Characteristics of Emergency Department Visits and Select Predictors of Hospitalization for Adults With Newly Diagnosed Cancer in a Safety-Net Health System

Abstract

PURPOSE:

There is little description of emergency department (ED) visits and subsequent hospitalizations among a safety-net cancer population. We characterized patterns of ED visits and explored nonclinical predictors of subsequent hospitalization, including time of ED arrival.

PATIENTS AND METHODS:

This was a retrospective cohort study of patients with cancer (excluding leukemia and nonmelanoma skin cancer) between 2012 and 2016 at a large county urban safety-net health system. We identified ED visits occurring within 180 days after a cancer diagnosis, along with subsequent hospitalizations (observation stay or inpatient admission). We used mixed-effects multivariable logistic regression to model hospitalization at ED disposition, accounting for variability across patients and emergency physicians.

RESULTS:

The 9,050 adults with cancer were 77.2% nonwhite and 55.0% female. Nearly one-quarter (24.7%) of patients had advanced-stage cancer at diagnosis, and 9.7% died within 180 days of diagnosis. These patients accrued 11,282 ED visits within 180 days of diagnosis. Most patients had at least one ED visit (57.7%); half (49.9%) occurred during business hours (Monday through Friday, 8:00 am to 4:59 pm), and half (50.4%) resulted in hospitalization. More than half (57.5%) of ED visits were for complaints that included: pain/headache, nausea/vomiting/dehydration, fever, swelling, shortness of breath/cough, and medication refill. Patients were most often discharged home when they arrived between 8:00 am and 11:59 am (adjusted odds ratio for hospitalization, 0.69; 95% CI, 0.56 to 0.84).

CONCLUSION:

ED visits are common among safety-net patients with newly diagnosed cancer, and hospitalizations may be influenced by nonclinical factors. The majority of ED visits made by adults with newly diagnosed cancer in a safety-net health system could potentially be routed to an alternate site of care, such as a cancer urgent care clinic.

INTRODUCTION

Adults with cancer generate high rates of emergency department (ED) visits, leading to hospitalization roughly 60% of the time—nearly four times the rate of the general population.^1,2 As a result, among Medicare enrollees, acute hospital care accounts for nearly half of all expenditures in the first 6 months after a diagnosis of advanced cancer.³ Yet for some cancers, oncologists estimate that nearly 20% of hospitalizations may be avoidable.⁴ Beyond health system expenditures and exposure to the harms of hospitalization, ED visits add to the out-of-pocket financial burden to the patient,⁵ and financial insolvency during cancer treatment may contribute to early mortality.⁶

Descriptions of several single-institution,^7-10 one statewide,² and one nationally representative administrative data set study¹ generally outline a high burden of ED visits for a commonly anticipated series of complaints (nausea/vomiting/dehydration, pain, shortness of breath/cough, and fever). Less is known about the frequency and preventability of ED visits and hospitalizations for adults with newly diagnosed cancer in safety-net populations. Prior studies have not used granular clinical details or examined nonclinical predictors of hospitalizations.

The purpose of this study was to describe the rates, times, and reasons for ED visits and identify predictors of subsequent hospitalization in a population-based safety-net health system. We did so with a particular eye toward identifying visits that could potentially be routed to an alternative site of care, such as a dedicated urgent care clinic for patients with cancer. We explored how often ED visits were for the commonly anticipated complaints related to cancer treatment and how often ED visits were initiated during business hours. We specifically examined whether time of arrival influenced the likelihood of hospitalization. If even these patients with higher clinical acuity have hospitalizations for nonclinical reasons, it could suggest broader health system–level effects on the delivery of cancer care.

METHODS

Population

We identified adults age 18 years and older in the cancer registry at the Parkland Health & Hospital System (Parkland), diagnosed between January 1, 2012 and December 31, 2016. Parkland is a population-based regional integrated safety-net health system in Dallas, Texas, comprising a 900-bed hospital and 12 community-based primary care and specialty clinics. Parkland is the sole safety-net provider of specialty care for the under- and uninsured in Dallas County. Low-income, uninsured residents are eligible for county-funded medical assistance, providing access to primary and specialty care, including diagnostic testing, evaluation, and treatment. All outpatient and inpatient care in the Parkland system is captured in its comprehensive electronic health record (EHR).

