University of Southern California Dissertations and Theses

Essays in opioid use and abuse

(USC Thesis Other)

Essays in opioid use and abuse

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ESSAYS IN OPIOID USE AND ABUSE

by
Sarah Anne Axeen

A dissertation submitted to the faculty of The Graduate School of the
University of Southern California
in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

Public Policy and Management
University of Southern California
May 2016

TABLE OF CONTENTS

Acknowledgements 1
Chapter 1: Introduction 3
Chapter 2: Describing the Use of Prescription Opioids by Medicare Beneficiaries 6
I. Introduction 6
II. Background 7
III. Data 11
IV. Empirical Strategy 15
V. Results 18
VI. Discussion 23
VII. Figures and Tables 26
Chapter 3: Assessing the Effectiveness of State Policies for Altering Opioid Use and
Misuse Among Medicare Beneficiaries
36
I. Introduction 36
II. Background 37
III. Data 45
IV. Empirical Strategy 49
V. Results 52
VI. Discussion 60
VII. Figures and Tables 63
Chapter 4: Emergency Department Contribution to the Prescription Opioid Epidemic 73
I. Introduction 73
II. Data and Sample 74
III. Empirical Strategy 78
IV. Results 80
V. Discussion 84
VI. Figures and Tables 87
References 93
Appendix to Chapter 2 102
Appendix to Chapter 3 108
Appendix to Chapter 4 116

1

ACKNOWLEDGEMENTS

I want to thank my dissertation chair and academic advisor, Dana Goldman, without
whom this dissertation and my progress through my doctoral program would not have been
possible. His support through access to rich data resources, research and teaching opportunities,
and timely, strategic advice made this journey rewarding and smooth. In particular, I would like
to thank him not only for his mentorship and guidance over the past five years, but also for his
continuing support and belief in my academic career.
I would also like to thank the rest of my dissertation committee—Geoff Joyce and Darius
Lakdawalla—for their thoughtful and constructive guidance on my dissertation project as well as
their mentorship throughout my time at USC. In particular, Darius’s sage advice on delivering an
effective presentation and Geoff’s encouragement to dig deeper into the Medicare claims data
improved both the content and delivery of my research and dissertation.
I am deeply indebted to my co-authors in the Department of Emergency Medicine,
Michael Menchine and Seth Seabury. Their prodding to pursue my interest in opioid prescribing
behavior led directly to our chapter of my dissertation and put me on the path to a better
understanding of the opioid epidemic in the United States.
Throughout my time at USC I have had the opportunity to work with and learn from a
number of faculty members at the Schaeffer Center. I would like to recognize John Romley, Julie
Zissimopolous, Neeraj Sood, Jason Doctor, and Erin Trish for their contributions. John ably
guided me through my first research project as a doctoral student; Julie ensured that my needs as
a student and research assistant at the Schaeffer Center were always met; Neeraj served as an
important member of my qualifying exam committee; Jason was a knowledgeable sounding
2

board for my questions about opioid use and addiction; and, Erin took the extra time to walk me
through the job search and interview process.
In addition to the faculty at the Schaeffer Center, my progress through graduate school
would not have been nearly as smooth without Patty St. Clair who taught me the ins and outs of
Medicare claims data, Jillian Wallis who coached me through the data acquisition and IRB
process in record time, and Cristina Wilson who managed to simplify the arcane grant
application process to a science. Also, I would like to recognize Sara Geiger, Briana White, and
Samantha Malisos who went above and beyond to ensure that meetings were made, deadlines
were met, and all my frantic emails were answered.
I would like to thank my friends and colleagues at Price for talking through endless
iterations of my charts and tables, learning more than they ever cared to know about opioids, and
supporting me through the doctoral process. I would also like to thank my parents for their
constant support and for believing in and valuing my doctoral education. Finally, I would like to
thank my husband Ben. His patience, encouragement, and assurances that everything would
work out made this process more rewarding and infinitely easier.
This dissertation project was made possible through funding from a USC Dissertation
Completion fellowship and an Agency for Healthcare Research and Quality Dissertation
Fellowship (1R36HS024251-01).
3

CHAPTER 1
INTRODUCTION

Abuse of legal prescription drugs, particularly opioids, is a growing problem in the
United States. This problem is driven by increases in the rate of prescribing of opioids as well as
increasing abuse of those drugs. The abuse leads to addiction, negative health outcomes,
increased disability costs, and death by overdose or poisoning. Despite growing attention to the
problem of prescription opioid use, misuse, and abuse in the United States, there are many
unanswered questions. Further, many policies aimed at curbing opioid use have been enacted
ahead of reliable evidence on the likely efficacy of such interventions.
The goal of my dissertation is to fill in major gaps in our understanding of the scope of
the opioid epidemic and the efficacy of policy responses to the crisis. Each paper adds critical
knowledge that deepens our understanding of the scope and trajectory of opioid utilization in the
United States, the effect of existing policies on both appropriate and inappropriate opioid
utilization, and the role of physicians in prescribing these drugs. Each paper relies on the analysis
of large datasets—both survey-based and claims-based—and statistical methods to add detail to
our understanding of this epidemic.
The first paper focuses on quantifying opioid utilization in Medicare. The majority of
research on opioid utilization and abuse has focused on the under-65 population. The few studies
that tackle the question of opioid use in Medicare provide single-year snapshots or report on a
limited number of outcomes. My results fill that gap by chronicling multiple measures of opioid
utilization in a 20 percent sample of Medicare beneficiaries from 2006 to 2012. I estimate
multivariate regression models to show patterns of use, misuse, and prescribing that control for
demographics, patient health, socio-economic status, and eligibility status. My results indicate
4

that, in contrast to the general population, opioid utilization in Medicare—while high—is fairly
stable. Despite stable use, opioid abuse in the Medicare population is on the rise; and, medical
diagnoses of abuse fail to capture substantial numbers of patients with utilization patterns
indicative of opioid misuse. The study also provides useful insights for future research and
policy—for example that claims-based estimates of misuse are important indicators and that
policies related to opioid use and abuse should explicitly consider the Medicare population as
they are large consumers of opioids.
The second paper leverages the sample of opioid users in Medicare to explore whether
three categories of state laws aimed at lowering improper utilization of opioids have any impact.
Previous analyses of the impact of state policies on opioid use have looked at relatively rare
outcomes like overdose, relied largely on survey data, and only explored the role of one type of
policy on these outcomes. I improve on that literature by considering the impact of three separate
types of laws, controlling for the existence of other opioid-related regulations, employing a large
and representative sample, exploring nine distinct outcomes, and estimating these effects in
administrative claims rather than survey data. In addition, I employ a difference-in-differences
framework to make my findings more robust than a simple comparison of states with and
without these laws. I find that controlling for individual characteristics, state and time fixed
effects, and pre-treatment trends by state, two of these three laws reduce measures of opioid
utilization, opioid abuse or misuse, or physician prescribing of opioids. Moreover, the impacts
are felt broadly across categories of patients. Thus my findings represent important evidence to
be considered in structuring more effective, targeted prescription drug regulations.
The third paper, co-authored with Seth Seabury and Michael Menchine, explores
differential prescribing of opioids by physician specialty. It was motivated by the development of
5

legislation to reduce opioid prescriptions from the emergency department (ED) by the American
College of Emergency Physicians despite the fact that the ED contribution to the opioid epidemic
is incompletely characterized. To fill this gap in the literature, we use multivariate regression
analysis to estimate the relative contribution of the ED to opioid prescribing from 1996 to 2012
using data from the Medical Expenditure Panel Survey (MEPS). A unique contribution of this
study is to look not only at increased counts of opioid prescriptions, but also at a decomposition
of those prescriptions into: new prescriptions, stronger prescriptions (as measured by conversion
to milligrams of morphine equivalence), or more frequent refills of prescriptions. We find that
the vast majority of the increase in prescribing observed in the MEPS data is due to prescriptions
originating in office-based rather than ED settings and to opioids that are refills of previous
prescriptions rather than incident prescriptions. Our findings indicate that the move to restrict
opioid prescriptions in the ED setting is unlikely to have a large impact on opioid prescribing.

6

CHAPTER 2
DESCRIBING THE USE OF PRESCRIPTION OPIOIDS
BY MEDICARE BENEFICIARIES

I. Introduction
The use and abuse of prescription opioids is a growing concern in the United States.
Despite the mounting evidence that the pace of both opioid use and abuse is climbing, relatively
little research has focused on the scope or trajectory of opioid use and abuse in the Medicare
population. Additionally, little is known about whether Medicare beneficiaries act in a similar
manner to other patients regarding opioid use, abuse, and misuse. This paper fills that gap by
explicitly examining the frequency with which Medicare beneficiaries use opioids, the length
and potency of those prescriptions, the prevalence and correlates of outcomes like opioid abuse
and overdose, as well as the patterns of opioid prescribing by physician specialty. The paper
proceeds as follows: in section 2, there is a review of the existing literature on opioid use and
abuse, the existing evidence on opioid use in Medicare, the role of physicians in prescribing
opioids, and existing knowledge on correlates of opioid use and abuse; in section 3 I outline my
data, sample, and analytical methods; in section 4 I share my empirical findings; and in section 5
I discuss the implications of these findings for future research on opioid use and abuse in
Medicare.

7

II. Background
II.I Use, Abuse, and Misuse of Prescription Opioids
Abuse of legal prescription drugs, particularly opioids, is a growing problem in the
United States. This problem is driven by increases in the rate of prescribing of opioids as well as
increasing abuse of those drugs. The abuse leads to addiction, negative health outcomes,
increased disability costs, and death by overdose or poisoning. Over the past decade opioid
prescriptions increased more than three-fold.(Prevention, 2013) These prescriptions are not only
for acute, short-term events; about 3% of adults with non-cancer pain are on long-term opioid
therapy regimens.(Dunn et al., 2010) Perhaps unsurprisingly, as opioid prescription rates have
increased, so has abuse of these drugs.(Cicero, Surratt, Inciardi, & Munoz, 2007) Drug treatment
programs have seen a nearly five-fold increase in admission for the abuse of legal, prescription
drugs and the incidence of overdoses, reported abuse, and enrollment in treatment is larger than
for illicit drugs.(Jena & Goldman, 2011) In 2007, opioid overdoses became the second leading
cause of unintentional death.(Gugelmann & Perrone, 2011) Further, opioid abuse that was not
fatal led to 130 ED visits per 100,000 among women.(Control & Prevention, 2013)
In addition to the medical costs, there are significant social and economic costs associated
with opioid abuse. The direct health care costs of patients abusing opioids are more than eight
times higher than those of non-abusers.(Strassels, 2009) Total costs for prescription opioid
abuse were nearly $73 billion in 2007.(Nora D. Volkow, Frieden, Hyde, & Cha, 2014) A
welfare analysis found that total societal costs of this abuse could total nearly $56
billion.(Birnbaum et al., 2011) The White House recently branded this increase in prescription
drug abuse a public health and safety crisis.(Policy & President, 2011) In response to this
epidemic, the Department of Health and Human Services has allocated significant resources for

