Advertisement

Opioid Prescribing for Opioid-Naive Patients in Emergency Departments and Other Settings: Characteristics of Prescriptions and Association With Long-Term Use

Open AccessPublished:September 25, 2017DOI:https://doi.org/10.1016/j.annemergmed.2017.08.042

      Study objective

      We explore the emergency department (ED) contribution to prescription opioid use for opioid-naive patients by comparing the guideline concordance of ED prescriptions with those attributed to other settings and the risk of patients’ continuing long-term opioid use.

      Methods

      We used analysis of administrative claims data (OptumLabs Data Warehouse 2009 to 2015) of opioid-naive privately insured and Medicare Advantage (aged and disabled) beneficiaries to compare characteristics of opioid prescriptions attributed to the ED with those attributed to other settings. Concordance with Centers for Disease Control and Prevention (CDC) guidelines and rate of progression to long-term opioid use are reported.

      Results

      We identified 5.2 million opioid prescription fills that met inclusion criteria. Opioid prescriptions from the ED were more likely to adhere to CDC guidelines for dose, days’ supply, and formulation than those attributed to non-ED settings. Disabled Medicare beneficiaries were the most likely to progress to long-term use, with 13.4% of their fills resulting in long-term use compared with 6.2% of aged Medicare and 1.8% of commercial beneficiaries’ fills. Compared with patients in non-ED settings, commercial beneficiaries receiving opioid prescriptions in the ED were 46% less likely, aged Medicare patients 56% less likely, and disabled Medicare patients 58% less likely to progress to long-term opioid use.

      Conclusion

      Compared with non-ED settings, opioid prescriptions provided to opioid-naive patients in the ED were more likely to align with CDC recommendations. They were shorter, written for lower daily doses, and less likely to be for long-acting formulations. Prescriptions from the ED are associated with a lower risk of progression to long-term use.

      Introduction

       Background

      Because of a 4-fold increase in opioid prescriptions since 1999, long-term opioid use has become major public health issues in the United States.
      • Paulozzi L.J.
      • Jones C.M.
      • Mack K.A.
      • et al.
      Vital signs: overdoses of prescription opioid pain relievers—United States, 1999–2008.
      • Jones C.M.
      • Mack K.A.
      • Paulozzi L.J.
      Pharmaceutical overdose deaths, United States, 2010.
      Because opioids are frequently prescribed to patients discharged from emergency departments (EDs), it is important to understand the relationship between ED opioid prescribing for opioid-naive individuals and their risk of progressing to recurrent opioid use.
      • Hoppe J.A.
      • Kim H.
      • Heard K.
      Association of emergency department opioid initiation with recurrent opioid use.
      Some policymakers and members of the public perceive EDs to be a significant source of overprescription of opioids.
      • Kea B.
      • Fu R.
      • Lowe R.A.
      • et al.
      Interpreting the National Hospital Ambulatory Medical Care Survey: United States emergency department opioid prescribing, 2006-2010.

      American College of Emergency Physicians. Emergency physicians use new tool to detect drug-seekers in the ER. Available at: http://newsroom.acep.org/2013-07-10-Emergency-Physicians-Use-New-Tool-to-Detect-Drug-Seekers-in-the-ER. Accessed January 24, 2017.

      Weiner SG. Emergency medicine’s role in prescription opioid abuse. Available at: http://www.acepnow.com/article/emergency-medicines-role-in-prescription-opioid-abuse/. Accessed July 26, 2017.

      This perception may stem from the fact that many ED visits involve chronic or acute pain; adult patients reported pain as the primary symptom in 45% of ED visits.
      • Chang H.Y.
      • Daubresse M.
      • Kruszewski S.P.
      • et al.
      Prevalence and treatment of pain in EDs in the United States, 2000 to 2010.
      With so many patients in pain, it is not surprising that recent studies have found that 17% to 21% of all ED discharges included a prescription for opioids.
      • Kea B.
      • Fu R.
      • Lowe R.A.
      • et al.
      Interpreting the National Hospital Ambulatory Medical Care Survey: United States emergency department opioid prescribing, 2006-2010.
      • Hoppe J.A.
      • Nelson L.S.
      • Perrone J.
      • et al.
      Opioid prescribing in a cross section of US emergency departments.
      What is already known on this topic
      Dependency is a risk of opioid pain management.
      What question this study addressed
      In opioid-naive patients, are emergency department (ED) prescriptions more or less likely to progress to long-term opioid use than those from other clinical settings?
      What this study adds to our knowledge
      In this claims database of 5.2 million prescriptions in opioid-naive patients, opioid prescriptions written in the ED—compared with other locations—were of lesser dose and duration, and were approximately half as likely to lead to long-term use.
      How this is relevant to clinical practice
      Opioid prescriptions from the ED appear less likely to lead to long-term use than those from other clinical settings.

       Importance

      Despite the public health consequences of nonmedical and long-term prescription opioid use,
      • Chou R.
      • Turner J.A.
      • Devine E.B.
      • et al.
      The effectiveness and risks of long-term opioid therapy for chronic pain: a systematic review for a National Institutes of Health Pathways to Prevention Workshop.
      • Goldenberg D.L.
      • Clauw D.J.
      • Palmer R.E.
      • et al.
      Opioid use in fibromyalgia: a cautionary tale.
      • Schuchat A.
      • Houry D.
      • Guy Jr., G.P.
      New data on opioid use and prescribing in the United States.
      • Califf R.M.
      • Woodcock J.
      • Ostroff S.
      A proactive response to prescription opioid abuse.
      short-term use of these medications is clinically indicated in select settings.
      • Lovich-Sapola J.
      • Smith C.E.
      • Brandt C.P.
      Postoperative pain control.
      • Hegmann K.T.
      • Weiss M.S.
      • Bowden K.
      • et al.
      ACOEM practice guidelines: opioids for treatment of acute, subacute, chronic, and postoperative pain.
      With some rare exceptions, health care professionals do not intend for an initial opioid prescription issued for an acute pain episode to result in indefinite repeated prescriptions.
      • Waljee J.F.
      • Li L.
      • Brummett C.M.
      • et al.
      Iatrogenic opioid dependence in the United States: are surgeons the gatekeepers?.
      Unfortunately, a dearth of information exists about the progression of intended short-term use to an unintended prolonged pattern of use,
      • Callinan C.E.
      • Neuman M.D.
      • Lacy K.E.
      • et al.
      The initiation of chronic opioids: a survey of chronic pain patients.
      • Stumbo S.P.
      • Yarborough B.J.
      • McCarty D.
      • et al.
      Patient-reported pathways to opioid use disorders and pain-related barriers to treatment engagement.
      which may occur in 1.5% to 27% of opioid-naive patients after they receive an initial prescription.
      • Alam A.
      • Gomes T.
      • Zheng H.
      • et al.
      Long-term analgesic use after low-risk surgery: a retrospective cohort study.
      • Barnett M.L.
      • Olenski A.R.
      • Jena A.B.
      Opioid-prescribing patterns of emergency physicians and risk of long-term use.
      • Brummett C.M.
      • Waljee J.F.
      • Goesling J.
      • et al.
      New persistent opioid use after minor and major surgical procedures in US adults.
      • Clarke H.
      • Soneji N.
      • Ko D.T.
      • et al.
      Rates and risk factors for prolonged opioid use after major surgery: population based cohort study.
      • Deyo R.A.
      • Hallvik S.E.
      • Hildebran C.
      • et al.
      Association between initial opioid prescribing patterns and subsequent long-term use among opioid-naive patients: a statewide retrospective cohort study.
      • Johnson S.P.
      • Chung K.C.
      • Zhong L.
      • et al.
      Risk of prolonged opioid use among opioid-naive patients following common hand surgery procedures.
      • Shah A.
      • Hayes C.J.
      • Martin B.C.
      Characteristics of initial prescription episodes and likelihood of long-term opioid use—United States, 2006-2015.
      • Sun E.C.
      • Darnall B.D.
      • Baker L.C.
      • et al.
      Incidence of and risk factors for chronic opioid use among opioid-naive patients in the postoperative period.
      This is critically important because intentional short-term use is emerging as a previously underrecognized segue to unintended prolonged opioid use.
      • Callinan C.E.
      • Neuman M.D.
      • Lacy K.E.
      • et al.
      The initiation of chronic opioids: a survey of chronic pain patients.
      • Stumbo S.P.
      • Yarborough B.J.
      • McCarty D.
      • et al.
      Patient-reported pathways to opioid use disorders and pain-related barriers to treatment engagement.
      • Alam A.
      • Gomes T.
      • Zheng H.
      • et al.
      Long-term analgesic use after low-risk surgery: a retrospective cohort study.
      • Barnett M.L.
      • Olenski A.R.
      • Jena A.B.
      Opioid-prescribing patterns of emergency physicians and risk of long-term use.
      • Brummett C.M.
      • Waljee J.F.
      • Goesling J.
      • et al.
      New persistent opioid use after minor and major surgical procedures in US adults.
      • Clarke H.
      • Soneji N.
      • Ko D.T.
      • et al.
      Rates and risk factors for prolonged opioid use after major surgery: population based cohort study.
      • Deyo R.A.
      • Hallvik S.E.
      • Hildebran C.
      • et al.
      Association between initial opioid prescribing patterns and subsequent long-term use among opioid-naive patients: a statewide retrospective cohort study.
      • Shah A.
      • Hayes C.J.
      • Martin B.C.
      Characteristics of initial prescription episodes and likelihood of long-term opioid use—United States, 2006-2015.
      • Sun E.C.
      • Darnall B.D.
      • Baker L.C.
      • et al.
      Incidence of and risk factors for chronic opioid use among opioid-naive patients in the postoperative period.
      • Hooten W.M.
      • St Sauver J.L.
      • McGree M.E.
      • et al.
      Incidence and risk factors for progression from short-term to episodic or long-term opioid prescribing: a population-based study.
      One of the 5 key questions proposed in the 2016 Centers for Disease Control and Prevention (CDC) guideline for prescribing opioids for chronic pain was to determine the effects of opioid therapy for acute pain on long-term opioid use.
      • Dowell D.
      • Haegerich T.M.
      • Chou R.
      CDC guideline for prescribing opioids for chronic pain—United States, 2016.
      The goal of these guidelines is to improve opioid prescribing practices to ensure patients have access to safer treatment while reducing the risks of nonmedical use and overdose.
      Limited research has been conducted to date on prescribing practices for acute pain that limit risk of long-term opioid use. Current recommendations are to prescribe the lowest effective dose and quantity needed for the expected duration of pain. With new guidelines and ED clinicians facing the challenge of patients seeking help for uncontrolled pain, it is natural to ask whether and how prescribing in the ED compares with that in other settings. The guidelines were not published until after the study period; our goal in using their recommendations is not to determine the adherence rates to the CDC guidelines per se, but rather to use them as a source of reasonable and evidence-based standards for comparing prescriptions attributed to different settings.

       Goals of This Investigation

      We used administrative claims data to compare characteristics of opioid prescriptions written for opioid-naive patients discharged from the ED and other settings and evaluated the risk of long-term use of prescription opioids by addressing these questions: To what extent are opioid prescriptions issued to opioid-naive patients in the ED or non-ED settings concordant with best practices on the number of days supplied, the daily dose of the prescription, and the number of prescriptions filled for long-acting or extended-release formulations? For opioid-naive patients, what is the difference in the rate of progression to long-term opioid use after an initial prescription in the ED compared with a non-ED setting?

      Materials and Methods

      We adhered to the Reporting of Studies Conducted Using Observational Routinely-Collected Health Data statement.
      • Benchimol E.I.
      • Smeeth L.
      • Guttmann A.
      • et al.
      The Reporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) statement.

       Study Design and Setting

      We conducted an analysis of administrative claims data from January 1, 2009, through December 31, 2015, from the OptumLabs Data Warehouse, a database composed of privately insured and Medicare Advantage enrollees throughout the United States; more than 35 million unique people had both medical and prescription drug coverage at some time during the study period.
      • Wallace P.J.
      • Shah N.D.
      • Dennen T.
      • et al.
      Optum Labs: building a novel node in the learning health care system.
      OptumLabs Data Warehouse contains longitudinal health information on enrollees from geographically diverse regions across the United States, with the greatest representation from the Southern and Midwestern states.

      Optum. Real world health care experiences from over 150 million unique individuals since 1993. Available at: https://www.optum.com/content/dam/optum/resources/productSheets/5302_Data_Assets_Chart_Sheet_ISPOR.pdf. Accessed June 6, 2017.