We excluded patients with only nonmelanoma skin cancer or leukemia. Patients with leukemia were excluded because of widely varying treatment and symptom patterns depending on unmeasured changes in acuity after diagnosis. We identified cancer type and categorized stage at diagnosis into advanced (stage IIIb and higher for lung cancer, stage III and higher for pancreatic, stage IV for all others except for brain) versus nonadvanced. The cancer registry also provided age, zip code, race/ethnicity, sex, comorbidities at diagnosis (organized into Charlson comorbidity index^11-13), and initial cancer treatment modalities (chemotherapy, radiation therapy, surgery, and/or immunotherapy).

ED Visits

After linking cancer registry patients to the EHR, we obtained dates and times of ED visits, primary presenting complaint, initial vital signs (temperature, heart rate, systolic blood pressure, respiratory rate), individual emergency physician seen, and subsequent disposition status (left against medical advice, death in the ED, observation stay, inpatient admission, or discharge home). We obtained primary payer at the time of ED visit. For patients with no ED visits in the 180 days after diagnosis, we assigned the modal payer for outpatient encounters occurring 4 months before and after cancer diagnosis.

We classified ED visits by days since diagnosis, hour of ED arrival, and whether the visit began during or after business hours (8:00 am to 4:59 pm, Monday through Friday). We separately grouped ED arrival times into 4-hour blocks.

From the EHR, we organized the primary presenting complaint for each ED visit into the following categories: nausea/vomiting/diarrhea/dehydration, fever/chills, pain/headache, extremity swelling/edema, medication refill, rash, shortness of breath/cough, and chest pain. Excluding chest pain, these low- to moderate-acuity complaints are frequently within the scope of the several oncology urgent care clinics described in the literature.^14-16 We based these textual categorizations on a previously validated ED chief complaint text processor.¹⁷ We also identified the presence or absence of abnormal initial vital signs (temperature < 96.8°F or > 100.4°F; heart rate > 90 beats per minute; systolic blood pressure < 90 or > 180 millimeters of mercury (mm Hg); respiratory rate > 20 breaths per minute) and whether the patient was referred to the ED from oncology clinic or another clinic.

Outcomes and Statistical Analysis

Our first outcome of interest was the rate of ED visits in the first 180 days after cancer diagnosis (January 1, 2012 to June 30, 2017). To account for patients who died within 180 days after diagnosis, we generated patient-months as the denominator. Every partial 30-day period the patient remained alive amounted to 1 patient-month, up to a maximum of 6 patient-months per patient. We used descriptive statistics to report ED visits per 1,000 patient-months, and stratified ED visits across a number of patient- and visit-level characteristics, including ED arrival time.

The second outcome of interest was the proportion of ED visits that resulted in a hospitalization (observation stay or inpatient admission). We excluded ED visits from the denominator if the patient died during the visit, left against medical advice, or left before being seen by a physician.

Next, we assigned remaining ED visits to a unique emergency physician identifier and tabulated the total number of ED visits per physician. On the basis of the literature, we expected substantial variation in hospitalization rates according to the individual physician seen.^18,19 Therefore, we described the unadjusted hospitalization rate at the physician level and anticipated that modeled subsequent hospitalizations would require adjustment for the correlation among hospitalization decisions generated by the same physician.

We used multilevel mixed-effects logistic regression modeling to identify predictors for hospitalization after ED visits, accounting for clustering of ED visits among patients and patients among emergency physicians. The prespecified primary predictors of interest were the time of ED arrival, presence of abnormal initial vital signs, and whether the patient was referred to the ED from the oncology outpatient clinic. These predictors were selected to contextualize a major clinical predictor of hospitalization (abnormal vital signs) with two nonclinical factors (time of ED arrival, oncology referral to ED) suggested to affect hospitalization decisions.^18,20

The multivariable model for hospitalization also adjusted for: patient sex, age, year of ED visit, Charlson comorbidity index at the time of cancer diagnosis, cancer type, whether advanced-stage cancer, and initial cancer treatment modalities. SAS 9.4 (SAS Institute, Cary, NC) and STATA/MP 15.1 STATA, College Station, TX) were used for statistical analyses.