8

direct intervention in this crisis as well as for studies to determine the best means to reduce this
epidemic in the future.(Nora D. Volkow et al., 2014)
Further, there is increasing evidence that opioids may not be effective in combatting
pain—particularly chronic pain. Patients with chronic, non-cancer pain report dissatisfaction
with opioid treatment; and, opioid treatment appears to have little effect on disability rates for
these patients.(Sehgal, Colson, & Smith, 2013) In meta-analyses of the efficacy of opioids for
back pain—a common form of chronic pain—researchers have found some small-term
improvements after the use of opioids, but little if any evidence of long-term
improvement.(Deshpande, Furlan, Mailis ‐Gagnon, Atlas, & Turk, 2007; Martell et al., 2007)
Additionally, physicians are raising concerns about the high risk of opioid abuse and
addiction.(Grady, Berkowitz, & Katz, 2011)
II.II Opioid Use in Medicare
Despite the abundance of evidence of the growth of opioid use in the general population,
there is no systematic study of the trajectory or characteristics of opioid use in the Medicare
population. A recent paper finds that increasing enrollment in Medicare Part D is associated with
an increase in total opioid prescriptions, but the authors cannot attribute those prescriptions to the
Medicare population alone. (Pacula, Powell, & Taylor, 2015) In one of the few studies of opioid
use in Medicare, the authors find that about 1.2 million beneficiaries fill opioid prescriptions in
2010; and, that many of those opioid users utilize multiple physicians to obtain opioids. (Jena,
Goldman, Weaver, & Karaca-Mandic, 2014) Unfortunately, because it is a single-year snapshot
of opioid use, the study tells us little about whether opioid use is increasing, decreasing, or stable
in the Medicare population. An examination of the use of opioids by disabled beneficiaries finds
that use is common and stable and that chronic use of opioids is rising. (Morden et al., 2014) An

9

analysis of opioid prescribing in Medicare found that unlike studies finding a small share of
inappropriate prescribers, that opioid prescribing is common and widespread in the Medicare
program. (Chen, Humphreys, Shah, & Lembke, 2016)
There is some evidence that Medicare beneficiaries are abusing or misusing opioids. The
Government Accountability Office found multiple examples of fraudulent utilization and misuse
in their audit of Medicare Part D claims data.(Office, 2011) A subsequent study by the Center for
Medicare and Medicaid Services (CMS) found that more than 20,000 Medicare beneficiaries
could be classified as outliers for their volume and frequency of opioid utilization.(Tudor, 2012)
There is also evidence that opioid use can lead to increased health risks for older adults
(Saunders et al., 2010; Solomon et al., 2010).
Despite the evidence that Medicare beneficiaries may abuse opioids, there is relatively
little research on the determinants and scope of abuse within this population as compared to the
general population. In particular, there is only one systematic study evaluating whether the same
characteristics that predict opioid abuse in the under-65 population do so in an older population.
But, that study focused solely on multiple prescribers as a measure of misuse rather than a
broader range of outcomes.(Jena et al., 2014) This study fills important gaps in our
understanding of trends and characteristics of opioid use and abuse in the Medicare population.
II.IIICharacteristics of Opioid Use and Abuse
While there are almost no studies on the characteristics of opioid use and abuse in
Medicare, there is a sizeable literature on correlates of opioid use and abuse in the general
population. Across the existing literature, studies find that larger and/or daily doses of opioids
are more likely to result in overdose.(Bohnert et al., 2011; Braden et al., 2010; Dunn et al., 2010;
Gomes, Mamdani, Dhalla, Paterson, & Juurlink, 2011) Men are more likely to overdose than are

10

women.(Dunn et al., 2010; Ives et al., 2006; Prevention, 2013) This finding is not universally
supported, with some studies finding that women are more likely to overdose.(Cerda et al., 2013)
People who begin using prescription drugs at a younger age are more likely to abuse those
drugs.(McCabe, Teter, Boyd, Knight, & Wechsler, 2005) Age generally has an inverse
relationship with the likelihood of abuse of prescription opioids (Edlund et al., 2010); however,
some studies have found that compared to other causes of overdose, opioid-associated overdoses
are more likely among older individuals.(Green, Grau, Carver, Kinzly, & Heimer, 2011) A study
of adolescents found that being a member of a low-income family is associated with a significant
increase in risk of opioid misuse.(Sung, Richter, Vaughan, Johnson, & Thom, 2005)
Many of the studies that seek to predict instances of opioid abuse focus on the role of
certain diagnoses as either predictive of or at least associated with opioid abuse. Among the most
frequently cited findings is that people with other mental disorders or a history of other substance
abuse are more likely to abuse opioids.(Braden et al., 2010; Dunn et al., 2010; Edlund et al.,
2010; Edlund, Steffick, Hudson, Harris, & Sullivan, 2007; Ives et al., 2006; Manchikanti et al.,
2007) A close second to a prior substance abuse diagnosis is the diagnosis of a mental health
disorder like a personality disorder. (Barth L. Wilsey et al., 2008) Recent studies have also found
that certain medical diagnoses, such as Hepatitis C and chronic pain disorders are highly
correlated with opioid abuse.(Robert Dufour, 2014) For example, a study of US veterans found
that pain diagnoses (back pain, arthritis, etc…) are correlated with opioid dependence or
abuse.(Edlund et al., 2007) However, there is mixed evidence of whether pain is associated with
increasing likelihood of abuse or overdose with some studies finding no significant association
and others finding that severity of pain increases the likelihood of adverse events.(Ives et al.,
2006; Pletcher, Kertesz, Kohn, & Gonzales, 2008)

11

II.IV Role of Physicians
Most research on physician prescribing of opioids focuses on the patient factors that drive
the decision to prescribe opioids. Among the findings of this literature are that a patient’s
manifestations of pain influence a decision to prescribe opioids(Turk & Okifuji, 1997), that
physicians are somewhat less likely to prescribe opioids to minority patients(J. H. Tamayo-
Sarver et al., 2003; Joshua H Tamayo-Sarver, Hinze, Cydulka, & Baker, 2003), and that
physicians confronted with the same clinical scenarios do not respond with similar prescribing
decisions.(Joshua H. Tamayo-Sarver, Dawson, Cydulka, Wigton, & Baker, 2004) Further, we
have good evidence that a large share of opioids that are abused or misused come from physician
prescriptions, rather than a black market. According to survey evidence, 20 percent of
individuals who abuse opioids receive opioids from physicians, rather than from family, friends,
or the black market. These nonmedical users are more likely to obtain opioids from physicians as
the length of their nonmedical use increases. (Christopher M Jones, Paulozzi, & Mack, 2014)
There is some evidence on the influence of physician specialty on opioid prescribing. In general,
office based physicians are the largest source of opioid prescriptions. Similarly, general
practitioners have been found to be much more likely to prescribe long-term opioids than other
physicians. There is also sizeable variation in the frequency of opioid prescribing by geographic
region.(Turk, Brody, & Okifuji, 1994) However, most of this research is based on physician
questionnaires rather than observation of physician practices.

III. Data
The purpose of this paper is to examine the extent and characteristics of opioid use and
abuse in a Medicare population as well as to explore physician prescribing of opioids to

12

Medicare beneficiaries. To generate these findings, two separate samples are created: a
beneficiary sample and a physician sample. The relevant inclusion and exclusion criteria are
explained below.
III.I Beneficiary Sample
To report and analyze the extent of opioid use, abuse, and prescribing, this paper starts
with the administrative claims for a 20 percent random sample of all Medicare beneficiaries from
2006 to 2012. That base sample is then restricted to those beneficiaries who have traditional, fee-
for-service Medicare for all months in a given year. These claims include Medicare Part A
(hospital care), Medicare Part B (outpatient and physician care), and Medicare Part D
(prescription drug) claims. They contain information on the timing and cost of all Medicare
claims, demographic information about the beneficiaries, and estimates of their health status. In
particular, we use information on the beneficiary’s sex, age, race or ethnicity, whether they are
dually eligible for both Medicare and Medicaid benefits, whether they receive a low income cost
sharing subsidy to help cover the cost of their Medicare Part D drug plan, whether they are
eligible for Medicare because of a qualifying disability, and whether they are at the end of their
life (if they died in that year). The paper also uses an estimate of the individual’s health status
based on the risk-adjustment calculations used by Medicare when reimbursing Part D plans—
called the RxHCC value—which is a function of the individual’s age, sex, category of eligibility,
and diagnostic history (Robst, Levy, & Ingber, 2007).
From this baseline of traditional Medicare beneficiaries, I then identify those
beneficiaries with a history of opioid use. Opioid use is defined as the existence of more than one
filled opioid prescription in a patient’s Medicare Part D claims in a given year. Opioids are
identified from beneficiaries’ Part D claims using national drug codes identified in the IMS

13

Health, First DataBank, or RxNorm drug databases. The analysis includes opioid agonists,
opioid partial agonists, and opioid combinations, but excludes antitussives as well as tramadol
products. The final beneficiary sample is representative of traditional Medicare beneficiaries
with multiple opioid prescriptions in a given year.
III.II Physician Sample
In addition to understanding the use of opioids by Medicare beneficiaries, this paper also
explores the prescribing patterns of opioids by physicians. This sample used for this analysis
starts with the Medicare Part D claims of the beneficiary sample of opioid users described above.
These claims are merged to an administrative file containing information about the prescribing
physician including medical specialty, certification, location, and more using a unique physician
identifier. The data is then collapsed to create a database of total physician prescribing of opioids
by year. Included in the database are variables for the average characteristics of a physician’s
patient population including the share who are white, average age, the share who are male,
average health risk score, and others that come from the patient sample described above. The
sample is roughly representative of physicians who prescribe opioids to Medicare beneficiaries.
III.III Outcome Measures
To describe opioid use and abuse, this paper relies on four separate outcome measures:
(1) the rate of receipt of opioids, (2) average days supply of opioid prescriptions, (3) the potency
of opioid prescriptions measured in milligrams of morphine equivalence, and (4) the rate of
receipt of high-dose opioids (>100 MME per day) conditional on opioid use. Two of the key
outcome measures come directly from the Medicare claims data—receipt of more than one
opioid prescription and the average “days supply” of opioids per beneficiary-year. However,
because opioid prescriptions vary not only in their length but also in their potency, this paper

14

also relies on estimates of those opioids morphine equivalence as a way to compare across
prescriptions. Following the methodology of the CONSORT research group as well as the Center
for Medicare and Medicaid Services’ guidance to health plans, morphine equivalence was
calculated according to a prescription’s active ingredient, dosage strength, and quantity of pills
distributed. For example, the active ingredient of oxycodone has a morphine equivalence of 1.5,
so a prescription for 30 pills of 20 milligrams of oxycodone would have a milligram morphine
equivalence (MME) of 1.5 x 20MG x 30 pills, or 900 MME. The active ingredient conversion
factors are available in the appendix.
Using these conversions, this paper also assesses the average milligrams of morphine
equivalence per beneficiary per year as well as the probability of receiving a “high-dose”
opioid—one with an MME per day that exceeds 100. Similarly, in the physician sample the
outcomes of interest are (1) average MME of a physician’s prescriptions, (2) average days
supply of a physicians’ prescriptions, (3) the average number of fills, and others.
III.IV Additional Explanatory Variables
In addition to the demographic and health variables identified above that might explain or
predict opioid use, this paper also examines these opioid users’ Part A and B claims for certain
diagnoses that have been linked to opioid use or misuse. Relying on ICD-9 codes that are
included on these patient claims, I created variables that flagged whether an individual had ever
been diagnosed with one of the following ailments in a given year: myalgia, cancer, opioid
abuse, abuse of other substances, alcohol abuse, opioid overdose, back pain, headache, fracture,
sprain or strain, major mental health diagnosis, hepatitis c, chronic pain, or arthritis. Complete
lists of ICD-9 codes are available in the appendix.