      It includes adjudicated claims for all health care services incurred by beneficiaries and submitted to the insurance company for payment. The included plans provide coverage for professional, facility, laboratory, and pharmacy claims. Administrative data include beneficiary sex, race or ethnicity, age, and dates of coverage. Medical claims include International Classification of Diseases, Ninth Revision (ICD-9) and ICD-10 procedure and diagnosis codes, Healthcare Common Procedure Coding System procedure codes, site of service codes, and provider specialty codes.
      • Bellolio M.F.
      • Sangaralingham L.R.
      • Schilz S.R.
      • et al.
      Observation status or inpatient admission: impact of patient disposition on outcomes and utilization among emergency department patients with chest pain.
      The commercial population covered by OptumLabs Data Warehouse is similar to the US population of commercially insured people in age, race or ethnicity, and sex. Further detail is provided in Appendix E1 (available online at http://www.annemergmed.com).
      This study was determined to be exempt from review by the Mayo Clinic Institutional Review Board.
      We identified opioid prescriptions filled between January 1, 2009, and December 31, 2015, by beneficiaries of all ages who had both medical and pharmacy coverage. To focus our attention on people for whom long-term opioid use is an important risk, we examined claims in the 3 months before each opioid fill and excluded prescription fills by beneficiaries who had either any hospice claims or at least 2 physician visits with a cancer diagnosis in those 3 months. (See Appendix E2, available online at http://www.annemergmed.com, for codes used to identify cancer and hospice beneficiaries.)
      We focused on opioid prescription fills among opioid-naive beneficiaries, defined as having no opioid fills in the previous 6 months. As such, we excluded fills for beneficiaries with less than 6 months of insurance enrollment before the index fill and those who had any opioid fills during those 6 months. We limited our cohort to the first opioid-naive fill for each beneficiary—the index fill. Our final cohort consists of 5.2 million index fills. A cohort flow chart is provided in Figure 1.
      We identified National Drug Codes for all opioids available during any part of the study period. For this study, we classified tramadol as an opioid. The complete list of medications classified as opioids for this study is provided in Appendix E3, available online at http://www.annemergmed.com.
      We used conversion factors provided by the CDC to convert the daily prescribed opioid dose to milligrams of morphine equivalents.

      Centers for Disease Control and Prevention. CDC compilation of opioid analgesic formulations with morphine milligram equivalent conversion factors, 2015 version. Available at: http://www.pdmpassist.org/pdf/BJA_performance_measure_aid_MME_conversion.pdf. Accessed May 3, 2016.

      Prescriptions written for 1,000 mg of morphine equivalents or more per day were excluded from the analysis (n=22,607; 0.04% of opioid fills by people with medical and prescription coverage) as extreme outliers.
      If multiple doses of the same opioid were filled on the same day with the same prescriber identification, we merged those fills and calculated a combined daily dose in milligrams of morphine equivalents. In rare cases (N=56,845 beneficiaries and 114,507 opioid-naive fills), beneficiaries filled prescriptions for multiple different opioids or the same opioid with different prescribers on the same day. In these cases, each fill was included separately in the analysis. For this reason, the opioid fill is the unit of analysis rather than the beneficiary. Complete details are provided in Appendix E4, available online at http://www.annemergmed.com.
      We used information from medical claims in the 30 days before and including the date of the index fill to determine the most likely source of the prescription. Detailed information is provided in Appendix E5, available online at http://www.annemergmed.com. We classified the most likely source of each index fill as ED visit only; non-ED visit only, which combines inpatient, outpatient, ambulatory surgery, and dental or accidental dental; and unknown source, which includes prescriptions in which both ED and non-ED services were provided on the same day (4.3% of all prescriptions), as well as prescriptions for which there was no medical claim in the 30 days leading up to and including the prescription fill date. A substantial proportion of prescriptions had no visits in the previous 30 days: 22% for the commercial population and 10% for the Medicare population; this rate is similar to that of a study using a different source of commercial claims, which found 28% of opioid fills unmatched, with a look-back period of 2 weeks.
      • Liu Y.
      • Logan J.E.
      • Paulozzi L.J.
      • et al.
      Potential misuse and inappropriate prescription practices involving opioid analgesics.
      Some of these prescriptions were likely written by dentists, who have been estimated to write 6.4% of opioid prescriptions.
      • Levy B.
      • Paulozzi L.
      • Mack K.A.
      • et al.
      Trends in opioid analgesic-prescribing rates by specialty, US, 2007-2012.
      We did not observe most dental visits because dentistry is not included in medical insurance benefits. In our sample of fills to opioid-naive patients, 7.0% of fills with a known prescriber specialty were written by a dentist or dental specialist. We present the results for prescriptions with unknown source throughout, but do not focus on the interpretation of this group of prescriptions.
      All patients in the study were opioid naive, with no insurance-paid opioid fills in the previous 6 months, which decreases the variability in the dose and duration of opioids that would be considered appropriate. It is likely that, regardless of the setting, people receiving a new opioid prescription with no previous recent fills were either experiencing acute pain or were experiencing an acute exacerbation of a chronic problem. Appropriate prescribing practices for acute pain or acute exacerbations of chronic problems are likely similar across settings.
      Administrative information was used to determine beneficiary age, race or ethnicity, sex, and type of insurance (commercial versus Medicare). We used type of insurance and beneficiary age to identify 3 key patient populations: commercially insured of all ages (commercial), people eligible for Medicare because of age (aged Medicare), and people with Medicare coverage who were younger than 65 years but qualified for Medicare because of long-term disability, end-stage renal disease, or other serious conditions (disabled Medicare).
      To assess patient illness burden, we used the Elixhauser comorbidity measures, a set of 31 measures indicating presence of comorbidities associated with increased risk of mortality.
      • Elixhauser A.
      • Steiner C.
      • Harris D.R.
      • et al.
      Comorbidity measures for use with administrative data.
      We used ICD-9 and ICD-10 codes to identify these comorbidities in the 6 months before the index fill.
      • Quan H.
      • Sundararajan V.
      • Halfon P.
      • et al.
      Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.
      We required a diagnosis to be present on 1 inpatient stay or 2 separate outpatient visits. Each index fill was categorized with flags for the Elixhauser comorbidities and whether the beneficiary had any claims in the 6 months before the fill.

       Outcome Measures

      We assessed concordance with best practice in opioid prescribing as summarized in the CDC guidelines.
      • Dowell D.
      • Haegerich T.M.
      • Chou R.
      CDC guideline for prescribing opioids for chronic pain—United States, 2016.
      One recommendation states that for acute pain, “[t]hree days or less will often be sufficient; more than seven days will rarely be needed.” The number of days supplied was extracted from the pharmacy claim. We coded binary variables for prescriptions for more than 3 days and more than 7 days’ supply.
      Another recommendation states that physicians should write prescriptions for the lowest effective dose. For patients using opioids for chronic pain, the CDC urged caution if increasing doses to greater than 50 mg of morphine equivalents per day; doses greater than 90 mg of morphine equivalents per day were suggested to be appropriate only for pain specialists to prescribe. Binary variables indicated whether an index fill was written for more than 50 or more than 90 mg of morphine equivalents per day. Because these would be exceptionally high doses for opioid-naive patients, we expected to identify a small proportion of fills in this dose range.
      A third recommendation from the guideline states that extended-release or long-acting opioids should not be used when opioid therapy is started. We anticipated that prescriptions for extended-release and long-acting formulations would be rare in a cohort of opioid-naive patients.
      To determine the risk of long-term use of opioids, we examined opioid fills in the 12 months after the index fill. Long-term opioid use was defined with the Consortium to Study Opioid Risks and Trends criteria: episodes of opioid prescribing lasting longer than 90 days and 120 or more total days’ supply or 10 or more prescriptions in the year after the index fill.
      • Von Korff M.
      • Saunders K.
      • Thomas Ray G.
      • et al.
      De facto long-term opioid therapy for noncancer pain.
      Only beneficiaries with at least 12 months of continuous enrollment after the index fill were included in this analysis. Because this analysis was performed on a different population than the analyses of guideline concordance (which was allowed to have less than 12 months of follow-up), we repeated all guideline concordance analyses using just the cohort with at least 12 months of follow-up. We found no difference in the results of these analyses. A full comparison of results is presented in Appendix E11, available online at http://www.annemergmed.com.

       Primary Data Analysis

      Distributions of demographic and fill characteristics were compared with χ2 goodness-of-fit tests, one-way ANOVA, and a Wald test for equality of coefficients after Poisson regression; see Appendix E7, available online at http://www.annemergmed.com, for further details. We report 95% confidence intervals (CIs) for selected results in the text; all CIs are presented in tables or appendices.
      Logistic regression was used to measure the association between the source of the initial prescription and study outcomes. Prescription source (ED, non-ED, and unknown) was interacted with beneficiary type (commercial, aged Medicare, and disabled Medicare) in the model. Covariates included year (continuous), age, age squared, age cubed, female sex, race or ethnicity, indicators for each Elixhauser comorbidity, and whether the beneficiary had any medical claims in the 6 months before the fill. Complete results from logistic regression models are provided in Appendix E8, available online at http://www.annemergmed.com. Adjusted probabilities of outcomes were calculated for each beneficiary type and prescription source. Risk ratios were generated from these probabilities, with 95% confidence intervals calculated with the delta method.
      • Dowd B.E.
      • Greene W.H.
      • Norton E.C.
      Computation of standard errors.
      Stata/MP (version 14.2; StataCorp, College Station, TX) was used for all analyses.
      StataCorp
      Stata: Release 14.
      We performed a supplementary analysis treating time as a categorical variable to allow estimation of time trends in the proportion of prescriptions concordant with guidelines. Results are described briefly in the text and presented in full in Appendix E11, available online at http://www.annemergmed.com.

      Results

      We identified 5,241,948 naive opioid fills that met inclusion criteria (Table 1). The rate of naive opioid fills per person-year was similar across the 3 populations, with 0.07 naive fills per person-year for the commercial, 0.07 for the aged Medicare, and 0.06 for the disabled Medicare populations (difference significant at P<.001, but a minor clinical difference). We report the total opioid fills per person-year (including both naive and non-naive fills, and including buprenorphine and methadone fills) for context. Although the rates of opioid-naive fills were similar across the 3 groups, the disabled Medicare population filled 7.15 total opioid prescriptions per person-year, 4 times more than the aged Medicare population and 8 times more than the commercial population.
      Table 1Characteristics of beneficiaries and opioid fills (unadjusted/crude rates).
      Beneficiary/Fill CharacteristicCommercialAged MedicareDisabled Medicare
      Value(95% CI)Value(95% CI)Value(95% CI)
      Beneficiary population
      Total person-years of coverage (all OLDW beneficiaries with medical and prescription coverage)69,664,46510,078,9481,369,427
      Unique beneficiaries (all OLDW beneficiaries with medical and prescription coverage)31,840,3823,679,349590,329
      Unique beneficiaries with a naive opioid fill4,447,662655,66780,982
      Opioid fills
      Number of naive opioid fills4,496,928662,57382,049
      Proportion of all naive opioid fills, %85.8(85.8–85.8)12.6(12.6–12.7)1.6(1.6–1.6)
      Naive opioid fills per covered person per year of insurance coverage0.07(0.07–0.07)0.07(0.07–0.07)0.06(0.06–0.06)
      Total opioid fills per covered person per year of insurance coverage (naive+non-naive fills)0.86(0.86–0.86)1.77(1.77–1.77)7.15(7.14–7.15)
      Characteristics of beneficiaries with naive opioid fills
      Sex, %
      Female52.7(52.6–52.7)57.2(57.0–57.3)51.2(50.9–51.6)
      Age, y
      Median (IQR)38(25–51)73(68–79)57(51–61)
      Race/ethnicity, %
      White71.9(71.9–72.0)74.7(74.6–74.8)64.6(64.3–65.0)
      Hispanic10.6(10.5–10.6)6.9(6.8–6.9)9.4(9.2–9.6)
      Black10.5(10.5–10.5)11.3(11.2–11.3)20.2(19.9–20.4)
      Asian3.7(3.7–3.7)2.4(2.4–2.5)1.7(1.6–1.8)
      Unknown3.3(3.3–3.3)4.8(4.7–4.8)4.1(4.0–4.2)
      Prescription source, %
      ED13.3(13.3–13.3)11.7(11.6–11.8)17.4(17.1–17.7)
      Not ED61.1(61.0–61.1)73.4(73.3–73.6)66.8(66.5–67.1)
      Unknown25.6(25.6–25.7)14.9(14.8–14.9)15.8(15.6–16.1)
      Insurance coverage after fill, %
      Covered 3 mo after fill90.3(90.3–90.3)93.0(92.9–93.0)93.9(93.8–94.1)
      Covered 6 mo after fill81.3(81.3–81.4)87.4(87.3–87.4)87.8(87.6–88.1)
      Covered 12 mo after fill68.2(68.1–68.2)79.4(79.3–79.5)79.2(78.9–79.5)
      Medical claims before fill, %
      Had any claims in 3 mo before fill85.5(85.5–85.6)95.5(95.5–95.6)95.1(94.9–95.2)
      Had any claims in 6 mo before fill90.6(90.6–90.6)97.5(97.5–97.6)97.2(97.1–97.4)
      Elixhauser comorbidities, 6 mo before fill
      Mean number of comorbidities0.25(0.25–0.25)1.39(1.38–1.39)1.43(1.42–1.44)
      Any comorbidity, %16.0(16.0–16.0)57.8(58.0–58.0)59.1(58.7–59.4)
      Characteristics of naive index opioid fills
      Drug filled (5 most common; further detail in Appendix E6, available online at http://www.annemergmed.com), %
      Hydrocodone SA58.9(58.9–59.0)49.2(49.1–49.4)49.7(49.3–50.0)
      Oxycodone SA18.8(18.8–18.8)16.6(16.6–16.7)19.4(19.1–19.7)
      Tramadol SA8.7(8.7–8.7)20.2(20.1–20.3)18.7(18.5–19.0)
      Codeine9.8(9.8–9.9)8.6(8.5–8.7)7.2(7.0–7.4)
      Propoxyphene2.3(2.3–2.3)2.4(2.4–2.5)2.0(1.9–2.1)
      DEA class, %
      Non–schedule II (short acting)73.0(72.9–73.0)71.5(71.4–71.6)69.5(69.1–69.8)
      Schedule II (short acting)26.5(26.5–26.5)27.6(27.5–27.8)28.6(28.3–28.9)
      Long acting (any schedule)0.5(0.5–0.5)0.9(0.8–0.9)1.9(1.8–2.0)
      Dose and supply, %
      >50 MME per day19.9(19.9–19.9)17.0(16.9–17.1)17.8(17.5–18.0)
      >90 MME per day6.0(5.9–6.0)5.2(5.1–5.2)5.6(5.5–5.8)
      >3 days’ supply57.4(57.4–57.5)68.0(67.9–68.1)67.5(67.2–67.8)
      >7 days’ supply14.4(14.4–14.4)30.6(30.4–30.7)33.9(33.6–34.3)
      Continued to long-term use, %
      Long-term opioid use (>10 fills or 120 days supplied in 1 y)1.8(1.8–1.8)6.2(6.1–6.3)13.4(13.1–13.6)
      CI, Confidence interval; OLDW, OptumLabs Data Warehouse; IQR, interquartile range; SA, sustained action; DEA, US Drug Enforcement Agency; MME, milligrams of morphine equivalents.
      See Appendix E7, available online at http://www.annemergmed.com, for details on calculation of CIs.