Sensitivity Analyses

We performed a sensitivity analysis by raising the threshold for abnormally low systolic blood pressure from 90 mm Hg to 100 mm Hg, to determine whether this changed the proportion of ED visits with abnormal vital signs and its impact on modeling subsequent hospitalizations. We also performed additional analyses for the multivariable model for hospitalization, adding an interaction term between abnormal vital signs and ED arrival time and an interaction term between ED arrival time and whether weekend or not.

Finally, to explore a cancer treatment team’s ability to divert ED visits, we identified ED visits occurring after diagnosis but before first contact with an oncologist or surgeon, excluding these when predicting the odds of hospitalization. The University of Texas Southwestern Medical Center institutional review board approved the study protocol (STU 112017-026).

RESULTS

We identified 9,050 adult patients diagnosed with cancers of interest between January 1, 2012 and December 31, 2016 (Table 1); 55.0% were female, and the median age was 56 years (interquartile range, 47 to 63 years). The cohort was 77.2% nonwhite, and most (79.8%) had county medical assistance or Medicaid as their primary payer. Nearly one-quarter of the patients (24.7%) had advanced-stage cancer at the time of diagnosis, and 9.7% died within 180 days of diagnosis. Table 1 lists full patient cohort characteristics, including breakdown by major cancer types and initial treatment modalities.

TABLE 1.

Characteristics of Adults With Newly Diagnosed Cancer at a Safety-Net Health System (N = 9,050)

This cohort of patients generated 11,282 ED visits within 180 days after their cancer diagnosis. Most patients (57.7%) had at least one ED visit. Nearly one-third (31.2%) had two or more ED visits. Overall, patients accrued ED visits at a rate of 227 visits per 1,000 patient-months (Table 2). Nearly six in 10 (57.5%) of these ED visits were for primary presenting complaints that included pain/headache, nausea/vomiting/diarrhea, shortness of breath/cough, swelling, rash, fever, and medication refills. Within the subset of ED visits generated by an oncology outpatient clinic referral, 34.7% were for these same primary complaints (results not shown).

TABLE 2.

Patterns and Characteristics of ED Visits Made by Adults With Newly Diagnosed Cancer at a Safety-Net Health System (N = 11,282)

Almost half of all ED visits (48.9%) recorded abnormal initial vital signs, and half (49.9%) of ED visits were initiated between 8:00 am and 4:59 pm, Monday through Friday (Table 2). After excluding 11 patients who died during the ED visit, and 477 visits where the patient left before being seen or against medical advice, 50.4% of ED visits led to a hospitalization (inpatient or observation stay). Table 2 lists the full characteristics for the ED visits, including disposition.

In unadjusted analyses, a similar proportion of patients were hospitalized whether they arrived to the ED during or after business hours (47.1% and 49.2%, respectively). We noted variation in hospitalization rate at the ED physician level, finding a median hospitalization rate of 50% that ranged from 11% to 70% at the 10th and 90th percentiles, respectively. This distribution did not substantively change even when we limited to physicians who had seen at least four, six, or eight ED visits (results not shown; available on request). The unadjusted physician-level hospitalization rate remained similar even when limited to business hours ED arrivals or for visits with complaints of pain/headache, nausea/vomiting/diarrhea, fever, edema, cough/shortness of breath, rash, and medication refill (Table 2).

Predictors of Hospitalization

We report prespecified predictors of interest for hospitalization in Table 3. In this fully adjusted mixed-effects model for hospitalization, patients arriving to the ED between 4:00 am and 7:59 am or between 8:00 am and 11:59 am had the lowest odds of hospitalization, at 0.69 (95% CI, 0.53 to 0.90) and 0.69 (95% CI, 0.56 to 0.84) decreased odds of hospitalization when compared with ED arrival between 8:00 pm and 11:59 pm. The remainder of ED arrival times did not have significantly different odds of hospitalization compared with 8:00 pm to 11:59 pm.

