Identifying Emergency Department Patients with Life-Limiting Illness May Be Getting a Little Easier

Emergency departments (EDs) are increasingly being called on to provide geriatric-specific care and are critical places for initiating goals of care conversations.¹ A majority of older adults, age 65 and older, will visit the ED in the last six months of life.² Some of these older patients have life-limiting illnesses and unmet palliative care needs. Some may benefit from a goals of care conversation, referrals for hospice, palliative care, or other supportive services. Yet, it can be challenging to identify and assist such patients in a fast-paced ED environment, and only 1 in 10 emergency clinicians state they currently use an effective strategy to screen and refer patients to palliative care.³ One barrier is the lack of pragmatic screening tools for identifying patients with life-limiting illness that are validated for the ED setting. Additionally, ED clinicians often care for 20–30 new undifferentiated patients per shift and may lack the time to use a screening tool.⁴ Finally, the need to provide immediate life-saving interventions may also prevent ED clinicians from initiating goals of care conversations and addressing unmet palliative care needs due to underlying life-limiting illnesses.

A new study by Haimovich et al.⁵ “Automatable end-of-life screening for older adults in the emergency department using electronic health records” provides a potential solution to the need for a validated screening tool. In their paper the authors propose and validate the Geriatric End-of-Life Screening Tool (GEST) to identify ED patients with life-limiting illness. They find their tool performs better than existing mortality-based prediction tools in identifying older adults in the ED at risk of six-month mortality, including the Charlson⁶ and the Elixhauser⁷ comorbidity indices. They performed a retrospective, observational cohort study of 449,147 patient encounters to develop and validate their model and included data from nine EDs across a large health system in New England. They find that among their cohort with a 30% six-month mortality risk, 70% were full code and only 6% had advance care planning (ACP) documentation, suggesting an opportunity for goals of care conversations. The area under the receiver operating characteristic curve (AUROC) for the GEST was very good at 0.82, compared to the Charlson (0.65) and Elixhauser (0.69). This study is highly relevant to geriatric emergency care and needed given the lack of validated tools for this unique setting.⁸ Some caveats to this study include that it was performed in an age-friendly health system where there may be informatics infrastructure that provides better documentation of EHR measures required for the screening tool calculation.

In their prediction model they included clinical and demographics variables, vital signs, and laboratory data. Top model coefficients included ED minimum systolic blood pressure, age, and history of lung cancer. GEST identified 10.5% of patients as having a risk of six-month mortality greater than 30%, while the Charlson and Elixhauser indices identified only around 2% of visits by older adults as having greater than 30% mortality risk. Thus, GEST identified 4 times as many patients at high risk with robust specificity. Unlike Elixhauser and Charlson, ED vital signs are incorporated in the GEST. Lessons learned from the authors’ approach include the importance of using not only historical and demographic risk factors, but also drawing on real-time data like vital signs and laboratory tests. This may be particularly important for prediction in patients with dynamic presentations and acute conditions such as those who present to EDs. To our knowledge, few health systems currently provide automated calculation of end-of-life indices as part of the EHR, and whether clinicians employ these tools, and they benefit patients remains understudied.

Alternative prognostic screening tools include the “surprise question”⁹ (Would you be surprised whether this patient died in the next 12 months?). This tool relies only on physician gestalt and thus has the potential for implementation without informatics support; however, it is non-automated and relies on clinician input, as well as clinical experience. In a study of 207 patients and 38 ED attendings, the surprise question demonstrated a sensitivity of 77%, specificity of 56%, positive predictive value of 32%, and negative predictive value of 90% at 12 months.⁹ Using the same question for prediction of 1 month mortality had an AUROC of only 0.73.¹⁰ Other ED screening tools exist, but rely on additional human resources above and beyond routine ED staff to perform screening, or are tailored to cancer, dementia, or specific populations of older adults.⁸

To be sure, an automated screening tool to identify ED patients with life-limiting illness alone will not directly result in patients receiving care that they desire. This tool can serve as a first step in identifying patients with a limited prognosis but does not identify whether these patients have unmet palliative care needs. Thus, it is still incumbent on ED staff to take these additional steps prior to offering primary (e.g., basic) palliative care or initiating referrals. Further barriers still exist to increase assessment of unmet palliative care needs among ED patients,⁸ and how screening tools can improve delivery of primary palliative care in the ED or referrals to palliative care services remains to shown. A common problem in EDs is that while patients arrive 24/7, other professionals may not have the capacity to consult at any time of day. Having a palliative care team available if patients screen positive on the GEST and would like palliative care referral would be ideal. Other models which include referral to palliative care after the ED visit on the inpatient service may be more feasible but will not address acute symptoms or other care needs quickly enough to influence ED disposition. Thus, patients may be hospitalized for inadequate pain control, even though their overall preferences for care are to avoid hospitalization. The ideal trajectory for patients whose suffering and symptom needs are unmet is to have them rapidly identified, addressed, and for patient to be able to return home rather than remaining in ED hallway beds - especially as our national ED boarding crisis continues to unfold.¹¹

Let’s examine benefits in three scenarios (see figure 1 for flow diagram), where an automated algorithm identifies patients with life-limiting illness and is available to support clinician decision making. For the ED discharged cohort, patients could receive ED referrals to hospice, palliative care, or other supportive services, which may prevent admission and facilitate patient access to care that aligns with their goals. For admitted patients, automated screening could help ED clinicians or social workers start discussions with inpatient teams and lead to earlier referrals to services. For patients in the ED with uncertain disposition, the ideal scenario may involve telehealth¹² or in person ED-based palliative care consultation. An in-ED palliative care consultation should strongly be considered if the disposition could be influenced by whether the patient and caregivers feel comfortable with going home with a better plan for pain and other symptom management in place. In this case, an admission could be avoided, if the ED clinician was planning to admit the patient solely to have palliative care needs addressed. These and other models of care will require fidelity testing, evaluation of effectiveness, and measurement of sustainability.¹³

FIGURE 1. Palliative Care Screening & Referral in the Emergency Department.

Three scenarios for palliative care screening and referral in the emergency department

Still, the availability of a valid screening tool that does not expend additional clinician effort is an important new development for older patients seen in the ED. Automated screening could prompt the ED clinician to assess palliative care needs when they previously did not, it could improve appropriate identification of patients near the end of life, and could substantiate ED operation team requests to create robust palliative care consult services by showing the mismatch between demand and supply. It could also positively influence research and policy. We could better determine prevalence of older ED patients with life-limiting illness, determine the care trajectories of these newly identified patients, and examine differences in their outcomes based on ED and post-ED care, including whether formal palliative care consultation was initiated or not. A future where older patients can rely on palliative care needs being addressed in the most appropriate healthcare setting is on the horizon, and leveraging our technology tools may be a feasible next step.