       Main Results

      The proportion of prescriptions attributed to the ED was 11.7% in aged Medicare, 13.3% in commercial, and 17.4% in the disabled Medicare populations (Table 1). The most common medication prescribed for naive fills across all treatment settings was hydrocodone (composing 58.9% of commercial fills, 49.2% of aged Medicare, and 49.7% of disabled Medicare) (Table 1).
      Guideline concordance of prescriptions is reported in Tables 1 and 2 and in Figure 2A to F. Unadjusted rates of guideline concordance by beneficiary population are reported in Table 1, adjusted rates of concordance by beneficiary population and treatment setting (ED, non-ED, and unknown) in Table 2, and adjusted risk ratios in Figure 2A to F. A table of risk ratios with 95% confidence intervals is provided in Appendix E10, available online at http://www.annemergmed.com.
      Table 2Regression-adjusted outcomes: proportions.
      OutcomeCommercialAged MedicareDisabled Medicare
      Adjusted Proportion, %(95% CI)Adjusted Proportion, %(95% CI)Adjusted Proportion, %(95% CI)
      Prescription for >3 days’ supply
      Non-ED65.9(65.9–66.0)74.6(74.4–74.7)76.8(76.4–77.1)
      Unknown47.2(47.1–47.3)56.3(56.0–56.6)61.9(61.1–62.7)
      ED37.0(36.9–37.1)41.6(41.3–41.9)36.7(36.0–37.5)
      Prescription for >7 days’ supply
      Non-ED19.1(19.0–19.1)36.7(36.5–36.8)42.5(42.1–42.9)
      Unknown7.7(7.7–7.8)20.4(20.1–20.7)28.2(27.4–28.9)
      ED3.1(3.1–3.1)4.5(4.3–4.6)3.9(3.6–4.2)
      Prescription for >50 MME
      Non-ED22.8(22.7–22.8)17.7(17.6–17.8)18.0(17.7–18.3)
      Unknown17.0(16.9–17.1)16.4(16.1–16.6)19.3(18.6–20.0)
      ED14.3(14.2–14.3)13.0(12.8–13.2)13.9(13.3–14.5)
      Prescription for >90 MME
      Non-ED7.2(7.2–7.2)5.5(5.4–5.5)5.8(5.6–6.0)
      Unknown4.4(4.4–4.5)4.8(4.6–4.9)5.9(5.5–6.3)
      ED3.3(3.3–3.4)3.7(3.5–3.8)3.3(3.0–3.6)
      Prescription for long acting/extended-release formulation
      Non-ED0.5(0.5–0.5)0.9(0.9–0.9)1.7(1.6–1.8)
      Unknown0.3(0.3–0.3)0.7(0.7–0.8)1.9(1.7–2.1)
      ED0.0(0.0–0.0)0.1(0.1–0.2)0.2(0.1–0.3)
      Progression to long-term use
      Non-ED2.0(2.0–2.0)6.9(6.8–7.0)14.8(14.4–15.1)
      Unknown1.3(1.2–1.3)4.8(4.7–5.0)13.3(12.6–13.9)
      ED1.1(1.1–1.1)3.1(2.9–3.2)6.2(5.7–6.6)
      Adjusted proportions calculated after logistic regression that included beneficiary category (commercial, aged Medicare, and disabled Medicare); year of fill (continuous); beneficiary age, age squared, and age cubed; indicators for each Elixhauser comorbidity and whether the beneficiary had any medical claims in the 6 months before the fill; female sex; and race/ethnicity.
      Figure thumbnail gr2
      Figure 2Risk ratios for outcomes by source of prescription. Risk ratios with non-ED prescription source as the reference category; bars indicate 95% CIs. Logistic regression with binary outcomes was performed, with independent variables representing beneficiary characteristics: beneficiary category (commercial, aged Medicare, disabled Medicare); year of fill (continuous); beneficiary age, age squared, and age cubed; indicators for each Elixhauser comorbidity and whether the beneficiary had any medical claims in the 6 months before the fill; female sex; and race/ethnicity. Adjusted proportions meeting each outcome were calculated for each beneficiary group with Stata’s marginal effects commands. Risk ratios were calculated from these adjusted proportions, with 95% CIs calculated with Stata’s nlcom command, which uses the delta method to produce standard errors. ER, Extended release; Com, commercial population; Mcr, Medicare; Disab., disabled.
      It was common for opioid-naive beneficiaries to fill prescriptions exceeding 3 days’ supply, with unadjusted proportions ranging from 57.4% among commercial to 68.0% among disabled Medicare beneficiaries (Table 1). Prescriptions from non-ED settings were the most likely to exceed 3 days: 65.9% among commercial beneficiaries, 74.6% for aged Medicare, and 76.8% for disabled Medicare (Table 2; regression adjusted). Prescriptions from the ED were 44% (commercial: adjusted risk ratio 0.56 [95% CI 0.56, 0.56]) to 52% (disabled Medicare: adjusted risk ratio 0.48 [95% CI 0.47, 0.49]) less likely to exceed 3 days’ supply than those from non-ED settings (Figure 2A).
      Prescriptions from the ED were also less likely to exceed 7 days’ supply compared with non-ED prescriptions. In the ED, adjusted proportions of prescriptions exceeding 7 days were 84% to 91% lower than in the non-ED setting (adjusted risk ratio comparing ED to non-ED was 0.16 [95% CI 0.16, 0.16] in the commercial population, 0.12 [95% CI 0.12, 0.13] in the aged Medicare population, and 0.09 [95% CI 0.08, 0.10] in the disabled Medicare population; adjusted percentages in Table 2; risk ratios in Figure 2B).
      Prescriptions for high doses of milligrams of morphine equivalents were common; 17.0% to 19.9% exceeded 50 mg of morphine equivalents per day, and 5.2% to 6.0% exceeded 90 mg of morphine equivalents per day (unadjusted proportions; see Table 1 for details). Prescriptions from the ED were 23% to 37% less likely to exceed 50 mg of morphine equivalents (Figure 2C; regression adjusted) and 33% to 54% less likely to exceed 90 mg of morphine equivalents than those attributed to non-ED settings (Figure 2D; regression adjusted).
      Prescriptions for long-acting or extended-release formulations were rare among opioid-naive beneficiaries, with unadjusted percentages ranging from 0.5% of commercial to 1.9% of disabled Medicare prescriptions.
      In the regression-adjusted analysis, prescriptions from the ED were 86% to 92% less likely to be written for long-acting or extended-release formulations than those attributed to non-ED settings (risk ratios ranged from 0.08 [0.07, 0.09] in the commercial population to 0.14 [95% CI 0.11, 0.17] in the aged Medicare population) (Figure 2E).
      In all beneficiary populations, prescriptions attributed to the ED were less likely to progress to long-term opioid use. For beneficiaries treated in the ED, 1.1% of commercial beneficiaries, 3.1% of aged Medicare, and 6.2% of disabled Medicare progressed to long-term use (Table 2). Commercial beneficiaries treated in the ED were 46% (adjusted risk ratio 0.54 [95% CI 0.53, 0.56]) less likely to progress to long-term use than commercial beneficiaries treated in non-ED settings. Aged Medicare beneficiaries were 56% (adjusted risk ratio 0.44 [95% CI 0.42, 0.46]) and disabled Medicare were 58% (adjusted risk ratio 0.42 [0.39, 0.45]) less likely to progress to long-term use if receiving a prescription in the ED compared with non-ED settings (Figure 2F).
      To determine whether guideline-concordant prescriptions were associated with a lower risk of progression to long-term use of opioids, we included a binary variable indicating whether the prescription met all guidelines in a regression of long-term use. Across nearly all care settings and beneficiary populations, a nonconcordant prescription was associated with a greater risk of progression to long-term opioid use (adjusted risk ratios range from 1.09 [95% CI 0.93, 1.26] for disabled Medicare beneficiaries treated in the ED to 5.42 [95% CI 4.79, 6.05] for aged Medicare beneficiaries treated in unknown settings; full regression results in Appendix E9, available online at http://www.annemergmed.com; all beneficiary/treatment setting comparisons showed statistically significant increases in risk except disabled Medicare/ED). See Table 3 for all risk ratios and confidence intervals.
      Table 3Risk ratios comparing risk of progression to long-term opioid use with nonconcordant versus fully concordant prescriptions.
      Risk of Progression to Long-Term UseCommercialAged MedicareDisabled Medicare
      Risk Ratio95% CI
      Delta method standard errors. Reference: prescription fully guideline concordant (≤3 days and ≤50 MME per day and not long acting).
      Risk Ratio95% CI
      Delta method standard errors. Reference: prescription fully guideline concordant (≤3 days and ≤50 MME per day and not long acting).
      Risk Ratio95% CI
      Delta method standard errors. Reference: prescription fully guideline concordant (≤3 days and ≤50 MME per day and not long acting).
      Nonconcordant prescriptions treatment setting
      Non-ED3.75(3.61–3.89)4.42(4.18–4.66)4.28(3.79–4.77)
      Unknown4.44(4.19–4.70)5.42(4.79–6.05)4.36(3.52–5.19)
      ED1.24(1.16–1.32)1.30(1.18–1.42)1.09(0.93–1.26)
      Delta method standard errors. Reference: prescription fully guideline concordant (≤3 days and ≤50 MME per day and not long acting).
      In every year of the study, across all 3 populations and all measures of guideline concordance, prescriptions attributed to the non-ED setting were more likely to exceed guideline limits than those attributed to the ED.
      The proportion of prescriptions exceeding 3 or 7 days and the proportion written for long-acting formulations were relatively stable during the study. However, the proportion of prescriptions written for large doses decreased from 2009 to 2011. In prescriptions exceeding 50 mg of morphine equivalents, this decrease ranged from 20% to 50%, depending on the treatment setting and beneficiary population. See Appendix E11, available online at http://www.annemergmed.com, for full details.
      The proportion of prescriptions progressing to long-term opioid use decreased during the study period in all beneficiary populations and treatment settings. The largest decrease in progression to long-term use was in the aged Medicare population treated in the ED, from 2.1% in 2009 to 1.2% in 2015 (–42%); the smallest decrease was in the commercial population treated in non-ED settings, from 2.7% in 2009 to 2.4% in 2015 (–11%).

