Sequential Organ Failure Assessment (SOFA) score on admission, 8 followed by reassessment at 48 hours and 120 hours to detect any new onset of the stated medical conditions that warrant exclusion and to calculate the change in SOFA score (eTables 3 and 4 in the Supplement).
For the primary analysis, the SOFA score was calculated from the first available variables during the first 24 hours after admission to the ICU (SOFA first 24). The respiratory system score in the SOFA assessment includes points for mechanical ventilation. Because no details are given in the New York State guidelines regarding how to incorporate mechanical ventilation into the initial assessment when mechanical ventilation has been initiated before assessment, we calculated the SOFA score with the regular inclusion of points for mechanical ventilation. Patients with a SOFA score of exactly 7 were not categorized by the guidelines as published (eTable 3 in the Supplement); thus, we assigned these patients to the highest priority group. Missing data were assumed to be normal and not imputed, as is common in assessments of SOFA score 8 and consistent with practice that would occur in real time regarding available data on a patient (eTable 5 in the Supplement). As a sensitivity analysis, we also recalculated the SOFA scores using the most extreme value for each variable in the first 24 hours after ICU admission (SOFA worst 24) and compared the classification of individuals by using the SOFA first 24 vs the SOFA worst 24.

White and Lo Guidelines
The original White and Lo guidelines included a similar approach of assessing comorbidities and SOFA score. There were 2 different categorizations of major comorbid conditions and severe life-limiting conditions (eTable 6 in the Supplement). We similarly mapped these conditions using available variables in the Philips eICU database (eTable 7 in the Supplement). However, this approach did not create exclusions for specific conditions; instead, a certain number of points were allocated for each condition and added to points from the SOFA score (eTables 8 and 9 in the Supplement). Although reassessment is recommended, no specific times for reassessment or explicit criteria for withdrawal of therapies are given; thus, we did not apply any reassessments with the White and Lo guidelines.
resuscitate, or other [further] limitations on life support or other care, such as "no blood products" or "no vasopressors or inotropes"). We used proportions for categorical data and mean (SD) values or median values (interquartile ranges) for continuous data, as appropriate. Next, we calculated individual exclusions and comorbidities. Some admissions met more than 1 exclusion criterion or had more than 1 major comorbidity. We calculated the SOFA scores for all admissions using the SOFA first 24 as the primary assessment and the SOFA worst 24 as a secondary assessment and applied the triage criteria of each guideline to assign each admission to a level of priority for receipt of mechanical ventilation (lowest, intermediate, and highest). We then assessed the characteristics, resource use, and outcomes of the group of admissions in each priority level. We calculated the proportion of total ventilator hours for all patients in the ICU that were taken up by the lowest priority group to understand how many ventilator hours would be potentially "freed up" by not offering mechanical ventilation to this group. We compared differences between the lowest and highest priority groups using standardized differences, with a difference of less than 0.1 considered a negligible imbalance. 10 Because we included all admissions (including multiple admissions by the same individual during the same hospitalization), we also performed a sensitivity analysis comparing agreement between the 2 sets of criteria but restricting the analysis to only the first ICU admission during the hospitalization.

Reassessment of Patients at 48 and 120 Hours
We applied the rules for reassessment of admissions at 48 and 120 hours using the New York State guidelines to determine the classification of admissions at these times for patients remaining in the ICU. We did not have an accurate way to determine new diagnoses or other changes in status that would allow someone to newly meet the exclusion criteria; thus, we did not apply this portion of the reassessment.

Comparison of Guidelines
We sought to determine whether these 2 different sets of triage criteria would identify the same or different admissions for allocation to the lowest priority group. We assessed the chance-corrected agreement between triage guidelines for assignment to the lowest priority category for mechanical ventilation using the κ statistic. All analyses were performed using SAS Enterprise Guide, version 7.1 (SAS Institute Inc), pgAdmin 4, version 3.2 (pgAdmin Development Team), and R studio, version 1.1.456 (R Project for Statistical Computing). The study was conducted in spring 2020.

Cohort
After exclusions, the cohort consisted of 40 439 admissions to the ICU for patients who received mechanical ventilation during their ICU stay (eFigure 1 in the Supplement); the mean (SD) age was 62.6 (16.6) years, 54.9% were male, and three-quarters (75.9%) were White individuals ( Table 1).
The mean (SD) APACHE IV predicted mortality was 27.9% (26.1%). Mechanical ventilation was initiated in 33.8% of admissions before the day of ICU admission and 66.2% on or after ICU admission, with a median (interquartile range) duration of mechanical ventilation of 46.7 (17.0-122. 9) hours. In the first 24 hours in the ICU, 87.9% of admissions were identified as receiving full therapy, 7.9% had a do-not-resuscitate order, and 4.2% had some other limitation on life support or other care. Overall hospital mortality (n = 8795) was 23.5%.

