Opioid-Related Emergency Department Encounters: Patient, Encounter, and Community Characteristics associated with Repeat Encounters

Abstract

Objectives.

We described the prevalence, trends, and factors associated with repeated emergency department (ED) encounters for opioid usage across multiple, independent hospital systems.

Methods.

A statewide regional health information exchange system provided ED encounters from 4 Indiana hospital systems for 2012 to 2017. In accordance with a series of International Classification of Diseases, Ninth Revision (ICD-9) and ICD-10 diagnosis codes for opioid abuse, adverse effects of opioids, opioid dependence and unspecified use, and opioid poisoning, we identified patients with an ED encounter associated with opioid usage (9,295 individuals; 12,642 encounters). Multivariate logistic regression models then described the patient, encounter, prescription history, and community characteristics associated with the odds of a patient’s incurring a subsequent opioid-related ED encounter.

Results.

The prevalence of repeated nonfirst opioid-related ED encounters increased from 9.0% of all opioid encounters in 2012 to 34.3% in 2017. The number of previous opioid-related ED encounters, unique institutions at which the individual had had encounters, the encounter’s being heroin-related, the individual’s having a benzodiazepine prescription filled within 30 days before the encounter, and being either Medicaid insured or uninsured compared with private insurance were associated with significantly greater odds of having a subsequent encounter.

Conclusions.

The ED is increasingly a site utilized as the setting for repeated opioid-related care. Characteristics of the individual, encounter, and community associated with repeated opioid-related encounters may inform real-time risk-prediction tools in the ED setting. Additionally, the number of institutions to which the individual has presented may suggest the utility of health information exchange data and usage in the ED setting.

Introduction

The emergency department (ED) is at the forefront of care in the opioid epidemic, with a 5.6% increase in opioid-related encounter rates quarterly from 2016 to 2017.¹ Among individuals visiting the ED for opioid-related reasons are a subset of patients who are repeated utilizers of ED services, defined as patients who experience initial and subsequent opioid-related (e.g., opioid withdrawal, opioid poisoning and overdose) encounters in a given period. For providers, repeated treatment of patients with substance use disorder can be a source of frustration.² Moreover, because these are preventable events, repeated utilization reflects missed opportunities of engaging patients in ongoing care, patient relapse, or treatment refusal.

Unfortunately, our understanding of factors associated with repeated encounters is scant and of limited applicability today. The existing literature on repeated encounters provides limited insight about how visits by repeating misusers differ, if at all, from general opioid-related encounters.^3–5 In addition, existing research on repeated opioid misuse ED visits predates both the implementation of expanded opioid-related diagnosis codes of International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10)^6,7 and critical changes in treatment and policies responding to the growing epidemic. For example, existing literature examined repeated visits before the widespread availability of naloxone,⁸ growth in prescription drug monitoring programs, and the general awareness of opioid prescribing rates.⁹ Other limitations include a reliance on claims data, which excludes the uninsured, including only Medicaid or commercially insured individuals, and cohorts that followed patients for short durations.

The objectives of this study are to describe trends in repeated opioid-related ED encounters; compare patient, encounter, and community characteristics between individuals with one encounter and those with multiple encounters; and identify characteristics associated with subsequent encounters. We address existing limitations by including a longer, updated study period, more recent data, and a more general sample. Findings from this study will help contribute to the increasing understanding of the opioid epidemic and provide insight into potentially relevant factors for predicting risk and preventing repeated visits.

Methods

Study Design

This retrospective cohort study will utilize an encounter-level analysis of a 6-year cohort of Indiana adults with ED encounters.

Sample & Setting

Our sample included all ED encounters by Indiana adults at four health systems in Indiana between 2012 and 2017. All index ED encounters in the sample had at least 9 months of post observation time according to the mean time between encounters for individuals with multiple encounters. Encounters in which an individual died during the admission were excluded from consideration as an index ED encounter.

