Interpreting the National Hospital Ambulatory Medical Care Survey: United States Emergency Department Opioid Prescribing, 2006–2010

Abstract

Objective

Prescription opioid overdoses are a leading cause of death in the United States. Emergency departments (EDs) are potentially high risk environments for doctor shopping and diversion. We hypothesized that opioid prescribing rates from the ED have increased over time.

Methods

We analyzed data on ED discharges from the 2006–2010 NHAMCS, a probability sample of all United States EDs. The outcome was documentation of an opioid prescription on discharge. The primary independent predictor was time. Covariates included severity of pain, a pain-related discharge diagnosis, age, gender, race, payer, hospital ownership, and geographic location of hospital. Up to three discharge diagnoses were available in NHAMCS to identify ‘pain-related’ (e.g. back pain, fracture, dental/jaw pain, nephrolithiasis) ED visits. We performed multivariate logistic regression to assess the independent associations between opioid prescribing and predictors. All analyses incorporated NHAMCS survey weights, and all results are presented as national estimates.

Results

Opioids were prescribed for 18.7% (95% CI: 17.7–19.7%) of all ED discharges, representing 18.8 million prescriptions per year. There were no significant temporal trends in opioid prescribing overall (adjusted p=0.93). Painful discharge diagnoses that received the top 3 highest proportion of opioids prescriptions included: nephrolithiasis (62.1%), neck pain (51.6%), and dental/jaw pain (49.7%). A pain-related discharge diagnosis, non-Hispanic white race, older age, male gender, uninsured status and Western region were associated with opioid prescribing (p<0.05).

Conclusions

We found with no temporal trend towards increased prescribing from 2006–2012. Our results suggest that problems with opioid over-prescribing are multifactorial and not solely rooted in the ED.

INTRODUCTION

Prescription drug misuse has been classified as an epidemic by the Centers for Disease Control and Prevention¹, the US Congress², and National Institute on Drug Abuse³. The United States economic burden due to the nonmedical use of prescription opioids in 2007 was estimated at $55.7 billion—$25.6 billion attributed to workplace costs, $25 billion in healthcare costs, and $5.1 billion in criminal justice costs⁴.

Emergency Departments (EDs) are perceived as high-risk contributors to the prescription opioid epidemic^2,5. The lack of provider continuity, pressure to rapidly turn over patients, and limitations in available medical records and pharmacy data create an environment vulnerable to doctor-shopping and diversion of prescribed opioids. As a result, multiple ED-targeted prescription drug guidelines have been developed by the American College of Emergency Physicians (ACEP),⁶ Washington State ACEP Chapter,⁷ New York City’s mayor’s office,⁸ and other organizations.^9,10

There are limited data on the nationwide prevalence and temporal trends for ED prescription opioid prescribing. Such information is critical for policy makers, public health officials, and clinicians to appropriately target high-risk settings. A recent paper suggested an increase in ED opioid prescribing over time¹¹; however, this study conflated opioid administration in the ED with opioid prescribing on discharge¹². Using nationally representative data, we describe prevalence, temporal trends, and patient predictors of ED opioid prescribing.

METHODS

Study Design and Population

This study was an analysis of publicly available cross-sectional survey data collected for National Hospital Ambulatory Medical Care Survey (NHAMCS) from 2006–2010. NHAMCS uses a four-stage probability design, collecting a nationally representative sample of ED visits based on non-institutional general and short-stay hospitals. At sampled hospitals, hospital staff members monitored by the US Census Bureau’s field agents, or the agent themselves, complete patient record forms for each sampled visit during a randomly assigned four week reporting period each year.^13,14,15 Details on the methodology of data collection, processing and validation are available from the National Center for Health Statistics (NCHS).¹⁴ The Ethics Review Board of the NCHS approves the NHAMCS annually, and the Oregon Health & Science University Institutional Review Board approved this study.

Prior to 2005, it was impossible to differentiate between administered versus prescribed medications in the NHAMCS as there was a single variable that included all medications administered in the ED and prescribed at discharge. Furthermore, drug coding changed from the National Drug Code Directory to Multum Lexicon in 2006. Thus, to minimize errors with drug coding changes and analyze data with our dependent variable of interest, opioid prescriptions at discharge, we studied all patients discharged after an ED visit from 2006 to 2010.