      Limitations

      A limitation of this study is the large number of prescriptions for which we were unable to assign a likely source. As noted, this issue has been found in a study using a different source of commercial claims.
      • Liu Y.
      • Logan J.E.
      • Paulozzi L.J.
      • et al.
      Potential misuse and inappropriate prescription practices involving opioid analgesics.
      We estimate that 7% to 10% of the prescriptions in this study were written by dentists, leaving 5% to 10% of the prescriptions unexplained. Further research is needed to clarify whether some of these prescriptions may indicate problematic prescribing practices in which a physician writes a prescription without seeing the patient.
      The data available in administrative claims did not allow us to attribute prescriptions to visits with complete certainty. Our method of attribution used the information available to assign a most likely source of prescriptions.
      This study is limited to prescription fills submitted for insurance payment and to a population of commercial and Medicare Advantage beneficiaries. We did not include the uninsured or people with Medicaid or fee-for-service Medicare.
      Given our study design, we were unable to evaluate whether the risk of long-term use was causally related to the prescription’s guideline concordance. Randomized controlled studies are unlikely to meet ethical guidelines for responsible research practice, given the weight of evidence on the risks of opioid use. The recent work of Barnett et al
      • Barnett M.L.
      • Olenski A.R.
      • Jena A.B.
      Opioid-prescribing patterns of emergency physicians and risk of long-term use.
      suggested that observational studies of provider variability may be a source of quasi-random variation that could be used to study the effect of prescription characteristics on patient outcomes.
      The CDC guidelines had not been released during the period we studied. Therefore, this study should not be understood as measuring physician adherence to the CDC guidelines, but rather as measuring physician practice in reference to evidence guiding best practices in prescribing.

      Discussion

      Compared with that in non-ED settings, opioid prescriptions provided to opioid-naive patients in the ED were more likely to align with CDC recommendations for duration of these prescriptions for acute pain. More than 40% of prescriptions filled by disabled Medicare patients treated in non-ED settings exceeded 7 days’ supply. In contrast, less than 5% of ED-attributed prescriptions exceeded 7 days in all 3 patient populations.
      ED prescription durations in our study were similar to those in the study by Weiner et al,
      • Weiner S.G.
      • Baker O.
      • Rodgers A.F.
      • et al.
      Opioid prescriptions by specialty in Ohio, 2010-2014.
      which used Ohio Prescription Drug Monitoring Program data to study ED prescribing patterns from 2010 to 2014. Observations from this study showed rates of prescriptions exceeding 3 days that were similar to those found in the ED prescriptions in our study: 34% across their study period compared with 38% (commercial), 42% (aged Medicare), and 37% (disabled Medicare) in our study.
      We expected a small number of prescriptions in excess of 50 or 90 mg of morphine equivalents to be issued to opioid-naive patients in any setting. These dose levels were selected by the CDC guideline writers as high doses for people with some degree of tolerance because of long-term opioid use. The high rates of prescriptions exceeding these levels were not anticipated. In non-ED settings, 1 in 6 prescriptions written for disabled Medicare patients and 1 in 5 for commercial patients were for more than 50 mg of morphine equivalents per day. The rates of prescriptions for more than 90 mg of morphine equivalents per day ranged up to 7% in the commercial population treated in non-ED settings. These high doses prescribed to people with no previous opioids in 6 or more months could be considered an indication that the individual had previously received opioids for a similar condition at a similar high dose. However, this may not be a safe practice for the majority of patients, in part because of the rapid resolution of opioid tolerance.
      • Strang J.
      • McCambridge J.
      • Best D.
      • et al.
      Loss of tolerance and overdose mortality after inpatient opiate detoxification: follow up study.
      • White J.M.
      • Irvine R.J.
      Mechanisms of fatal opioid overdose.
      As expected, we found low overall rates of prescriptions for long-acting and extended-release opioids. However, more than 2% of prescriptions written for the disabled Medicare population were for these formulations. This is a safety concern for opioid-naive patients because initiating opioid treatment with long-acting or extended-release formulations increases in the risk of overdose compared with immediate-release opioids.
      • Miller M.
      • Barber C.W.
      • Leatherman S.
      • et al.
      Prescription opioid duration of action and the risk of unintentional overdose among patients receiving opioid therapy.
      The rate of continued use observed in this Medicare Advantage population is higher than that reported by Barnett et al,
      • Barnett M.L.
      • Olenski A.R.
      • Jena A.B.
      Opioid-prescribing patterns of emergency physicians and risk of long-term use.
      who followed Medicare fee-for-service beneficiaries who were opioid naive in the previous 6 months and received an opioid prescription in the ED. They found that 1.2% to 1.5% of these patients progressed to long-term use, defined as 180 days supplied in 12 months after the initial prescription, but excluding the 30 days after the prescription. More than 3% of our Medicare Advantage beneficiaries treated in the ED progressed to long-term use; however, our definition of long-term use was not as strict as the definition used by Barnett et al
      • Barnett M.L.
      • Olenski A.R.
      • Jena A.B.
      Opioid-prescribing patterns of emergency physicians and risk of long-term use.
      (120 days rather than 180 days in their study), and we did not exclude the 30 days after the prescription.
      Although we did not set out to compare guideline concordance across beneficiary populations, we found that the disabled Medicare population was more likely to receive prescriptions exceeding 3 and 7 days, and was more likely to receive long-acting formulations compared with the aged Medicare and commercial populations. Future investigations comparing these populations are needed.
      In conclusion, opioid prescriptions attributed to the ED for opioid-naive patients were more likely to adhere to best practices for opioid prescribing to opioid-naive patients compared with those attributed to non-ED settings. ED prescriptions were shorter in duration, written for lower doses, and less likely to be for long-acting or extended-release formulations. These prescriptions had a lower risk of progression to long-term opioid use.
      Across all treatment settings and patient populations, guideline concordance was associated with a lower risk of long-term use. Among opioid-naive patients, greater guideline concordance in the ED may have been an important driver that helped mitigate the progression to long-term opioid use. Future research may explore why ED prescriptions for opioid-naive patients are more likely to be guideline concordant, with the hope of replicating that success in other settings in which opioid-naive patients are treated.
      The authors acknowledge the Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery for making this work possible.

      Appendix E1

       Comparison of demographics: OptumLabs Data Warehouse and US commercially insured and Medicare Advantage

      Table E1Demographic comparison: commercial population.
      CharacteristicPrivately Insured Population <65 Years, %
      US Privately InsuredOLDW Commercial
      Race/ethnicity
      Source: Kaiser Family Foundation/Urban Institute analysis of CPS ASEC, 2011 data.
      White7273
      Asian65
      Hispanic1211
      Black1011
      American Indian<6NA
      Women
      Source: Annual Social and Economic Supplement of the Current Population Survey (CPS ASEC), 2015 data.
      5049
      Age, y
      Source: Annual Social and Economic Supplement of the Current Population Survey (CPS ASEC), 2015 data.
      0–172424
      18–241110
      25–341616
      35–441518
      45–541718
      55–5998
      60–6476
      OLDW commercial includes people with both medical and prescription coverage; excludes people with unknown race/ethnicity, year of birth, or sex.
      Source: Kaiser Family Foundation/Urban Institute analysis of CPS ASEC, 2011 data.
      Source: Annual Social and Economic Supplement of the Current Population Survey (CPS ASEC), 2015 data.
      Table E2Demographic comparison: Medicare Advantage population.
      CharacteristicMedicare Advantage, %
      US Medicare AdvantageOLDW Medicare Advantage
      Race/ethnicity
      White7676
      Black1112
      Hispanic89
      Other63
      Women5557
      OLDW Medicare Advantage includes people with both medical and prescription coverage; excludes people with unknown race/ethnicity or sex. US Medicare Advantage source: Kaiser Family Foundation, 2015 data.

      Appendix E2

       Codes used to identify exclusions for cancer and hospice care

      To limit the analysis to patients with an important risk of long-term opioid use, we excluded patients with active cancer diagnoses or hospice service use.
      For the cancer population, we examined claims for evaluation and management services in the 3 months before the index opioid fill. We identified patients with at least 2 claims on separate days that included a cancer diagnosis. We used Elixhauser comorbidity diagnoses for metastatic cancer or for solid tumors without metastases.
      • Quan H.
      • Sundararajan V.
      • Halfon P.
      • et al.
      Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.
      Evaluation and management codes: 99024, 99201, 99202, 99203, 99204, 99205, 99211, 99212, 99213, 99214, 99215, 99217, 99218, 99219, 99220, 99221, 99222, 99223, 99224, 99225, 99226, 99231, 99232, 99233, 99234, 99235, 99236, 99238, 99239, 99241, 99242, 99243, 99244, 99245, 99251, 99252, 99253, 99254, 99255, 99261, 99262, 99263, 99271, 99272, 99273, 99274, 99275, 99281, 99282, 99283, 99284, 99285, 99289, 99290, 99291, 99292, 99293, 99294, 99299, 99300, 99301, 99302, 99303, 99304, 99305, 99306, 99307, 99308, 99309, 99310, 99311, 99312, 99313, 99315, 99316, 99318, 99321, 99322, 99323, 99324, 99325, 99326, 99327, 99328, 99331, 99332, 99333, 99334, 99335, 99336, 99337, 99339, 99340, 99341, 99342, 99343, 99344, 99345, 99347, 99348, 99349, 99350, 99351, 99352, 99353, 99356, 99357, 99358, 99359, 99360, 99361, 99362, 99366, 99367, 99368, 99371, 99372, 99373, 99374, 99375, 99376, 99377, 99378, 99379, 99380, 99415, 99416, 99438, 99441, 99442, 99443, 99444, 99446, 99447, 99448, 99449, 99466, 99467, 99471, 99472, 99475, 99476, 99478, 99479, 99480, 99487, 99488, 99489, 99490, 99495, 99496, 99497, 99498, 99499, 0188T, 0189T, S0257, S0260, S0310
      To identify patients receiving hospice services, we looked for at least one claim with either a hospice procedure code or a hospice revenue code in the 3 months before the index fill.
      Procedure codes: 99377, 99378, G0182, G0337, G9474, G9475, G9476, G9477, G9478, G9479, G9524, Q5001, Q5002, Q5003, Q5004, Q5005, Q5006, Q5007, Q5008, Q5009, Q5010, S0255, S9126, T2042, T2043, T2044, T2045, T2046
      Revenue codes: 115, 0125, 0135, 0145, 0155, 0235, 0650, 0651, 0652, 0653, 0654, 0655, 0656, 0657, 0658, 0659

      Appendix E3

       Opioid drugs included or excluded

      We identified all opioid drugs present in the table of NDC codes in OptumLabs Data Warehouse. For the purposes of this analysis, we classified tramadol as an opioid. We excluded DEA schedule 5 drugs (codeine cough syrups).
      To limit the sample to drugs intended for home use, we excluded any injected or infused drug—those for which the dosage form was vial, syringe, ampule, cartridge, intravenous solution, etc.
      We included only drugs that had a defined dose unit such as a tablet, pill, and milligrams per milliliter. This excluded drugs in powder or bulk form.
      We included both single drug formulations and combinations of drugs. Table E1 includes all opioid drug combinations found in the table of NDC codes. Both long- and short-acting formulations were included.
      Buprenorphine, methadone, and drug combinations including naloxone may be used for both pain management and medication-assisted therapy for opioid use disorder. However, in an opioid-naive population, it is highly likely that these medications are being used for medication-assisted therapy. We excluded these drugs from the analysis of opioid-naive prescriptions. However, we did include them when determining whether a person was opioid naive and in calculating the risk of progression to long-term use.
      Table E3Opioid drugs and combinations included.
      OpioidDrug Combinations IncludedNot Eligible for Index FillLong ActingShort Acting
      BuprenorphineBuprenorphineXX
      Buprenorphine/naloxoneXX
      ButorphanolButorphanolX
      CodeineCodeineX
      Codeine/acetaminophenX
      Codeine/acetaminophen/butabarbitalX
      Codeine/acetaminophen/butalbitalX
      Codeine/aspirinX
      Codeine/aspirin/butalbital/caffeineX
      Codeine/aspirin/carisoprodolX
      Codeine/aspirin/phenacetin/caffeineX
      DihydrocodeineDihydrocodeine/acetaminophen/caffeineX
      Dihydrocodeine/aspirin/caffeineX
      FentanylFentanylXX
      HydrocodoneHydrocodoneXX
      Hydrocodone/acetaminophenX
      Hydrocodone/acetaminophen/dietary supplement no. 11X
      Hydrocodone/aspirinX
      Hydrocodone/ibuprofenX
      HydromorphoneHydromorphoneXX
      LevomethadylLevomethadylX
      LevorphanolLevorphanolX
      MeperidineMeperidine/acetaminophenX
      Meperidine/promethazineX
      MethadoneMethadoneXX
      MorphineMorphine sulfateXX
      Morphine sulfate/naltrexoneX
      OpiumOpiumX
      Opium/belladonnaX
      OxycodoneOxycodoneXX
      Oxycodone/acetaminophenX
      Oxycodone/aspirinX
      Oxycodone/ibuprofenX
      OxymorphoneOxymorphoneXX
      PentazocinePentazocine/acetaminophenX
      Pentazocine/aspirinX
      Pentazocine/naloxoneXX
      PropoxyphenePropoxypheneX
      Propoxyphene/acetaminophenX
      Propoxyphene/aspirin/caffeineX
      TapentadolTapentadolXX
      TramadolTramadolXX
      Tramadol/acetaminophenX
      Tramadol/dietary supplement no. 11X
      Table E4Example prescription information.
      PatientPrescriberDateDrugFormulationTotal MMEDays’ Supply
      1101/1/2011HydrocodoneSA502
      1101/1/2011HydrocodoneSA10010
      1151/1/2011HydrocodoneLA505
      2101/1/2011OxycodoneSA505
      2151/1/2011OxycodoneSA5010