Data Collection and Processing

This study used health information exchange data from the Indiana Network for Patient Care. The network compiles electronic health record data from various institutions across the state into a standardized structure. The data used included all encounter-level data occurring within 4 hospital systems in Indiana and linked encounters for patients across different settings and systems.¹⁰ The Indiana Network for Patient Care, supplemented with Surescripts, provided all clinical, encounter, demographic, and prescription fill data. The United States Census Bureau’s American Fact Finder 2012 to 2016 5-year averages were used for community-level characteristics. These were matched at the patient zip code level.

Our dependent variables included an individual’s having at least one opioid-related encounter and the occurrence of a repeated opioid-related encounter for that individual during the study period. Opioid-related encounters were defined by the Healthcare Costs and Utilization Project’s list of ICD-9/10 codes across any of the diagnosis fields in the data.⁷ This list includes 122 codes within 4 categories of opioid-related codes: “opioid abuse,” “adverse effects of opioids,” “opioid dependence and unspecified use,” and “opioid poisoning.”

Risk factors included patient demographic characteristics, encounter characteristics, prescription history, and community characteristics. Patient characteristics included sex; age; race; insurance status; and the number of previous opioid-related ED encounters before the current encounter, any ED encounters before the current encounter, and unique institutions the individual had visited for any ED encounter, including the current encounter. We described the encounter by measures of day/night, weekend/weekday, season, year, institution, disposition at the end of the encounter, having a heroin diagnosis at the encounter, and Elixhauser comorbidity score.¹¹ Prescription data included the individual’s having an opioid or benzodiazepine prescription filled within 30 days before the encounter. Benzodiazepines were included because of their known interaction with opioids in ED encounters and overdoses.¹² Community-level covariates included total population in thousands, median household income in thousands of dollars, and percentage of the population that was white, uninsured, unemployed, and older than 25 years, with less than a high school degree.

Primary Data Analysis

We first describe trends in opioid-related ED encounters and repeat encounters between 2012–2017. Descriptive statistics were also used to compare frequencies and percentages of individuals with only one opioid-related encounter during the study period with those with multiple encounters.

We modeled patient and encounter characteristics associated with an individual’s having a subsequent ED encounter, using an encounter-level multivariable logistic regression with clustered standard errors at the patient level to account for repeated observations. Independent variables included patient, encounter, prescription, and community-level characteristics described earlier. The same time restriction of the encounter was used according to the mean time between encounters for patients with multiple encounters. In addition, encounters in which an individual died were removed from this model because they were ineligible for having a future encounter (n=14).

A series of 5 analyses was run as sensitivity tests. The first excluded all encounters from 2017 to allow a longer period for a subsequent encounter to occur and thus provides more conservative estimates. The second used community characteristics as quartiles because there were many observations within each zip code and therefore with the same community characteristics. The third sensitivity analysis used an individual-level negative binomial approach to model the count of repeated encounters. An individual who had only one opioid-related encounter therefore had zero repeated encounters. This model considered the individual and encounter characteristics at the first encounter to model the count of repeated encounters. To accommodate the differences in follow-up period for individuals within the study, we conducted an Andersen-Gill recurrent event survival analysis as the fourth sensitivity analysis. Finally, because our sample size was limited by missing data primarily in the insurance and disposition fields, we ran a sensitivity analysis in which observations with these variables missing were included as an additional category of “missing” to ensure that our findings were not being driven by this missingness.

This study was approved by the Indiana University Institutional Review Board. All analyses were conducted in Stata (version 13.1; StataCorp, College Station, TX).

Results

Between 2012–2017, there were 12,642 opioid-related ED encounters across 9,295 patients. Nearly one-quarter (22.1%) of patients had 2 or more opioid-related encounters (mean days between encounters = 272). Both the number of opioid-related ED encounters and the proportion that were repeated encounters increased over time (Figure 1). In 2012, 9.0% (95% confidence interval [CI] 7.1% to 11.4%) of opioid-related ED encounters were repeated encounters; by 2017, 34.3% (95% CI 32.7% to 36.0%) were repeated encounters.

Figure 1.