Measures and Variables

The outcome was documentation of an opioid prescription on discharge, using Multum Lexion level 3 therapeutic drug categories [code 060 (CNS; Analgesics; Narcotic) and 191 (CNS; Analgesics; Narcotic analgesic combinations)]. In order to determine individual opioid trends, we further identified opioids classified by the Drug Enforcement Agency (DEA) as schedule II, III, IV, and V opioids and coded them using already available NCHS DRUG ID codes (generic ingredient drug code) and/or Medcode ID (drug names mentioned in record). Drug combinations such as hydrocodone-acetaminophen were collapsed into their primary opioid class such as hydrocodone. Long-acting, extended-release and/or sustained-release opioids (as classified by the DEA) were also individually identified using the NHAMCS drug ID and/or Medcode ID.

The primary predictor variable was year of ED visit with additional predictors including a pain-related discharge diagnosis, level of pain, age, gender, race/ethnicity, payer, hospital ownership, and geographic location of hospital. Analysis was restricted to discharged patients. Missing values of age, gender, race and ethnicity were imputed by NCHS¹⁶. We combined race and ethnicity into a new four-category variable that would most likely identify minority groups experiencing health disparities¹³: non-Hispanic White, non-Hispanic Black, Hispanic, and non-Hispanic other. Non-Hispanic others included Asian, Native Hawaiian/other Pacific Islanders, American Indian/Alaska Native, and patients with more than one race reported. From 2009–2010 level of pain was measured on a 0–10 scale and from 2006–2008 pain was categorized as: “none” (0), “minimal” (1–3), “moderate” (4–6), and “severe” (7–10). In this analysis, we used the four pain levels coded in NHAMCS 2006–2008, with subsequent years converted to these categories using the methodology recommended by NHAMCS.¹⁷ Up to three discharge diagnoses were reported for each ED visit, and we categorized the three diagnoses using the hierarchical coding of Agency for Healthcare Research and Quality Clinical Classifications Software (CCS), which collapses more than 13,000 ICD-9 CM diagnosis codes into a smaller set of meaningful categories.¹⁸ Categories were designated as “pain-related” based on existing literature and clinical judgment of the study team (back pain, headache, abdominal pain, neck pain, chest pain, cholelithiasis, nephrolithiasis, pelvic pain, dental/jaw pain, fractures, non-fracture injuries (such as sprains and strains, or concussion), arthritis/joint pain, joint pain, sickle cell anemia, and cancer-related). See supplement for a full list of ICD-9 codes collapsed into these pain-related CCS diagnostic categories.

Statistical Analysis

We report unweighted raw ED visits (true number of patient record forms collected) and weighted national representative proportions of opioid prescription. We performed multivariable logistic regression to assess the independent associations between opioid prescribing and predictors. The variable year was treated as a discrete variable to determine differences between different years for opioid prescribing. Test of trend for individual opioids and overall opioid prescriptions was done by treating year as a continuous variable. Additionally, pain-related CCS diagnostic categories were treated as a dichotomous variable (yes/no), and cancer-related pain was included in the analysis to account for its impact on ED opioid prescribing. These analyses incorporated NHAMCS complex survey design features including cluster, strata and probability weights to produce nationally representative estimates. The probability weights take into account the probability of visit selection and other post-survey adjustments, e.g. for nonresponse. As recommended by McCaig, et al,¹⁹ estimates based on less than 30 sample or unweighted records are considered unreliable thus sub-analyses on buprenorphine, opium, fentanyl, methadone, papaverine, meperidine, and morphine were not considered in this study. All analyses were conducted using SAS 9.3 (SAS Institute Inc., Cary, NC, USA).

RESULTS

Of the 502.4 million ED discharges represented by the 2006–2010 NHAMCS data, 94.0 million (SD 5.5 million) received an opioid prescription, an average of 18.8 million prescriptions per year, representing 18.7% (95% CI: 17.7–19.7) of all ED discharges from 2006–2010 (Table 1). The most common categories of ED visitors prescribed opioids were aged 25–44yo (42%), female (55.8%), non-Hispanic White race (66.5%), privately insured (36.2%), from the South (44.5%), and seen in voluntary, non-profit hospitals (74.9%) (Table 1).

Table 1.

Association between ED discharge visit characteristics and opioid prescription; NHAMCS 2006–2010 [N=139,256 unweighted visits, weighted n=502.4 million (SD 22.5)].