TABLE 3.

Mixed-Effects Multivariable Model for Hospitalization After ED Visit

We also stratified arrival time results on the basis of whether the visit occurred on weekdays or weekends. Results were not substantively different on weekdays, and even on weekends, patients arriving between 8:00 am and 11:59 am had 0.55 lower odds (95% CI, 0.32 to 0.95) of hospitalization.

Our clinical predictors all reported higher odds of hospitalization: abnormal initial ED vital signs had 1.79 higher odds (95% CI, 1.55 to 2.06), ED referral from oncology clinic had 1.74 higher odds (95% CI, 1.29 to 2.35), and patients with advanced-stage cancer had more than twice the adjusted odds of hospitalization (2.24; 95% CI, 1.83 to 2.75). Full modeling results including covariate are available in Appendix Table A1 (online only).

Sensitivity analysis of raising the threshold of abnormally low systolic blood pressure from less than 90 mm Hg to less than 100 mm Hg increased the odds of hospitalization from 1.79 to 1.99; adding an interaction term between ED arrival time and abnormal vital signs yielded no significant interaction terms and decreased the abnormal vital signs odds to 1.70. Adding an interaction term between time of ED arrival and whether weekend or weekday visit yielded no significant interaction terms and only slightly affected ED arrival time odds (0.69 to 0.58 for 4:00 am to 7:59 am arrivals and 0.69 to 0.74 for 8:00 am to 11:59 am arrivals). Finally, we identified that 10.6% of ED visits (1,196 out of 11,282) occurred after the date of diagnosis but before first contact with a surgeon or oncologist; even after excluding these ED visits, the adjusted odds ratios of our primary predictors were not substantively changed (4:00 am to 7:59 am ED arrival: 0.68; 8:00 am to 11:59 am ED arrival: 0.73; abnormal vital signs in ED: 1.82; referred from oncology clinic: 2.03; advanced-stage cancer: 2.00).

DISCUSSION

In this population-based study of 9,050 adults with active cancer in a safety-net health system, we found that most patients (57.7%) had at least one ED visit within 180 days of diagnosis, at a rate of 227 visits per 1,000 patient-months. Half of patients arrived during business hours, slightly less than half of patients had an abnormal vital sign on ED arrival, and almost six in 10 visits were for low- to moderate-acuity primary presenting complaints that are common among patients newly diagnosed with cancer.

Our mixed-effects logistic modeling of subsequent hospitalizations found that patients who arrived to the ED in the mornings (8:00 am to 11:59 am) were consistently more likely to be discharged home, even after adjusting for key clinical factors. This adjusted model also found more than two times higher odds of hospitalization for patients who had advanced-stage cancer at the time of diagnosis, and approximately 1.8 higher odds of hospitalization if patients were referred to the ED from the oncology clinic or had abnormal initial vital signs in the ED.

Overall, our findings confirm and extend reports in the literature showing the predominance of ED visits during business hours,² as well as high rates of hospitalization after an ED visit.^1,2 Although few prior studies limited ED visits to those occurring 180 days after cancer diagnosis, making direct comparisons of ED visit rates difficult, we found similar patterns of ED visit timing and hospitalization across patients with a range of payer types.

Furthermore, we noted considerable variation in hospitalization rates across individual emergency physicians. Although physician-level variation is well described across many different medical decisions,^21-24 it was notable to observe marked physician-level variation in hospitalization rate among vulnerable patients with cancer who had an overall hospitalization rate of nearly 50%.

Several hospital systems have developed dedicated oncology urgent care clinics that have a scope of practice that includes pain (excluding chest pain), nausea/vomiting/dehydration, shortness of breath/cough, fever, and rash.^14-16 Including the 2.2% of visits that were for medication refills, our findings suggest that more than half of ED visits in a safety-net population might be for such urgent care–sensitive cancer conditions. Consistent with other recent studies, there may be substantial opportunity to treat²⁵ these conditions in a less-expensive and more patient-centric site of care.