Trends & prevalence of opioid and subsequent opioid-related ED encounters

Comparisons of characteristics across individuals with multiple opioid-related encounters during the study period with those with only one were conducted with frequencies and percentages (Table 1). These comparisons included only individuals with complete data (n= 5,129) and considered the encounter-level characteristics of the first opioid-related encounter the individual had during the study period (the index encounter). First encounter and individual characteristics for patients with one encounter compared with those with multiple encounters differed by encounter year, race, having an opioid dispensed before the encounter, having a heroin-related diagnosis code, insurance status, and all community-level characteristics. Age also differed significantly, with a mean age of 33.6 years (95% CI 33.0 to 34.2 years) for patients with multiple encounters compared with 37.3 years (95% CI 36.9 to 37.7 years) for those with a single encounter.

Table 1.

Individual & Encounter characteristics of individuals with single compared to multiple opioid-related encounters

	Patient had multiple encounters		Patient had single encounter		Total
	n	%	n	%	n	%
Encounter Year
2012	149	11.7	477	12.4	626	12.2
2013	144	11.3	368	9.5	512	10.0
2014	126	9.9	278	7.2	404	7.9
2015	278	21.8	608	15.8	886	17.3
2016	475	37.2	1577	40.9	2052	40.0
2017	105	8.2	550	14.3	655	12.8
Season
winter	342	26.8	1157	30.0	1499	29.2
spring	310	24.3	846	21.9	1156	22.5
summer	272	21.3	827	21.4	1099	21.4
fall	353	27.6	1028	26.6	1381	26.9
Time of Day
nighttime	629	49.3	1872	48.5	2501	48.7
daytime	648	50.7	1986	51.5	2634	51.3
Day of Week
weekday	934	73.1	2816	73.0	3750	73.0
weekend	343	26.9	1042	27.0	1385	27.0
Disposition
visit ended normally, discharged home	1163	91.1	3507	90.9	4670	90.9
left AMA	32	2.5	85	2.2	117	2.3
transferred to discharged to another facility/department	12	0.9	78	2.0	90	1.8
other	70	5.5	188	4.9	258	5.0
Gender
female	587	46.0	1733	44.9	2320	45.2
male	690	54.0	2125	55.1	2815	54.8
Race
white	1039	81.4	2819	73.1	3858	75.1
black or african american	108	8.5	374	9.7	482	9.4
all other	130	10.2	665	17.2	795	15.5
Opioid Rx before enct
no fill	1211	94.8	3558	92.2	4769	92.9
filled	66	5.2	300	7.8	366	7.1
Benzodiazepine Rx before enct
no fill	1239	97.0	3758	97.4	4997	97.3
filled	38	3.0	100	2.6	138	2.7
Opioid and Benzodiazepine Rx before enct
One or no fills	1254	98.2	3787	98.2	5041	98.2
both filled	23	1.8	71	1.8	94	1.8
Elixhauser Level
0	703	55.1	1965	50.9	2668	52
1	486	38.1	1580	41	2066	40.2
2	79	6.2	272	7.1	351	6.8
3+	9	0.7	41	1.1	50	1.0
Heroin Diagnosis
no heroin diagnosis	808	63.3	2859	74.1	3667	71.4
heroin diagnosis	469	36.7	999	25.9	1468	28.6
Insurance Status
Private/employer-based	179	14.0	658	17.1	837	16.3
Medicaid	655	51.3	1789	46.4	2444	47.6
Medicare	83	6.5	495	12.8	578	11.3
self-pay	334	26.2	856	22.2	1190	23.2
other	26	2.0	60	1.6	86	1.7

