The n e w e ng l a n d j o u r na l of m e dic i n e

Special Article

Opioid-Prescribing Patterns of Emergency

Physicians and Risk of Long-Term Use
Michael L. Barnett, M.D., Andrew R. Olenski, B.S.,
and Anupam B. Jena, M.D., Ph.D.​​

A BS T R AC T

BACKGROUND
Increasing overuse of opioids in the United States may be driven in part by physician From the Department of Health Policy
prescribing. However, the extent to which individual physicians vary in opioid pre- and Management, Harvard T.H. Chan
School of Public Health (M.L.B.), the De-
scribing and the implications of that variation for long-term opioid use and adverse partment of Health Care Policy, Harvard
outcomes in patients are unknown. Medical School (A.R.O., A.B.J.), the Divi-
sion of General Internal Medicine and
METHODS Primary Care, Department of Medicine,
Brigham and Women’s Hospital (M.L.B.),
We performed a retrospective analysis involving Medicare beneficiaries who had and the Department of Medicine, Massa-
an index emergency department visit in the period from 2008 through 2011 and chusetts General Hospital (A.B.J.), Boston,
had not received prescriptions for opioids within 6 months before that visit. After and the National Bureau of Economic Re-
search, Cambridge (A.B.J.) — all in Massa-
identifying the emergency physicians within a hospital who cared for the patients, chusetts. Address reprint requests to Dr.
we categorized the physicians as being high-intensity or low-intensity opioid prescrib- Barnett at the Department of Health Pol-
ers according to relative quartiles of prescribing rates within the same hospital. We icy and Management, Harvard T.H. Chan
School of Public Health, Kresge Bldg.,
compared rates of long-term opioid use, defined as 6 months of days supplied, in 4th Fl., 677 Huntington Ave., Boston, MA
the 12 months after a visit to the emergency department among patients treated 02115, or at ­mbarnett@​­hsph​.­harvard​.­edu.
by high-intensity or low-intensity prescribers, with adjustment for patient charac- This article was updated on February 16,
teristics. 2017, at NEJM.org.

RESULTS N Engl J Med 2017;376:663-73.

DOI: 10.1056/NEJMsa1610524
Our sample consisted of 215,678 patients who received treatment from low-inten- Copyright © 2017 Massachusetts Medical Society.
sity prescribers and 161,951 patients who received treatment from high-intensity
prescribers. Patient characteristics, including diagnoses in the emergency depart-
ment, were similar in the two treatment groups. Within individual hospitals, rates
of opioid prescribing varied widely between low-intensity and high-intensity pre-
scribers (7.3% vs. 24.1%). Long-term opioid use was significantly higher among
patients treated by high-intensity prescribers than among patients treated by low-
intensity prescribers (adjusted odds ratio, 1.30; 95% confidence interval, 1.23 to 1.37;
P<0.001); these findings were consistent across multiple sensitivity analyses.
CONCLUSIONS
Wide variation in rates of opioid prescribing existed among physicians practicing
within the same emergency department, and rates of long-term opioid use were in-
creased among patients who had not previously received opioids and received
treatment from high-intensity opioid prescribers. (Funded by the National Insti-
tutes of Health.)

n engl j med 376;7 nejm.org  February 16, 2017 663

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

R 
ates of opioid prescribing and Me thods
opioid-related overdose deaths have qua-
drupled in the United States over the past Study Population
three decades.1-3 This epidemic has increasingly Using the Centers for Medicare and Medicaid Ser-
affected the elderly Medicare population, among vices carrier files for a 20% random sample of
whom rates of hospitalization for opioid overdoses beneficiaries from January 1, 2008, through De-
quintupled from 1993 through 2012.4-6 The risks cember 31, 2011, we identified index emergency
of opioid use are particularly pronounced among department visits at acute care hospitals by Medi-
the elderly, who are vulnerable to their sedating care beneficiaries. We defined an index visit as the
side effects, even at therapeutic doses.7 Multiple earliest visit at which a beneficiary had an evalu-
studies have shown increased rates of falls, frac- ation and a management claim by an emergency
tures, and death from any cause associated with medicine physician with a place-of-service desig-
opioid use in this population.8-11 Even short-term nation in the emergency department. Emergency
opioid use may confer a predisposition to these medicine physicians were defined as physicians
side effects and to opioid dependence.12 with an emergency medicine specialty who
It is frequently argued that the prescribing be- billed 90% or more of claims with an emergency
havior of physicians has been a driver of the opioid department place of service. We included only
epidemic.13,14 Prescribing has increased to the point one index visit to the emergency department per
that in 2010, enough opioids were prescribed in beneficiary and excluded all visits to the emer-
the United States to provide every American adult gency department that resulted in a hospital ad-
with 5 mg of hydrocodone every 4 hours for a mission.
month.1,14 This growth may be driven in part by We limited our analyses to beneficiaries who
high variability in physician prescribing of opi- had been continuously enrolled in Medicare Part D
oids; this variability may reflect overprescribing for 18 months or more, including at least the
beyond what is required for appropriate pain period from 6 months before the index visit to
management.1,15,16 This inconsistency is not sur- 12 months afterward. We included only beneficia-
prising, because few clinical guidelines exist, and ries who had not had an opioid prescription filled
there is limited evidence to direct the appropri- in the 6 months before the index visit. In addition,
ate use of opioids.17,18 However, few studies have we excluded beneficiaries with hospice claims or a
investigated the extent to which individual phy- cancer diagnosis between 2008 and 2012.
sicians vary in opioid prescribing and the impli- We assigned each index emergency department
cations of that variation for long-term opioid use visit to a physician according to the billing Na-
and related adverse outcomes in patients. tional Provider Identifier (NPI) and then linked
To examine the extent to which emergency phy- each visit to a hospital by matching to facility
sicians within the same hospital varied in rates of claims in the outpatient department file according
opioid prescribing, we studied a national sample of to the date and beneficiary. To ensure that we had
Medicare beneficiaries who received treatment in an adequate sample size at the physician and hos-
an emergency department and who had not used pital level, we excluded physicians with fewer than
prescription opioids within 6 months before the five emergency department visits and hospitals
index visit to the emergency department. In order with fewer than five physicians billing for emer-
to understand how initial exposure to an opioid gency department visits in our sample. If physi-
relates to subsequent outcomes, we identified high- cians practiced in more than one hospital, they
intensity and low-intensity opioid prescribers with- were assigned to the hospital at which they had the
in each hospital and examined rates of long-term most visits, and any visits at other facilities were
opioid use and future hospitalizations among pa- excluded.
tients treated by these two groups of prescrib- This study was approved by the institutional
ers.19,20 To address the challenge of selection bias, review board at Harvard Medical School, which
we relied on the fact that patients are unlikely to waived the requirement for informed consent
choose an emergency department physician once since the data were deidentified and only aggre-
they have chosen a facility. gate results would be reported.