15

Finally, for the estimation of opioid abuse, this paper explores the viability of using non-
medical indicators of opioid abuse to identify individuals who are likely abusers or misusers of
opioids. These measures are composites of indicators for whether a beneficiary obtains
prescriptions from multiple prescribers, whether beneficiaries fill those prescriptions at multiple
pharmacies, and whether those opioid prescriptions are overlapping (e.g. a prescription from a
different doctor that is filled before a prior prescription has elapsed). Following existing
literature on doctor-shopping these misuse indicators will combine increasingly restrictive
definitions of multiple prescribing, pharmacies and overlapping fills to determine the best non-
medical indicators of abuse.

IV. Empirical Strategy
IV.I Opioid Use
This paper uses multivariate regression analysis to estimate the outcomes of interest
noted above, controlling for characteristics like age, race/ethnicity, gender, eligibility status, low-
income status, diagnoses of interest, and health status. Health status is measured using
Medicare’s formula for risk-adjustment which considers patient demographics alongside medical
and diagnosis history.
(1.1) y
it
= δ
t
+ βX
it
+ ε
it

In equation (1.1), above, t represents the year, δ
t
is a year fixed effect that accounts for
the general time trend, X is a set of individual characteristics of interest noted above, and y is one
of the opioid-related outcomes of interest. Regressions with a binary outcome are estimated
using multivariate logit models while those with continuous outcomes are estimated by OLS;
each outcome is estimated separately. To deal with the skewness in the distribution of “days

16

supply” and MME, those outcomes are estimated using log-linear OLS models. This paper
implements the smearing methodology from Duan (1983) to account for the differences in the
log and normal distribution functions when transforming the results to estimate mean “days
supply” and MME.(Duan, 1983)
The analysis of physician prescribing will follow a similar methodology as in the
beneficiary analysis above. The slight differences stem from the use of physician-level, rather
than beneficiary level data. In equation (1.2) p indexes the physician rather than the beneficiary
and X is a set of physician-level characteristics, including the physician’s specialty as well as the
characteristics of the patients he or she serves (e.g. average health status, share that are non-
white, average age, etc…). Further, all outcomes explained above are averaged across all
prescriptions a physician writes to compare usefully between physicians with different volumes
of patients reporting pain.
(1.2) y
pt
= δ
t
+ βX
pt
+ ε
pt
The means of these outcomes will be estimated using linear multivariate regression
analysis. Again, to deal with the skewness in the distribution of average MME, those outcomes
are estimated using log-linear OLS models and re-transformed using the smearing methodology
from Duan (1983).
IV.II Opioid Abuse
Following the same methodology as in equation (1.1), I estimate the unadjusted and
adjusted rates of diagnosis of (1) opioid overdoses and (2) opioid substance abuse as noted in
these beneficiaries’ medical claims. The paper pays particular attention to diagnoses and patient
characteristics that may be associated with opioid use and abuse in the Medicare population to
determine whether they are similar to the findings for the general population.

17

In addition to estimating the scope and determinants of medically diagnosed opioid abuse
or overdose, which are rare (for example, less than 1 percent of inpatient claims in this data are
for opioid overdoses) I also estimate the likelihood of other indicators of opioid misuse. One
such outcome is whether the patient exhibits doctor-shopping behavior.
Existing estimates of doctor-shopping generally identify patients with multiple
prescribers, multiple pharmacies, and overlapping prescription fills. However, these estimates
have not yet been validated against patient outcomes (see for examples: (Buurma et al., 2008;
Jena et al., 2014; McDonald & Carlson, 2013; Peirce, Smith, Abate, & Halverson, 2012; Pradel,
Delga, Rouby, Micallef, & Lapeyre-Mestre, 2010; Schloff, Rector, Seifeldin, & Haddox, 2004;
B. L. Wilsey et al., 2011)). Thus, I further estimate regressions of the form in (1.3) to “validate”
these administrative estimates of opioid abuse against the medically-diagnosed estimates. These
regressions are identical to those in equation (1.1) except that they include an additional variable,
status as a doctor-shopper, defined in various ways. The outcome of interest will be whether the
patient has an opioid overdose or a diagnosis of opioid abuse
(1.3) y
it
= δ
t
+ βX
it
+ λDrShop
it
+ ε
it

It is important to note that doctor shopping in equation (1.3) is endogenous as abuse may
lead to doctor shopping or vice versa, making any claims of causality impossible to establish.
Therefore, I emphasize that the role of equation (1.3) is to determine which estimates of doctor
shopping are most correlated with medical diagnoses of opioid abuse, controlling for as many
other factors as practical.

18

V. Results
From 2006 to 2012, more than one-third (36.8%) of Medicare beneficiaries filled a
prescription for an opioid each year while about a quarter (24.3%) of the full sample of
traditional Medicare beneficiaries filled multiple opioid prescriptions. Over the same time period
more than 5 percent of all prescriptions filled by Medicare for this sample were for opioids. As
shown in Table 1, opioid users are slightly younger, in worse health, more likely to have a low
socio-economic status (dual eligible for Medicare and Medicaid or receiving a low-income
subsidy to afford Medicare Part D coverage), more likely to have a history of disability, and
slightly more likely to be female than are the full sample of traditional Medicare beneficiaries.
V.I Trends in Utilization of Opioids
Contrary to national estimates of opioid use, there has been relatively little change in the
share of Medicare beneficiaries using opioids—across many measures of use. In fact, there has
been a slight slowdown in opioid use in 2011 and 2012. However, the dosage strength of those
opioids and the length of opioid prescriptions grew from 2006 to 2012. Among users—those
beneficiaries who fill multiple opioid prescriptions in a year--beneficiaries fill an average of 9
prescriptions per year for an average of 193 days of opioids, and about 20 percent of
beneficiaries receive a high dose (defined as more than 100 milligrams of morphine-equivalent
opioids per day). The share receiving high dose prescriptions declined markedly after 2010. In
addition, about 14 percent of opioid users receive 365 or more days supply in a given calendar
year.
In general, younger beneficiaries, those with worse health, those eligible due to disability,
and low-income beneficiaries are significantly more likely to use opioids and have stronger,

19

lengthier prescriptions. Full regression results used to generate these estimates are available in
the Appendix.
Despite the stable nature of opioid use among Medicare beneficiaries, the distribution of
opioid users by diagnosis has shifted from 2006 to 2012. For example, the share of opioid users
receiving a diagnosis of cancer (18%), bone fracture (14%), or a sprain or strain (18%) remained
stable over the period, but those with a diagnosis of back pain (26% in 2006) grew 4 percentage
points and those with a diagnosis of chronic pain (1.7% in 2006) grew more than 20 percentage
points. Further, there was a 5 percentage point jump in major mental health diagnoses and a 7
percentage point jump in diagnoses of non-opioid drug abuse among opioid users.
In general, diagnoses that are often cited as relatively more appropriate uses of opioids:
cancer pain and acute pain diagnoses like fractures, sprains, or strains are associated with lower
levels of MME and shorter prescriptions. Diagnoses of opioid abuse, chronic pain, and back pain
are associated with significantly higher levels of MMEs and lengthier prescriptions. Opioid
abuse, drug abuse, and opioid overdoses are also associated with higher levels of MME, longer
prescriptions, and a higher probability of receiving a high daily dose of opioids—confirming the
findings of earlier studies. Notably, a diagnosis of alcohol abuse is associated with significantly
lower levels of MME and shorter prescriptions, in contrast to diagnoses of other drug abuse.
Only alcohol abuse is associated with a significantly lower probability of having high, daily
doses of opioids (>100 MME per day); all other diagnoses are associated with an elevated
likelihood of receiving such prescriptions.
In addition to trends in the distribution of opioid users by diagnosis, I also exploit the
panel nature of my data to determine the persistence of opioid use in the Medicare population.
Between 10 and 15 percent of Medicare beneficiaries fill multiple opioids in consecutive years

20

from 2006 to 2012. Among opioid users, nearly 60 percent fill multiple opioid prescriptions in
consecutive years and the average user fills multiple prescriptions for 3.7 years. Among users
whose use spans consecutive years, these beneficiaries fill 11.5 prescriptions for more than 254
days on average. More than 20 percent of these consecutive year users receive high daily opioid
doses. The number of fills and days supply for these consecutive year users is notably higher
than for all opioid users. Further, as shown in Figure 2, as years of opioid use increases from one
to seven, the average number of fills quintuples from 3 to 15, the share receiving a high dose
opioid increases from 14 to 25 percent and the average day supply increases from 40 to 333—
nearly a full year’s supply. It appears that instead of using opioids for short periods of acute pain,
many beneficiaries use opioids for chronic pain with escalating lengths and strengths of these
drugs.
V.II Trends in Physician Prescribing of Opioid in Medicare
In Medicare, the largest share of prescriptions for opioids is written by physicians with
training in general medicine, family medicine, or internal medicine. On average, physicians who
prescribe opioids do so an average of 14 times per year; but, the top 1 percent of prescribers
provides an average of more than 150 prescriptions for opioids per year. Notably, as shown in
Figure 4, from 2007 to 2012 the total number of prescriptions written by nurses grew by 150%;
and, the share of total prescriptions written by nurses doubled from three to six percent.
Prescribing by all specialties grew by only 14 percent over the same time period.
As shown in Table 4, those physicians who explicitly deal with pain or addiction
medicine prescribe opioids more frequently than other physicians. Additionally, physicians with
a specialty of anesthesia, pain medicine, and addiction medicine prescribe significantly higher
average MMEs than general medicine physicians. They also prescribe stronger and longer

21

prescriptions than their counterparts. Emergency medicine physicians, on the other hand,
prescribe comparatively shorter prescriptions and are about half as likely as a general medicine
physician to write a high dose prescription. These patterns indicate that it is those physicians
best trained to deal with pain and addiction who account for the largest share of opioid
prescribing in Medicare. Full regression results are available in the appendix.
V.III Trends in Opioid Abuse in Medicare
About 2.2 percent of opioid users from 2006 to 2011 were diagnosed as opioid abusers in
their Medicare claims. Controlling for the effect of relevant covariates, the likelihood of opioid
abuse is higher for younger, male, Hispanic, sicker, and disabled beneficiaries. These findings
are qualitatively similar to findings for the general population. It is interesting to note that unlike
in the findings on opioid use, dual eligibility for Medicare and Medicaid does not increase the
probability of abusing opioids. Perhaps this result points to a Medicaid-specific policy like “lock-
in” that impacts the likelihood of abusing opioids in the dually-eligible population. Cancer
diagnoses and muscle sprain or strain diagnoses are associated with significantly lower
likelihood of opioid abuse while all other diagnoses were associated with an increased likelihood
of opioid abuse. The strongest associations with opioid abuse, all else equal, were with diagnoses
of other drug abuse, opioid overdose, and chronic pain.
From 2006 to 2012 an average of 0.4 percent of Medicare beneficiaries experienced an
opioid-related overdose. All else equal, female, white, disabled, and sicker beneficiaries had
higher odds of opioid related overdose than their counterparts. Again, unlike for opioid use, dual-
eligible beneficiaries had no higher rates of overdose than their counterparts. Similar to the
results for opioid abuse, cancer and sprain or strain diagnoses had no significant association with