According to the main logistic regression model, if the patient at the index ED encounter had a benzodiazepine prescription filled (odds ratio [OR] 1.91; 95% CI 1.17 to 3.10) or the encounter was heroin related (OR 1.44; 95% CI 1.26 to 1.65), then the odds of a subsequent ED encounter were greater (Table 2). Being insured through Medicaid (OR 1.49; 95% CI 1.24 to 1.78) or uninsured (OR 1.42; 95% CI 1.15 to 1.77) was also associated with greater odds of having a subsequent encounter. The number of previous opioid-related ED encounters was associated with a higher odds of having subsequent encounters (OR 1.47; 95% CI 1.39 to 1.55), as was the number of unique institutions at which individuals had experienced an ED encounter of any type, including for that encounter (OR 1.17; 95% CI 1.10 to 1.25). The odds of a subsequent encounter were lower if the index encounter ended with a transfer as opposed to a routine discharge home (OR 0.53; 95% CI 0.32 to 0.86) and if the patient was a nonblack minority compared with white (OR 0.67; 95% CI 0.55 to 0.81). Odds of incurring a subsequent encounter were lower for each age group compared with the youngest age group, 18 to 24 years. In addition, the magnitude of the odds ratio decreased for each increasing age group but one (e.g., OR for 25 to 34 years compared with < 25 years 0.76, 95% CI 0.65 to 0.88; OR for 35 to 44 years compared with < 25 years 0.58, 95% CI 0.49 to 0.70), suggesting that the odds of a subsequent encounter decrease consistently across age categories. At the zip code level, percentage of the community uninsured (OR 0.97; 95% CI 0.95 to 1.0002) and percentage white (OR 0.99; 95% CI 0.99 to 0.99) were associated with significantly lower odds of a future encounter, whereas population in thousands (OR 1.01; 95% CI 1.001 to 1.01) and the percentage of patients with less than a high school degree (OR 1.04; 95% CI 1.01 to 1.06) were associated with greater odds of a future encounter.

Table 2.

Logistic regression results for encounter-level analysis of the individual having a future opioid-related encounter

	Odds Ratio	95% Confidence Interval		p-Value
		lower bound	upper bound
Encounter Year (ref=2012)
2013	0.98	0.76	1.27	0.89
2014	1.07	0.80	1.43	0.66
2015	1.13	0.87	1.48	0.36
2016	0.78	0.62	0.99	0.04
2017	0.56	0.42	0.73	<0.01
Season (ref=winter)
spring	1.05	0.88	1.24	0.60
summer	0.92	0.78	1.09	0.35
fall	0.88	0.74	1.04	0.14
Encounter Daytime (ref= night)	1.00	0.90	1.12	0.99
Encounter Weekend (ref=weekday)	0.93	0.82	1.05	0.26
Disposition (ref=discharged home/normally)
left AMA	1.09	0.78	1.53	0.61
transferred/discharged to another dept/facility	0.53	0.32	0.86	0.01
other	0.83	0.64	1.08	0.18
Patient Number Previous Opioid ED Encounters	1.47	1.40	1.55	<0.01
Patient Number Previous Any ED Encounters	1.00	1.00	1.01	0.55
Patient Count of Institutions Visited	1.17	1.09	1.24	0.00
Male (ref=female)	1.08	0.96	1.20	0.20
Race (ref=white)
black or african american	0.92	0.74	1.14	0.44
all other	0.67	0.55	0.82	<0.01
Age at Encounter	0.98	0.98	0.99	<0.01
Opioid Rx Filled before Encounter	1.04	0.79	1.37	0.77
Benzodiazepine Rx Filled before Encounter	1.98	1.22	3.22	0.01
Opioid & Benzodiazepine Filled before Encounter	0.58	0.29	1.15	0.12
Elixhauser Level (ref=0)
1	0.94	0.82	1.07	0.36
2	1.22	0.96	1.55	0.11
3+	0.83	0.43	1.59	0.58
Heroin Diagnosis Code	1.43	1.26	1.64	<0.01
Insurance (ref=private/employer)
Medicaid	1.45	1.21	1.73	<0.01
Medicare	1.09	0.83	1.43	0.54
self-pay	1.37	1.11	1.70	<0.01
other	1.55	1.00	2.41	0.05
Community % Uninsured	0.97	0.95	1.00	0.05
Community % Unemployed	1.01	0.98	1.04	0.55
Community Median Household Income	1.01	1.00	1.01	0.06
Community % White	0.99	0.99	0.99	0.01
Community Total Population	1.01	1.00	1.01	0.01
Community % Less than HS Degree	1.04	1.01	1.06	<0.01
Constant	0.48	0.20	1.15	0.10

Note: model also adjusted for institution but results not shown.

