Temporal Factors Associated With Opioid Prescriptions for Patients With Pain Conditions in an Urban Emergency Department

Key Points Question Have emergency department clinicians responded to the opioid epidemic through altering opioid prescription rates? Findings In this cross-sectional study of 556 176 emergency department patient encounters and 70 218 opioid prescriptions within a single emergency department, yearly prescriptions decreased by 66.3% between 2013 and 2018. This decrease was associated with a 71.1% reduction in the number of opioid prescriptions for musculoskeletal pain (back, limb, joint, and neck pain) and lesser, but still marked, decreases for fractures and kidney stones. Meaning Reductions in yearly opioid prescriptions across varying indications appear to be aligned with recognition of the opioid crisis in addition to national, state, and departmental education guidelines.


Introduction
Heightened attention to the prescription of opioids for the treatment of pain has been a central goal in medicine over the past decade.Opioid misuse was associated with 68% of US drug overdose deaths in 2017 and more than 400 000 deaths from 1999 to 2017. 1,26][7] A 2018 study 8 suggested that emergency department (ED)   prescriptions following new Centers for Disease Control and Prevention guidelines 9 show little association with long-term opioid use, although up to 13.4% of Medicare patients in the study went on to receive long-term opioid therapy.In any case, a 2015 study reported that 17.1% of all ED patients were discharged with an opioid prescription during the week of data collection, 10 and a 2017 study demonstrated equal efficacy for certain pain treatment in the ED with nonopioid analgesics. 11It is challenging for prescribers to discern the benefits and risks of opioid prescribing within an encounter for acute pain, [12][13][14][15][16] but with up to two-thirds of all ED patients seeking treatment for pain, [17][18][19] a 22.2% nationwide reduction in all opioid prescriptions ordered from 2013 to 2017, 20 and guidelines recommending judicious opioid prescribing, 9,21 it is important to discern whether emergency medicine is reducing opioid prescribing for the treatment of pain.
The aim of this study was to evaluate temporal changes in overall opioid prescribing and prescriptions for specific pain conditions in an urban academic ED between 2009 and 2018.In addition, the temporal pattern of opioid prescribing at the individual clinician level was examined, as previous studies have indicated that the decrease in opioid prescription counts may be dependent on a subset of clinicians decreasing opioid prescribing, while others maintain high-intensity prescribing, regardless of specialty 22 and including ED clinicians. 5,23,24We also examined demographic factors that may be associated with opioid prescribing to assess the possibility of underlying opioid prescription bias within the ED.Opioid prescriptions were manually selected by name of the drug and are included in eTable 1 in the Supplement.Hydrocodone plus acetaminophen was the primary agent, representing 97.1% of all of the prescriptions.Diagnostic conditions were defined using ICD-9 and ICD-10 codes and are presented in eTable 2 in the Supplement.

All patient encounters in the
Twelve diagnostic conditions-back pain, joint pain, limb pain, neck pain, fracture, sprain, contusion, other unspecified injury, abdominal pain, kidney stone, respiratory distress, and pharyngitis-were selected for analyses because they had the highest opioid prescription volume.
Patients with these conditions accounted for 59.4% of all opioids prescribed and allowed for distinct and convenient grouping of patients based on pain sources (Figure 1).Encounters from 2009 to 2014 had an ICD-9 code defined as primary, identifying the likely condition for which an opioid was prescribed within the encounter.After 2014, ICD-10 codes were implemented and primary codes were no longer delineated within the data set obtained.To ensure that the opioid was given for the specific condition, patients within a singular ICD-10 code were included for selection into a condition.
Although data on certain patients may be lost using this criterion, yearly patient counts in each condition remained relatively consistent with the years using ICD-9 coding, demonstrating few exclusions.Patients with multiple ICD-10 codes within the same diagnostic group only, most notably fractures, were also included.Any patients with an ICD-10 code for other unspecified injury were included within this diagnostic group, as this was likely a secondary code in the ICD-10 system and kept yearly patient counts similar to ICD-9 years.Because the aim was to look at changes over time, changes from 2009 to 2014 will have consistency within the ICD-9 system, and those from 2015 to 2018 will have consistency within the ICD-10 system.From these conditions, patients were categorized into 3 groups: musculoskeletal pain (back, joint, limb, and neck pain), musculoskeletal trauma (fracture, sprain, contusion, and other unspecified injury), and nonmusculoskeletal pain (abdominal pain, kidney stone, respiratory distress, and pharyngitis).These groupings define the source of the pain, identify the observation of objective pathologic factors by the clinician (pain vs trauma), and delineate opioid prescriptions between musculoskeletal and nonmusculoskeletal conditions.Any patient with a fracture, sprain, and/or contusion ICD-10 code in addition to an other unspecified injury diagnosis code was not double counted in the musculoskeletal trauma grouping.

