Patterns and Results of Triage Advice Before Emergency Department Visits Made by Patients With Cancer

PURPOSE:

Patients with cancer undergoing treatment frequently visit the emergency department (ED) for commonly anticipated complaints (eg, pain, nausea, and vomiting). Nearly all Medicare Oncology Care Model (OCM) participants prioritized ED use reduction, and the OCM requires that patients have 24-hour telephone access to a clinician, but actual reductions in ED visits have been mixed. Little is known about the use of telephone triage for acute care.

METHODS:

We identified adults aged 18+ years newly diagnosed with cancer, linked to ED visits from a single institution within 6 months after diagnosis, and then analyzed the telephone and secure electronic messages in the preceding 24 hours. We coded interactions to classify the reason for the call, the main ED referrer, and other attempted management. We compared the acuity of patient self-referred versus clinician-referred ED visits by modeling hospitalization and ED visit severity.

RESULTS:

From 2011 to 2018, 3,247 adults made 5,371 ED visits to the university hospital and self-referred to the ED 58.5% of the time. Clinicians referred to outpatient or oncology urgent care for 10.3% of calls but referred to the ED for 61.3%. Patient self-referred ED visits were likely to be hospitalized (adjusted Odds Ratio [aOR], 0.89, 95% CI, 0.64 to 1.22) and were not more severe (aOR, 0.75, 95% CI, 0.55 to 1.02) than clinician referred.

CONCLUSION:

Although patients self-referred for six of every 10 ED visits, self-referred visits were not more severe. When patients called for advice, clinicians regularly recommended the ED. More should be done to understand barriers that patients and clinicians experience when trying to access non-ED acute care.

INTRODUCTION

Patients with cancer undergoing active treatment frequently visit the emergency department (ED) for common disease- and treatment-related complaints (eg, pain, nausea, and vomiting).^1-3 Understanding this, nearly all participants in the Centers for Medicare and Medicaid Innovation Oncology Care Model (OCM) alternative payment program prioritized ED use reduction, with many emphasizing patient access to alternative sites of acute care. Despite these efforts, actual reductions in ED visits have been mixed.⁴ The OCM mandates that participants provide 24-hour telephone access to a clinician who has access to medical records.⁵ Little is known about the frequency and quality of telephone triage for acute care.

The Harold C. Simmons Comprehensive Cancer Center (SCCC), an OCM participant, created an oncology urgent care clinic (OUCC) in May 2012 with the goal of reducing ED visit growth, although OUCC was only used by one out of every eight patients who used the ED.⁶ To understand the reasons for OUCC underuse, we analyzed the triage interactions as documented in the electronic health record (EHR) in the 24 hours preceding an ED visit. Our main hypothesis was that patients with cancer underutilize existing sick care triage advice, and that a higher level of clinical acuity would not explain why patients self-referred to the ED.

METHODS

We identified adults aged 18+ years with new cancer diagnoses from December 1, 2011, to February 28, 2018, in the SCCC at the University of Texas Southwestern Medical Center (University) tumor registry. We excluded nonmelanoma skin cancer and leukemias, the latter in line with other research⁷ that excluded acute leukemias because of inpatient treatment and hospitalizations frequently due to disease relapse, as well as differential acute care needs in the early phase of chronic leukemia. The tumor registry provided demographic details, as well as cancer type, stage, date of diagnosis, and treatment modalities. We collected other chronic medical conditions at the time of cancer diagnosis into a Charlson Comorbidity Index.⁸ We linked patients to their EHR data (EHR—Epic, Verona, WI), identified ED visits within 6 months (180 days) of diagnosis, and analyzed the telephone and secure electronic patient portal messaging interactions in the 24 hours preceding an ED visit. The EHR is integrated across the inpatient and outpatient settings and includes care records from all specialties.