22

overdoses while all other diagnoses were positively associated with opioid-related overdose.
Full regression results are available in the appendix.
While medically-diagnosed opioid abuse and opioid-related overdoses are fairly rare,
there is increasing evidence that other behavioral patterns may indicate opioid misuse (if not
abuse). One such pattern is “doctor-shopping,” whereby patients visit multiple providers in a
short period of time to obtain duplicative prescriptions. For example, as shown in Figure 3, a
large share of Medicare beneficiaries visit multiple providers and pharmacies to fill opioids.
Unfortunately, there is no administrative standard that defines this behavior, so it must be
inferred from administrative claims. Following the literature, indicators of doctor shopping
consist of a combination of evidence that a patient has visited multiple physicians, has
overlapping prescriptions from multiple physicians, and has visited multiple pharmacies to fill
those prescriptions.(Buurma et al., 2008; Cepeda, Fife, Chow, Mastrogiovanni, & Henderson,
2013; Jena et al., 2014; McDonald & Carlson, 2013; Peirce et al., 2012; Pradel et al., 2010;
Schloff et al., 2004; Webster & Webster, 2005; White, Birnbaum, Schiller, Tang, & Katz, 2009)
It is important to note that these behaviors—especially as observed in administrative claims
data—could also be signals for fragmented care rather than drug-seeking behavior. As such,
none of these indicators will be perfect predictors of opioid abuse or overdose.
The goal of identifying additional, administrative metrics of opioid abuse or misuse is
based in a desire to recognize patterns that may lead to opioid abuse, but also in a desire to
identify opioid-related outcomes that are not reliant on medical diagnoses as those are rare events
and are not be included in recent Medicare claims data.(Frakt & Bagley, 2015) To that end, the
selection of appropriate abuse indicators relies on balancing their prevalence with their
explanatory power in a regression context. For example, while an indicator for having an

23

overlapping prescription improves the fit of the models of abuse or overdose more than other
indicators, it is fairly common and as such is likely too inclusive to be the best non-medical
indicator of misuse. Similarly, combining all of the indicators creates a sample that is similar in
size to the population with opioid abuse diagnoses, but improves the fit of the overdose and
abuse models the least. As such these two indicators may not be the best additional indicators of
opioid abuse or misuse.
Thus, it appears that the individual indicators of whether a patient has more than three
opioid prescribers, fills those prescriptions at more than two pharmacies, and has more than a
year’s supply of opioids are fairly good indicators of likely opioid abuse as is the standard doctor
shopping indicator (combination 0, 1, 2). As noted earlier, they are not perfect indicators of
abuse and may capture fragmented care alongside misuse of opioids. As a result, as shown in
Table 5, I estimate that between 4 and 14 percent of opioid users are possibly misusing those
prescriptions.

VI. Discussion
While the descriptive nature of this study is a key limitation to drawing causal
relationships about the role of the explanatory variables on opioid outcomes of interest, it does
provide an important baseline for understanding opioid use, abuse, and misuse in the Medicare
population. Opioid use in Medicare is fairly constant from 2006 to 2012, but over that period
opioid abuse, average prescription strengths, and average prescription lengths are increasing.
Similarly, while the share of beneficiaries who use opioids who have an acute pain diagnosis has
changed very little over the period, the share with chronic pain diagnoses has increased
dramatically. Further, among users with multiple years of opioid use, that utilization increases

24

monotonically with the duration of that use. Additionally, depending on the metric in use,
between 4 and 14 percent of Medicare beneficiaries exhibit some utilization pattern that can be
categorized as misuse of opioids. And, unsurprisingly, physicians who deal explicitly with pain
and addiction are prescribers of the most and most potent opioid prescriptions for Medicare
beneficiaries.
The Medicare population exhibits many similarities and a few key dissimilarities in their
use of opioids when compared to findings for the U.S. population. First, there has been no clear
trend in the utilization of opioids by the Medicare population. While studies by the CDC and
others have shown a huge increase in the number of patients receiving opioids, a fairly stable
share of Medicare beneficiaries fill prescriptions for opioids. A possible explanation for this
stable trend is that while in the general population it is true that pain went undertreated, that such
a phenomenon was less true in the Medicare population. Additional data predating the 2006
introduction of Medicare Part D might shed some light on whether this stability is a short- or
long-term trend for Medicare beneficiaries.
In addition, while this study confirms that younger Medicare beneficiaries, like younger
patients in general are more likely to use opioids, it is the case that conditional on use, increasing
age is associated with longer, stronger prescriptions. While this finding does not directly
contradict existing research, it suggests that the Medicare beneficiaries most at risk for potential
misuse of or addiction to opioids may be older rather than younger beneficiaries.
Unlike prior studies, many of which have lacked proxies for socio-economic status, this
study shows that low socio-economic status beneficiaries are more likely to use opioids, more
likely to use high-dose opioids, and have longer, stronger prescriptions than their counterparts.
Interestingly, there is little evidence that these individuals, especially when dually-eligible, are

25

more likely to abuse opioids or to overdose on opioids. It is possible that their status as dually-
eligible beneficiaries means that they are subject to Medicaid policies that lessen this risk. If such
policies do exist, it would be interesting to explore whether they are the cause of this disconnect
between use and abuse or overdose in this population.
While this study provides a useful baseline for understanding opioid use in Medicare, it
provides many avenues for additional research to explore why Medicare beneficiaries exhibit
dissimilar behavior from the general population. It also raises the question of whether any policy
interventions over the period from 2006 to 2012 are the source of these findings.

26

VII. Figures and Tables
Figure 1. Share of Opioid Users with a Selected Diagnosis by Year, 2007 - 2012

Notes: Opioid abuse diagnoses were redacted in 2012.

Cancer
Opioid Abuse
Back Pain
Fracture
Mental Health
Chronic Pain
0%
5%
10%
15%
20%
25%
30%
35%
40%
2007 2008 2009 2010 2011 2012

27

Figure 2. Intensity of Opioid Use by Years of Opioid Use

3.1
5.0
6.8
8.5
9.9
12.0
14.7
0
2
4
6
8
10
12
14
16
Average Count of Opioid Fills
40
82
125
166
205
265
333
0
50
100
150
200
250
300
350
Average Days Supply
14%
16%
17%
19%
20%
21%
25%
0%
5%
10%
15%
20%
25%
30%
1 2 3 4 5 6 7
Share High Dose Users
Count of Years with Multiple Opioid Fills

28

Figure 3. Use of Multiple Providers by Opioid Users

38.4
33.9
14.2
6.4
3.1
1.6
0.9
0.5
0.3
0.2
0.4
66.5
22.8
6.3
2.3
1.0
0.5
0.2
0.1
0.1
0.0
0.1
0
10
20
30
40
50
60
70
1 2 3 4 5 6 7 8 9 10 >10
Percent of Opioid Users
Count of Unique Opioid Providers
Prescribers
Pharmacies

29

Figure 4. Growth in the Number of Opioid Prescriptions by Specialty, 2007 – 2012

13.9%
9.5%
16.2%
16.9%
83.4%
148.8%
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
All
Specialties
General
Medicine
Emergency
Medicine
Surgery Pain
Medicine
Nursing
Growth in Total Prescriptions 2007 - 2012

30

Table 1. Summary Statistics
Opioid Users (>1 Opioid Rx per year) Full
Year 2006 2007 2008 2009 2010 2011 2012 Sample
Obs. (mill) 0.47 0.72 0.75 0.77 0.80 0.80 0.82 5.14
Age 66.3 67.1 66.9 66.7 66.3 65.7 65.7 69.9
Risk Score 1.30 1.28 1.29 1.31 1.32 1.35 1.35 1.11
% Any Dual 63.5 49.5 49.3 50.0 50.2 50.5 49.4 38.7
% Any LIS 3.7 8.0 7.7 7.4 7.6 7.3 6.8 7.1
% Disabled 48.9 45.8 46.6 47.7 49.1 51.1 51.1 31.0
% Male 32.4 33.8 34.5 34.9 35.5 36.9 37.4 35.3
% White 74.8 78.1 77.7 76.8 76.5 76.1 76.4 76.5

31

Table 2. Regression Adjusted Opioid Utilization, 2006 – 2012
2006 2007 2008 2009 2010 2011 2012 Mean SD
Full Sample
+

Any Use 37.6% 36.8% 37.3% 37.6% 37.7% 36.0% 35.3% 36.8% 0.21
>1 Opioid 25.1% 24.1% 24.5% 24.9% 25.1% 23.8% 23.2% 24.3% 0.20
>30 Days 19.0% 18.1% 18.8% 19.2% 19.7% 18.7% 18.3% 18.8% 0.19
Conditional on >1 Opioid Rx in a Year
+

Fills* 8.6 8.6 8.8 9.0 9.1 9.3 9.1 8.9 3.7
Days* 162 169 182 192 204 214 212 193 143
MME* 9,138 9,417 10,018 11,283 11,744 10,621 10,151 10,428 10,781
High
Dose ^
26.0% 25.1% 23.6% 22.9% 20.6% 12.8% 12.4% 20.0% 0.10

Notes: All results are on a per-year basis.
+
Findings control for differences in age, sex, race, dual eligibility, low income status, disability status, ESRD status,
end-of-life status, various diagnoses, and health status.
^High Dose users receive more than 100 milligrams of morphine equivalence per day for some period of time.
*These estimates are predicted using the smearing estimator in Duan (1983)

32

Table 3. Regression Adjusted Measures of Persistence of Opioid Use, 2006 - 2012

2006-
2007
2007-
2008
2008-
2009
2009-
2010
2010-
2011
2011-
2012
Mean
>1 Opioid Fills in Consecutive Years
+

Share of Full Sample 9.6% 14.4% 15.1% 15.4% 14.8% 14.4% 13.9%
Share of Users 40.0% 58.6% 60.7% 61.4% 62.1% 61.9% 52.4%
Conditional on >1 Opioid Fills in Consecutive Years
+

Fills* 11.7 11.3 11.4 11.5 11.5 11.4 11.5
Days* 237.2 240.0 247.8 259.1 265.8 269.8 254.9
MME* 13,580 13,794 15,412 15,918 14,259 14,122 14,589
High Daily Dose^ 26.0% 24.7% 24.2% 22.3% 15.7% 15.3% 20.9%

Notes:
+
Findings control for differences in age, sex, race, dual eligibility, low income status, disability status, ESRD
status, end-of-life status, various diagnoses, and health status.
^High Dose users receive more than 100 milligrams of morphine equivalence per day for some period of time.
*These estimates are predicted using the smearing estimator in Duan (1983)

33

Table 4. Physician Results

Mean Fills
Mean Days
Supply
Mean MME per
Day
Share High
Dose Rxs
General, Internal, or
Family Medicine
22 19 47.3 6.9%
Dental 5 4 41.4 2.8%
Emergency 11 5 50.3 3.4%
Nursing 11 15 47.6 6.5%
Surgery 12 8 60.8 7.1%
Anesthesia 17 20 61.3 10.9%
Hospice or Palliative
Care
12 13 63.3 14.2%
Pain Medicine 90 31 70.4 13.2%
Addiction Medicine 21 22 106.1 37.4%
Other or Unknown 8 10 50.2 6.8%
Total 14 12 49.7 6.4%

Notes: All findings are on a per physician-year basis. Days supply and MME per day are calculated on a per-
prescription basis.
Findings control for differences in age, sex, race, dual eligibility, low income status, disability status, ESRD
status, end-of-life status, various diagnoses, and health status.

34

Table 5. Abuse Results
Outcome 2006 2007 2008 2009 2010 2011 2012
Opioid Abuse 1.6% 1.6% 1.8% 2.1% 2.5% 3.1% --
Opioid Overdose 0.3% 0.3% 0.3% 0.4% 0.4% 0.4% 0.4%
(1) Overlapping Rx 25.7% 25.4% 26.5% 27.0% 28.0% 28.5% 28.0%
(2) >3 Prescribers 12.8% 12.4% 13.3% 13.4% 13.8% 14.0% 13.9%
(3) >2 Pharmacies 9.3% 9.6% 10.3% 11.0% 10.9% 11.3% 11.4%
(4) >365 Day Supply 12.5% 12.9% 14.1% 14.5% 15.1% 15.6% 14.9%
(1-3) Combination 4.1% 4.0% 4.4% 4.6% 4.6% 4.7% 4.4%
(1-4) Combination 1.9% 1.9% 2.2% 2.3% 2.3% 2.2% 2.0%

Notes: All results are on a per-year basis. An overlapping rx indicates that the individual has more than one opioid
prescription at the same time from different physicians.
Findings control for differences in age, sex, race, dual eligibility, low income status, disability status, ESRD status,
end-of-life status, various diagnoses, and health status.
Due to data redaction of substance abuse claims, there are no claims for opioid abuse diagnoses in 2012.