664 n engl j med 376;7 nejm.org  February 16, 2017

The New England Journal of Medicine

Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 Opioid Prescribing and Risk of Long-Term Use

Definitions of Opioid Prescriptions the main exposure. We chose 180 days as a spe-
and Intensity of Physicians’ Prescribing cific marker for clinically significant long-term
We identified prescription claims corresponding opioid use beyond the common duration of 90 days
to an opioid (excluding methadone) according to described in previous literature.16,22,23 Therefore,
the National Drug Code in the Medicare Part D this outcome captures the extent to which other
database.16 We attributed an opioid prescription to physicians prescribe opioids for the subsequent
an index emergency department visit and the as- 12 months after a patient’s index emergency de-
sociated physician if it was filled by the patient partment visit.
within 7 days after the date of the emergency de- Secondary outcomes were rates of hospital
partment visit; in a sensitivity analysis, we restrict- encounters (emergency department visits, hospi-
ed this duration to 3 days. This attribution method talizations, or both), including those potentially
was necessary because prescriber NPI informa- related to adverse effects of opioids and those
tion is not available in the Part D database (see associated with a selection of medical conditions
the Methods and Results section in the Supple- that were unlikely to be influenced by opioid use,
mentary Appendix, available with the full text of in the 12 months after an index emergency depart-
this article at NEJM.org). For this and subsequent ment visit (definitions are provided in the Methods
opioid prescriptions, we extracted the number of and Results section in the Supplementary Appen-
days for which opiates were supplied and calcu- dix).8-10,24 To assess for possible undertreatment of
lated the morphine equivalents dispensed, using pain by low-intensity prescribers that could have
standard conversion tables.21 led to repeat emergency care, we also measured
We defined the main exposure as treatment rates of repeat emergency department visits at 14
by a “high-intensity” or “low-intensity” opioid pre- and 30 days that resulted in the same primary di-
scribing physician within the same hospital. For agnosis as the initial emergency department visit,
each physician, we first calculated the proportion classified according to Clinical Classifications Soft-
of all emergency department visits after which an ware (CCS) groups (categorizations of codes in the
opioid prescription was filled. We then grouped International Classification of Diseases, 9th Revision).25
physicians into quartiles of rates of opioid pre-
scribing within each hospital and classified physi- Patient Covariates
cians as being in the top (high-intensity) or bot- We collected information on the patients’ age, sex,
tom (low-intensity) quartile of prescribing rates. In race or ethnic group, dual eligibility for Medicaid
93 hospitals, because of a high number of pre- and Medicare coverage, and disability status.26 Us-
scribers who did not prescribe opioids, there were ing the Chronic Conditions Warehouse database,
fewer than four separate groups of prescribers to we also captured the presence of any of 11 chronic
assign to quartiles; therefore, the highest and low- conditions (Table 1) as well as the number of coex-
est prescribers in a hospital were assigned to the isting chronic conditions that a patient had at the
high-intensity and low-intensity groups. We also time of an index emergency department visit.
defined an alternative exposure, classifying phy-
sicians according to the median dose (in morphine Statistical Analysis
equivalents) of prescriptions filled after an emer- Our strategy to reduce selection bias relied on the
gency department visit (“high-dose intensity” and assumption that within the same hospital, patients
“low-dose intensity”). do not choose specific emergency physicians, and
therefore patients treated by physicians of varying
Outcomes opioid prescribing intensity may be similar with
Our primary outcome was long-term opioid use, respect to both observable and unobservable char-
which we defined as 180 days or more of opioids acteristics. To assess this approach, we compared
supplied in the 12 months after an index emer- the characteristics of patients who saw high-inten-
gency department visit, excluding prescriptions sity prescribers with those who saw low-intensity
within 30 days after the index visit. We applied this prescribers. We assessed balance in the case mix
exclusion because otherwise this outcome, by de- by comparing rates of visits classified according to
sign, would be correlated with our definition of the top 25 CCS groups, as well as by plotting the

n engl j med 376;7 nejm.org  February 16, 2017 665

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Table 1. Characteristics of the Patients, According to Opioid Prescribing Intensity of Physician Seen.*