ED Visits

Among patients who made ED visits within 6 months of new cancer diagnosis, we identified the date and time of arrival, whether the patient was hospitalized (observation or inpatient admission), and the Emergency Severity Index (ESI) of the ED visit.⁹ The ESI is a validated triage algorithm that generates clinically relevant stratifications of patients with high reproducibility, from one (most urgent—requires immediate life-saving intervention) to five (least urgent). The ESI was recorded in the EHR by clinical staff for each ED arrival as part of standard care.

Triage Interactions and Coding

After collecting the telephone and EHR patient portal messaging notes preceding ED visits, we excluded interactions that were unrelated to an acute complaint (eg, routine prescription refill requests and appointment rescheduling). We referred to the remaining notes as triage interactions.

From the free-text documentation of these triage interactions, two coders (A.S.H. and H.C.) abstracted information on (A) the primary complaint(s), including instances when the clinician called the patient due to a critically abnormal lab value or imaging result; (B) whether the final triage recommendation was made by the patient/caregiver, nurse, advanced practice provider (APP), or physician, and whether the patient expressed a preference for or against the ED; (C) whether the call was made during clinic hours (M-F, 8 am to 5 pm excluding holidays) or after hours. After-hour calls bypass triage nursing and are routed to an on-call physician in the relevant specialty; (D) whether any nonvisit interventions were recommended (eg, new prescriptions), and whether instructions were provided for when to present to the ED (eg, temperature > 100.4°F); and (E) whether the clinician attempted to schedule the same-day outpatient clinic or OUCC visit, and if this was successfully scheduled (OUCC was staffed by an APP, offering rapid laboratories and limited intravenous therapies, during clinic business hours). Any coding clarifications were resolved by consensus.

We organized the primary complaints of phone triage interactions and ED visits into clinical categories, which were adapted from a validated ED chief complaint text processor.¹⁰

Statistical Analysis

We generated descriptive characteristics for the patient cohort, ED visits, pre-ED clinical contacts, and the coded information from the triage interactions. We used chi-squared tests to compare patients who never called before an ED visit with patients who called at least once before an ED visit. For the most common major complaint categories, we calculated the proportion of patient ED self-referral, the proportion hospitalized, and the proportion of ED visits with the two most severe ESI levels.

We then used mixed-effect multivariate logit regression, clustering ED visits at the patient level, to model clinical severity, using hospitalization (binary) and the ESI (ordinal). Because of the low frequency of the most extreme ESI levels (n = 45 for ESI level 1 and n = 14 for ESI level 5), we collapsed the five-level scale into a three-level scale (two most emergent levels—1 and 2, an intermediate level—3, and the two least emergent levels—4 and 5). To model hospitalizations, we excluded ED visits where the patient left before being seen or left against medical advice (n = 29). To model ED visit severity, we excluded visits with no ESI recorded (n = 24).

Our prespecified primary variable of interest was the primary person who made the decision to refer to the ED: patient, clinician, both clinician and patient (unable to distinguish), and unclear (limited or no documentation on whether any ED recommendation was made). We also examined the association of the outcomes for patients who called after clinic business hours. We included an interaction term between these two main independent variables.

Multivariable models included the following as covariates: patient's age, sex, race/ethnicity, Charlson index, insurance type, and cancer type (lung, colorectal, breast, melanoma, head and neck, brain, kidney, lymphoma, prostate, ovarian-uterine-vaginal, pancreas, other gastrointestinal, unknown, and others); whether it is advanced-stage cancer (stages IIIB and higher for lung cancer, stages III and higher for pancreatic cancer, and stage IV for all others except brain cancer);^6,11 initial treatment modalities (chemotherapy, surgery, radiation therapy, and immunotherapy); whether or not outpatient management instructions were given during triage; whether or not the same-day outpatient clinic appointment was scheduled; and whether or not an OUCC visit was scheduled.

Finally, for added interpretability, we applied marginal effect analyses to generate the adjusted likelihood of hospitalization and ED severity across the different primary ED referrer categories and whether the call was made during or after clinic business hours.