      Appendix E4

       Definition of an opioid fill

      An opioid fill was defined as a drug dispensed on a single day to an individual beneficiary by a single prescriber. Here, drug is defined by the opioid ingredient and formulation type: for example, all short-acting hydrocodone prescriptions filled on the same day with the same prescriber identification for the same individual would be counted as one fill. The total MME amount dispensed for each drug/person/prescriber combination is summed and divided by the maximum days’ supply across the prescriptions. Examples:
      Patient 1 has 2 fills on 1/1/2011:
      • 1.
        150 MME of SA hydrocodone with a day’s supply of 10=15 MME per day
      • 2.
        50 MME of LA hydrocodone with a day’s supply of 5=10 MME per day
      Patient 2 has 2 fills on 1/1/2011:
      • 1.
        50 MME of SA oxycodone with a day’s supply of 5 from prescriber 10=10 MME per day
      • 2.
        50 MME of SA oxycodone with a day’s supply of 10 from prescriber 15=5 MME per day

      Appendix E5

       Determining prescription source

      To determine the source of an opioid fill, we attempted to link a prescription claim to a medical claim representing the encounter in which the beneficiary received the prescription. Because opioids are scheduled drugs, we expect that in most cases, the prescriber will consult with the patient in person before writing the prescription. As of 2013, 41 states and Washington, DC, had state laws requiring physical examinations in relation to prescriptions for a controlled substance.

      Office for State, Tribal, Local and Territorial Support. Prescription drug physical examination requirements. Available at: https://www.cdc.gov/phlp/docs/pdpe-requirements.pdf. Accessed July 3, 2017.

      To find the visit that generated a prescription, we look for all medical claims in the 30 days before and including the date a prescription was filled. We used revenue and procedure codes identified by the National Committee for Quality Assurance to find inpatient, outpatient, and ED visits (except code 92888, used for physician consultation with emergency medical services).

      NCQA. 2015 Quality Rating System (QRS) HEDIS value set directory. Available at: http://store.ncqa.org/index.php/2015-qrs-hedis-valueset-directory.html. Accessed May 15, 2017.

      We used revenue codes to identify ambulatory surgery services and Current Procedural Terminology codes to identify dental services. All other procedures and visits that were not classified as inpatient, outpatient, ED, ambulatory surgery, or dental services were captured and labeled “other” services. This category includes laboratory tests, imaging, physical therapy, and chiropractic care.
      We identified inpatient, outpatient, ED, ambulatory surgery, or dental visit in the 30 days up to and including the index fill date. We attempted to match the provider identification for these visits to the prescriber identification on the pharmacy claim. If we found a matching visit, we assigned that visit as the most likely source of the prescription (N=1,590,929; 30.3% of fills with any visit within 30 days). Because of limitations in the OLDW when the study was completed, we were unable to match prescriber identifications to medical claims physician identifications for Medicare Advantage beneficiaries.
      If no visits were found with a provider identification matching the prescription, we assigned the most proximal visit as the source of the prescription. If no visits were found within 30 days of the index fill, but other services were present (for example, laboratory tests or imaging), we considered the prescription to have an unknown source. Of all prescriptions with a known source, 72% were filled on the same day as the visit, 85% within 3 days, and 90% within 1 week.
      Once we identified the visit considered the most likely source of the fill, we classified the most likely source of each index fill as ED visit only; non-ED visit only, which combines inpatient, outpatient, ambulatory surgery, and dental/accidental dental; and unknown source. The unknown source category included both fills with no visit in the previous 30 days and fills in which the beneficiary had both an ED visit and a non-ED visit on the same day.
      A substantial proportion of prescriptions fall into the unknown source category: 26% for the commercial population and 15% to 16% for the Medicare population; most of these fills were classified as unknown because of having no visit in the previous 30 days. Another study using a different source of commercial claims found a similarly high rate of prescriptions that could not be matched to a visit: 28% unmatched with a look-back period of 2 weeks (vs 30 days for this study).
      • Liu Y.
      • Logan J.E.
      • Paulozzi L.J.
      • et al.
      Potential misuse and inappropriate prescription practices involving opioid analgesics.
      Some of these prescriptions were likely written by dentists, who have been estimated to write 6.4% of opioid prescriptions.
      • Levy B.
      • Paulozzi L.
      • Mack K.A.
      • et al.
      Trends in opioid analgesic-prescribing rates by specialty, US, 2007-2012.
      We did not observe most dental visits because dentistry is not included in medical insurance benefits. In our sample of fills to opioid-naive patients, 7.0% of fills with a known prescriber specialty were written by a dentist or other dental specialist. We present the results for prescriptions with unknown source throughout, but do not focus on the interpretation of this group of prescriptions.

      Appendix E6

       Drug filled—complete list

      Table E5Drugs filled by population.
      OpioidsCommercialAged MedicareDisabled Medicare
      % of fills95 CI% of fills95 CI% of fills95 CI
      Hydrocodone SA58.9(58.9,59.0)49.2(49.1,49.4)49.7(49.3,50.0)
      Oxycodone SA18.8(18.8,18.8)16.6(16.6,16.7)19.4(19.1,19.7)
      Tramadol SA8.7(8.7,8.7)20.2(20.1,20.3)18.7(18.5,19.0)
      Codeine9.8(9.8,9.9)8.6(8.5,8.7)7.2(7.0,7.4)
      Propoxyphene2.3(2.3,2.3)2.4(2.4,2.5)2.0(1.9,2.1)
      Morphine SA<0.1(<0.1,<0.1)1.4(1.4,1.5)0.5(0.4,0.5)
      Hydromorphone SA0.4(0.4,0.4)0.6(0.5,0.6)0.8(0.8,0.9)
      Oxycodone LA0.2(0.2,0.2)0.3(0.3,0.3)0.4(0.4,0.5)
      Morphine LA0.1(0.1,0.1)0.1(0.1,0.1)0.5(0.4,0.5)
      Meperidine0.4(0.4,0.4)0.1(0.1,0.1)0.1(0.1,0.2)
      Fentanyl LA<0.1(<0.1,<0.1)0.3(0.3,0.3)0.3(0.3,0.3)
      Tramadol LA0.1(0.1,0.1)0.1(0.1,0.1)0.1(0.1,0.2)
      Tapentadol SA0.1(0.1,0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Oxymorphone LA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)0.1(0.1,0.1)
      Pentazocine<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Butorphanol<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Opium<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Tapentadol LA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Oxymorphone SA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Dihydrocodeine<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Hydromorphone LA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Levorphanol<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Hydrocodone LA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)
      Fentanyl SA<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)<0.1(<0.1,<0.1)

      Appendix E7

       Calculation of confidence intervals in Table 1

      Table E6Calculation methods.
      MeasureCI Calculation/Statistical Tests
      Naive opioid fills per covered person per year of insurance coverage

      Total opioid fills per covered person per year of insurance coverage

      Mean number of comorbidities
      Mean calculated with Poisson exact CI
      Sex

      Race/ethnicity

      Prescription source

      Insurance coverage after fill (3, 6, and 12 mo)

      Had any claims in (3 or 6 mo) before fill

      Any comorbidity

      Drug filled

      DEA class

      Greater than (50 or 90 MME, 3 or 7 days)

      Long-term opioid use
      CIs for proportions calculated with logit transform
      AgeNA: median with interquartile range

      Appendix E8

       Complete logistic regression results for analyses presented in Table 2

      Table E7Logistic regression results.
      Odds Ratios ReportedPrescription Characteristics
      (1)

      >3 Days
      (2)

      >7 Days
      (3)

      >50 MME
      (4)

      >90 MME
      (5)

      Long Acting
      (6)

      Long-Term Use
      Beneficiary population
      Commercial111111
      (1–1)(1–1)(1–1)(1–1)(1–1)(1–1)
      Aged Medicare0.9671.0220.9290.7710.7561.701
      (0.956–0.978)(1.010–1.034)(0.917–0.941)(0.754–0.788)(0.713–0.803)(1.653–1.751)
      Disabled Medicare1.2971.9420.7540.8821.7004.465
      (1.270–1.323)(1.908–1.977)(0.737–0.771)(0.851–0.915)(1.585–1.822)(4.332–4.601)
      Treatment setting
      Non-ED111111
      (1–1)(1–1)(1–1)(1–1)(1–1)(1–1)
      Unknown source0.4590.3500.6900.5930.6510.613
      (0.456–0.461)(0.347–0.353)(0.685–0.694)(0.586–0.600)(0.624–0.680)(0.597–0.630)
      ED0.3000.1320.5570.4380.07990.535
      (0.298–0.302)(0.130–0.134)(0.553–0.561)(0.432–0.445)(0.0702–0.0911)(0.518–0.553)
      Treatment setting×population
      Unknown source×aged Medicare0.9491.2451.3071.4511.2761.105
      (0.935–0.964)(1.222–1.268)(1.281–1.333)(1.401–1.503)(1.165–1.397)(1.057–1.154)
      Unknown source×disabled Medicare1.0601.4821.5781.7121.7591.430
      (1.017–1.105)(1.419–1.549)(1.501–1.659)(1.572–1.863)(1.512–2.047)(1.334–1.534)
      ED×aged Medicare0.7940.5921.2351.4881.7160.785
      (0.780–0.807)(0.570–0.615)(1.206–1.265)(1.425–1.553)(1.333–2.208)(0.740–0.831)
      ED×disabled Medicare0.5670.3881.3181.2611.3970.686
      (0.544–0.590)(0.355–0.425)(1.249–1.390)(1.138–1.396)(0.912–2.141)(0.627–0.751)
      Women0.9370.9330.9881.1220.7200.923
      (0.934–0.941)(0.928–0.937)(0.984–0.993)(1.113–1.130)(0.702–0.739)(0.911–0.936)
      Race/ethnicity
      White111111
      (1–1)(1–1)(1–1)(1–1)(1–1)(1–1)
      Black1.1651.2640.8400.8950.7251.017
      (1.158–1.172)(1.254–1.274)(0.834–0.846)(0.884–0.906)(0.692–0.759)(0.996–1.038)
      Hispanic1.2301.3990.8150.8940.7490.853
      (1.222–1.237)(1.388–1.411)(0.809–0.822)(0.882–0.906)(0.712–0.787)(0.832–0.875)
      Asian1.0851.0320.8440.7930.7720.456
      (1.074–1.096)(1.018–1.047)(0.834–0.855)(0.775–0.811)(0.712–0.837)(0.431–0.481)
      Unknown race/ethnicity1.0031.0080.9520.9181.0690.872
      (0.993–1.013)(0.995–1.022)(0.941–0.964)(0.899–0.937)(1.003–1.139)(0.841–0.904)
      Year1.0191.0350.9070.7670.9460.952
      (1.018–1.020)(1.033–1.036)(0.906–0.908)(0.766–0.769)(0.940–0.952)(0.949–0.955)
      Age0.9430.9010.9080.8821.0751.317
      (0.942–0.944)(0.900–0.902)(0.906–0.909)(0.880–0.883)(1.064–1.087)(1.305–1.330)
      Age

      Office for State, Tribal, Local and Territorial Support. Prescription drug physical examination requirements. Available at: https://www.cdc.gov/phlp/docs/pdpe-requirements.pdf. Accessed July 3, 2017.