35

Table 6. Choosing Between Administrative Abuse Metrics
Pseudo R-Squared Value
DV = Opioid Abuse
Pseudo R-Squared Value
DV = Opioid Overdose
Baseline 0.2841 0.1922
Overlapping Rx (0) 0.2920 0.1961
>3 Prescribers (1) 0.2880 0.1936
>2 Pharmacies (2) 0.2914 0.1936
>365 Day Supply (3) 0.2928 0.1958
Combination (0, 1, 2) 0.2895 0.1930
Combination (0, 1, 2, 3) 0.2884 0.1930

Notes: Findings control for differences in age, sex, race, dual eligibility, low income status, disability status, ESRD
status, end-of-life status, various diagnoses, and health status.

36

CHAPTER 3
ASSESSING THE EFFECTIVENESS OF STATE POLICIES FOR ALTERING OPIOID
USE AND MISUSE AMONG MEDICARE BENEFICIARIES

I. Introduction
As outlined in detail in Chapter 2, abuse of legal prescription drugs, particularly opioids,
is a growing problem in the United States. In response to the health and economic costs of this
epidemic, both federal and state governments have acted in an attempt to lessen the opioid crisis
in the United States. The most common governmental response has been the implementation of
regulations aimed at restricting access to controlled substances. There are various national laws,
like the Controlled Substances Act, that set the parameters of legal and illegal use of prescription
drugs, but physicians and pharmacists are largely regulated at the state level. As a result, states
have experimented with various laws to tackle the widespread problem of opioid misuse and
abuse. The most common form of experimentation is the implementation of prescription drug
monitoring programs (PDMPs) to track prescriptions of controlled substances.
But, given the escalation in opioid use over the past decade--even the most stringent of
these PDMPs have done little to solve the problem. There is another set of state-level regulations
aimed at reducing abuse of controlled substances that has not yet been rigorously evaluated.
These laws include: tamper-resistant prescription pads, pain management clinics, physical exams
before prescribing, patient identification laws, and others.(Prevention) Anecdotal evidence has
shown fewer prescriptions of opioids in the wake of these laws, but there has not yet been a
systematic analysis of their impact.(Office of the Attorney General, 2014) This chapter fills that
gap.

37

The paper proceeds as follows: section II contains background on the types of laws I am
evaluating as well as highlights the contribution of this paper in relation to existing evidence on
the efficacy of these laws, section III explains my data sources, section IV outlines my
estimation strategy, section V provides the findings of my analysis, and section VI concludes
with policy recommendations.

II. Background
Both federal and state governments have experimented with a wide variety of laws aimed
at regulating the prescription and use of controlled substances. In particular, states have
implemented seven distinct categories of laws which include: stricter regulation of pain
management clinics, introduction of tamper-resistant prescription pads, prescription drug
monitoring programs, mandatory identification laws, doctor shopping laws, overdose-related
Good Samaritan laws, and physical examination laws. All of these laws are aimed at restricting
or reducing the availability of controlled substances like opioids. In this paper, I focus on the
impact of pain management clinic laws, tamper-resistant prescriptions pads, and prescription
drug monitoring programs on a wide variety of opioid outcomes. I chose these laws because they
were implemented during my observation period, while many of the other laws pre- or post-date
my data range; they have previously been shown to impact opioid outcomes at least anecdotally;
they are cited as promising policies for directly combatting opioid misuse rather than broad
policies aimed at regulating prescription drug sales; and they employ distinct policy mechanisms
to achieve the end goal of a reduction in opioid misuse.
Pain management clinic laws are aimed at preventing inappropriate prescribing of
controlled substances. They aim to rein in so-called “pill mills” because of their high volume of

38

opioid prescriptions. The regulations vary but in general they set required levels of physician
involvement, inspection processes, licensing procedures, or standards of care at clinics. If a clinic
fails on a required dimension, it can be shuttered. The laws define pain management clinics as
sources of care where the majority of patients are being treated for pain and/or are receiving
controlled substances to treat that pain. From 2006 through 2011, three states implemented pain
clinic regulations.
These increased regulations on pain clinics serve as effective supply constraints on the
easy availability of prescription opioids. For example, if a patient who used to obtain
prescriptions at one of these clinics is no longer able to, he or she will be forced to seek them
elsewhere—and perhaps face more resistance from a new provider. This purported mechanism is
even stronger if it is the case that the most indiscriminate clinics are those most affected by the
law. However, policies working solely through supply constraints can have negative spillovers.
For example, the hypothesized patient above might engage in doctor shopping to obtain multiple
opioid prescriptions to replace those prescriptions easily obtained at a less-regulated pain clinic.
Existing reports have found a sizeable effect of these regulations; however, they have
been largely anecdotal in nature. An evaluation of Florida’s pain clinic law and prescription
monitoring program found decreased reports of diversion after the implementation of both
laws.(Surratt et al., 2014) Further, Florida news media report that more than 1,200 pain clinics
have been closed and that controlled-substance related overdose deaths have fallen in the wake
of their law.(Anderson, 2015) Despite their anecdotal promise and increasing implementation, I
am aware of no rigorous analyses of their causal impact to date. This study fills a hole in the
existing literature as it directly estimates the impact of these laws on a wide variety of opioid
outcomes.

39

Tamper-resistant prescription pads were designed to limit the creation or use of
fraudulent prescriptions for controlled substances. While the details vary, such laws require the
use of security features like serialized prescription pads, watermarks showing “void” when
photocopied, or pre-printed check boxes for quantity and dosage. Any prescriptions lacking such
security features may not be filled at a pharmacy. There are two major categories of laws within
tamper-resistance—those that require the use of such pads for all prescriptions and those that
apply only to Medicaid prescriptions. Because my analysis tests the impact of these laws in the
Medicare population, I only consider those laws that apply to all prescriptions in the state. From
2006 through 2011, nine states began requiring the use of tamper-resistant prescription pads for
all controlled substance prescriptions.
Tamper-resistant prescription pads could lower opioid prescribing or misuse through one
of two policy mechanisms. The most straightforward pathway through which such a policy could
impact opioid misuse is to raise the effective cost of engaging in fraud. As prescriptions can less
easily be copied or altered, it becomes more difficult or time-consuming for patients to engage in
fraud and as a result might limit such behavior. Another possible route would be a behavioral
“nudge” for providers. After the implementation of the policy, prescribers had to order
specialized prescription pads for these prescriptions. Thus, each time a prescriber wants to write
an opioid prescription, he must select a distinct prescription pad which could induce him to be
more restrained in his prescribing practices by prescribing a lower quantity or fewer refills.
Either mechanism could reduce opioid use or misuse, but the effect would be seen in different
outcomes. By employing a wide variety of opioid outcomes I am able to determine which of
these pathways is the most likely.

40

While studies have pointed to the introduction of these laws as an important tool in
combatting fraudulent controlled substance use, to the best of my knowledge their direct impact
has not been estimated.(Fishman, Papazian, Gonzalez, Riches, & Gilson, 2004; Jayawant &
Balkrishnan, 2005) Two studies that considered the impact of prescription drug monitoring
programs on opioid outcomes concluded that states with such programs and tamper-resistant
prescription pads saw larger reductions or statistically significant reductions in opioid use or
misuse when compared to states without both laws.(Paulozzi, Kilbourne, & Desai, 2011;
Paulozzi & Stier, 2010) However, due to their study design, neither could determine whether
those effects were due to the combination of laws or in fact to the tamper-resistant prescription
laws rather than the prescription drug monitoring program. This study fills a hole in the existing
literature on policy efforts to limit opioid misuse by estimating the direct effect of these laws as
well as working to understand how those effects are achieved.
Prescription drug monitoring programs are one of the most wide-spread tools used to
combat inappropriate opioid use. These laws require a physician or pharmacist to report all
prescriptions of covered drugs to a centralized database. This database can then be queried by
prescribers or pharmacists prior to issuing prescriptions if they suspect that a patient may be
fraudulently obtaining opioids. These laws vary in the frequency and mode of reporting, what
classes of drugs are covered, who can query the system, and whether a physician is required to
query the system before issuing a prescription. Due to the wide variety of such laws, I only
consider the implementation of electronic prescription drug monitoring programs in my
analysis—where data on prescriptions is shared electronically rather than via paper forms. From
2006 through 2011, 19 states implemented or upgraded to electronic prescription monitoring
programs. While other programs may be in place, the lag in their responsiveness due to the need

41

to collect paper forms makes them less likely to have an impact. Therefore, I focus on these
enhanced versions of prescription drug monitoring programs as they are most likely to produce
an impact.
Prescription drug monitoring programs rely on the assumption that the provision of
information or access to information can help eliminate the information asymmetry between
patients and physicians. Patients have private information about their true level of pain, their
existing prescriptions for opioids, and their proclivity to misuse these prescriptions that
physicians cannot access. The PDMP helps at least partially solve that problem by giving
physicians a means to access data on their existing prescriptions for opioids. While not
completely resolving the information asymmetry, responsive and well-utilized PDMPs can
lessen it.
Compared to other laws regulating controlled substances, there is a literature assessing
the impact of prescription drug monitoring programs on a wide variety of opioid-related
outcomes like drug diversion, overdoses, doctor-shopping, and opioid abuse. I am not aware of
any studies that comprehensively evaluate the impact of monitoring programs on opioid use and
prescribing, just misuse. Among the most important findings of studies of this kind are that
proactive rather than reactive programs have a larger impact on drug diversion and doctor
shopping, that states without programs that border ones with programs see an increase in drug
diversion, and that programs make the prosecution of drug diversion cases by law enforcement
faster. In sum, rather than lowering opioid misuse, what most evaluations find is that prescription
monitoring programs make it easier to monitor prescription drug diversion.
A few of these studies utilize reliable datasets and rigorous methods like difference-in-
differences analysis or comparison of a treatment and control group. One of these studies on the

42

impact of PDMPs by Simeone and Holland shows that their implementation does decrease illegal
drug diversion. In this paper, the authors compared states with PMPs and states without to see if
supply and abuse of Schedule II drugs differed.(Simeone & Holland, 2006) In Simeone and
Holland’s paper, they note that those programs that are proactive rather than reactive (relying on
the physician to query a patient’s prescription history) are more effective. A study by Reisman,
et al had similar findings.(Reisman, Shenoy, Atherly, & Flowers, 2009) In a comparison of
bordering states, New York and Pennsylvania, the authors find that New York’s PDMP has been
more successful in part because New York performs proactive analyses of the collected data to
identify irregularities, something Pennsylvania only began doing recently (Paulozzi & Stier,
2010). In a national study, Paulozzi and colleagues find no effect on overdoses and small effects
on consumption.(Paulozzi, Kilbourne, et al., 2011) However, the vast majority of these studies
rely on observational methods or the use of convenience samples, making it difficult to
generalize their findings.(Baehren et al., 2010) My focus on a large, representative, individual-
level data set of a population that has high levels of opioid use and the use of econometric
estimates to produce plausibly causal estimates provides an improvement on the existing
literature. Further, I look at their impact on a larger range of outcomes, not just opioid abuse,
overdose or diversion.
In addition to empirical studies on opioid abuse, there is a growing theoretical literature
on the economics of addiction that considers which forms of medical or policy intervention are
most likely to impact the consumption of addictive goods, like opioids, and break cycles of
addictive behavior. Many of the theories derived to explain addiction agree that punitive policies
like prohibitions or criminalization are unlikely to help reduce addictive behavior.(Bernheim &
Rangel, 2005) They also tend to agree that taxation or higher prices on addictive goods are

