It is now widely accepted that opioid medications taken as directed for therapeutic indications can contribute to the development of opioid use disorder (OUD).1–3 Opioid prescriptions, even in limited duration, are associated with later long-term use,4–7 and individuals suffering from OUD often recall that their first exposure to opioids was through a prescription for pain.8,9 Yet, the relationship between therapeutic opioid exposure and subsequent OUD remains largely unexplored.1,8,10,11 For example, we do not know the degree to which later use is due to OUD, ongoing chronic pain, or new painful conditions. We also do not know the risk of OUD among similar individuals without therapeutic exposure. Emergency departments are a common source of therapeutic opioid exposure, but prescriptions are short term, low dose, and potentially unavoidable.5,6,8,12–14 Development of OUD is a complex and extended process,1,15–17 so the contribution of EDs to OUD incidence, if any, is presumably only indirect in the case of initial prescriptions or perhaps somewhat more direct in the case of repeat prescriptions. Information about trajectories of pain and opioid use after emergency care that could inform such questions is growing but remains limited.18–20
ImportanceThe opioid crisis has been declared a national emergency,21,22 with 130 deaths daily.23 Because of increased awareness and legislation, opioid prescriptions have decreased; yet, opioid overdoses continue to climb.23,24 Moreover, overdose deaths are likely underreported and do not capture the extensive morbidity and mortality resulting from complications of OUD and injection drug use.25,26 Even a small improvement in the contribution of therapeutic opioid use to incident OUD would be a significant benefit given the millions exposed each year.
Goals of this investigationThis investigation was undertaken using available data to preliminarily explore the association between ED opioid prescriptions and subsequent OUD. We report OUD incidence at 1 year based on health record documentation among those who did and did not receive an ED opioid prescription, number of ED opioid prescriptions relative to other medical sources, time to OUD after an ED encounter, and modeling to evaluate development of OUD when controlling for demographics, mental health history, and receipt of prior opioid prescriptions.
METHODS Study design and settingThis retrospective cohort study was conducted using electronic health records (EHR) from a single academic health system encompassing 2 Midwestern EDs (1 urban and 1 suburban). Combined ED volume is approximately 115,000 visits for 70,000 unique patients annually. The urban ED is a tertiary, level I trauma center that serves as the primary safety-net hospital in the region.
Selection of participantsWe selected all adult patients (over 18 years of age) discharged to home from either of the EDs in 2015, identifying their first ED visit as the index visit. Any patient with existing OUD at index, or in the year prior, was excluded.
MeasurementsDiagnostic codes (International Classification of Diseases, Ninth Revision and Tenth Revision) and medication lists were obtained from the EHR for all health system encounters for January 2014–December 2016 (year before and after the index ED encounter in 2015) to identify markers for OUD, mental health diagnosis, and non-opioid substance use. We identified all patients with an opioid prescription at index ED visit and the source of any additional opioid prescriptions from the health system for all patients in the year before and after the index visit. OUD was defined by composite definition that included any of the following coded diagnoses: OUD, opioid withdrawal, dependence, heroin use, or unintentional opioid overdose; or any prescription for buprenorphine or naloxone.
OutcomesWe describe OUD incidence within 1 year of the ED encounter and time to first indicator of OUD. We then use a Cox proportional hazards model measuring the effect of ED opioid prescription at the index encounter on later OUD among those with a repeat postindex health system encounter. Secondary outcomes were sources of therapeutic opioid exposure and frequency of risk factors associated with OUD among those who developed OUD.