Findings from all 5 of the sensitivity analyses generally remained consistent with the main analysis in terms of direction and significance, aside from a few slight differences mainly concentrated in the community characteristics. Similar findings from the final sensitivity analysis in which the missing insurance and disposition observations were included as an additional category in particular help to alleviate concerns that this missingness drove results. The odds ratio estimate on other/missing insurance remained significantly greater than 1 (OR 1.27; 95% CI 1.05 to 1.53) because it was in the original model, and the estimates on the other insurance indicators did not change either. The few differences between the main model and sensitivity analyses suggest our results are robust to these measurement decisions, missingness in the most prevalent variables, and model specifications, and strengthen our findings.

Limitations

Data may have limited generalizability as the sample is reflective of EDs in one Midwestern state that has been particularly affected by the opioid epidemic. However, the use of data from multiple EDs identified a broad population. In addition, our data are limited in that we captured only encounters beginning in 2012, which may miss individual’s having a ‘first’ encounter previously. Although we showed comparisons of the sample for individuals with only one encounter compared with those with multiple encounters, our main, multivariate regression results were structured to identify encounter-level characteristics associated with having a subsequent encounter rather than an individual’s having one or multiple opioid-related encounters. Similarly, we were unable to identify individuals who may have died during our study period outside of ED encounters. Our data also exhibit limitations in terms of missingness and rely on consistent entry into an electronic health record and some fields that are free text. Next, our study was able to identify only ED encounters that had opioid-related diagnoses appropriately coded in the medical record. This may have missed encounters that were not coded as specifically opioid-related but rather as a more general substance-related encounter. Finally, we were unable to establish the causal reasons for subsequent encounters because not all substance misuse results in ED encounters and our methods were not structured to identify causal factors. Although our study does have limitations, it improves on previous research in a wider population with more recent data and updated ICD-10 codes and is robust to variations in analyses.

Discussion

The prevalence of repeated opioid-related ED encounters is high and increasing. We identified a proportion of individuals experiencing repeated encounters substantially higher than previous estimates (22% versus 7%), although on average our study followed individuals for a longer period.^4,5 The increasing prevalence of opioid-related encounters may be attributable to coding changes, institutional and governmental policy changes, and increased provider awareness of the epidemic. However, the increase in subsequent encounters may also be an indicator of the lifesaving effect of increased availability and use of naloxone rather than an increasing burden in repeated encounters.

Findings suggest potential target areas for intervention within ED settings and across the health care system. The clear association with the count of previous encounters, the count of previous hospital systems where the patient presented, and heroin-related encounters suggests the value of incorporating health information exchange and basic risk-stratification processes into the ED. Patient information from multiple EDs would help identify encounters occurring between institutions and therefore give a more accurate indication of risk. Additionally, simple rules-based risk stratification (e.g., a patient has had 3 previous encounters) could be easily implemented to identify candidates for case management or specific interventions. Outside of the ED setting, the increased risk from concurrent benzodiazepine prescriptions reinforces the need for widespread use of prescription drug monitoring programs by ambulatory care providers and pharmacists. Among many other possibilities that are beyond the scope of this study and data, the significant relationship between having a heroin encounter and a future opioid-related encounter may reflect differences in treatment, provider attitudes, or access to appropriate follow-up care. Further work is needed to understand this difference.

Additionally, the increased odds of a future encounter associated with the number of institutions at which the individual has experienced any ED encounter may reflect the value of having health information exchange data that span institutions both for research and for practice. Without having access to a health information exchange, the care team at an ED may be unable to access previous patient medical history in a timely manner. Although we found this association significant, it is a proxy for availability of past patient information in the health information exchange and is not reflective of providers accessing the health information exchange during an encounter.

Repeated encounters related to opioids can serve as an important indicator for the ongoing surveillance of the opioid epidemic. For example, given the burden on the workforce and the clear missed opportunity of effective substance use disorder management since the index encounter, repeated encounters could serve as overall community measures for the burden of the epidemic. In addition, the ED is not the most effective setting for the ongoing management and support required to effectively manage substance use disorders. Therefore, repeated encounters could serve as an outcome measure for intervention evaluations. In addition to patient implications, ED care is costly and reductions in repeated encounters may be an area of potential savings and efficiency gains that may be of interest to hospital systems, providers, and payers.