      1.0011.0031.0021.0020.9990.995
      (1.001–1.001)(1.002–1.003)(1.002–1.002)(1.002–1.003)(0.999–0.999)(0.995–0.996)
      Age

      NCQA. 2015 Quality Rating System (QRS) HEDIS value set directory. Available at: http://store.ncqa.org/index.php/2015-qrs-hedis-valueset-directory.html. Accessed May 15, 2017.

      1.0001.0001.0001.0001.0001.000
      (1.000–1.000)(1.000–1.000)(1.000–1.000)(1.000–1.000)(1.000–1.000)(1.000–1.000)
      Elixhauser comorbidities (binary)
      CHF1.0361.0510.9481.0490.8441.165
      (1.014–1.058)(1.030–1.072)(0.926–0.969)(1.012–1.087)(0.782–0.911)(1.119–1.214)
      Arrhythmia1.0971.0111.1921.1291.4320.930
      (1.081–1.112)(0.997–1.026)(1.174–1.210)(1.102–1.158)(1.357–1.510)(0.901–0.960)
      Valvular disease1.1740.9941.1371.0420.9060.780
      (1.148–1.201)(0.973–1.016)(1.111–1.163)(1.004–1.081)(0.834–0.984)(0.742–0.820)
      Pulm. circ. dis.1.1741.1580.9530.9291.1121.196
      (1.132–1.217)(1.119–1.199)(0.918–0.989)(0.874–0.987)(0.992–1.247)(1.117–1.281)
      Peripheral vascular dis.1.2091.0411.1911.1711.1711.211
      (1.182–1.237)(1.020–1.063)(1.164–1.219)(1.129–1.214)(1.088–1.261)(1.163–1.262)
      Uncomp. HTN1.2171.2221.1301.1361.5471.203
      (1.208–1.225)(1.213–1.231)(1.121–1.139)(1.121–1.151)(1.494–1.602)(1.182–1.225)
      Comp. HTN1.1161.0800.9790.9531.0260.961
      (1.087–1.145)(1.053–1.107)(0.952–1.007)(0.910–0.998)(0.930–1.132)(0.911–1.014)
      Other neuro.1.1031.1641.0231.0851.1981.239
      (1.079–1.126)(1.139–1.190)(0.999–1.047)(1.046–1.126)(1.107–1.296)(1.185–1.296)
      Chronic pulm. dis.1.1731.1851.1091.1281.5691.379
      (1.160–1.187)(1.171–1.200)(1.096–1.124)(1.106–1.151)(1.495–1.648)(1.342–1.417)
      Peptic ulcer1.4661.3590.9101.0051.0721.614
      (1.368–1.571)(1.275–1.448)(0.848–0.976)(0.899–1.123)(0.861–1.333)(1.435–1.814)
      Uncomp. diabetes1.0411.1030.9370.9640.9461.232
      (1.031–1.051)(1.091–1.114)(0.927–0.948)(0.946–0.982)(0.903–0.992)(1.205–1.260)
      Comp. diabetes1.0371.0330.9270.9330.9781.125
      (1.014–1.061)(1.011–1.056)(0.904–0.950)(0.895–0.973)(0.893–1.072)(1.076–1.175)
      Paralysis1.3781.6131.0851.1571.8462.237
      (1.309–1.450)(1.542–1.687)(1.034–1.139)(1.076–1.244)(1.628–2.093)(2.066–2.422)
      Renal failure0.9781.0020.9300.9330.8671.020
      (0.956–1.001)(0.980–1.024)(0.906–0.954)(0.895–0.973)(0.795–0.946)(0.975–1.067)
      Solid tumor w/o mets0.9990.7151.4661.2391.3660.814
      (0.976–1.022)(0.697–0.733)(1.432–1.500)(1.192–1.288)(1.250–1.493)(0.768–0.864)
      Liver dis.1.2141.1011.1051.0821.2151.300
      (1.182–1.247)(1.070–1.132)(1.075–1.136)(1.033–1.134)(1.102–1.340)(1.225–1.381)
      Met. cancer1.5881.3291.4951.2922.3141.852
      (1.476–1.708)(1.243–1.422)(1.404–1.593)(1.168–1.428)(1.947–2.751)(1.603–2.139)
      HIV/AIDS1.1151.1040.9910.9791.2751.076
      (1.053–1.180)(1.033–1.181)(0.927–1.059)(0.871–1.100)(0.979–1.660)(0.921–1.257)
      Rheumatoid arthritis1.4972.0170.9610.9922.2072.304
      (1.462–1.532)(1.973–2.061)(0.937–0.985)(0.951–1.034)(2.038–2.390)(2.209–2.402)
      Hypothyroid1.1121.0971.1771.1211.6201.025
      (1.098–1.127)(1.081–1.112)(1.161–1.194)(1.095–1.147)(1.535–1.710)(0.992–1.060)
      Lymphoma1.1221.1501.0771.0702.0441.626
      (1.067–1.180)(1.094–1.209)(1.022–1.136)(0.981–1.168)(1.759–2.374)(1.471–1.798)
      Coagulopathy1.1911.0651.3511.1341.4101.063
      (1.157–1.227)(1.035–1.097)(1.313–1.390)(1.082–1.188)(1.282–1.550)(0.996–1.135)
      Obesity1.2721.1521.7442.1151.9921.095
      (1.252–1.292)(1.132–1.172)(1.717–1.771)(2.067–2.164)(1.877–2.114)(1.050–1.141)
      Weight loss1.1701.2341.1521.3581.6681.655
      (1.131–1.210)(1.196–1.274)(1.114–1.192)(1.291–1.427)(1.524–1.827)(1.556–1.761)
      Fluid/electrolyte dis.1.3011.1691.2511.1761.6621.205
      (1.278–1.324)(1.149–1.190)(1.229–1.274)(1.143–1.211)(1.566–1.764)(1.161–1.251)
      Blood loss anemia1.1671.0091.4351.2751.3961.061
      (1.113–1.225)(0.961–1.059)(1.371–1.503)(1.186–1.372)(1.197–1.628)(0.956–1.178)
      Deficiency anemia1.1051.1001.0360.9891.1681.166
      (1.076–1.135)(1.071–1.131)(1.007–1.066)(0.944–1.036)(1.057–1.290)(1.101–1.235)
      Alcohol abuse1.2451.2141.1941.1471.3511.529
      (1.201–1.291)(1.167–1.263)(1.149–1.240)(1.075–1.223)(1.198–1.523)(1.417–1.650)
      Drug abuse1.0811.3301.0411.0652.3322.318
      (1.030–1.134)(1.259–1.406)(0.986–1.098)(0.973–1.166)(2.015–2.700)(2.096–2.562)
      Psychoses0.9080.9610.8841.0311.0131.272
      (0.874–0.943)(0.925–0.998)(0.846–0.923)(0.964–1.102)(0.900–1.139)(1.188–1.361)
      Depression1.0441.1911.0421.0882.0981.853
      (1.030–1.058)(1.173–1.209)(1.026–1.058)(1.061–1.116)(1.991–2.210)(1.796–1.911)
      Any claim 6 mo before fill1.0781.1851.3171.4472.0851.110
      (1.070–1.086)(1.168–1.202)(1.303–1.331)(1.417–1.477)(1.909–2.276)(1.066–1.156)
      Constant1.16e–169.59e–316.99e+841.12e+2318.69e+441.53e+39
      (1.86e–17–7.27e–16)(8.28e–32–1.11e–29)(7.41e+83–6.59e+85)(1.80e+229–6.93e+232)(2.02e+39–3.761e+50)(1.682e+36–1.386e+42)
      Observations5,243,4985,243,4985,243,4985,243,4985,243,4983,658,393
      CHF, Congestive heart failure; HTN, hypertension.
      Exponentiated coefficients; 95% CIs in parentheses.

      Appendix E9

       Complete results for analysis of guideline concordance predicting long-term use

      Table E8Logistic regression results.
      Odds Ratios Reported(1)

      Long-Term Use
      Prescription guideline concordance
      Fully concordant (≤3 days’ supply, ≤50 MME per day, short-acting formulation)0.259
      (0.250–0.269)
      Treatment setting
      Non-ED1
      (1–1)
      Unknown source0.764
      (0.743–0.786)
      ED0.489
      (0.469–0.511)
      Guideline concordance×treatment setting
      Fully concordant×non-ED1
      (1–1)
      Fully concordant×unknown source0.851
      (0.794–0.913)
      Fully concordant×ED3.099
      (2.883–3.331)
      Enrollee population
      Commercial1
      (1–1)
      Aged Medicare1.733
      (1.683–1.785)
      Disabled Medicare4.472
      (4.336–4.613)
      Guideline concordance×enrollee population
      Fully concordant×commercial1
      (1–1)
      Fully concordant×aged Medicare0.801
      (0.748–0.857)
      Fully concordant×disabled Medicare0.743
      (0.653–0.844)
      Treatment setting×enrollee population
      Non-ED×commercial1
      (1–1)
      Non-ED×aged Medicare1
      (1–1)
      Non-ED×disabled Medicare1
      (1–1)
      Unknown source×commercial1
      (1–1)
      Unknown source×aged Medicare1.053
      (1.005–1.103)
      Unknown source×disabled Medicare1.308
      (1.214–1.409)
      ED×commercial1
      (1–1)
      ED×aged Medicare0.818
      (0.758–0.883)
      ED×disabled Medicare0.666
      (0.587–0.755)
      Guideline concordance×treatment setting×enrollee population
      Fully concordant×non-ED×commercial1
      (1–1)
      Fully concordant×non-ED×aged Medicare1
      (1–1)
      Fully concordant×non-ED×disabled Medicare1
      (1–1)
      Fully concordant×unknown source×commercial1
      (1–1)
      Fully concordant×unknown source×aged Medicare0.975
      (0.840–1.131)
      Fully concordant×unknown source×disabled Medicare1.165
      (0.909–1.493)
      Fully concordant×ED×commercial1
      (1–1)
      Fully concordant×ED×aged Medicare1.183
      (1.038–1.349)
      Fully concordant×ED×disabled Medicare1.518
      (1.225–1.881)
      Year0.957
      (0.953–0.960)
      Age1.316
      (1.304–1.328)
      Age

      Office for State, Tribal, Local and Territorial Support. Prescription drug physical examination requirements. Available at: https://www.cdc.gov/phlp/docs/pdpe-requirements.pdf. Accessed July 3, 2017.

      0.995
      (0.995–0.996)
      Age

      NCQA. 2015 Quality Rating System (QRS) HEDIS value set directory. Available at: http://store.ncqa.org/index.php/2015-qrs-hedis-valueset-directory.html. Accessed May 15, 2017.

      1.000
      (1.000–1.000)
      Elixhauser comorbidities
      CHF1.169
      (1.122–1.217)
      Arrhythmia0.919
      (0.891–0.949)
      Valvular disease0.758
      (0.721–0.797)
      Pulm. circ. dis.1.182
      (1.104–1.267)
      Peripheral vascular dis.1.185
      (1.138–1.235)
      Uncomp. HTN1.158
      (1.138–1.179)
      Comp. HTN0.948
      (0.899–1.000)
      Other neuro.1.239
      (1.184–1.296)
      Chronic pulm. dis.1.346
      (1.310–1.383)
      Peptic ulcer1.565
      (1.392–1.760)
      Uncomp. diabetes1.226
      (1.199–1.254)
      Comp. diabetes1.124
      (1.075–1.175)
      Paralysis2.230
      (2.058–2.416)
      Renal failure1.029
      (0.984–1.076)
      Solid tumor w/o mets0.802
      (0.756–0.851)
      Liver dis.1.256
      (1.183–1.334)
      Met. cancer1.729
      (1.497–1.999)
      HIV/AIDS1.093
      (0.936–1.277)
      Rheumatoid arthritis2.193
      (2.103–2.287)
      Hypothyroid1.006
      (0.973–1.040)
      Lymphoma1.617
      (1.462–1.788)
      Coagulopathy1.034
      (0.968–1.104)
      Obesity1.036
      (0.993–1.080)
      Weight loss1.630
      (1.532–1.734)
      Fluid/electrolyte dis.1.158
      (1.116–1.202)
      Blood loss anemia1.027
      (0.925–1.140)
      Deficiency anemia1.157
      (1.093–1.226)
      Alcohol abuse1.504
      (1.393–1.623)
      Drug abuse2.304
      (2.083–2.548)
      Psychoses1.323
      (1.236–1.416)
      Depression1.854
      (1.797–1.913)
      Any claim within 6 mo before fill1.057
      (1.014–1.101)
      Women0.920
      (0.908–0.933)
      Race/ethnicity
      White1
      (1–1)
      Black0.998
      (0.977–1.019)
      Hispanic0.829
      (0.808–0.850)
      Asian0.452
      (0.428–0.478)
      Unknown race/ethnicity0.875
      (0.844–0.907)
      N3,658,393
      Exponentiated coefficients; 95% CIs in brackets.