Sensitivity Analysis

To test the robustness of our findings, we matched patient addresses and classified them at the census tract level (2009 American Community Survey) to identify low education (≧10% of individuals with less than high school education) and high poverty (≧25% of households below the federal poverty level) and to add validated measures that adjusted for income and education level.^12-14 We also reran analyses after reclassifying the triage interactions with limited or no documentation and those with patient and clinician agreement, with all being clinician-referred ED visits.

SAS v9.4 (SAS Institute, Cary, NC) was used for data management and STATA/MP 15.1 (StataCorp, College Station, TX) for statistical analyses. The UT Southwestern Institutional Review Board approved this study (STU 112017-026, 122017-042).

RESULTS

We identified 5,371 ED visits, with fewer than half (2,500, 46.5%) preceded by a patient call or message seeking clinical advice. Compared to patients who never called before ED visits, patients who called before any ED visit were more often non-Hispanic White (71.6% v 58.5%, P < .001), more often commercially insured (53.3% v 45.9%, P < .001), more often received chemotherapy (59.4% v 46.7%, P < .001) or surgery (56.3% v 46.5%, P < .001) as part of their initial regimen, and tended to have a higher proportion of ED visits within 6 months after diagnosis (9.7% v 3.1% had four or more ED visits, P < .001). See Table 1 for complete demographic details.

TABLE 1.

Characteristics of Patients With Newly Diagnosed Cancer Who Visited the ED Within 6 Months (180 Days) of Diagnosis, 2012-2018

Reasons for ED Visits

Figure 1 shows descriptive statistics stratified by ED visit complaint. The most frequent reasons for ED visits included pain, fever, device/line or ostomy problems, or unspecified complaint, shortness of breath, and nausea/vomiting/diarrhea/constipation—collectively representing 55.6% of the ED visits made. Pain, chest pain, shortness of breath, dehydration, malaise/fatigue, and hypotension or hypertension all had high proportions of patient self-referral (> 60%), whereas fever, nausea/vomiting/diarrhea/constipation, and bleeding had lower proportions of patient self-referral (< 30%). We also noted a number of complaints that had lower ESI severities but a much higher proportion of hospitalization (pain, nausea/vomiting/diarrhea/constipation, malaise/fatigue/failure to thrive, and edema).

Fig 1.

Most frequent emergency department (ED) primary complaints among patients with newly diagnosed cancer, 2012-2018. ESI, Emergency Severity Index.

Epidemiology of Triage Interactions

Just under half of the calls preceding an ED visit (46.5%) occurred after clinic hours. After accounting for the calls where patients were merely informing clinicians of ED arrival (5.2% of calls) hoping to expedite ED evaluation or another unspecified reason, nearly 6 in 10 of ED visits (58.5%) were patient self-referred. Of the 2,500 triage, 14.4% had limited or no documentation, which included instances when calls or messages were not returned before the ED visit.

Figure 2 summarizes the main triage interaction findings. When a patient sought sick care triage advice from a clinician (physicians 53.4% of the time, nurse 44.0% of the time, and APP 2.5% of the time), the clinician recommended going to the ED 61.3% of the time (physicians 63.7% of the time, nurses 60.4% of the time, and APP 83.7% of the time). Only 22.2% of the clinician interactions explored non-ED options (including 10.3% same-day outpatient or OUCC visit attempted or scheduled), and patients still decided to go to the ED after 10.9% of the interactions, such as when clinicians offered nonvisit management such as a new prescription. For 7.2% of the calls, clinicians did not directly refer the patient to the ED but gave parameters on what should trigger a future ED visit.

Fig 2.

Outcomes¹ of 2,500 triage interactions in the 24 hours before an emergency department (ED) visit, 2012-2018.