43

ineffective, in contrast to the policy implications of the theory of rational addiction. (Becker,
Grossman, & Murphy, 1991, 1994; Grossman & Chaloupka, 1998)However, depending on the
particulars of the models—namely assumptions regarding preferences and how to measure
welfare—there are slight variations in which form of policy is seen as best for lowering
addiction. In general, this literature concludes that policies that provide opportunities for better
self-regulation will be most successful.(Bernheim & Rangel, 2005) Laibson, for example, argues
that policies that reduce the cues that can trigger desires to engage in addictive behaviors will be
most effective. Therefore, he calls for regulations that lower these cues, like no smoking in
public buildings, as a way to lessen addiction.(Laibson, 2001) Other theories predict that policies
that provide for better decision-making through the imposition of a mechanism like a
consumption cap, binding decisions, or decisions derived from a narrower choice set are likely to
reduce the probability of addiction or overconsumption.(Brocas & Carrillo, 2008; Loewenstein,
2000; Loewenstein & O’Donoghue, 2004) These theories also agree that information alone is not
sufficient to reduce or eliminate addictive behavior.(Bickel, Jarmolowicz, Mueller, &
Gatchalian, 2011)
Following these theories of addiction, the laws I study focus on reducing the availability
of opioids, of creating higher barriers to obtaining an opioid prescription, and to providing
information to opioid providers. Based on the theoretical predictions of these economic models
of addiction, the greater regulation of pain management clinics plausibly reduces the supply of
opioids—and in doing so also reduces the choice set of opioid providers. As noted earlier, these
laws could also lead to perverse outcomes like increased doctor-shopping for prescriptions as a
patient’s provider choice set is narrowed. Nonetheless, these theories of addiction would predict
some effect of these laws. The use of tamper-resistant prescription pads lowers or eliminates the

44

possibility that patients can fraudulently copy prescriptions—again providing an opportunity for
better decision-making as it makes fraud more difficult. Additionally, it is plausible that the need
for providers to use a different pad could nudge their prescribing behavior. While the link
between theoretical predictions of reduction in addictive behavior and tamper resistant pads is
weaker it is still plausible that there might be an impact. As PDMPs focus on provider rather than
patient behavior, these theories provide little insight into their likely efficacy.
However another important policy component to consider is whether these policies are
likely to be effectively enforced, not just implemented. I hypothesize that pain management
clinic regulations and tamper-resistant prescription pads are likely to have an impact as these
laws have clear embedded enforcement mechanisms. While I cannot be sure whether they are
perfectly enforced, the fact that these policies rely on existing regulatory checks (e.g. requiring
pharmacists to determine the veracity of a prescription) or new ones (e.g. the creation of
licensing and inspection procedures for pain clinics) increase the likelihood that they will have
an impact on opioid outcomes. Prescription drug monitoring programs, on the other hand, only
create a voluntary database for physicians to check. Without an enforcement mechanism like a
requirement that prescribers access such information with regularity, there is a smaller likelihood
that PDMPs will have a large impact on opioid outcomes. While I do not believe that PDMPs
will increase opioid use, hypotheses based on theories of addiction or the quality of the policy’s
enforcement mechanism favor pain management clinic laws and tamper-resistant prescription
pad laws over PDMPs.

45

III. Data
My analysis relies on two major datasets—a compilation of the effective dates of seven
different categories of state laws aimed at reducing inappropriate opioid use and a 20 percent
random sample of Medicare inpatient, outpatient, and prescription drug claims. The dataset on
the effective dates of laws is an original dataset I compiled based on data from the CDC’s Public
Health Law Program combined with a search of legal databases and online state registers. To put
together the most accurate set of effective dates I first determined which states had implemented
one of these laws as categorized by the CDC Public Health Law Program. Then, as the CDC
only recorded the effective year of the law where available, I followed a three-step process. First,
I searched Lexis-Nexis for the effective dates of the laws and provisions of interest. When Lexis-
Nexis was unclear or the law itself did not include an effective date, I searched the state register
of these states to find the resulting rulemaking or regulations which included effective dates.
Finally, where possible, I corroborated those effective dates in the regulations with reports in the
news media or by interest groups. Additionally, for the effective dates of prescription monitoring
programs I relied on a dataset from the National Association of Model State Drug Laws which
lists the date of first electronic receipt of data to a state’s prescription drug monitoring program.
That date is a good proxy for the implementation of an electronic prescription monitoring
program.
To test the impact of these laws on opioid-related outcomes, I employ a 20 percent
random sample of administrative claims for all Medicare beneficiaries from 2006 to 2011. That
base sample is then restricted to those beneficiaries who have traditional, fee-for-service
Medicare for all months in a given year so that I can observe all of their medical and prescription
utilization. These claims include Medicare Part A (hospital care), Medicare Part B (outpatient

46

and physician care), and Medicare Part D (prescription drug) claims. They contain information
on the timing and cost of all Medicare claims, demographic information about the beneficiaries,
and estimates of their health status. In particular, I use information on the beneficiary’s sex, age,
race or ethnicity, whether they are dually eligible for both Medicare and Medicaid benefits,
whether they receive a low income cost sharing subsidy to help cover the cost of their Medicare
Part D drug plan, whether they are eligible for Medicare because of a qualifying disability, and
whether they are at the end of their life (if they died in that year). The paper also uses an estimate
of the individual’s health status based on the risk-adjustment calculations used by Medicare
when reimbursing Part D plans—called the RxHCC value—which is a function of the
individual’s age, sex, category of eligibility, and diagnostic history(Robst et al., 2007).
From this baseline of traditional Medicare beneficiaries, this paper then identifies those
beneficiaries with a history of opioid use. Opioid use is defined as filled opioid prescriptions in a
patient’s Medicare Part D claims. Opioids are identified from beneficiaries’ Part D claims using
national drug codes identified in the IMS Health, First DataBank, or RxNorm drug databases.
The analysis includes opioid agonists, opioid partial agonists, and opioid combinations, but
excludes opioid antagonists, antitussive opioid combinations, as well as tramadol products.
I create a variety of opioid outcomes for analysis to offer a more complete assessment of opioid
utilization as well as to determine on which margin these policies have an impact. These
outcomes can be grouped broadly into the categories of opioid utilization or opioid misuse/abuse.
Opioid utilization outcomes I report include: (1) probability of being an opioid user, (2) count of
fills of opioid in a quarter conditional on use of opioids, and (3) receipt of a high dose opioid
prescription in a quarter (high dose is defined as a prescription that provides more than 100
MME per day). Notably, I define an opioid user as an individual with an opioid fill in a quarter

47

who also fills more than one opioid prescription during a year. This definition increases the
likelihood that the beneficiaries under study are overusing or inappropriately using opioids as
compared to an individual with just a single opioid fill.
Two of these key outcome measures come directly from the Medicare claims data—
receipt of more than one opioid prescription and the count of opioid fills per beneficiary-quarter.
However, because opioid prescriptions vary not only in their quantity but also in their potency,
this paper also relies on estimates of those opioids’ morphine equivalence as a way to compare
across prescriptions. Following the methodology of the CONSORT research group as well as the
Center for Medicare and Medicaid Services’ guidance to health plans, morphine equivalence was
calculated according to a prescription’s active ingredient, dosage strength, and quantity of the
medication distributed. For example, the active ingredient of oxycodone has a morphine
equivalence of 1.5, so a prescription for 30 pills of 20 milligrams of oxycodone would have a
milligram morphine equivalence (MME) of 1.5 x 20MG x 30 pills, or 900 MME. The active
ingredient conversion factors are available upon request. Thus, the third key outcome is
calculated using these equivalence calculations and data points like days supply and strength of
prescriptions that comes directly from the claims data.
Opioid misuse and abuse outcomes that I measure include: (1) prescriber shopping, (2)
prescriber and pharmacy shopping, and (3) diagnosis of opioid abuse or overdose. The
diagnosis-related outcome measures come directly from Medicare claims—whether they had a
diagnosis of opioid abuse or overdose in a given year. I note that my time period precedes the
redaction of substance-abuse related claims by CMS so I am still capturing all diagnoses of
substance abuse.(Frakt & Bagley, 2015) However, there is concern that these diagnosis-related
measures are missing a large share of opioid users who may be misusing or over-utilizing

48

opioids. Following the literature on doctor-shopping for opioids, I define misuse through a
variety of measures of prescriber-shopping (having overlapping opioid fills from different
prescribers; using 3 or more prescribers in a quarter), pharmacy-shopping (using 2 or more
pharmacies in a quarter), and overutilization (receiving more than 90 days supply in a quarter).
The ultimate measure I employ for prescriber and pharmacy shopping is an indicator variable for
a patient with: overlapping fills from different physicians, using 3 or more prescribers, and using
2 or more pharmacies in a given quarter. Prescriber shopping is defined as an indicator variable
for a patient with overlapping fills from different physicians who uses 3 or more prescribers per
quarter.
This paper also estimates the impact of these laws on physician prescribing. The sample
used for this analysis starts with the sample of opioid users described above. These claims are
merged to an administrative file containing information about the prescriber including medical
specialty, certification, and location using a unique prescriber identifier. The data is then
collapsed to create a database of total physician prescribing of opioids by quarter and year.
Included in the database are variables for the average characteristics of a physician’s patient
population including: the share who are white, average age, the share who are male, average
health risk score, and others that come from the patient sample described above. The sample is
roughly representative of prescribers who prescribe opioids to Medicare beneficiaries. However,
I only keep those physicians who report a primary state of practice in a given year. While this
limits the generalizability of my sample, it is necessary for the identification of my models.
Prescribing measures that I estimate include (1) count of fills per quarter per prescriber,
(2) average MME per day of a prescriber’s prescriptions, and (3) the share of a prescriber’s

49

prescriptions that are high-dose (more than 100 MME per day). All measures are conditional on
a prescriber prescribing at least one opioid in a given quarter.
It is important to note that because my data comes from administrative claims, I cannot
account for any opioid prescriptions that are filled and paid for in cash rather than through the
use of Medicare benefits. Therefore, my outcome measures are conservative estimates of
utilization and prescribing. Further, as I only observe filled prescriptions, all outcomes are
measured per-prescription written; I cannot verify whether these prescriptions were fully
consumed nor whether they were consumed by the beneficiary in question. As such, my
prescription-based outcome measures reflect opioids prescribed rather than opioids consumed.
While these are clear limitations of my data, by employing a rich dataset and a wide
range of outcome measures of opioid use, misuse, and prescribing, I am able to offer a more
complete assessment of the impact of these laws on opioid use and misuse. I can estimate not
only their impact on the probability of being an opioid user, but also conditional on being a user
the intensity of use as measured by number of fills or propensity to receive high dose opioids. It
is also the first assessment I have seen of the impact of such policies on opioid prescribing.
Additionally, the use of both medically diagnosed abuse outcomes as well as misuse as observed
in prescription utilization patterns provides a clearer understanding of the impact of these laws
on their target population—opioid misusers.

