Discussion

In this Delphi study, experts on health-care triage and emergency planning reached consensus on a set of clinically and ethically relevant and appropriate patient information items needed to support triage team decision-making in crisis capacity settings during the COVID-19 pandemic. In developing this list, experts worked to optimize and balance multiple goals by selecting information items that would support accurate prognostication and consistency while avoiding factors associated with implicit biases. A final set of patient information items included patient preferences for care, a list of select severe or end-stage chronic comorbidities (chronic lung, heart, liver, oncologic, and kidney diseases), the reason for hospitalization, duration of ICU care, specific information on the severity of current illness, clinical trajectory, and age. Furthermore, our results support the feasibility of collecting these patient information items from an institutional electronic health record.

Delphi participants grappled with multiple—and often conflicting—priorities, including the need to provide sufficient information to support accurate prognostication, careful avoidance of potentially biasing information, as well as pragmatic considerations around time efficiency in developing a final list of patient information items needed to support triage. Balancing these priorities involved complex ethical, clinical, and operational considerations, and experts expressed a range of diverse perspectives about the most appropriate approach.

Despite multiple existing guidelines and approaches to disaster triage advocating the use of physiology-based scoring systems, ^{Reference Raith, Udy and Bailey21,Reference Grissom, Brown and Kuttler22} deliberation among Delphi participants aligned with emerging concerns about the limitations and pitfalls of using these composite scores for emergency triage reflected in the broader literature. First, these scores may not provide sufficiently granular prognostic predictions to differentiate between seriously ill patients to be of value in disaster triage. ^{Reference Raith, Udy and Bailey21,Reference Grissom, Brown and Kuttler22} Furthermore, these scores have not been validated to predict mortality for patients with COVID-19 or other future viral pandemics. ^{Reference Ferreira, Bota and Bross23–Reference Raschke, Agarwal and Rangan25} Finally, these scores may incorporate both recognized and implicit biases. ^{Reference Cleveland Manchanda, Couillard and Sivashanker11,Reference Sederstrom and Wiggleton-Little26,Reference Schmidt, Roberts and Eneanya27} This last point is especially pertinent because health-care emergencies often exacerbate existing health disparities, ^{Reference Solomon, Wynia and Gostin28,Reference Webb Hooper, Nápoles and Pérez-Stable29} and while the triage process is unlikely to effectively correct for the many inequities in the health-care system, it should at a minimum seek to avoid worsening inequities. ^{Reference Cleveland Manchanda, Couillard and Sivashanker11,Reference Tolchin, Hull and Kraschel13,Reference Solomon, Wynia and Gostin28,Reference Cleveland Manchanda, Sanky and Appel30,Reference Bruce and Tallman31}

The question of whether to include age as a factor in triage team decision-making generated substantial debate among Delphi participants, which is also reflected in the broader literature. ^{Reference Farrell, Ferrante and Brown32} Age is a strong predictor of mortality for critically ill patients, including for patients with COVID-19, ^{Reference Raschke, Agarwal and Rangan25} and because older adults, on average, have a lower expected longevity, consideration of age in triage would result in a disproportionate number of older adults being denied treatment. ^{Reference Montero-Odasso, Hogan and Lam33} However, there is substantial variability in survival among individuals of similar age, ^{Reference de Rooij, Abu-Hanna and Levi34,Reference Brunner-Ziegler, Heinze and Ryffel35} and prevalent negative implicit biases directed at older adults—as well as the very young—risk shaping triage decisions to an extent beyond the prognostic relevance of this factor. ^{Reference FitzGerald and Hurst36,Reference Chapman, Kaatz and Carnes37} Our results reflect agreement about the prognostic value of age, but also reinforce the importance of ongoing real-time monitoring for the impact of implicit biases throughout the triage process.

Consistent with prior work identifying factors associated with ICU mortality, ^{Reference Zilberberg and Epstein38–Reference Chelluri, Im and Belle40} our findings reinforce the relevance of severe or end-stage chronic lung, heart, liver, oncologic, and kidney diseases in predicting prognosis. Delphi experts also considered conditions that may have a somewhat weaker association with prognosis, but are also known to be strongly shaped by health inequities and social determinants of health (eg, diabetes and hypertension). ^{Reference Harrell, Hall and Taliaferro41,Reference Walker, Strom Williams and Egede42} These factors may also lack commonly agreed-upon definitions. The group ultimately determined that these factors did not—on balance—warrant inclusion on a final patient information form. Clinical frailty and malnutrition, which may have prognostic value for critically ill patients, were also excluded because of concerns about a lack of a consistent application and risk of bias against people living with disabilities. ^{Reference Solomon, Wynia and Gostin28,Reference Chelluri, Im and Belle40} The inclusion of free text boxes on an information form was suggested as an opportunity for clinical teams to provide other valuable clinical context but this option was similarly believed to involve a high risk of introducing irrelevant and potentially biasing information into the triage team process. Ultimately, relatively few patient information items were included in a final patient information form, perhaps reflecting a default to exclude potentially biasing or ambiguous information. The development of specific definitions of each information item may serve to further support consistency and limit introduction of implicit biases in data collection and interpretation.

Past consensus processes identified several discrete clinical conditions as having a uniquely poor prognosis, including high body surface area burns, ^{Reference Saffle, Gibran and Jordan43} severe trauma, and persistent vegetative state. These conditions have historically served as exclusionary criteria in some triage algorithms, including for WA state. ¹⁴ Our results suggest that this very specific and limited set of conditions may appropriately constitute a separate “striped” triage category. People affected by these conditions would be prioritized lower than other groups in crisis capacity settings, but rather than being categorically excluded from consideration for scarce resources (as in existing algorithms), they would receive therapy as possible, guided by the patients’ preferences. This approach supports proportionality, that is, withholding resources only to the extent necessitated by scarcity.

These findings lay the foundation for additional studies to precisely define, operationalize, and determine the effectiveness of this set of patient information items in supporting triage team functioning. Further evaluation of this process through triage team simulations as well as additional input from public stakeholders, especially those representing underserved populations, would strengthen the ethical foundation of this set of patient information items to be used for allocating scarce life-sustaining resources in a health-care crisis. ^{Reference Cleveland Manchanda, Sanky and Appel30}