AnalysisWe report the proportion of individuals who develop OUD within 1 year of index ED visit overall and stratify by whether or not an opioid was prescribed at index ED encounter. For patients who developed OUD, we describe summary statistics on systemwide opioid prescribing before and after their index ED encounter, the time from index ED encounter to first documentation of criteria meeting the study definition of OUD, and the frequency of risk factors associated with OUD (mental health condition and non-opioid substance use). Specified outcomes were reported with their respective 95% confidence intervals (CIs). Statistical analyses were conducted using R [version 4.0.2]27 in R Studio,28 and the epiR29 package.30 Figures also were built in R using RStudio, using the ggplot2 package.31
Finally, we constructed a Cox proportional hazards model to control (to the degree possible) for other factors that might influence development of OUD.32,33 This approach requires exclusion of any patients who did not have a visit somewhere in the health system in the 12 months post index. Model covariates included age, race, sex, mental health conditions (apart from substance use disorder), non-opioid substance use disorders, opioids prescribed (systemwide) in the 12 months before index ED visit, and opioids prescribed (systemwide) between index ED visit and end of the censored time period, defined as last system encounter or first encounter meeting study criteria for OUD.
RESULTS Characteristics of study subjectsIn 2015, there were 115,886 ED visits representing 71,460 unique patients. Patients were excluded for age <18 years (1775; 2%), ED disposition other than discharge home (18,074; 25%), and pre-existing OUD (1707; 2%). Of the 49,904 patients included for analysis, 55% were white, 54% were women, median age was 39 years, and 5176 (10.4%) received an opioid prescription at their index ED encounter (Table 1). Of the entire included cohort, 669 (1.3%; 95% CI, 1.2–1.4) developed OUD in the 12 months after index ED visit. The proportion developing OUD at 12 months was 1.5% (95% CI, 1.2–1.9) for patients prescribed an opioid at index and 1.3% (95% CI, 1.2–1.4) for patients who were not.
TABLE 1 Demographics and characteristics of sample
| Study population | Opioid prescribed index ED visit | No Opioid prescribed index ED visit | Developed OUDa | No OUDb | ||||||
| N = 49,904 (%) | N = 5176 (%) | N = 44,728 (%) | N = 669 (%) | N = 49,235 (%) | ||||||
| Demographics | ||||||||||
| Age, y, median (IQR) | 39 (28–53) | 42 (30–54) | 39 (27–53) | 36 (29–50) | 39 (28–53) | |||||
| Sex | ||||||||||
| Female | 26,801 | (53.7) | 2669 | (51.6) | 24,132 | (54.0) | 333 | (49.8) | 26,468 | (53.8) |
| Male | 23,103 | (46.3) | 2507 | (48.4) | 20,596 | (46.0) | 336 | (50.2) | 22,767 | (46.2) |
| Race | ||||||||||
| White/Caucasian | 27,618 | (55.3) | 3368 | (65.1) | 24,250 | (54.2) | 129 | (19.3) | 27,091 | (55.0) |
| Black/African American | 18,659 | (37.4) | 1517 | (29.3) | 17,142 | (38.3) | 527 | (78.8) | 18,530 | (37.6) |
| Hispanic | 1,504 | (3.0) | 116 | (2.2) | 1388 | (3.1) | 6 | (0.9) | 1498 | (3.0) |
| Other | 2,080 | (4.2) | 170 | (3.3) | 1910 | (4.3) | 7 | (1.0) | 2073 | (4.2) |
| Unknown/Missing | 43 | (0.1) | 5 | (0.1) | 38 | (0.1) | – | – | 43 | (0.1) |
| Risk factors | ||||||||||
| History of mental health diagnosis | 6,064 | (12.2) | 549 | (10.6) | 5515 | (12.3) | 179 | (26.80) | 5885 | (12.0) |
| History of non-opioid substance use (any) | 6,290 | (12.6) | 542 | (10.5) | 5748 | (12.9) | 160 | (23.90) | 6130 | (12.5) |