Relevant Comorbidities
As applied to this cohort, 5.4% (95% CI, 5.2%-5.6%) of admissions met at least 1 exclusion criterion in the New York State triage guidelines ( contributed points to the triage priority score. The most common comorbidities were chronic lung disease and dementia (Table 2).

SOFA Scores
The mean (SD) SOFA first 24 score for the cohort was 4.5 (3.7), with a median of 4 and a range from 0 to 21 (eFigure 2 in the Supplement). When recalculated using the worst value in the first 24 hours after ICU admission, the mean (SD) was 5.5 (3.9), with a median of 5 and a range from 0 to 22. Sixty percent of the cohort changed to a different SOFA score when using SOFA worst 24 rather than SOFA first 24, with 18.6% changing by 2 or more points.  (Figure).

Characteristics and Hospital Outcomes
The

Discussion
When applied retrospectively to a cohort of critically ill patients without COVID-19, 2 different ventilator allocation guidelines identified 4.3% and 8.9% of admissions as meeting the lowest level of priority to receive mechanical ventilation. Patients who were assigned the lowest level of priority by these guidelines received 2 to 3 days of ventilation, and both guidelines identified a slightly higher percentage of non-White patients as lowest priority. However, the 2 triage criteria identified substantially different patients for initial consideration for withholding (or very early withdrawal) of mechanical ventilation.
Although many triage criteria exist, few studies have assessed the application of these criteria to actual patients. After the H1N1 pandemic in 2009, a few small studies assessed triage scores or tools for triage. [11][12][13] However, those studies focused on predicting the need for mechanical ventilation rather than assessing which patients would be identified as lowest priority and determining their outcomes. In the present study, we identified challenges with attempting to apply these criteriaparticularly SOFA scores-to patients retrospectively admitted to ICUs given that some life-sustaining interventions had already been applied. Our finding that the use of the worst value of a variable to calculate a SOFA score led to different scores for patients highlights the fluid nature of these scores and the challenges associated with hard cutoffs; a SOFA score may change markedly from before to after intubation in particular, with either deterioration or stabilization after intubation. Moreover, SOFA scores have been primarily validated for general intensive care or sepsis populations and not for isolated respiratory disease. 14,15 The New York State criteria overall identified a sicker cohort. Because the overall goal of these triage criteria is to minimize allocation of a scarce resource to those who seem least likely to benefit, the New York State criteria appeared closer to achieving this goal. However, many admitted patients identified as lowest priority by both sets of criteria survived to at least hospital discharge. Moreover, we are unable to determine whether the choice to forgo mechanical ventilation in favor of treatments, such as prone positioning, may result in better outcomes for some patients, as has been raised as an area of discussion with regard to patients with COVID-19 and their treatment

Limitations
This study has major limitations. We applied these criteria to a non-COVID-19 cohort of admissions who received mechanical ventilation. We recognize that the demographic characteristics and severity of illness (eg, SOFA scores) of patients admitted with COVID-19 may have a different distribution. In particular, patients with COVID-19 may represent a more homogeneous population that would allow for better agreement between different guidelines when applied and, owing to single organ failure, may more frequently have a low SOFA score. However, potential application of triage criteria would not be isolated to patients with COVID-19 but would be as relevant for patients represented in this cohort. Future work should assess more contemporary patients to determine what the application of these criteria would look like in a period with primarily COVID-19 admissions and how SOFA or other scores may need to be recalibrated for a COVID-19 population. We could not fully determine patients' preferences for ventilation and other invasive support or the clinical criteria used for the decision to place a patient on a ventilator because we were limited by the use of retrospective data. We recognize that there can be substantial variability in these decisions that influence the cohort we assessed. 18 We also had to approximate some of the criteria for comorbidities based on available data, and we had some missing data, particularly regarding APACHE IV, that we chose not to impute. We tried to be conservative in our approach to err on the side of undertriage rather than overtriage of individuals to the lowest priority categories. Of the many triage protocols proposed, we chose to assess only 2 that could be most easily operationalized using retrospective cohort data, 19 and we examined only US criteria and patients. A preliminary triage proposal in Ontario, Canada, has a very different approach, which eschews the use of any severity of illness score in favor of a more detailed assessment of clinical condition and underlying diseases to determine who is at high risk of death. 20

Conclusions
This study assessed the application of 2 different ventilator triage criteria retrospectively to critically ill patients without COVID-19, showing the complexity of this application and the disparate choices that might be made depending on the triage criteria chosen. Allocation of life-saving resources in a pandemic is a challenging concept, and many competing interests would be involved in decisions to withhold or withdraw mechanical ventilation. We recognize that the criteria put forth in these guidelines were done so as suggestions rather than as hard-and-fast rules to be adopted across a cohort of patients by using administrative data. This study highlights the importance of assessing these approaches using actual patient data because it allows for determination of the ways that guidelines may be difficult to operationalize, whether certain groups may be disproportionately affected, and what further work is needed to refine these approaches for clinical use. In particular, the lack of agreement between the assessed guidelines highlights that these guidelines approach this allocation challenge in somewhat unique ways and lead to different choices.