Patients Treated by Patients Treated by

Low-Intensity Prescriber High-Intensity Prescriber
Characteristic (N = 215,678)† (N = 161,951)† P Value‡
Age (yr) 68.8±16.3 68.2±16.4 <0.001
Female sex (%) 64.5 64.5 0.75
White race (%)§ 76.6 75.9 0.03
Medicare–Medicaid dual eligibility (%) 49.3 49.6 0.19
Disabled (%) 36.4 37.6 <0.001
Chronic conditions (no.) 3.6±2.3 3.5±2.3 <0.001
Presence of chronic illness (%)
Acute myocardial infarction 49.8 48.5 <0.001
Alzheimer’s dementia 17.5 16.0 <0.001
Atrial fibrillation 13.6 12.7 <0.001
Cerebrovascular disease 17.3 16.3 <0.001
Chronic kidney disease 20.6 20.1 <0.001
Chronic obstructive pulmonary disease 28.1 27.2 <0.001
Congestive heart failure 30.5 29.0 <0.001
Depression 39.5 39.5 0.94
Diabetes 37.8 37.7 0.69
Hyperlipidemia 69.5 69.1 0.09
Hypertension 78.1 77.4 <0.001
Census region (%)
Northeast 19.8 20.9 0.03
Midwest 24.4 23.2 0.01
South 38.4 38.6 0.82
West 17.3 17.3 0.97

* Plus–minus values are means ±SD.

† Within each hospital, high-intensity and low-intensity prescribers were defined as the top (high) or bottom (low) quar-
tile of emergency physician prescribers in terms of opioid prescribing. Since many physicians did not prescribe opioids,
some hospitals had many physicians with prescribing rates equal to zero in the low-intensity prescriber group, making
it larger overall than the high-intensity group. Low-intensity prescribers included 8297 physicians; the total number (±SD)
of patients per physician was 26.0±18.6. High-intensity prescribers included 6133 physicians; the total number of pa-
tients per physician was 26.4±18.5.
‡ Unadjusted P values were estimated with the use of Student’s t-test for means or the z-test for comparison of proportions.
§ Race or ethnic group was self-reported.

cumulative distribution of the primary-diagnosis the occurrence of primary or secondary outcomes.

CCS group between high-intensity and low-inten- The key explanatory variable was a binary indica-
sity prescribers. Differences in distributions be- tor for whether a patient was treated by a high-
tween groups were assessed with the use of the intensity or low-intensity prescriber (the middle
Kolmogorov–Smirnov test.27 two quartiles of prescribers were not included in
We then computed unadjusted rates and odds these models). Other covariates included the
ratios for each outcome, stratified according to patients’ age, sex, race or ethnic group, Medicaid
treatment by a high-intensity or low-intensity eligibility, and disability status, as well as the
opioid prescriber. To account for residual differ- presence of 11 different chronic conditions. We
ences between patient populations, we estimated prespecified several subgroup analyses to assess
adjusted odds ratios with patient-level multivari- for heterogeneity (Table 2). All models account-
able logistic regression. Dependent variables were ed for grouping of patients within hospitals with

666 n engl j med 376;7 nejm.org  February 16, 2017

The New England Journal of Medicine

Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 Opioid Prescribing and Risk of Long-Term Use