	No. of ED Unweighted Visits Recorded	Weighted No. of Total ED Discharge Records in Millionsa, (SD)	Proportion of Weighted ED Discharge Records, (95% CI)	Weighted ED Discharge Records Prescribed Opioid in Millionsb, (SD, row % c, column % d)	Adjusted OR (95% CI)e
Year
2006	28,548	95.7 (5.1)	19.1 (17.5–20.6)	17.2 (1.1, 18.0, 18.3)	Reference
2007	27,685	92.3 (6.6)	18.4 (16.7–20.0)	17.0 (1.6, 18.4, 18.0)	1.0 (0.9–1.1)
2008	26,700	98.3 (5.6)	19.6 (18.1–21.0)	18.6 (1.38, 17.8, 19.8)	1.1 (0.9–1.2)
2009	28,467	111.9 (7.6)	22.3 (20.5–24.0)	20.7 (1.9, 18.5, 22.0)	1.0 (0.9–1.1)
2010	27,856	104.2 (5.8)	20.7 (18.9–22.5)	20.5 (1.4, 19.7, 21.9)	1.0 (0.9–1.1)
Age*
Under 15 years	31,205	111.1 (6.2)	22.1 (20.9–23.2)	3.9 (0.3, 3.5, 4.2)	Reference
15–24 years	24,301	88.9 (4.3)	17.7 (17.3–18.1)	18.5 (1.1, 20.8, 19.7)	4.5 (4.0–4.9)
25–44 years	41,705	150.7 (7.0)	30.0 (29.4–30.6)	39.4 (2.5, 26.2, 42.0)	5.8 (5.2–6.4)
45–64 years	27,558	98.7 (4.3)	19.7 (19.2–20.1)	24.2 (1.5, 24.5, 25.8)	5.8 (5.2–6.5)
65–74 years	6,443	23.5 (1.1)	4.7 (4.5–4.9)	4.4 (0.3, 18.6, 4.7)	5.3 (4.6–6.2)
75 years and over	8,044	29.4 (1.3)	5.9 (5.5–6.2)	3.6 (0.2, 12.1, 3.8)	3.6 (3.1–4.2)
Sex*
Male	63,588	226.3 (9.8)	45.0 (44.6–45.5)	41.5 (2.3, 18.3, 44.2)	Reference
Female	75,668	276.1 (12.8)	54.9 (54.5–55.4)	52.5 (3.2, 19.0, 55.8)	0.9 (0.9–0.9)
Race*
Non-Hispanic White	80,388	300.4 (14.5)	59.8 (57.1–62.5)	62.5 (3.6, 20.8, 66.5)	Reference
Non-Hispanic Black	32,098	113.6 (9.6)	22.6 (19.9–25.4)	18.5 (2.0, 16.3, 16.7)	0.8 (0.7–0.9)
Hispanic	20,341	70.7 (6.0)	14.1 (11.9–16.2)	10.4 (1.0, 14.8, 11.1)	0.8 (0.8–0.9)
Non-Hispanic Othersf	6,429	17.7 (1.8)	3.5 (2.8–4.2)	2.6 (0.3, 14.7, 2.8)	0.8 (0.7–0.9)
Payer*
Medicaid	39,028	131.8 (6.8)	26.2 (24.9–27.5)	18.3 (1.2, 13.9, 19.5)	Reference
Private insurance	46,973	171.7 (7.8)	34.2 (33.0–35.3)	34.1 (2.1, 19.9, 36.2)	1.1 (1.0–1.2)
Medicare	16,587	61.4 (2.7)	12.2 (11.6–12.8)	10.4 (0.7, 17.0, 11.1)	1.0 (0.9–1.1)
Uninsured	21,330	80.5 (4.1)	16.0 (15.0–16.9)	19.7 (1.3, 24.5, 21.0)	1.2 (1.2–1.3)
Otherg	7,173	26.6 (2.1)	5.3 (4.6–5.9)	6.1 (0.5, 23.0, 6.5)	1.1 (1.0–1.3)
Unknown/Blank	8,165	30.4 (4.3)	6.0 (4.6–7.5)	5.3 (0.9, 14.5, 5.7)	1.1 (0.9–1.2)
Region*
West	26,883	96.4 (10.7)	19.2 (15.5–22.9)	19.6 (2.7, 20.3, 20.8)	Reference
Northeast	33,305	91.3 (4.3)	18.2 (16.1–20.3)	13.2 (0.8, 14.5, 14.0)	0.6 (0.6–0.7)
Midwest	29,044	107.8 (12.8)	21.5 (17.2–25.7)	19.5 (2.7, 18.0, 20.7)	0.8 (0.7–0.9)
South	50,024	207.0 (14.5)	41.2 (36.9–45.5)	41.8 (3.8, 20.2, 44.5)	1.0 (0.8–1.2)
Hospital Type¥
Voluntary, non-profit	101,798	374.4 (19.8)	74.5 (69.9–79.2)	70.4 (4.6, 18.8, 74.9)	Reference
Government, non-federal	23,885	71.2 (8.9)	14.2 (10.8–17.6)	11.7 (1.6, 16.4, 12.4)	0.7 (0.6–0.9)
Proprietary	13,573	56.8 (10.0)	11.3 (7.7–14.9)	11.9 (2.3, 20.9, 12.7)	1.1 (0.9–1.2)