      Appendix E10

       Risk ratios and confidence intervals for results presented in Figure 2

      Table E9Risk ratios with CIs.
      OutcomeTreatment SettingCommercialAged MedicareDisabled Medicare
      Risk RatioLower CIUpper CIRisk RatioLower CIUpper CIRisk RatioLower CIUpper CI
      >3 days’ supplyNon-ED (ref.)1
      Dashes indicate reference category.
      11
      Unknown source0.720.710.720.760.750.760.810.790.82
      ED0.560.560.560.560.550.560.480.470.49
      >50 MME/dayNon-ED (ref.)111
      Unknown source0.750.740.750.920.910.941.071.031.11
      ED0.630.620.630.730.720.750.770.740.81
      >7 days’ supplyNon-ED (ref.)111
      Unknown source0.410.400.410.560.550.560.660.640.68
      ED0.160.160.160.120.120.130.090.080.10
      >90 MME/dayNon-ED (ref.)111
      Unknown source0.620.610.630.870.840.901.010.931.09
      ED0.460.460.470.670.640.690.570.520.63
      Long-acting/extended releaseNon-ED (ref.)111
      Unknown source0.650.630.680.830.770.901.140.981.30
      ED0.080.070.090.140.110.170.120.070.16
      Long-term opioid useNon-ED (ref.)111
      Unknown source0.620.600.640.700.670.720.900.850.95
      ED0.540.530.560.440.420.460.420.390.45
      Risk ratios were calculated from marginal effects after logistic regression; 95% CIs were calculated for the ratio with the delta method.
      Dashes indicate reference category.

      Appendix E11

       Supplementary analyses—time trends in guideline concordance and progression to long-term use

      To understand trends during the 7 years of our study period, we repeated the main analyses but included time as a categorical variable fully interacted with beneficiary population and treatment setting. We calculated marginal effects (predicted probability of exceeding 3 days, 7 days, 50 MME, 90 MME, of prescribing a long-acting formulation, and of progression to long-term opioid use) and graphed them by beneficiary population and time. We also included an “as observed” analysis showing the average across the entire population.
      Figure thumbnail fx1
      Figure E1A, Probability of prescription greater than 3 days’ supply. B, Probability of prescription greater than 7 days' supply. C, Probability of prescription greater than 50 MME per day. D, Probability of prescription greater than 90 MME per day. E, Probability of prescription for long-acting/extended release formulation. F, Probability of progression to long-term use.
      Table E10Probability of prescription greater than 3 days supply, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (>3 Days)Lower CIUpper CIPr (>3 Days)Lower CIUpper CIPr (>3 Days)Lower CIUpper CI
      2009
      Not ED0.6800.6790.6810.6570.6520.6620.7240.7120.735
      ED0.3650.3620.3680.2820.2720.2920.3070.2840.330
      Unknown0.4590.4570.4610.4500.4390.4610.5510.5230.579
      2010
      Not ED0.6750.6740.6760.6450.6410.6490.7200.7090.730
      ED0.3700.3660.3730.2860.2780.2950.2780.2580.298
      Unknown0.4620.4600.4640.4300.4200.4390.5490.5240.574
      2011
      Not ED0.6680.6670.6700.6840.6800.6890.7540.7430.764
      ED0.3700.3670.3730.3190.3100.3270.3050.2850.325
      Unknown0.4670.4650.4690.4480.4390.4560.5840.5600.608
      2012
      Not ED0.6630.6620.6650.6680.6640.6720.7270.7180.737
      ED0.3810.3770.3840.3300.3210.3380.3060.2870.325
      Unknown0.4630.4610.4650.4350.4270.4440.5490.5270.571
      2013
      Not ED0.6760.6750.6780.6660.6620.6690.7280.7180.737
      ED0.4000.3960.4030.3350.3270.3430.3240.3060.342
      Unknown0.4770.4740.4790.4250.4170.4330.5370.5160.559
      2014
      Not ED0.6830.6820.6850.6870.6830.6910.7410.7310.751
      ED0.4050.4020.4090.3470.3390.3550.3310.3130.349
      Unknown0.4870.4840.4890.4760.4680.4840.5540.5320.575
      2015
      Not ED0.6960.6950.6980.6900.6870.6940.7290.7190.740
      ED0.4030.4000.4070.3520.3440.3600.3460.3260.365
      Unknown0.5010.4990.5040.4970.4890.5040.5510.5290.572
      Table E11Probability of prescription greater than 7 days’ supply, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (>7 Days)Lower CIUpper CIPr (>7 Days)Lower CIUpper CIPr (>7 Days)Lower CIUpper CI
      2009
      Not ED0.2090.2080.2100.2140.2100.2170.3490.3370.360
      ED0.0320.0300.0330.0200.0170.0220.0280.0210.035
      Unknown0.0750.0740.0760.1000.0950.1050.1990.1800.218
      2010
      Not ED0.2120.2100.2130.2120.2090.2150.3460.3360.356
      ED0.0340.0320.0350.0190.0170.0210.0240.0180.030
      Unknown0.0750.0740.0760.1010.0970.1050.2060.1890.224
      2011
      Not ED0.2140.2130.2150.2250.2220.2270.3600.3500.370
      ED0.0340.0330.0350.0210.0190.0230.0290.0220.035
      Unknown0.0770.0750.0780.0950.0910.0990.2310.2130.248
      2012
      Not ED0.2200.2190.2220.2280.2250.2310.3550.3460.365
      ED0.0370.0360.0390.0220.0200.0240.0250.0190.031
      Unknown0.0790.0780.0800.0870.0840.0900.1980.1830.214
      2013
      Not ED0.2280.2260.2290.2300.2270.2330.3580.3490.367
      ED0.0390.0370.0400.0240.0220.0250.0300.0250.036
      Unknown0.0820.0810.0840.0870.0840.0900.1830.1680.197
      2014
      Not ED0.2360.2350.2380.2410.2380.2440.3460.3360.355
      ED0.0370.0360.0390.0240.0220.0260.0300.0240.036
      Unknown0.0830.0810.0840.1090.1060.1130.1800.1660.194
      2015
      Not ED0.2480.2460.2490.2440.2410.2470.3540.3440.364
      ED0.0370.0360.0380.0250.0230.0270.0260.0200.031
      Unknown0.0850.0830.0860.1150.1110.1180.1890.1740.203
      Table E12Probability of prescription greater than 50 mg of morphine equivalents per day, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (>50 MME)Lower CIUpper CIPr (>50 MME)Lower CIUpper CIPr (>50 MME)Lower CIUpper CI
      2009
      Not ED0.2790.2770.2800.2830.2790.2880.2410.2300.252
      ED0.1770.1740.1790.2540.2430.2650.2010.1800.222
      Unknown0.1910.1900.1930.2460.2360.2560.2330.2100.255
      2010
      Not ED0.2680.2660.2690.3290.3250.3340.2690.2590.279
      ED0.1720.1700.1750.2860.2770.2960.2160.1970.235
      Unknown0.1850.1830.1860.3030.2950.3120.2450.2240.265
      2011
      Not ED0.2130.2120.2150.1750.1720.1790.1580.1500.166
      ED0.1410.1390.1430.1280.1210.1350.1370.1210.152
      Unknown0.1480.1460.1490.1560.1500.1630.1540.1380.170
      2012
      Not ED0.2150.2140.2160.1890.1860.1920.1570.1500.165
      ED0.1350.1330.1380.1320.1250.1390.1190.1050.133
      Unknown0.1490.1470.1500.1650.1590.1720.1730.1570.189
      2013
      Not ED0.2000.1990.2020.1820.1790.1860.1600.1530.167
      ED0.1190.1170.1210.1230.1170.1290.1160.1040.129
      Unknown0.1370.1350.1380.1610.1550.1670.1530.1390.168
      2014
      Not ED0.1950.1940.1970.1670.1640.1700.1450.1380.152
      ED0.1140.1110.1160.1050.1000.1110.0990.0870.111
      Unknown0.1330.1310.1350.1290.1240.1340.1550.1400.169
      2015
      Not ED0.1970.1950.1980.1690.1660.1720.1330.1260.141
      ED0.1040.1010.1060.0970.0920.1030.0920.0800.104
      Unknown0.1340.1320.1360.1290.1240.1340.1340.1200.148
      Table E13Probability of prescription greater than 90 mg of morphine equivalents per day, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (>90 MME)Lower CIUpper CIPr (>90 MME)Lower CIUpper CIPr (>90 MME)Lower CIUpper CI
      2009
      Not ED0.1280.1270.1290.1070.1040.1100.1150.1070.123
      ED0.0590.0570.0600.0960.0900.1030.0840.0690.099
      Unknown0.0740.0730.0750.0800.0750.0860.0910.0760.106
      2010
      Not ED0.1150.1140.1160.1360.1330.1390.1380.1310.146
      ED0.0520.0510.0540.1160.1090.1220.0940.0800.107
      Unknown0.0670.0650.0680.1190.1130.1250.1120.0970.126
      2011
      Not ED0.0610.0600.0610.0390.0380.0410.0520.0470.057
      ED0.0250.0240.0260.0220.0190.0240.0290.0210.036
      Unknown0.0290.0280.0300.0290.0270.0320.0490.0390.058
      2012
      Not ED0.0600.0590.0610.0360.0340.0370.0440.0400.049
      ED0.0240.0230.0250.0180.0160.0200.0230.0160.029
      Unknown0.0280.0280.0290.0260.0240.0280.0540.0440.063
      2013
      Not ED0.0500.0490.0500.0340.0330.0350.0430.0390.047
      ED0.0220.0210.0230.0140.0120.0160.0170.0120.023
      Unknown0.0240.0230.0250.0230.0210.0260.0370.0290.044
      2014
      Not ED0.0430.0430.0440.0250.0240.0260.0290.0250.032
      ED0.0260.0250.0270.0070.0050.0080.0100.0060.014
      Unknown0.0220.0210.0220.0160.0140.0180.0270.0210.034
      2015
      Not ED0.0410.0410.0420.0190.0180.0200.0250.0220.029
      ED0.0200.0190.0210.0060.0050.0070.0070.0030.010
      Unknown0.0200.0190.0210.0110.0100.0130.0200.0140.026
      Table E14Probability of prescription for a long-acting or extended release formulation, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (LA/ER)Lower CIUpper CIPr (LA/ER)Lower CIUpper CIPr (LA/ER)Lower CIUpper CI
      2009
      Not ED0.0070.0070.0070.0070.0060.0070.0120.0100.015
      ED0.0000.0000.0000.0010.0000.0010.0010.0000.002
      Unknown0.0040.0030.0040.0040.0030.0050.0100.0060.014
      2010
      Not ED0.0060.0060.0070.0050.0050.0060.0130.0110.015
      ED0.0000.0000.0010.0010.0000.0010.0020.0000.004
      Unknown0.0030.0030.0040.0030.0020.0040.0070.0050.010
      2011
      Not ED0.0070.0060.0070.0050.0050.0060.0140.0120.016
      ED0.0000.0000.0010.0010.0000.0010.0010.0000.003
      Unknown0.0030.0030.0030.0030.0030.0040.0120.0090.016
      2012
      Not ED0.0070.0070.0070.0050.0050.0050.0120.0100.014
      ED0.0010.0000.0010.0000.0000.0010.0010.0000.002
      Unknown0.0030.0030.0030.0030.0020.0040.0140.0100.017
      2013
      Not ED0.0070.0070.0070.0040.0030.0040.0090.0080.011
      ED0.0000.0000.0010.0010.0000.0010.0000.0000.001
      Unknown0.0030.0030.0030.0020.0020.0030.0070.0050.010
      2014
      Not ED0.0070.0060.0070.0040.0040.0050.0080.0070.010
      ED0.0000.0000.0010.0000.0000.0010.0010.0000.002
      Unknown0.0030.0020.0030.0020.0020.0020.0070.0040.009
      2015
      Not ED0.0050.0040.0050.0040.0040.0040.0080.0070.010
      ED0.0000.0000.0010.0000.0000.0010.0010.0000.001
      Unknown0.0020.0020.0020.0020.0020.0030.0040.0020.006
      Table E15Probability of progression to long-term opioid use, with 95% confidence intervals.
      CommercialAged MedicareDisabled Medicare
      Pr (Long Term)Lower CIUpper CIPr (Long Term)Lower CIUpper CIPr (Long Term)Lower CIUpper CI
      2009
      Not ED0.0270.0270.0280.0500.0480.0520.1180.1100.126
      ED0.0140.0130.0150.0210.0180.0230.0450.0360.055
      Unknown0.0150.0140.0150.0310.0270.0340.0810.0670.094
      2010
      Not ED0.0260.0250.0260.0460.0440.0470.1100.1030.117
      ED0.0120.0110.0130.0170.0140.0190.0400.0310.048
      Unknown0.0130.0120.0140.0290.0260.0320.0800.0680.092
      2011
      Not ED0.0270.0260.0270.0470.0450.0480.1120.1060.119
      ED0.0130.0120.0140.0190.0170.0210.0410.0330.048
      Unknown0.0130.0130.0140.0270.0250.0290.0890.0770.100
      2012
      Not ED0.0260.0250.0260.0420.0410.0430.1030.0970.109
      ED0.0130.0120.0140.0170.0150.0190.0360.0290.042
      Unknown0.0120.0120.0130.0210.0190.0220.0760.0660.086
      2013
      Not ED0.0250.0240.0260.0410.0390.0420.0930.0880.099
      ED0.0110.0100.0120.0160.0140.0180.0350.0280.041
      Unknown0.0120.0120.0130.0180.0170.0200.0660.0570.075
      2014
      Not ED0.0230.0230.0240.0370.0350.0380.0850.0800.091
      ED0.0100.0090.0110.0130.0110.0150.0300.0240.036
      Unknown0.0110.0100.0110.0200.0190.0220.0600.0510.068
      2015
      Not ED0.0240.0230.0250.0320.0310.0330.0810.0750.086
      ED0.0120.0110.0130.0120.0110.0130.0280.0220.034
      Unknown0.0100.0100.0110.0180.0170.0200.0560.0480.064