the use of robust standard errors clustered at the low-intensity prescribers and high-intensity pre-
hospital level.28 scribers varied by a factor of 3.3 within the same
The hypothetical long-term effect of filling hospital (7.3% vs. 24.1% of emergency department
an initial opioid prescription after an emergency visits, P<0.001) (Table 2 and Fig. 1A). Across all
department visit versus not filling a prescription subgroups, prescribing rates among high-intensi-
was estimated to approximate the number of pa- ty prescribers were triple those among low-inten-
tients who would need to be prescribed an initial sity prescribers. The highest average rate was seen
opioid for one patient to become a long-term user. among patients who visited the emergency de-
A definition of “number needed to harm” is pro- partment with an injury (23.7%) (Table 2). There
vided in the Methods section in the Supplemen- was minimal correlation between physicians’ pre-
tary Appendix.29 scribing rates and the median initial dose of an
We performed additional sensitivity analyses opioid prescription that was filled (r = −0.08) (Fig.
to address the possibility of selection bias and S3 and Table S1 in the Supplementary Appendix)
sensitivity to design assumptions (details are pro- or type of opioid prescribed (Table S2 in the Sup-
vided in the Methods section in the Supplementary plementary Appendix).
Appendix). These analyses included an alternative Overall, long-term opioid use at 12 months
exposure in which intensity was defined accord- was significantly higher among patients treated
ing to the median dose of opioid prescription filled by high-intensity prescribers than among patients
after an emergency department visit, as described treated by low-intensity prescribers (1.51% vs.
above. Replicating our results with an alternative 1.16%; unadjusted odds ratio, 1.31; 95% confi-
definition of exposure that operates through a dence interval [CI], 1.24 to 1.39) (Table 2 and
similar causal pathway (increased opioid exposure) Fig. 2). After adjustment, there was minimal
could argue against selection bias, particularly if change in this difference (adjusted odds ratio,
the two exposures (dose and frequency of opioid 1.30; 95% CI, 1.23 to 1.37). This finding corre-
prescribing) are minimally correlated. sponds to a number needed to harm of 48 pa-
All analyses were performed with the use of tients receiving an opioid prescription to theo-
Stata software, version 14.1 (StataCorp). The 95% retically lead to 1 excess long-term opioid user.
confidence intervals around reported estimates We observed a stepwise increase in long-term
reflect 0.025 in each tail, or P values no higher opioid use with exposure to physicians in each
than 0.05. quartile of opioid prescribing frequency. As com-
pared with the first (low-intensity) quartile, pa-
tients treated by physicians in the second quar-
R e sult s
tile had an adjusted odds ratio for long-term opioid
Our sample consisted of 215,678 patients treated use of 1.10 (95% CI, 1.04 to 1.16) and patients
by a low-intensity opioid prescriber and 161,951 treated by physicians in the third quartile had an
patients treated by a high-intensity opioid pre- adjusted odds ratio of 1.19 (95% CI, 1.12 to 1.25)
scriber during an index emergency department (Fig. 1B, and Table S3 in the Supplementary Ap-
visit. Overall, the characteristics of patients treated pendix). Differences in long-term opioid use be-
by high-intensity opioid prescribers were similar tween patients treated by high-intensity prescribers
to those of patients treated by low-intensity pre- and those treated by low-intensity opioid prescrib-
scribers, although several differences were signifi- ers were consistent across subgroups, with mini-
cant given the large sample size (Table 1). Diag- mal change after multivariable adjustment (Fig. 2).
noses in patients seen by high-intensity prescribers Rates of opioid-related hospital encounters and
and those seen by low-intensity prescribers were encounters for fall or fracture were significantly
similar (P = 0.87 by the Kolmogorov–Smirnov test higher in the 12 months after an index emergency
for differences in the distribution of 300 CCS department visit among patients treated by high-
groups, according to prescriber group [Fig. S1 in intensity opioid prescribers than among patients
the Supplementary Appendix; for distribution of treated by low-intensity opioid prescribers (rates of
the top 25 of 300 CCS diagnosis groups, according any opioid-related encounter, 9.96% vs. 9.73%;
to prescriber group, see Fig. S2 in the Supplemen- adjusted odds ratio, 1.03; 95% CI, 1.00 to 1.05;
tary Appendix]). P = 0.02; rates of encounters for fall or fracture,
On average, rates of opioid prescribing between 4.56% vs. 4.28%; adjusted odds ratio, 1.07; 95% CI,

n engl j med 376;7 nejm.org  February 16, 2017 667

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Table 2. Rate of Filling a Prescription for Opioids within 7 Days after Emergency Department Visit and Rate of Long-Term Use,
According to Opioid Prescribing Intensity of Physician Seen.*

Initial Rate of Opioid Prescription

Variable in Emergency Department Rate of Long-Term Use
Average Low-Intensity High-Intensity Low-Intensity High-Intensity
Rate Prescriber Prescriber Prescriber Prescriber
percent
Overall rate of opioid prescribing 14.7 7.3 24.1 1.16 1.51
Rate of prescribing according to patient characteristic
Age
<65 yr 17.9 8.8 28.9 2.09 2.82
65–74 yr 16.4 8.4 26.4 0.86 1.00
75–84 yr 12.3 6.0 20.4 0.68 0.89
≥85 yr 8.9 4.3 15.6 0.86 1.01
Sex
Male 15.1 7.6 24.7 1.22 1.53
Female 14.4 7.1 23.8 1.13 1.51
Race
White 14.6 7.5 23.7 1.16 1.50
Black 14.9 6.8 25.0 1.38 1.87
Medicare–Medicaid dual eligibility
No 14.4 7.4 23.4 0.71 0.87
Yes 15.0 7.2 24.8 1.62 2.17
No. of chronic conditions†
0 18.7 9.5 29.9 1.29 1.73
1 or 2 16.9 8.5 27.4 1.07 1.52
≥3 13.3 6.5 22.0 1.17 1.48
Alzheimer’s disease
No 15.9 8.0 25.9 1.15 1.52
Yes 8.5 4.0 14.7 1.19 1.49
Disabled
No 13.4 6.7 22.1 0.71 0.87
Yes 17.0 8.3 27.5 1.95 2.59
Depression
No 14.8 7.4 24.2 0.78 1.01
Yes 14.5 7.1 23.9 1.74 2.28
Census region
Northeast 11.9 5.2 20.5 0.78 1.11
Midwest 14.0 6.9 23.4 1.27 1.45
South 15.9 8.4 25.3 1.33 1.81
West 16.2 7.7 26.7 1.07 1.44
Emergency department visit for injury‡
No 13.7 6.7 22.7 1.16 1.53
Yes 23.7 12.6 35.8 1.14 1.42