Abbreviations: NHAMCS, National Hospital Ambulatory Medical Care Survey; ED, emergency department; SD, standard deviation; OR, odds ratio; CI, confidence interval.

^aNumber of weighted ED discharge visit records, includes both visits unprescribed and prescribed opioids.

^bNumber of weighted ED visit records for those prescribed an opioid at discharge.

^cWeighted Row Proportion: Among all ED discharge records for a specific subcategory, this is the proportion prescribed an opioid at discharge. Weighted proportions are estimated incorporating survey weights.

^dWeighted Column Proportion: Among all ED discharge records prescribed an opioid, this is the proportion prescribed an opioid for a specific subcategory. Weighted proportions are estimated incorporating survey weights.

^eAdjusted for year, age, sex, race, payer, region, hospital type, pain severity, and pain-related CCS

^fNon-Hispanic others include: Asian, Native Hawaiian/other Pacific Islanders, American Indian/Alaska Native, and patients with more than 1 race reported.

^gOther payer includes worker’s compensation, no charge/charity, and other sources of payment such as Veteran Affairs.

^*Overall chi-square p-value for multivariable analysis <0.001

^¥Overall chi-square p-value for multivariable analysis <0.05

Boldface type highlights variables with the highest odds ratio or where the reference was greater than the other variables.

Trend over time

The number of ED visits rose from 95.7 million in 2006 to 104.2 million in 2010, with an associated increase from 17.2 million to 20.5 million in the number of patients prescribed opioids. However, the proportion of patients prescribed opioids remained nearly constant from 2006 (18.0%) to 2009 (18.5%), rising only slightly in 2010 (19.7%) (unadjusted overall p=0.09). After adjusting for gender, age, race, insurance, region, hospital type, pain severity, pain-related CCS diagnostic category, there remained no significant difference in the proportions of ED patients receiving opioid prescriptions between different years (overall p=0.72).

Figure 1 also shows no temporal change in the proportion of patients prescribed opioids during the time period (testing for trend, adjusted OR 0.99, 95% CI: 0.97–1.02, p=0.93). When testing trend for specific opioid medications, there was an increase over time in prescribing of oxycodone (adjusted OR 1.1, 95% CI: 1.1–1.2, p=0.04), a decrease in prescribing of propoxyphene (adjusted OR: 0.8, 95% CI 0.8–0.9, p<0.001), and no temporal changes in prescribing of codeine (adjusted OR: 1.0, 95% CI: 0.9–1.0, p=0.08), hydrocodone (adjusted OR 0.99, 95% CI: 0.97–1.03, p=0.85) and hydromorphone (adjusted OR: 1.0, 95% CI: 0.9–1.2, p=0.78), Figure 1. This decrease in propoxyphene prescribing is consistent with propoxyphene being pulled from the market in 2010.

Figure 1.

Proportion of ED discharges prescribed specific opioid.

Characteristics Associated with Opioid Prescribing

In the multivariable analysis, patients with a pain-related CCS diagnostic category were more likely to receive an opioid prescription (OR 2.4; 95% CI 2.3–2.6, p<0.001). Other factors that were independently associated with increased opioid prescribing include non-Hispanic white race, older age, male gender, uninsured status and Western region (Table 1, p<0.05).

Figure 2 illustrates specific pain-related diagnoses associated with ED prescription opioids. Among patients prescribed opioids, the six most common diagnoses were non-fracture injuries (28.9%), back pain (10.5%), fractures (9.5%), abdominal pain (8.3%), dental/jaw pain (6.0%), and headache (4.0%).

Figure 2.

Association of pain-related CCS Diagnostic category and opioid prescribed at ED discharge, row proportions; and proportion of all ED opioid discharges with any pain-related CCS category, column proportion.