At the same time, the primary reason for presentation in the EHR may not be the only factor to determine suitability for an urgent care setting. It is also not clear how often a cancer urgent care clinic definitively manages these complaints to avoid an ED referral. Nevertheless, compared to a comprehensive redesign of cancer care delivery, our findings suggest that narrowly targeting nonemergent acute care may be a fruitful pathway for safety-net health systems seeking to reduce ED visit rates. This may be most relevant for practices participating in one of several oncology value-based payment demonstration programs launched by the Center for Medicare & Medicaid Innovation, such as the Oncology Care Model.^26,27

Downstream from the ED visit, our hospitalization modeling results suggest that there are likely nonclinical factors affecting physician decisions to hospitalize patients, including the time of day of ED arrival. Although patients might present with lower-acuity complaints during daytime hours, our findings persisted after accounting for this potential explanation. This may reflect other sources of variation in hospitalization rates, such as ED waiting times or shift-related cycles of decision making²⁸ and hospital bed availability. The literature outlines several more specific nonclinical influences, such as differences in physician-level hospitalization rates according to personal risk aversion²² and system-level factors such as patient volume during a shift.²⁹ It is unclear to us why patients arriving in the morning were more likely to be discharged home, even on weekends, and this deserves additional quantitative and qualitative study.

The major strengths of our study include the ability to accurately identify important clinical details such as cancer type, stage, comorbidities, and days since cancer diagnosis. Furthermore, we were able to classify ED visits by primary presenting complaint, which is quite different from categorizing by the diagnosis codes generated after the end of the ED visit.³⁰ Finally, this study takes advantage of EHR-specific data at the ED visit, including hour of arrival, vital signs, and individual emergency physician seen, to explore important clinical and nonclinical predictors of hospitalization.

Our findings should be interpreted in light of certain limitations. Although our patient cohort was population based and spans 5 years, it represents the experience of one safety-net health system, so the generalizability of our findings is unknown. However, there are similar systems in nearly every metropolitan area, and the care of vulnerable patients with cancer is under-reported in the literature.

Relatedly, by relying on EHR data, our description of ED visits and hospitalizations does not include visits made to other health systems. As a result, our findings may underestimate the frequency of ED visits and hospitalizations, with unclear implications for the predictors of hospitalization we examined.

In addition, there may be unmeasured confounders in our regression analysis predicting hospitalization, such as whether the patient is homeless, the strength of social supports at home, and how each emergency physician assesses these factors. Although we do not identify these factors specifically, we do account for correlations in the likelihood of hospitalization using a random patient effect in our modeling and nesting within individual emergency physician.

Adults with newly diagnosed cancer in a safety-net health system commonly seek acute care through the ED. The majority of visits are for commonly expected reasons and might feasibly be treated in a different setting, such as a cancer urgent care clinic. This is particularly relevant for health systems responsible for total expenditures, including safety-net health systems and those participating in alternative payment demonstration projects. Cancer care delivery research should explore whether the identified lower-acuity ED visits could safely be addressed in an alternate setting, while reducing the overall number of ED visits. Such work could more directly identify potential urgent care–sensitive cancer conditions. Research should also seek to further understand hospitalization decision making at the individual physician level. Understanding nonclinical predictors of hospitalization, such as time of day of arrival to the ED, may clarify avenues to reduce unplanned hospitalizations.

Appendix

TABLE A1.

Mixed-Effects Multivariable Model for Hospitalization After ED Visit

AUTHOR CONTRIBUTIONS

Conception and design: All authors

Financial support: Ethan A. Halm

Administrative support: Valorie Harvey

Collection and assembly of data: Arthur S. Hong

Data analysis and interpretation: Arthur S. Hong, Navid Sadeghi, Simon Craddock Lee, 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

Characteristics of Emergency Department Visits and Select Predictors of Hospitalization for Adults with Newly Diagnosed Cancer in a Safety-Net Health System

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. 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/jop/site/ifc/journal-policies.html.

Arthur S. Hong

Consulting or Advisory Role: Janssen (I), AbbVie (I), Celgene (I)

Travel, Accommodations, Expenses: Janssen (I), Celgene (I), AbbVie (I)

No other potential conflicts of interest were reported.