Factors Associated With ED Visit Severity and Hospitalization After an ED Visit

We found no significant difference in adjusted odds of hospitalization between patient self-referred and clinician-referred ED visits (aOR, 0.89, 95% CI, 0.64 to 1.22). Modeling ESI levels showed a nonsignificant difference between patient self-referred and clinician-referred ED visit severity, although there was a trend toward lower severity of patient self-referred ED visits (aOR, 0.75, 95% CI, 0.55 to 1.02). Full mixed-effect modeling of hospitalization after an ED visit and ED ESI level is given in the Appendix (online only).

If the patient initiated a call after clinic business hours, compared with calling during business hours, the adjusted odds of hospitalization were significantly lower (aOR, 0.72, 95% CI, 0.56 to 0.93), along with higher odds of high severity ED visit (aOR, 1.26, 95% CI, 1.00 to 1.60). The interaction term between after-hour call and ED referring party had only one significant but large magnitude term: when the call was made after business hours and both patient and clinician had decided on ED referral (aOR, 8.39, 95% CI, 1.03 to 68.21).

Marginal effect analysis generated nonsignificant differences in the adjusted likelihood of hospitalization based on the primary referring party, but the likelihood of hospitalization was 66.9% (95% CI, 64.2 to 69.6) for a triage interaction during clinic hours, significantly higher than a 60.2% likelihood of hospitalization (95% CI, 57.3 to 63.1) for a triage interaction after clinic hours. Similarly, referring clinician did not significantly distinguish the likelihood of the ED visit having the highest ESI severities, but an after-hour triage interaction had somewhat higher likelihood of high ESI compared with a business-hour call (49.2%, 95% CI, 46.2 to 52.2; v 43.2%, 95% CI, 40.5 to 46.0, respectively).

Older age was significantly associated with slight increase in hospitalization odds (aOR, 1.02, 95% CI, 1.01 to 1.03), although not significantly associated with higher ED visit severity.

Sensitivity Analyses

We noted other significant factors associated with hospitalization including advanced-stage cancer (aOR, 1.45, 95% CI, 1.11 to 1.88) and Black and Hispanic race/ethnicity (aOR, 0.69, 95% CI, 0.52 to 0.92, and aOR, 0.54, 95% CI, 0.38 to 0.75, respectively). However, these were no longer significant after adding low-education and high-poverty adjustments to the model. Similarly, while interactions where both the patient and the clinician agreed on ED visit overall and during an after-hour call (in the interaction term) were associated with significant differences in odds of hospitalization (aOR, 0.32, 95% CI, 0.11 to 0.87 and aOR, 8.39, 95% CI, 1.03 to 68.21, respectively), these results were not robust to additional adjustments for poverty and low education. Low education and high poverty had no independently significant association with hospitalization, and their addition had no substantive impact on our two predictors of interest. Reclassifying all triage interactions with limited or no documentation, as well as those with patient and clinician agreement, as clinician-referred ED visits did not substantively impact our modeling results. See the Appendix for full model and sensitivity analysis results.

DISCUSSION

Our study presents several important findings on the epidemiology of pre-ED visit management and triage of acute complaints of adults newly diagnosed with cancer. First, ED visits to the cancer-treating hospital showed substantial patient self-referral without the patient contacting their cancer treatment team, and a small proportion of patient calls were merely to inform the treatment team of ED arrival, amounting to a true patient self-referral rate of 58.5%. Second, patient self-referred ED visits did not appear to be of higher clinical severity and were no more likely to be hospitalized than those where the clinician referred the patient to the ED. These results suggest, at a minimum, that patients may be overestimating their need for the ED and hospital-based care. Since patients and caregivers in general have the least clinical training of everyone involved in their care, this seems unsurprising. However, given this second result, it is concerning that very few patients reach out for advice in the first place, despite the availability of 24-hour phone access to a clinician who can review their medical records. Whether this underuse is due to poor prior experiences or simply being unaware of the service should be further investigated.