^αUp to three diagnoses were recorded per visit record and then categorized by the Agency for Healthcare Research and Quality Clinical Classification Software (CCS), which collapses more than 13,000 ICD-9 CM diagnoses codes into a smaller set of meaningful categories. Selected categories were designated as “pain-related” or potential for abuse based on existing literature and clinical judgment of the study team.

We also determined the pain-related CCS diagnostic categories for which opioids were most frequently prescribed. Of patients with nephrolithiasis, 62.1% received an opioid prescription, as did 51.6% of patients with neck pain, 49.7% of those with dental pain, 48.7% of those with fractures, 48.3% of those with cholelithiasis, and 45.2% of those with back pain.

DISCUSSION

Over 18 million US ED visits per year resulted in an opioid prescription. Although there is no evidence of a temporal increase in ED opioid prescribing rates from 2006–2010, we believe that the absolute number of opioid prescriptions justifies continued efforts to restrain potentially inappropriate prescribing in ED settings. We provide novel insights about specific pain-related diagnoses that might be the focus of future interventions.

Our results provide novel insights. Prior studies have focused on inadequate pain management in the ED associated with disparities in race,^{13,20,21,22,23} drug-related ED visits,^24,25 and the administration and prescribing rather than prescribing alone of opioids from the ED^11,13,26 (due to use of a variable that did not distinguish the two). ED prescribing studies have been limited to pediatric populations,²⁶ or described findings from a single year²⁷. Our study adds to the literature by reporting opioid prescribing trends using a multi-year, nationally representative survey of ED visits.

Our findings differ from a recent study¹¹ using the same data, which reported a nearly 50% relative increase in ED opioid prescribing from 2001–2010. Prior to 2006, however, NHAMCS did not distinguish between opioid administered in the ED vs prescribed at discharge. By restricting our analysis to 2006–2010 when NHAMCS collected data specifically on opioid prescribing, we believe that our results provide a more accurate description of temporal trends.

Our findings help contextualize the role of emergency physicians in the opioid prescription crisis. Other studies suggest that emergency physicians prescribe less than 5% of all opioids, behind family physicians, internists, dentists, and orthopedic surgeons⁶. The overall sales of all prescription opioids have increased steadily from 1999–2010²⁸. Our findings suggest that the emergency medicine contribution to opioid prescribing is related to increases in total ED visits, rather than changes in provider prescribing behavior.

However, we believe that the high risk environment of the ED and the absolute number of opioid prescriptions justify continued efforts to reduce inappropriate opioid prescriptions in conjunction with efforts to curtail these numbers in the primary care setting, as this is where the majority of opioid prescriptions are written. Only by recognizing that the burden of the opioid epidemic is multi-factorial and that all specialties have a role in this epidemic, can we pose to make a substantial difference in opioid misuse prevention.

For the ED, EM providers are well-position to contribute these efforts and the problem of the “habitual” ED visitor seeking opioid pain medications has been well documented over the past two decades^29–31. A proposal by the Washington State Medicaid program to deny hospital and provider payments for “inappropriate ED visits,” since rescinded, was motivated in part by the perception that many ED encounters were related to prescription opioid seeking³². For example, a prior study of ED visitors with a pain-related complaint revealed a high number of opioid prescriptions filled the prior year (median 7, range 0–128)³³.

Although our study cannot assess the “appropriateness” of opioid prescribing, we do identify specific pain-related diagnoses that often result in opioid prescriptions. These findings could influence clinical care. For example, prescription drug monitoring programs (PDMP) are databases authorized by 49 states that track dispensing of controlled substances. Providers may consider a reduced threshold to query a PDMP for conditions such as dental pain and back pain. Future studies should assess whether these identified conditions may be risk-factors for inappropriate use.

LIMITATIONS

First, the NHAMCS survey does not allow the assessment of the “appropriateness” of each prescription, which would likely require clinical data and longitudinal pharmacy records. Second, clinically relevant data such as dosage and morphine milligram equivalents are not recorded by NHAMCS. Third, NHAMCS chart abstraction may be associated with documentation errors³⁴. However, NHAMCS abstraction methodology did not change during our study time frame and is unlikely to invalidate our analyses of temporal trends. Finally, changes in NHAMCS drug coding and data element preclude exploration of ED opioid prescribing prior to 2006.

CONCLUSIONS

Although the perception of increasing ED opioid prescribing is not confirmed in our study, our findings support legislative and professional efforts towards reducing inappropriate opioid prescribing and better documentation of indications of opioid prescribing – in ED settings and elsewhere in the medical community.