       Supplementary analyses—comparison of results when limiting analysis to beneficiaries with at least 1 year of follow-up

      Analyses of prescription guideline concordance by treatment setting included all qualifying prescriptions, regardless of the amount of follow-up time available for the beneficiary. Analysis of the risk of progression to long-term use was limited to patients with at least 1 year of follow-up, as required by the definition of long-term use (≥120 days or ≥10 fills during 12 months). To determine whether the results of the guideline concordance analyses were affected by the difference in the population included, we repeated all guideline concordance analyses, limiting them to people with at least 1 year of follow-up.
      We present the results as forest plots, one for each beneficiary population (commercial [Figure E2], aged Medicare [Figure E4], and disabled Medicare [Figure E4]). The risk ratios comparing the non-ED setting with the ED setting and the unknown setting are presented for each of the guideline concordance outcomes. The 1-year follow-up population is presented in red, whereas the population not limited by follow-up time is presented in blue. There were no statistically significant differences across the 2 populations.
      Figure thumbnail fx2
      Figure E2Comparison of main results using population with 1 year of follow-up to those with any length of follow-up, commercial beneficiaries.
      Figure thumbnail fx3
      Figure E3Comparison of main results using population with 1 year of follow-up to those with any length of follow-up, aged Medicare beneficiaries.
      Figure thumbnail fx4
      Figure E4Comparison of main results using population with 1 year of follow-up to those with any length of follow-up, disabled Medicare beneficiaries.

      References

        • Paulozzi L.J.
        • Jones C.M.
        • Mack K.A.
        • et al.
        Vital signs: overdoses of prescription opioid pain relievers—United States, 1999–2008.
        MMWR Morb Mortal Wkly Rep. 2011; 60: 1487-1492
        • Jones C.M.
        • Mack K.A.
        • Paulozzi L.J.
        Pharmaceutical overdose deaths, United States, 2010.
        JAMA. 2013; 309: 657-659
        • Hoppe J.A.
        • Kim H.
        • Heard K.
        Association of emergency department opioid initiation with recurrent opioid use.
        Ann Emerg Med. 2015; 65: 493-499.e4
        • Kea B.
        • Fu R.
        • Lowe R.A.
        • et al.
        Interpreting the National Hospital Ambulatory Medical Care Survey: United States emergency department opioid prescribing, 2006-2010.
        Acad Emerg Med. 2016; 23: 159-165
      1. American College of Emergency Physicians. Emergency physicians use new tool to detect drug-seekers in the ER. Available at: http://newsroom.acep.org/2013-07-10-Emergency-Physicians-Use-New-Tool-to-Detect-Drug-Seekers-in-the-ER. Accessed January 24, 2017.

      2. Weiner SG. Emergency medicine’s role in prescription opioid abuse. Available at: http://www.acepnow.com/article/emergency-medicines-role-in-prescription-opioid-abuse/. Accessed July 26, 2017.

        • Chang H.Y.
        • Daubresse M.
        • Kruszewski S.P.
        • et al.
        Prevalence and treatment of pain in EDs in the United States, 2000 to 2010.
        Am J Emerg Med. 2014; 32: 421-431
        • Hoppe J.A.
        • Nelson L.S.
        • Perrone J.
        • et al.
        Opioid prescribing in a cross section of US emergency departments.
        Ann Emerg Med. 2015; 66: 253-259.e1
        • Chou R.
        • Turner J.A.
        • Devine E.B.
        • et al.
        The effectiveness and risks of long-term opioid therapy for chronic pain: a systematic review for a National Institutes of Health Pathways to Prevention Workshop.
        Ann Intern Med. 2015; 162: 276-286
        • Goldenberg D.L.
        • Clauw D.J.
        • Palmer R.E.
        • et al.
        Opioid use in fibromyalgia: a cautionary tale.
        Mayo Clin Proc. 2016; 91: 640-648
        • Schuchat A.
        • Houry D.
        • Guy Jr., G.P.
        New data on opioid use and prescribing in the United States.
        JAMA. 2017; 318: 425-426
        • Califf R.M.
        • Woodcock J.
        • Ostroff S.
        A proactive response to prescription opioid abuse.
        N Engl J Med. 2016; 374: 1480-1485
        • Lovich-Sapola J.
        • Smith C.E.
        • Brandt C.P.
        Postoperative pain control.
        Surg Clin North Am. 2015; 95: 301-318
        • Hegmann K.T.
        • Weiss M.S.
        • Bowden K.
        • et al.
        ACOEM practice guidelines: opioids for treatment of acute, subacute, chronic, and postoperative pain.
        J Occup Environ Med. 2014; 56: e143-e159
        • Waljee J.F.
        • Li L.
        • Brummett C.M.
        • et al.
        Iatrogenic opioid dependence in the United States: are surgeons the gatekeepers?.
        Ann Surg. 2016; 265: 728-730
        • Callinan C.E.
        • Neuman M.D.
        • Lacy K.E.
        • et al.
        The initiation of chronic opioids: a survey of chronic pain patients.
        J Pain. 2017; 18: 360-365
        • Stumbo S.P.
        • Yarborough B.J.
        • McCarty D.
        • et al.
        Patient-reported pathways to opioid use disorders and pain-related barriers to treatment engagement.
        J Subst Abuse Treat. 2017; 73: 47-54
        • Alam A.
        • Gomes T.
        • Zheng H.
        • et al.
        Long-term analgesic use after low-risk surgery: a retrospective cohort study.
        Arch Intern Med. 2012; 172: 425-430
        • Barnett M.L.
        • Olenski A.R.
        • Jena A.B.
        Opioid-prescribing patterns of emergency physicians and risk of long-term use.
        N Engl J Med. 2017; 376: 663-673
        • Brummett C.M.
        • Waljee J.F.
        • Goesling J.
        • et al.
        New persistent opioid use after minor and major surgical procedures in US adults.
        JAMA Surg. 2017; 152: e170504
        • Clarke H.
        • Soneji N.
        • Ko D.T.
        • et al.
        Rates and risk factors for prolonged opioid use after major surgery: population based cohort study.
        BMJ. 2014; 348: g1251
        • Deyo R.A.
        • Hallvik S.E.
        • Hildebran C.
        • et al.
        Association between initial opioid prescribing patterns and subsequent long-term use among opioid-naive patients: a statewide retrospective cohort study.
        J Gen Intern Med. 2017; 32: 21-27
        • Johnson S.P.
        • Chung K.C.
        • Zhong L.
        • et al.
        Risk of prolonged opioid use among opioid-naive patients following common hand surgery procedures.
        J Hand Surg. 2016; 41: 947-957.e3
        • Shah A.
        • Hayes C.J.
        • Martin B.C.
        Characteristics of initial prescription episodes and likelihood of long-term opioid use—United States, 2006-2015.
        MMWR Morb Mortal Wkly Rep. 2017; 66: 265-269
        • Sun E.C.
        • Darnall B.D.
        • Baker L.C.
        • et al.
        Incidence of and risk factors for chronic opioid use among opioid-naive patients in the postoperative period.
        JAMA Intern Med. 2016; 176: 1286-1293
        • Hooten W.M.
        • St Sauver J.L.
        • McGree M.E.
        • et al.
        Incidence and risk factors for progression from short-term to episodic or long-term opioid prescribing: a population-based study.
        Mayo Clinic Proc. 2015; 90: 850-856
        • Dowell D.
        • Haegerich T.M.
        • Chou R.
        CDC guideline for prescribing opioids for chronic pain—United States, 2016.
        MMWR Recomm Rep. 2016; 65: 1-49
        • Benchimol E.I.
        • Smeeth L.
        • Guttmann A.
        • et al.
        The Reporting of Studies Conducted Using Observational Routinely-Collected Health Data (RECORD) statement.
        PLoS Med. 2015; 12: e1001885
        • Wallace P.J.
        • Shah N.D.
        • Dennen T.
        • et al.
        Optum Labs: building a novel node in the learning health care system.
        Health Aff (Millwood). 2014; 33: 1187-1194
      3. Optum. Real world health care experiences from over 150 million unique individuals since 1993. Available at: https://www.optum.com/content/dam/optum/resources/productSheets/5302_Data_Assets_Chart_Sheet_ISPOR.pdf. Accessed June 6, 2017.

        • Bellolio M.F.
        • Sangaralingham L.R.
        • Schilz S.R.
        • et al.
        Observation status or inpatient admission: impact of patient disposition on outcomes and utilization among emergency department patients with chest pain.
        Acad Emerg Med. 2017; 24: 152-160
      4. Centers for Disease Control and Prevention. CDC compilation of opioid analgesic formulations with morphine milligram equivalent conversion factors, 2015 version. Available at: http://www.pdmpassist.org/pdf/BJA_performance_measure_aid_MME_conversion.pdf. Accessed May 3, 2016.

        • Liu Y.
        • Logan J.E.
        • Paulozzi L.J.
        • et al.
        Potential misuse and inappropriate prescription practices involving opioid analgesics.
        Am J Manag Care. 2013; 19: 648-665
        • Levy B.
        • Paulozzi L.
        • Mack K.A.
        • et al.
        Trends in opioid analgesic-prescribing rates by specialty, US, 2007-2012.
        Am J Prev Med. 2015; 49: 409-413
        • Elixhauser A.
        • Steiner C.
        • Harris D.R.
        • et al.
        Comorbidity measures for use with administrative data.
        Med Care. 1998; 36: 8-27
        • Quan H.
        • Sundararajan V.
        • Halfon P.
        • et al.
        Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.
        Med Care. 2005; 43: 1130-1139
        • Von Korff M.
        • Saunders K.
        • Thomas Ray G.
        • et al.
        De facto long-term opioid therapy for noncancer pain.
        Clin J Pain. 2008; 24: 521-527
        • Dowd B.E.
        • Greene W.H.
        • Norton E.C.
        Computation of standard errors.
        Health Serv Res. 2014; 49: 731-750
        • StataCorp
        Stata: Release 14.
        StataCorp LLC, College Station, TX2015
        • Weiner S.G.
        • Baker O.
        • Rodgers A.F.
        • et al.
        Opioid prescriptions by specialty in Ohio, 2010-2014.
        Pain Med. 2017; (http://dx.doi.org/10.1093/pm/pnx027)
        • Strang J.
        • McCambridge J.
        • Best D.
        • et al.
        Loss of tolerance and overdose mortality after inpatient opiate detoxification: follow up study.
        BMJ. 2003; 326: 959-960
        • White J.M.
        • Irvine R.J.
        Mechanisms of fatal opioid overdose.
        Addiction. 1999; 94: 961-972
        • Miller M.
        • Barber C.W.
        • Leatherman S.
        • et al.
        Prescription opioid duration of action and the risk of unintentional overdose among patients receiving opioid therapy.
        JAMA Intern Med. 2015; 175: 608-615