IV. Empirical Strategy
To identify the impact of these state laws on the opioid outcomes listed above, I exploit
the variation in timing of the implementation of the laws across states. In particular, I compare
those states that implement the laws between 2006 and the end of 2011 to those states that do

50

not--controlling for other observable factors, unobservable state characteristics, and
unobservable time characteristics. By netting out the impact of observable and unobservable
differences between the treated and untreated states, I draw a plausible causal link between the
laws and changes in opioid-related outcomes in the treated states.
To net out the effect of observed and unobserved characteristics, I estimate multivariate
regressions of the form in equation (1).

y
ist
= δ
t
+ α
s
+ θ
1
PC
st
+ θ
2
TR
st
+ θ
3
PMP
st
+ η
1
ID
st
+ η
2
OD
st
+ η
3
SHOP
st
+ βX
ist
+ ε
ist
(1)

In the equations above, i indexes the individual beneficiary, s the state, and t the quarter-
year. Y
ist
is the opioid outcome of interested, δ
t
is a year fixed effect, α
s
is a state fixed effect, and
X
ist
is a vector of patient-level characteristics. Characteristics include age, race, gender, health
status, Medicare eligibility status, and whether the individual died in that year. η
n
LAW
st
is a
dummy variable for whether the state had one of the three non-focus laws (ID requirements,
overdose-related good Samaritan laws, and specific doctor shopping laws) in place in that quarter
and year. PC (pain clinic regulations), TR (tamper resistant prescription pads), and PMP
(prescription drug monitoring program) are indicator variables that equal one if that state has that
law of interest in place in that quarter and year, otherwise they equal zero. The θs, then, are the
parameters of interest as they estimate the difference in outcomes between treatment and control
states. I expect the θs to be negative or zero as these laws should either reduce or have no effect
on the opioid outcomes of interest.
For the prescriber analysis I estimate multivariate regressions of a similar form where p
indexes the prescriber and X
pst
represents the average characteristics of the prescriber’s patients

51

(e.g. average age, share male, share white, average health status score, share with various
Medicare eligibility statuses).

Y
pst
= δ
t
+ α
s
+ θ
1
PC
st
+ θ
2
TR
st
+ θ
3
PMP
st
+ η
1
ID
st
+ η
2
OD
st
+ η
3
SHOP
st
+ βX
pst
+ ε
pst
(2)

For count outcomes like opioid fills, total MMEs, and days supply, y
ist
or y
pst
is modeled
as the natural log of that outcome variable and I estimate log-linear regressions. As a robustness
check I also estimate these outcomes using a fixed effects poisson model. Other outcomes—
probability of use, probability of high-dose prescriptions, and probability of misuse, abuse, or
overdose—are estimated using linear probability models. Following the literature, I estimate all
regressions using robust standard errors clustered at the state level.(Bertrand, Duflo, &
Mullainathan, 2004) I also estimate the models above by different diagnosis categories as a way
to examine whether the policy impact is concentrated in individuals with certain types of pain
diagnoses (e.g. acute versus chronic, malignant versus non-malignant).
For proper identification of the parameters of interest, it is necessary that implementation
of these laws is uncorrelated with unobserved determinants of opioid related outcomes.
Therefore, a common threat to identification in difference-in-differences estimates is the
possibility of policy endogeneity—whereby some unobserved factor is correlated both with the
implementation of these laws and the opioid outcomes.(Besley & Case, 2000) The included state
fixed effects will remove the effect of any time-invariant unobserved factor and the included
quarter-year fixed effect will remove any national trends. For example, in 2010, propoxyphene—
a commonly abused opioid was removed from the market and a tamper resistant form of
OxyContin was introduced; the fixed effects will absorb the impact of these national

52

changes.(Dart et al., 2015; Larochelle, Zhang, Ross-Degnan, & Wharam, 2015) These
adjustments account for most, but not all categories of problematic unobserved factors.
What these adjustments do not control for is some state-specific trend in the outcome
measures that is correlated with the law’s implementation. To help reduce the possibility of bias
from such unobserved correlation, I include dummies for the implementation of the three
additional control laws. I also estimate robustness checks which include state-specific linear
time-trends, which controls for any pre-treatment trends in a state.(Besley & Burgess, 2004) I
tried estimating my models using placebo or fuzzy implementation dates, and found some
evidence of anticipation of the policy. Therefore, as a robustness check of my quarterly results, I
also estimate these models at an annual level which allows the implementation date to shift
slightly. While I cannot test whether I have removed all sources of bias from my estimates, the
inclusion of other related state laws and state-specific time trends lessens the possibility that my
parameter estimates are heavily biased.

V. Results
Opioid use is common among Medicare beneficiaries. As shown in Table 1A, about 19
percent of Medicare beneficiaries are opioid users in a given quarter—that is, individuals who
fill an opioid in a quarter and fill multiple opioids within a calendar year. These users are
younger, in worse health, more likely to be dually eligible for Medicare and Medicaid, receive a
subsidy (LIS) to afford Part D coverage, live in the South, and have been disabled at some point
than the average Medicare beneficiary. These users receive an average of three prescriptions per
quarter, more than 60 days supply per quarter, and 15 percent receive a high dose prescription
per quarter. In addition, about 3 percent of users are diagnosed as abusers and more exhibit

53

doctor or pharmacy shopping tendencies. Opioid prescribers are largely general or internal
medicine physicians. As shown in Table 1B, the average opioid prescriber prescribes 24
prescriptions to Medicare beneficiaries per quarter, 10 percent of which are high dose
prescriptions.
V.I. Opioid Utilization
Figure 1 shows the unadjusted mean of opioid misuse or abuse outcomes by state and
quarter. Bars labeled “C,” or control, reflect the average, unadjusted outcome for states and
quarters without the law. Bars labeled “T,” or treatment reflect the average, unadjusted outcome
for states and quarters after the law’s implementation. The bars in blue correspond to pain
management clinic regulations, orange to tamper-resistant prescriptions, and red to prescription
drug monitoring programs. Looking across the first row of charts, all three laws appear to
slightly lower the average number of fills and lower the probability that a patient receives a high
daily dose of opioids. However, only the introduction of tamper-resistant prescriptions appears to
result in a lower probability of opioid use; the other laws appear to be associated with an increase
in the probability of using. While these results do not control for a myriad of confounding
factors, they provide some evidence that these laws could be having an impact on opioid
utilization.
Table 2 shows the result of the difference-in-differences models in equation (1) for opioid
utilization measures. Specification 1 includes only the laws of interest and state and time fixed
effects, specification 2 adds controls for the additional laws aimed at opioid prescribing, and
specification 3 adds controls for beneficiary demographics, eligibility status, and health status.
Tamper-resistance laws significantly lower the probability of being an opioid user by 0.35
percentage points (p<0.05). There is a similar effect of pain clinic regulations and prescription

54

monitoring programs though the effect is less significant (p<0.1). Interestingly, tamper-resistant
prescription pads lower the average number of fills among opioid users by 1.4 percent (p<0.05),
but have no effect on the probability of receiving a high dose prescription. However, pain clinic
management regulations appear to lower the probability of receiving a high dose prescription by
3.4 percentage points (p<0.05). The results related to the impact of prescription monitoring
programs on measures of utilization, while negative, are indistinguishable from zero.
These results imply that tamper resistant prescription pads are associated with lower
quantity of opioids prescribed to patients and pain management clinics with lower strength. I also
estimated count models using fixed effects poisson models as robustness checks. The results are
qualitatively similar when estimated with log-linear or count models. Further, the results are
largely robust to the inclusion of state-specific linear time trends, although the effect of tamper
resistance laws and PDMPs on the probability of opioid use fades and the effect of pain clinic
laws on that outcome is more significant. When I estimate these results at the annual rather than
the quarterly level, the effect of PDMPs on opioid use fades as does the effect of tamper
resistance laws on opioid fills. Based on these sensitivity analyses, I am hesitant to conclude that
PDMPs have any effect on any of these utilization outcomes; but, the results related to tamper
resistant pads and pain clinic laws are supported in all or most of the additional estimation
strategies. Table 6 contains an accounting of the impact of my sensitivity analyses; full results
are available in the appendix. In addition, I employed different definitions of opioid use—the one
employed in table 2, the receipt of any opioid in a quarter, and the receipt of two opioids in a
quarter—the results are similar regardless of definition. These additional results are available in
the appendix.

55

V.II. Opioid Abuse or Misuse
Figure 1 shows the unadjusted mean of opioid abuse or misuse outcomes by state and
quarter. Bars labeled “C,” or control, reflect the average, unadjusted outcome for states and
quarters without the law. Bars labeled “T,” or treatment reflect the average, unadjusted outcome
for states and quarters after the law’s implementation. The bars in blue correspond to pain
management clinic regulations, orange to tamper-resistant prescriptions, and red to prescription
drug monitoring programs. Looking across the second row of charts, the implementation of these
laws appear to have small or no effect on the opioid users targeted by these laws: abusers and
misusers. There appears to be no effect on abuse or overdose diagnoses of any of the laws, a
sizeable effect of prescriber shopping of pain clinic and tamper resistance laws, and a small
decrease in prescriber and pharmacy shopping with the introduction of tamper resistant
prescriptions and pain clinic regulations. None of these estimates, though, control for the effect
of confounding factors.
Table 3 shows the difference-in-differences estimates of the models related to measures
of opioid abuse or misuse. Again I focus on specification 3 which includes quarter-year and state
fixed effects, controls for characteristics of the beneficiaries, and controls for the introduction of
other state laws. Consistent with the unadjusted findings in Figure 1, there is relatively little
effect of these laws on measures of opioid abuse or misuse. Pain clinic laws are associated with a
0.3 percentage point reduction (p<0.05) in the probability of an abuse or overdose diagnosis. The
introduction of tamper-resistant prescription pads is associated with a 0.3 percentage point
reduction (p<0.05) in the probability of a patient exhibiting doctor- and pharmacy-shopping
behavior. None of the laws appear to significantly impact prescriber shopping—in fact the

56

coefficients on pain clinic laws and PDMPs are all positive. Again, prescription drug monitoring
programs have no discernable association with reductions in measures of opioid abuse or misuse.
In models that add state-specific time trends or models estimated at the annual level, the
results for tamper-resistant laws are qualitatively similar. Notable differences include a
significant reduction of 0.1 percentage points (p<0.05) in opioid abuse or overdose associated
with the introduction of tamper resistance laws in models that include state-specific linear time
trends. Sensitivity analyses of the effect of pain clinic laws on the probability of abuse or
overdose both find effects that are impossible to distinguish from zero. Therefore, I am hesitant
to conclude that pain clinic laws affect abuse or overdose.
Additionally, the inclusion of state-specific linear time trends produce effects that show a
statistically significant increase in prescriber shopping as well as in physician and pharmacy
shopping associated with the introduction of pain clinic laws (p<0.05). The latter results are
particularly interesting as they imply that the reduction in supply resulting from stricter
regulation may have forced patients to increase their doctor shopping as the laxest outlets for
opioids may have been eliminated. In general, while there are some associations between the
introduction of these laws and lower rates of opioid abuse or misuse, with the exception of
reducing shopping behavior, they are relatively ineffective at lowering outcomes that these
policies were designed to target.
V.III. Physician Prescribing of Opioids
Figure 1 shows the unadjusted mean of opioid utilization outcomes by state and quarter.
Bars labeled “C,” or control, reflect the average, unadjusted outcome for states and quarters
without the law. Bars labeled “T,” or treatment reflect the average, unadjusted outcome for states
and quarters after the law’s implementation. The bars in blue correspond to pain management