* Long-term opioid use was defined as 6 months of days of opioids supplied in the 12 months after an index emergency department visit,
­excluding the first 30 days after the index emergency department visit.
† The number of chronic conditions among 11 possible conditions is shown. These conditions are the following: acute myocardial infarction,
Alzheimer’s dementia, atrial fibrillation, cerebrovascular disease, chronic kidney disease, chronic obstructive pulmonary disease, congestive
heart failure, depression, diabetes, hyperlipidemia, and hypertension.
‡ An emergency department visit for an injury was defined as any emergency department visit with an “E” code associated with an injury
­(according to the codes in the International Classification of Diseases, 9th Revision). Of patients who visited an emergency department for a
reason other than an injury, 195,651 saw low-intensity prescribers and 144,895 saw high-intensity prescribers. Of patients who visited an
emergency department because of an injury, 20,027 saw low-intensity prescribers and 17,056 saw high-intensity prescribers.

668 n engl j med 376;7 nejm.org  February 16, 2017

The New England Journal of Medicine

Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 Opioid Prescribing and Risk of Long-Term Use

1.03 to 1.11; P<0.001) (Table 3). There was no this claim are lacking. Our results provide evi-
significant difference in 12-month rates of over- dence that this mechanism could drive initiation
all hospital encounters or non–opioid-related en- of long-term opioid use through either increased
counters. Assessment of rates of short-term emer- rates of opioid prescription or prescription of a
gency department revisits for possible evidence high, versus a low, dose of opioid. Although cau-
of undertreated pain showed that rates of 14-day sality cannot be established from this observa-
and 30-day repeat emergency department visits
with the same primary diagnosis as the index visit
A
were similar in the two prescriber groups (P>0.07)
30
(Table S4 in the Supplementary Appendix).
We replicated the main analysis described above

Percent of Emergency Department

Visits with an Opioid Prescription
25
in multiple other sensitivity analyses, including one
that used a propensity score–matched cohort (see 20
the Results section in the Supplementary Appen-
dix), although the finding of increased hospital 15
encounters for falls or fractures with the use of
propensity score matching (P = 0.05) and with the 10
use of CCS category fixed effects (P = 0.12) was
not significant (Table S6 in the Supplementary 5

Appendix). In analyses in which physicians were

0
categorized as being high or low “dose intensity” 1 2 3 4
prescribers, we found increased odds of long-term Prescriber Quartile
opioid use among patients treated by high dose-
intensity prescribers (odds ratio, 1.32; 95% CI, B
1.19 to 1.46) (Table S6 and Fig. S4 in the Supple- 1.4
mentary Appendix).
Odds Ratio for Long-Term Opioid Use

1.3
Discussion
In a large, national sample of patients enrolled in
1.2
Medicare Part D who received care in an emer-
gency department and who had not used prescrip-
tion opioids in the 6 months before the visit to
1.1
the emergency department, we found substantial
variation in the opioid prescribing patterns of
emergency physicians within the same hospital. 1.0
The intensity of a physician’s opioid prescribing 1 2 3 4
(Low intensity, (High
was positively associated with the probability that reference) intensity)
a patient would become a long-term opioid user Prescriber Quartile
over the subsequent 12 months. Although our
study was observational, we sought to minimize Figure 1. Prescribing Rates and Adjusted Odds Ratios
for Long-Term Opioid Use, According to Quartile of
selection bias by comparing the characteristics of Physician Opioid Prescribing.
patients seen by different emergency physicians Panel A shows rates of opioid prescribing by emergen-
within the same hospital. The association between cy physicians according to within-hospital quartile.
physician prescribing rates and increased long- I bars represent 95% confidence intervals. Panel B
term opioid use was consistent across numerous shows the adjusted odds ratios and corresponding
subgroups and across all quartiles of physician 95% confidence intervals for rates of long-term opioid
use, according to quartile of physician opioid prescrib-
prescribing in a dose–response pattern. ing. Physicians in each quartile were compared with
It is commonly thought that opioid dependence those in the lowest prescribing quartile. Odds ratios
often begins through an initial, possibly chance, were estimated with the use of logistic-regression
exposure to a physician-prescribed opioid, al- models.
though data from studies to empirically evaluate