Statistical Analysis
Baseline demographic characteristics and characteristics of patient subsets were determined using descriptive analyses.Absolute and relative opioid prescription changes were descriptively evaluated as a function of time, condition group (ie, musculoskeletal trauma, musculoskeletal pain, and other pain), and conditions within condition groups.Proportions and their SEs were calculated with normal approximations (ie, SE = [p(1 − p)/n] 1/2 ).Following descriptive evaluation of the data, 2013 was chosen as the reference year for continuous and controlled estimates of the effects of time in our population because that is when opioid prescribing peaked.Opioid prescription counts were determined by sex (male, female), race/ethnicity (white, black, Hispanic, Asian, and other), insurance status (private, Medicare, Medicaid, and self-pay), and age (0-15,16-30, 31-65, and >65 years) for all encounters and in conditions of interest.
Following descriptive evaluation of the data, inferential statistics were carried out to further examine temporal opioid prescribing.Specifically, univariable and multivariable logistic regression models were constructed, with each modeling whether an opioid was prescribed within an encounter as the dependent variable and year as the primary independent variable.Multivariable models incorporated adjustments for age, sex, race/ethnicity, and insurance status (stratified as described in the Methods section); age 31 to 65 years, male, white race, and private insurance were chosen as reference categories because they represented the highest proportion of opioid prescriptions among the patient subgroups.Odds ratios (ORs), adjusted ORs (aORs), and their 95% CIs were calculated.All ORs and aORs represent the odds of receiving an opioid relative to the prior year, with 2013 being the intercept.In addition, a multivariable logistic regression with interaction terms between year and race/ethnicity, year and age, year and sex, and year and insurance status were used to examine whether differences in the reduction of opioid prescriptions from 2013 to 2018 existed within patient subgroups.
Twelve clinicians were selected for having more than 10 000 encounters from 2009 to 2017.
Clinician-level data-but not other data-from 2018 were not available, so this year was excluded for clinician-level analyses.These 12 clinicians were chosen because they represented the upper tercile of ED prescribers by opioid prescription numbers during this period and saw a representative caseload in a year over most years, allowing for temporal analysis.Multivariable logistic regression models, which adjusted for patient age, sex, race/ethnicity, and insurance status, were used for individual clinicians to examine their opioid prescribing over time while controlling for patient demographic characteristics.Adjusted ORs and 95% CIs for each clinician were calculated.All data were processed using SAS, version 9.4 (SAS Institute Inc).Findings were considered significant at 2-sided, 2-tailed P = .05.
The peak opioid prescription rates for each clinician in any single year between 2012 and 2015 ranged from 15.1 to 19.9 opioid prescriptions per 100 encounters.All physicians decreased the number of opioid prescriptions, such that in 2017, no single physician of the 12 included in the analysis prescribed more than 8.8 opioids per 100 encounters, which was associated with a 44.7% to 61.9% decrease from 2013 to 2017.The decrease in opioid prescribing was substantial and relatively similar in magnitude across 11 of 12 clinicians when controlling for patient demographic characteristics (Table 3).

Discussion
Much attention has been given to the prescribing of opioids for pain by US physicians in response to the opioid epidemic.In a study of opioid prescribing within an urban academic ED, our analysis notes the expected temporal changes given the nationwide attention to opioid prescribing while providing details of prescription patterns by physicians for patients within certain conditions and demographic    treatment of musculoskeletal pain have minimal effect on pain and disability, 25 high opioid burden, 17,26,27 increased adverse effects, 28 and possible increased likelihood of repeated use from a single opioid prescription. 6,29Likewise, many of the patients diagnosed with back, joint, limb, and neck pain have this pain chronically and present to the ED for acute pain episodes with regularity. 17,30,31Guidelines recommend against opioid prescription in these cases. 21With up to 10% to 16% of patients presenting to the ED with chronic pain, 32,33 these musculoskeletal conditions are an important diagnostic group to target for nonopioid pharmacologic interventions.