| Alcohol | 1,956 | (3.9) | 145 | (2.8) | 1811 | (4.0) | 40 | (6.00) | 1916 | (3.9) |
| Marijuana | 4,250 | (8.5) | 382 | (7.4) | 3868 | (8.6) | 84 | (12.60) | 4166 | (8.5) |
| Cocaine | 596 | (1.2) | 46 | (0.9) | 550 | (1.2) | 29 | (4.30) | 576 | (1.2) |
| Hallucinogen | 15 | (0.0) | 1 | (0.0) | 14 | (0.0) | 1 | (0.10) | 14 | (0.0) |
| Stimulants | 50 | (0.1) | 4 | (0.1) | 46 | (0.1) | 3 | (0.40) | 47 | (0.1) |
| Sedatives | 75 | (0.2) | – | – | 75 | (0.2) | 6 | (0.90) | 69 | (0.1) |
| Inhalants | 2 | (0.0) | – | - | 2 | (0.0) | – | – | 2 | (0.0) |
| Other drug | 405 | (0.8) | 28 | (0.5) | 377 | (0.8) | 47 | (7.00) | 358 | (0.7) |
| Opioid prescribing (RX) | ||||||||||
| Opioid RX at index | 5176 | (10.4) | 5176 | (100) | – | – | 80 | (12.00) | 5096 | (10.40) |
| Opioid RX within 12 months before index – any | 4277 | (8.6) | 986 | (19.0) | 3291 | (7.4) | 126 | (18.80) | 4151 | (8.40) |
| Opioid RX within 12 months before index - Non ED | 2876 | (5.8) | 596 | (11.5) | 2280 | (5.1) | 89 | (13.30) | 2787 | (5.70) |
| Opioid RX within 12 months before index – ED | 1,971 | (3.9) | 540 | (10.4) | 1431 | (3.2) | 65 | (9.70) | 1906 | (3.90) |
| Prescribed opioid post indexc - Non ED | 2,108 | (4.2) | 613 | (11.8) | 1495 | (3.3) | 81 | (12.10) | 2027 | (4.10) |
| Prescribed opioid post indexc - ED | 1,578 | (3.2) | 520 | (10.0) | 1058 | (2.4) | 54 | (8.10) | 1524 | (3.10) |
| Outcomes (within 12 months post index) | ||||||||||
| OUD composite definitiond | 669 | (1.3) | 80 | (1.5) | 589 | (1.3) | – | – | – | – |
| OUD diagnosis | 427 | (0.9) | 51 | (1.0) | 376 | (0.8) | – | – | – | – |
| Dependence | 230 | (0.5) | 26 | (0.5) | 204 | (0.5) | – | – | – | – |
| Opioid withdrawal | 27 | (0.1) | 4 | (0.1) | 23 | (0.1) | – | – | – | – |
| Heroin use | 95 | (0.2) | 9 | (0.2) | 86 | (0.2) | – | – | – | – |
| Unintentional opioid overdose | 50 | (0.1) | 8 | (0.2) | 42 | (0.1) | – | – | – | – |
| Naloxone prescription | 124 | (0.2) | 24 | (0.5) | 100 | (0.2) | – | – | – | – |
| Buprenorphine prescription | 58 | (0.1) | 30 | (0.6) | 28 | (0.1) | – | – | – | – |
Based on first encounter with indicator or OUD.
Last follow-up visit in 12 months (no OUD/Dep, at least 1 subsequent visit), or 12 months postindex visit (no OUD and no subsequent visit).
Number of prescriptions between index and first onset of OUD or last follow up visit within 12 months (no OUD)
dd; OUD classified as (1) ICD-9/10 diagnosis of: OUD, opioid dependence, opioid withdrawal, heroin use, or unintentional opioid overdose; or (2) prescription for naloxone or buprenorphine.
Abbreviations: ED, emergency department; ICD-9/10, International Classification of Diseases, Ninth Revision/Tenth Revision; IQR, interquartile range; OUD, opioid use disorder.
Patients who developed OUD in the 12 months after index ED visitFor the 669 patients who developed OUD in this study, 80 (12.0%; 95% CI, 9.6–14.7) were prescribed an opioid at index. The majority of patients (423, 63.2%) who developed OUD did so within 6 months after index visit and over a third (259, 38.7%) did so within 3 months (Figure 1). Median time from ED index visit to onset of OUD was 4.5 months overall (interquartile range [IQR] 1.6–7.6, range 0.0–11.9), 4.1 months (IQR 1.8–7.0, range 0.1–11.9) for those prescribed an opioid, and 4.5 (IQR 1.6–7.7, range 0.0–11.9) for those who were not (Figure 1). There were 45 (6.7%) patients who met criteria for OUD within 1 week of ED index visit, approximately evenly distributed between exposure groups. Sensitivity analysis excluding these patients did not change study findings.