n engl j med 376;7 nejm.org  February 16, 2017 669

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Subgroup Odds Ratio

Unadjusted Adjusted (95% CI)
Overall opioid use at 12 mo 1.31 1.30 (1.23–1.37)
Age
<65 yr 1.36 1.37 (1.27–1.48)
65–74 yr 1.16 1.17 (1.03–1.32)
75–84 yr 1.32 1.33 (1.15–1.54)
≥85 yr 1.18 1.18 (0.98–1.42)
Sex
Male 1.26 1.25 (1.14–1.37)
Female 1.34 1.33 (1.24–1.43)
Race
White 1.29 1.28 (1.20–1.37)
Black 1.36 1.35 (1.19–1.53)
Census region
Northeast 1.43 1.42 (1.23–1.64)
Midwest 1.14 1.12 (1.00–1.27)
South 1.37 1.36 (1.25–1.48)
West 1.35 1.34 (1.17–1.53)
No. of chronic conditions
None 1.35 1.36 (1.15–1.61)
1 or 2 1.43 1.41 (1.26–1.58)
≥3 1.26 1.25 (1.17–1.34)
Medicare–Medicaid dual eligibility
No 1.22 1.20 (1.08–1.33)
Yes 1.35 1.34 (1.26–1.44)
Depression
No 1.30 1.29 (1.18–1.41)
Yes 1.32 1.30 (1.21–1.40)
Emergency department visit for injury
No 1.32 1.30 (1.23–1.38)
Yes 1.25 1.25 (1.05–1.49)
Alzheimer’s disease
No 1.32 1.31 (1.23–1.39)
Yes 1.26 1.26 (1.09–1.44)
Disabled
No 1.23 1.24 (1.13–1.36)
Yes 1.33 1.34 (1.25–1.44)
0.8 1.2 1.6 2.0
Odds Ratio (95% CI)

Figure 2. Unadjusted and Adjusted Odds Ratios for Long-Term Opioid Use, According to Treatment by High-Intensity
or Low-Intensity Opioid Prescriber.
All unadjusted odds ratios were estimated with the use of bivariate logistic regression with the occurrence of long-
term opioid use as the dependent variable and exposure to a high-intensity provider as the key explanatory variable.
All adjusted models had further adjustment for the patients’ age, sex, race or ethnic group, Medicare–Medicaid
dual eligibility, and disability status and the presence of 11 chronic conditions.

tional study, if our results represent a causal re- to take hold. Conversion to long-term use may be
lationship, for every 48 patients prescribed a new driven partly by clinical “inertia” leading outpa-
opioid in the emergency department who might tient clinicians to continue providing previous
not otherwise use opioids, 1 will become a long- prescriptions. Such clinical inertia may affect only
term user; this is a low number needed to harm a narrow segment of the population; this could
for such a common therapy. explain why rates of initial opioid prescribing may
Of course, prescriptions provided by other phy- vary by a factor of three, whereas long-term use
sicians in the months after an emergency depart- varies by only approximately 30%. It is impor-
ment visit are necessary for long-term opioid use tant to acknowledge that patients treated by high-

670 n engl j med 376;7 nejm.org  February 16, 2017

The New England Journal of Medicine

Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 Opioid Prescribing and Risk of Long-Term Use

Table 3. Hospital Encounters within 12 Months after an Index Emergency Department Visit to a Low-Intensity or High-
Intensity Opioid Prescriber.*

Adjusted Odds Ratio

Type of Hospital Encounter 12-Mo Encounter Rate (95% CI)† P Value†

Patients Treated Patients Treated

by Low-Intensity by High-Intensity
Prescriber Prescriber
(N = 215,678) (N = 161,951)

% of patients
Any hospital encounter 60.5 60.3 0.99 (0.97–1.00) 0.13
Any hospitalization 46.1 45.8 0.99 (0.97–1.00) 0.15
Any emergency department visit 57.4 57.1 0.99 (0.97–1.00) 0.07
Any opioid-related hospital encounter 9.73 9.96 1.03 (1.00–1.05) 0.02
Fall or fracture 4.28 4.56 1.07 (1.03–1.11) <0.001
Constipation 4.16 4.11 0.99 (0.96–1.02) 0.44
Respiratory failure 2.04 2.01 0.98 (0.94–1.03) 0.46
Opioid poisoning 0.07 0.10 1.40 (1.12–1.74) <0.001
Any selected non–opioid-related 11.77 11.75 1.00 (0.98–1.02) 0.85
hospital encounter
Hyperglycemia 0.24 0.24 0.99 (0.87–1.14) 0.93
Urinary tract infection 1.08 1.13 1.04 (0.98–1.11) 0.17
Atrial fibrillation 6.48 6.39 0.98 (0.95–1.01) 0.24
Stroke 4.12 4.08 0.99 (0.96–1.02) 0.52

* A hospital encounter refers to a hospitalization or an emergency department visit. Definitions are provided in the
Methods section in the Supplementary Appendix. CI denotes confidence interval.
† Adjusted odds ratios and P values were estimated with the use of logistic-regression models with occurrence of an opi-
oid-related hospitalization in the 12 months after an emergency department visit as the dependent variable. The key co-
variate was an indicator for being seen by a high-intensity or a low-intensity prescriber. All models were adjusted for
age, sex, race and ethnic group, Medicare–Medicaid dual eligibility, and the presence of 11 chronic conditions.