JAMA Network Open | Emergency Medicine
All patient demographic subgroups saw a decrease in opioids prescribed for them following the peak of opioid prescribing in 2013.Comparing racial subgroups, black race was associated with the greatest decrease in opioid prescribing, as well as the lowest odds of receiving opioids across the entire decade.This finding is consistent with data reporting lower doses of analgesics provided to patients of minority racial/ethnic groups predating the recognition of the opioid crisis, as opposed to white patients who have historically had the highest likelihood of receiving opioids. 34,35Patients with Medicaid had the lowest odds of receiving an opioid-a group in which a prior study noted a high burden of opioid prescriptions in the ED for acute pain. 27In terms of patient age and in contrast to a nationwide study of ED opioid prescribing, there was no statistically significant difference in opioid prescribing between patients aged 31 to 65 and older than 65 years. 36 an individual clinician level, all analyzed physicians were associated with markedly and similarly reduced prescription rates from 2013 to 2017.0][41][42] Furthermore, in 2017, a quality-control program was implemented within our ED, in which quarterly prescribing patterns were reviewed by the individual clinicians who were compared with their peers. 43The consistency of reduction in opioid prescribing among clinicians demonstrates that treatment decisions are made not only on an individual level, but also within the larger context of the medical environment in which physicians are influenced by guidelines and departmental policy.

Limitations
This study has several limitations.The use of ICD codes for conditions does not necessarily mean the patient was given the opioid for that condition, although steps were made to diminish this possible factor.As always, a drug prescribed for a patient does not guarantee consumption.Pharmacotherapy using nonopioid alternatives does not necessarily improve an individual outcome, and given that this study was conducted in an ED, long-term outcomes (repeat visits, repeat prescriptions, and opioid use disorder) are difficult to analyze.Data on the severity of pain were not available and comorbidities (eg, cancer) were not analyzed, although this information likely would not change the overall conclusion.In addition, the change from ICD-9 to ICD-10 diagnosis codes in 2015 created discrepancies between the number of patients in that year compared with the other years, so caution should be used in examining 2015 data independently from the overall pattern during the study period.Another limitation is that this study did not have robust data for quantity and dose of the opioid used-this information is important because higher morphine milligram equivalents are associated with long-term opioid use and death, [44][45][46] and the clinician analyses in prior studies included this factor to define high-and low-intensity prescribing patterns in clinicians. 5,22These data points were intermittent owing to interruptions in data collection at the Enterprise Data Warehouse from various electronic health record changes.In addition, we recognize that the generalizability of this study, given that it focuses on a single department with a single set of physicians, is limited.This study reports, however, an association between a targeted reduction in opioid prescriptions for musculoskeletal pain conditions, such as back, joint, limb, and neck pain, and a major decrease in opioid prescribing, including a collective decrease in opioid prescriptions across all clinicians within the ED.
Northwestern Memorial Hospital ED and Northwestern Memorial Hospital Feinberg Mezzanine Emergency Room, Chicago, Illinois, between January 1, 2009, and December 31, 2018, were selected from the Northwestern Medicine Enterprise Data Warehouse.An encounter was defined by a unique patient (identified by a unique patient identifier) having a unique time and date entered into the Enterprise Data Warehouse database from the electronic health record.An encounter included the self-identified age, sex, race/ethnicity, payer status, opioid prescribed, deidentified physician prescriber, and International Classification of Diseases, Ninth Revision (ICD-9), and International Statistical Classification of Diseases, 10th Revision (ICD-10), diagnosis codes for each patient.To fully anonymize the data, the Enterprise Data Warehouse assigned each patient and physician a randomized unique identifier, had visit dates shifted within a 10-day window, and grouped patient age within 5 years to properly deidentify the data set.This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies.Exclusion criteria included any encounter without an ICD diagnosis and encounters not labeled as emergency.The study was approved by the institutional review board at Northwestern University.All data were deidentified and a waiver of informed consent was granted by the institutional review board.

Figure 1 .
Figure 1.Temporal Opioid Prescribing Within Diagnosis Groups

(4. 9 )
Abbreviation: NR, not reported.a Temporal opioid use within the Northwestern Memorial Hospital emergency department, including all patients seen and within demographic subgroups for 2-year increments.

Table 2 .
20mporal Opioid Prescription and Odds of Opioid Prescription Compared With the Prior Year by Condition subgroups over time.From 2013 to 2018, the ED experienced a 66.3% decrease in opioid prescriptions-a much greater reduction than the national decrease of 22% from 2013 to 2017.20This reduction exceeds the 54% decrease in initial treatment in nationwide opioid prescribing for opioidnaive patients and is markedly greater than the 16% decrease for all patients (naive and non-naive)

Table 3 .
Temporal Opioid Use and Odds of Opioid Use Compared With the Prior Year for Individual Clinicians a