It is notable that clinicians also referred patients to the ED more than half of the time (61.3% of interactions) and infrequently attempted to access alternative sites of care (10.3% of interactions). Clinicians may have their own sets of barriers, concerns, or lack of awareness to using the same-day clinic or cancer urgent care visits. For clinicians and for patients, the role of video visits to help assess patient acuity should be investigated as well, given rapid recent increases in its use.

Finally, it is interesting that patient calls triaged by an on-call physician after clinic hours had significantly higher ED visit severity but lower odds of hospitalization. Patients may have called after hours for issues that were more emergent rather than waiting until the next day, yet these may also be acute conditions that are readily managed in the ED. This may at least be partially explained by chest pain ED visits, which seemed to have higher ESI severity (72.6% of visits were in top two ESI levels) than proportion hospitalized (53.4%).

Efforts at cancer care delivery innovation have demonstrated that patient-centered redesign of care,¹⁵ including a robust acute care triage system, can help reduce unplanned ED care.¹⁶ In turn, OCM has spurred a broader range of providers to focus on establishing alternate sites of acute care such as the same-day outpatient clinic and OUCC.^4,17-21 Patients will invariably experience illness that requires ED-level care, and it is critical to ensure that triage protocols are safe and reliable. Yet, the key point of entry to these ED alternatives is appropriate triage, largely managed through the telephone, which depends on patients or caregivers to initiate the call.

Our results suggest that patients and caregivers underused available clinician advice and had difficulty estimating the need for ED and hospital-level care. It seems more likely that this is due to patients and caregivers assuming that nothing could be done short of ED treatment, but this suggests under-recognition of acute care options and merits future investigation. Future work should also explore how education and outreach for patients and clinicians could impact ED use by increasing the use of triage and ED alternatives.

The variable rates of patient self-referral according to primary complaint suggest specific services that could be targeted for improved referral to OUCCs by patients and clinicians or even services that could be appropriately added to the scope of care.

Our findings should be interpreted with some limitations. Chiefly, not all patient contacts with clinicians may have been recorded in the EHR. Patients may call their physicians on personal phones, and these encounters may not be systematically documented in the EHR. Patients who do not receive primary or specialty care at our institution may have triage interactions that were not available to us. We also relied on what was documented of the triage interaction, not a direct recording. However, these missed instances of clinician triage are expected to be very infrequent, and the documented details would be expected to be reflective of the interaction. We also noted that some interactions documented advice from unaffiliated clinicians.

We also note that a number of after-hour triage interactions lacked enough documentation on the nature of the discussion and advice. However, when we ran a sensitivity analysis making the most conservative assumption that all these encounters had the clinician referring the patient to the ED, our results were not substantively different.

Finally, we did not capture triage decisions leading to ED visits to other hospitals. It is not clear whether these visits would systematically higher or lower in clinical severity, but this should be a direction of future study.

Our results highlight the range of triage advice sought out by patients newly diagnosed with cancer, as well as the difficulty patients, caregivers, and clinicians face when deciding to use the ED. Health systems, particularly the growing number in the Medicare OCM focused on improving the overall value of cancer care delivery, should incorporate this information to optimize the provision of clinician advice and alternative sites of acute care.

Appendix

TABLE A1.

Full Mixed-Effects Modeling Results With SA—and Marginal Effects Analysis

AUTHOR CONTRIBUTIONS

Conception and design: Arthur S. Hong, Simon J. Craddock Lee

Financial support: Ethan A. Halm

Administrative support: Ethan A. Halm

Collection and assembly of data: Arthur S. Hong, Hannah Chang, Hannah Fullington

Data analysis and interpretation: Arthur S. Hong, D. Mark Courtney, Simon J. Craddock Lee, John W. Sweetenham, Ethan A. Halm

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Patterns and Results of Triage Advice Before Emergency Department Visits Made by Patients With Cancer

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I =Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/op/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).