intensity opioid prescribers may have dispropor- though there were small differences in character-
tionately required appropriate opioid therapy, istics between patients treated by the two pre-
although some of the variation we observed in scriber groups, adjustment for these characteris-
rates of opioid prescribing may also indicate over- tics did not change our results with respect to
use. If differences in appropriate use were a major long-term opioid use. We also replicated our re-
driver of variation in prescribing, we may have sults in a range of sensitivity analyses, and the case
expected increased rates of short-term emergency mix of emergency department visits across 300
department revisits due to inadequately treated diagnosis categories was statistically indistinguish-
pain among the low-intensity prescriber groups. able between groups. In addition, we replicated our
However, we did not find a difference in such rates results with an alternative exposure definition
between prescriber groups; this suggests that in- based on opioid dose; this exposure was mini-
creased opioid prescribing did not prevent revis- mally correlated with the frequency of opioid pre-
its to the emergency department. scribing by physicians.
Our results are unlikely to be explained by se- Our study has several limitations. Most impor-
lection bias for several reasons. First, our analysis tant, this is an observational study and cannot be
focused on variation in opioid prescribing of emer- interpreted as causal, although our findings were
gency physicians within the same hospital; these robust in several sensitivity analyses addressing
physicians are unlikely to select or attract system- selection bias. Second, since we could not ob-
atically different patient populations. Second, even serve whether an opioid prescription was appro-

n engl j med 376;7 nejm.org  February 16, 2017 671

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

priate, our ability to quantify the extent of over- Finally, our statistical tests do not account for the
use of opioids was limited. Third, because we false positive rate associated with multiple second-
focused on Medicare patients with Part D enroll- ary analyses; therefore, P values should be regard-
ment and emergency department visits, our re- ed as exploratory.
sults may not be generalizable to other popula- In conclusion, we found variation by a factor
tions. However, the growing prevalence of opioid of more than three in rates of opioid prescribing
misuse among the elderly makes this an impor- by emergency physicians within the same hospital
tant group to study.16,24 Fourth, the association and increased rates of long-term opioid use among
between high-intensity opioid prescribers and opi- patients treated by high-intensity opioid prescrib-
oid-related hospital encounters within 12 months ers. These results suggest that an increased likeli-
after the index emergency department visit was hood of receiving an opioid for even one encounter
small and not significant in some sensitivity analy- could drive clinically significant future long-term
ses. In addition, for outcomes with a significant opioid use and potentially increased adverse out-
association, the absolute difference in rates of comes among the elderly. Future research may
hospital encounters between groups was small explore whether this variation reflects overpre-
(e.g., an absolute difference of 0.23 percentage scription by some prescribers and whether it is
points for any opioid-related encounter) (Table 3). amenable to intervention.
Therefore, we have weaker evidence to support
Supported by a grant from the Office of the Director, Na-
this association than our results on long-term opi- tional Institutes of Health (NIH) (NIH Director’s Early Indepen-
oid use, and if it were causal, the clinical magni- dence Award, 1DP5OD017897-01, to Dr. Jena).
tude of this association would be small. Fifth, Dr. Barnett reports serving as medical advisor to and holding
stock in Ginger.io; and Dr. Jena, receiving consulting fees from
we were unable to unequivocally attribute an Pfizer, Hill-Rom Services, Bristol-Myers Squibb, Novartis Phar-
opioid prescription to an emergency physician; maceuticals, Vertex Pharmaceuticals, and Precision Health Eco-
however, our analyses were robust with respect nomics. No other potential conflict of interest relevant to this
article was reported.
to a stricter threshold of 3 days to fill a prescrip- Disclosure forms provided by the authors are available with
tion for opioids from an emergency physician.16 the full text of this article at NEJM.org.