Experts suggest that emergency department (ED) opioid prescriptions may be associated with new opioid use disorder (OUD). In this analysis of 49,904 ED patients, the incidence of new onset OUD at 12 months was similar between patients who did and did not receive ED opioid prescriptions (adjusted hazard ratio, 1.1; 95% confidence interval, 0.9-1.4%). Non-opioid substance use disorder, mental health conditions, and prior opioid prescriptions were stronger predictors of 12-month OUD.
Enlarge this image.FIGURE 1. Days from index ED encounter to first indicator of opioid use disorder in health records, based on ED prescription at index encounter disorder in health records, based on ED prescription at index encounter. There were 669 patients who developed OUD, each dot represents an individual patient. Abbreviations: ED, emergency department; OUD, opioid use disorder
Before the index encounter, 126 (18.8%) were prescribed an opioid by any prescriber type, about half of which (9.7%) originated from an ED. Post index, 135 (20.1%) were prescribed an opioid by any prescriber type and of these, 54 (8%) were from an ED (Figure 2). Among those who received health system prescriptions, repeat prescriptions both before and after the index ED encounter were common (Table 1).
Enlarge this image.FIGURE 2. Timing and source of opioid prescriptions before and after index ED visit among patients who developed opioid use disorder. Each dot represents a prescription event for a person who ultimately developed opioid use disorder after an index ED encounter, only a subset of whom received an opioid prescription at the index ED encounter. Prescription events after development of opioid use disorder not shown. Opioid prescriptions given at the index ED encounter not shown. Abbreviations: ED, emergency department
In the Cox proportional hazards model, we excluded 17,782 (35.6%) patients who did not have a follow-up visit in the system in the 12 months after index ED visit, leaving 32,124 patients included. Opioid prescription at index ED visit did not have a meaningful effect on OUD (HR, 1.1; 95% CI, 0.9–1.4). The model did find that non-opioid substance use disorder (HR, 2.1; 95% CI, 1.8-2.6), a mental health condition (HR, 1.7; 95% CI, 1.4-2.0), and at least 1 opioid prescription in the 12 months before index (HR, 1.5; 95% CI, 1.2–1.8) increased the risk of developing OUD in the year after an ED visit. The meaning of study findings did not change according to whether opioid prescriptions were included as a continuous or binary variables; we report the binary variable as easier to interpret.
LIMITATIONSAlthough advancing current literature in important respects, our results should be interpreted in context with several important limitations. Most important, the inaccuracies inherent to retrospective records review and especially diagnosis coding are well known.34-36 It is logical to assume these inaccuracies are even greater for behavioral health diagnoses, screening for which is not a traditional or structured focus in emergency care. Consequently, misclassification of patients in terms of OUD, both before and after the index encounter, is likely in this study design. This misclassification bias, likely leading to underreporting of OUD, could bias the overall proportion of persons with OUD toward the null. Moreover, our composite definition of OUD is unlikely to correlate precisely with OUD; we do not have data on non-medical opioid exposure over the same study period. The problem of data quality in existing records also applies to other patient characteristics that may affect the relationship between therapeutic opioid exposure and eventual OUD. As we included electronic health record data from only 2 hospitals within a health system, the patterns of opioid exposure we report may not generalize beyond the single system we investigated, although the nature of any causal relationship between therapeutic opioid exposure and subsequent OUD within any single person is unlikely to vary by treatment location. Finally, the authors acknowledge the data are dated, yet adhere the time for analysis was extensive given the size and complexity of the project, and we do not believe that there would be any change over time this period of time in the degree to which opioid therapy leads to OUD.