References
1. Vital signs:​overdoses of prescription Reid MC. Management of persistent pain targeting-the-opioid-drug-crisis-a-health
opioid pain relievers — United States, in the older patient: a clinical review. -and-human-services-initiative/​).
1999–2008. MMWR Morb Mortal Wkly JAMA 2014;​312:​825-36. 14. Department of Health and Human
Rep 2011;​60:​1487-2. 8. Miller M, Stürmer T, Azrael D, Levin Services Behavioral Health Coordinating
2. Prescription opioid overdose data. At- R, Solomon DH. Opioid analgesics and Committee. Addressing prescription drug
lanta:​Centers for Disease Control and the risk of fractures in older adults with abuse in the United States: current activi-
Prevention, 2016 (http://www​.cdc​.gov/​ arthritis. J Am Geriatr Soc 2011;​59:​430-8. ties and future opportunities. 2014 (https:/​/​
drugoverdose/​data/​overdose​.html). 9. Solomon DH, Rassen JA, Glynn RJ, www​.cdc​.gov/​drugoverdose/​pdf/​hhs
3. National Institute on Drug Abuse. Lee J, Levin R, Schneeweiss S. The com- _prescription_drug_abuse_report_09​
America’s addiction to opioids:​heroin and parative safety of analgesics in older .2013​.pdf).
prescription drug abuse. May 14, 2014 adults with arthritis. Arch Intern Med 15. Kuo Y-F, Raji MA, Chen N-W, Hasan
(https:/​/​w ww​.drugabuse​.gov/​about-nida/​ 2010;​170:​1968-76. H, Goodwin JS. Trends in opioid prescrip-
legislative-activities/​testimony-to 10. Buckeridge D, Huang A, Hanley J, et tions among Part D Medicare recipients
-congress/​2016/​americas-addiction-to al. Risk of injury associated with opioid from 2007 to 2012. Am J Med 2016;​129(2):​
-opioids-heroin-prescription-drug-abuse). use in older adults. J Am Geriatr Soc 2010;​ 221.e21-221.e30.
4. Healthcare Cost and Utilization Proj- 58:​1664-70. 16. Jena AB, Goldman D, Karaca-Mandic
ect. Hospital inpatient utilization related 11. O’Neil CK, Hanlon JT, Marcum ZA. Ad- P. Hospital prescribing of opioids to
to opioid overuse among adults, 1993- verse effects of analgesics commonly used Medicare beneficiaries. JAMA Intern Med
2012. August 2014 (https:/​/​w ww​.hcup-us​ by older adults with osteoarthritis: focus on 2016;​176:​990-7.
.ahrq​.gov/​reports/​statbriefs/​sb177 non-opioid and opioid analgesics. Am J 17. Poon SJ, Greenwood-Ericksen MB.
-Hospitalizations-for-Opioid-Overuse​.jsp). Geriatr Pharmacother 2012;​10:​331-42. The opioid prescription epidemic and the
5. Vestal C. Older addicts squeezed by 12. Butler MM, Ancona RM, Beauchamp role of emergency medicine. Ann Emerg
opioid epidemic. Stateline. July 26, 2016 GA, et al. Emergency department pre- Med 2014;​64:​490-5.
(http://pew​.org/​2asv3TO). scription opioids as an initial exposure 18. Tamayo-Sarver JH, Dawson NV, Cy-
6. Dufour R, Joshi AV, Pasquale MK, et preceding addiction. Ann Emerg Med dulka RK, Wigton RS, Baker DW. Vari-
al. The prevalence of diagnosed opioid 2016;​68:​202-8. ability in emergency physician decision
abuse in commercial and Medicare man- 13. Frank R. Targeting the opioid drug making about prescribing opioid analge-
aged care populations. Pain Pract 2014;​ crisis:​a Health and Human Services ini- sics. Ann Emerg Med 2004;​43:​483-93.
14(3):​E106-E115. tiative. Health Affairs Blog. April 3, 2015 19. Pletcher MJ, Kertesz SG, Kohn MA,
7. Makris UE, Abrams RC, Gurland B, (http://healthaffairs​.org/​blog/​2015/​04/​03/​ Gonzales R. Trends in opioid prescribing

672 n engl j med 376;7 nejm.org  February 16, 2017

The New England Journal of Medicine

Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.
 Opioid Prescribing and Risk of Long-Term Use

by race/ethnicity for patients seeking care Emergency department visits among re- 27. National Institute of Standards and
in US emergency departments. JAMA cipients of chronic opioid therapy. Arch Technology. Kolmogorov-Smirnov Good-
2008;​299:​70-8. Intern Med 2010;​170:​1425-32. ness-of-Fit Test. 2013 (http://www​.itl​.nist​
20. Hoppe JA, Nelson LS, Perrone J, Wein- 24. Jena AB, Goldman D, Weaver L, Kara- .gov/​div898/​handbook/​eda/​section3/​
er SG. Opioid prescribing in a cross sec- ca-Mandic P. Opioid prescribing by mul- eda35g​.htm).
tion of US emergency departments. Ann tiple providers in Medicare: retrospective 28. White H. A heteroskedasticity-consis-
Emerg Med 2015;​66(3):​253-259.e1. observational study of insurance claims. tent covariance matrix and a direct test for
21. Dowell D, Haegerich TM, Chou R. BMJ 2014;​348:​g1393. heteroskedasticity. Econometrica 1980;​48:​
CDC guideline for prescribing opioids for 25. Healthcare Cost and Utilization Proj- 817-38.
chronic pain — United States, 2016. ect. Clinical Classifications Software (CCS) 29. Angrist JD, Pischke J-S. Mostly harm-
JAMA 2016;​315:​1624-45. for ICD-9-CM (https:/​/​w ww​.hcup-us​.ahrq​ less econometrics: an empiricist’s com-
22. Dunn KM, Saunders KW, Rutter CM, .gov/​toolssoftware/​ccs/​ccs​.jsp). panion. Princeton, NJ:​Princeton Univer-
et al. Opioid prescriptions for chronic 26. Meara E, Horwitz JR, Powell W, et al. sity Press, 2008.
pain and overdose: a cohort study. Ann State legal restrictions and prescription- Copyright © 2017 Massachusetts Medical Society.
Intern Med 2010;​152:​85-92. opioid use among disabled adults. N Engl
23. Braden JB, Russo J, Fan M-Y, et al. J Med 2016;​375:​44-53.

Tiger, Kanha National Park, India Anil Sain, M.D.

n engl j med 376;7 nejm.org  February 16, 2017 673

The New England Journal of Medicine
Downloaded from nejm.org on May 17, 2017. For personal use only. No other uses without permission.
Copyright © 2017 Massachusetts Medical Society. All rights reserved.