DISCUSSIONThe question of whether and how short-term opioid therapy is causally related to later OUD and the degree to which this can be avoided is central for policy and practice in episodic care settings such as the ED. Discovering the answer to such questions requires longitudinal study that accounts for (1) inclusion of exposed and unexposed groups, (2) better characterization of therapeutic exposure source, timing, and intensity, (3) concurrent understanding of both therapeutic and non-medical exposure, and (4) precise classification distinguishing episodic use, chronic use/dependence, and OUD at study entry and conclusion of follow-up.
This investigation, although limited to existing data and exceedingly preliminary, is to our knowledge the first attempt to disentangle these factors in a longitudinal study design that goes beyond associating an initial prescription and later use of unknown cause and without a concurrent control. We found that 1.3% of all patients developed indicators of OUD within 12 months of an ED visit, with the median time to documentation of OUD being 4.5 months. Opioid prescriptions were common in the year before and year after the ED visit for individuals developing OUD. As expected based on existing literature,32,33 the frequency of mental health conditions and non-opioid substance use was high in the group that developed OUD. We urge caution when interpreting our primary finding that an ED opioid prescription at the index visit did not increase risk for subsequent diagnosis of OUD within the following year. Although this report should provide guidance for future prospective study, the quality and completeness of data available in existing health records are insufficient for definitive conclusion.
Individuals may have already been suffering from unrecognized OUD at the index visit and newly developed OUD may have been unrecognized in follow-up. Known limitations in diagnosis coding would affect both measures, as would the lack of data from other health systems, lack of follow-up in any health system, and no data source to characterize other non-medical sources of opioid exposure. The significant number of individuals who met criteria for OUD immediately after an index visit is a convincing illustration of the potential for misclassification in this study design.
There is also the potential for individuals without OUD to have been classified as such inappropriately. For example, naloxone could be prescribed to individuals without opioid use disorder, although this was anecdotally uncommon in this health system during the study period. Also, we chose to include diagnoses of opioid dependence in the composite definition given inaccuracies in diagnosis coding and difficulty disentangling OUD from opioid dependence, even though patients with physiologic opioid dependence may not meet criteria for OUD.37,38
Our analysis of therapeutic opioid exposure was limited to opioid prescription and did not consider opioid administration or patients who were hospitalized after the ED encounter. A single episode of opioid administration is even more limited in dose and duration than a prescription, but administration may still contribute to trajectories leading to OUD. For example, repeated dosing occurs within and across health encounters, neuropsychological effects may be even more pronounced with parenteral administration, and some individuals might theoretically be unusually primed for rapidly triggered addiction or relapse while in ongoing recovery. Administration and prescription of opioids in hospital settings is not directly controlled by the ED. However, ED practice might have influence in setting ongoing patient and practitioner expectations or even influencing the degree of future pain. If this is not the case, there is still the analogous question of whether and how prescriptions for opioids at hospital discharge contribute to OUD incidence.
An additional factor, unexplored by this study, is the potential selection bias introduced by the prescription itself. It is possible for clinicians to differentially limit opioid therapy according to perceived risk of opioid use disorder. Although understandable and even desirable, the resulting selection bias would inhibit the ability to detect an effect of the opioid prescription and further illustrates the need to adequately control for differences between groups in other characteristics influencing the development of OUD.
In summary, a small but meaningful proportion of the ED population will develop OUD within 1 year even without ED opioid prescription. We found no association between an initial ED opioid prescription and later OUD. However, accurate understanding of patient characteristics and evolution of those characteristics overtime is exceptionally difficult in a retrospective approach, given the complexity factors influencing OUD incidence and opioid exposure. Although this study is an important initial advance in the investigation of whether and how short-term opioid exposure is causally related to later OUD, a prospective longitudinal study design with a control group will be essential to accurately classify participants with respect to exposure, outcome, and other contributory factors.
AUTHOR CONTRIBUTIONSBEP, RMA, and MSL contributed to the conception and the design of the study. RMA conducted the statistical analyses. All of the authors contributed to the interpretation of the results.
© 2021. This work is published under http://creativecommons.org/licenses/by-nc-nd/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.