57

clinic regulations, orange to tamper-resistant prescriptions, and red to prescription drug
monitoring programs. Looking across the last row of charts, only tamper resistant prescriptions
appear to have an impact on the average number of fills per physician, but all three laws,
particularly pain management clinic regulation, appear to lower the average MME per day
prescribed. There is a similar effect of these laws on the share of prescriptions that are high dose.
Table 4 shows estimates of the impact of the introduction of these laws on various
prescriber-level outcomes. I report estimates from specification 3 which controls for the average
characteristics of a prescriber’s patient population, the prescriber’s specialty, additional laws,
quarter-year fixed effects, and state fixed effects. Consistent with the findings from Figure 1, the
introduction of pain management clinic regulations is associated with a significant, negative
reduction in measures of physician prescribing. Their introduction is associated with a 1.4
percent reduction in prescriptions written (p<0.05), a 9.2 percent reduction in the average MME
per day prescribed (p<0.01), and a 3.4 percentage point reduction in the probability of
prescribing a high-dose opioid (p<0.01). Interestingly, though the introduction of tamper-
resistant prescription pads appeared to affect the quantity of opioids received by patients, it had
no significant association with the quantity of opioids prescribed by providers. Again, the
introduction of prescription drug monitoring programs appears to have no impact on physician
prescribing.
These findings are robust to the inclusion of state-specific linear time trends. The findings
related to MME per day and the share of fills that are high dose are also robust to estimation
using count models and/or to estimation at the annual rather than quarterly level. However, the
effect of pain clinic laws on the count of opioid fills per prescriber is indistinguishable from zero
when the models are estimated with fixed effect poisson models or are estimated at the annual

58

level. These findings lend further credence—this time on the provider side—that stricter
regulation of pain clinics has an effect not on the frequency of opioid use, but instead on the
strength of opioids prescribed. It is plausible that prior to these regulations these clinics were
more likely to provide such high dose opioids and that by constraining the supply of these clinics
that the strength of opioid being prescribed was lowered.
V.IV. Impact by Patient Diagnosis
In Table 5 I show the impact of these laws by a patient’s diagnostic history. It is plausible
that these policies might not have a large impact in the general opioid-using population, but that
their impact might be restricted to certain kinds of patients that are at higher risk for abuse or
misuse, for example. To investigate that possibility, I separately estimate the previous models by
patient diagnoses. In particular I show the impact on patients with a diagnosis of cancer or an
acute fracture as well as the impact on patients with a chronic pain diagnosis (chronic pain or
back pain). As chronic pain sufferers are more likely to be opioid abusers than are malignant or
acute pain sufferers, and these laws purportedly aim to reduce problematic use of opioids, I
would expect to see reductions on measures of utilization largely for chronic pain sufferers and
not malignant or acute pain sufferers. As the table shows, though, tamper-resistant prescription
pads are associated with significant reductions in the average number of fills across many types
of diagnoses and pain management clinic regulations are associated with significant reductions in
the probability of receiving a high dose prescription across many diagnosis types. Aside from
opioid fills, the impact appears larger among cancer and fracture patients than chronic pain
sufferers.

59

V.IV. Impact by Patient Eligibility Status
In table 6 I show the impact of these laws by a patient’s initial eligibility status for
Medicare—whether they were disabled or elderly. Much like these policies having a differential
impact by patient diagnosis, it is also plausible that the policies might have a differential impact
on patients by their eligibility status as they represent fairly distinct patient populations—most
notably, the disabled population is much younger than the population eligible due to old age.
Additionally, recent expansions to the Social Security Disability Insurance program made it
easier for patients suffering from chronic pain to gain Medicare coverage. As these patients are
likely to be opioid users, these policies might be targeted at the disabled Medicare beneficiaries
rather than the aged beneficiaries.
As shown in table 6, there are a few notable differences in the policies’ effects when
decomposed by eligibility status. First, prescription drug monitoring programs appear to have a
small, statistically significant impact on the probability of being an opioid user among disabled
Medicare beneficiaries. While there is still no evidence of an impact of these policies on other
measures of use and misuse, by narrowing in on the disabled Medicare population I find some
small impact of these policies that is absent in the broader Medicare population.
Additionally, it appears that where tamper resistant prescription pads are in place that
their effect is concentrated in the disabled rather than the non-disabled population. The reduction
in the probability of opioid use, in opioid fills, and in the probability of doctor and pharmacy
shopping associated with the implementation of tamper-resistant prescription pad laws occurs
only among disabled beneficiaries and opioid users. These laws do not appear to alter the
utilization or misuse behavior of non-disabled Medicare beneficiaries. To the extent that disabled
Medicare beneficiaries may be considered higher-risk opioid users, these findings provide some

60

evidence that tamper resistant prescription pad policies might be better-targeted interventions as
they do not appear to affect patient care for aged, non-disabled beneficiaries.
In contrast, the impact of pain management clinic laws on measures of opioid use is more
widespread. For example, the reduction in the probability of receiving a high dose opioid
prescription is similar in magnitude and level of statistical significance for both disabled and
non-disabled beneficiaries. However, the effect of pain management clinic laws on measures of
opioid misuse is more mixed. The impact of these policies on the probability of an abuse
diagnosis or overdose is concentrated in the disabled population while the impact on doctor and
pharmacy shopping occurs only in the non-disabled Medicare population. Unlike the tamper-
resistant prescription pad laws, the supply constraints implemented by pain management clinic
regulations are widespread and appear to have similar effects regardless of eligibility status. To
the extent that policies aim to reduce opioid use and misuse across the board rather than in
particular subpopulations, these laws appear to be broader in their effects.

VI. Discussion
As shown in Section V, the introduction of pain management clinic regulations and the
required use of tamper resistant prescription pads are associated with a variety of reductions in
opioid utilization, misuse, and prescribing; prescription drug monitoring programs appear to
have little or no impact on opioid use or misuse in the Medicare population. Tables 5 and 6
presents a summary of these findings related to utilization and misuse. It shows the percent
change in mean outcomes associated with the implementation of these laws.
These results reinforce that these laws may have an impact on opioid utilization or
misuse, but on different margins and through different mechanisms. Tamper-resistant

61

prescription pads required that physicians order and utilize new pads when prescribing narcotics.
It is plausible that this behavioral change led physicians to reconsider their decision to write
prescriptions; but, the lack of an effect of these policies on measures of prescribing undermines
that conclusion. Instead, the reduction in prescriber and pharmacy shopping points instead to the
likelihood that the reductions were due to the increased cost of engaging in fraud. Such a
conclusion is also supported by the concentration of the effect in quantity rather than in strength
as duplicating prescriptions became more difficult. Further, these regulations appear to be
relatively more targeted with the bulk of their effect occurring on disabled rather than aged
Medicare beneficiaries.
Pain management clinic regulations were designed to decrease the high volume or
indiscriminate prescribing occurring at pain management clinics—to reduce supply. Consistent
with that focus was a reduction in utilization on the margin of opioid strength rather than on
quantity. That finding is also bolstered by evidence that pain management clinics may be
associated with a small increase in shopping for opioids as users had to look elsewhere following
a reduction in supply. Similarly, as it is plausible that patients who were more likely to be
misusers and/or abusers sought out prescriptions at these relatively lax clinics, it follows that
these laws were associated with a small reduction in abuse and misuse and other laws were not.
The impacts of these laws were felt broadly, with similar reductions among multiple
subpopulations of Medicare beneficiaries.
Prescription monitoring programs appear to have little to no impact on opioid utilization,
misuse, or physician prescribing despite the attention and resources that have been directed
toward their implementation. It is plausible that this lack of impact has more to do with the
possibility that although such programs are in place, relatively few prescribers are utilizing the

62

resource.(Haffajee, Jena, & Weiner, 2015) Unlike the tamper-resistant and pain management
clinic regulations which set strict requirements with embedded enforcement mechanisms,
prescription monitoring programs are largely “opt-in” policies that require physicians to set up
accounts and proactively seek out information about their patients. This lack of an enforcement
mechanism is a plausible explanation for the small impact on opioid outcomes. To probe the
importance of embedded enforcement mechanisms, future research could explore whether the
subset of prescription drug monitoring programs which proactively report patient utilization to
physicians or that require prescribers to check the registry prior to prescribing have a larger
impact on these measures of use, misuse, and prescribing.
Taking all of my results into consideration, it seems that prescription drug monitoring
programs in their current setup are unlikely to be an effective policy for reducing opioid use or
misuse. I would urge the reformulation of prescription monitoring program so that they have an
effective enforcement mechanism like the other, more impactful policies. While tamper resistant
prescription drug pads show promise in lowering some outcomes related to opioid misuse and
the quantity of opioids received by patients, they have no impact on prescribing of opioids. The
introduction of pain management clinic regulations appear to be the most impactful of these
policies in magnitude, but there is some evidence that they may lead to negative effects like
increased shopping behavior or might have too broad of an impact. Therefore, a more evidence-
based policy agenda would favor the implementation of multiple laws aimed at the margins of
both quantity and potency of opioids. In particular, pairing pain management clinic regulations
with tamper-resistant prescription pads could attack the problem from multiple angles as well as
limit the negative effects of supply constraints. Future research could work to determine the
optimal combination of laws to tackle the problem of high and rising opioid use and misuse.

63

VII. Figures and Tables
Table 1A. Summary Statistics of Patient Sample by Opioid User Status, 2006 – 2011

Full Sample Opioid Users
Observations (Beneficiar-Quarters) 64,208,328 12,284,478
Demographics & Health Status
White 77.1% 77.0%
Black 10.6% 13.2%
Asian 3.1% 1.2%
Hispanic 7.0% 6.1%
Other Race/Ethnicity 2.2% 2.5%
Age (Years) 70.2 65.7
Male
37.4% 34.3%
Health Status (Score) 1.1 1.3
Died in year 3.2% 4.6%
Eligibility Status
Ever Dual 39.1% 53.9%
Ever LIS 5.9% 7.5%
Ever Disabled 31.0% 52.3%
ESRD 1.5% 2.7%
Location of Medicare Beneficiaries
West 17.2% 17.2%
South 39.1% 45.1%
Northeast 18.6% 13.4%
Midwest 25.1% 24.3%
Selected Diagnoses of Opioid Users
Cancer 16.5%
Fracture 13.5%
Back Pain 30.4%
Chronic Pain 16.0%
Outcomes of Interest
Opioid Use 19.1% 100%
Fills 3.1
Days 61.3
MME 3,619
High Dose Prescription 15.5%
Abuse or Overdose 2.9%
Doctor Shopping (1) 16.5%
Pharmacy Shopping (2) 13.1%
>90 Days 19.1%
Combo of (1) & (2) 3.6%

64

Table 1B. Summary Statistics of Opioid Prescriber Sample, 2006 – 2011

Opioid Prescribers
Observations (Prescriber-Quarters)

5,980,677
Average Demographics & Health Status of Patients

Share White

75.9%

Age (Years)

64.4

Male

35.9%

Health Status (Score)

1.3
Died in year 4.7%
Average Eligibility Status of Patients

Ever Dual

51.0%

Ever LIS

6.9%

Ever Disabled

51.0%

ESRD

3.4%
Location of Prescribers

West

20.1%

South

37.1%

Northeast

18.0%

Midwest

24.7%
Outcomes of Interest per Prescriber
Fills 24.3
Days 14.2
MME 710.9
MME per day 52.1
High Dose 10.1%
Prescriber Specialty
General Medicine 44.3%
Dental 6.0%
ER 7.5%
Nurse 4.5%
Surgery 12.3%
Anesthesia 0.5%
Palliative Care or Hospice Care 0.1%
Pain Medicine 1.1%
Addiction Medicine 0.1%
Other/Unknown 23.6%

65

Figure 1. Unadjusted Outcomes by Control (C) and Treatment (T) States and Law

Probability of Any Opioid Use
by Law
Count of Opioid Fills by Law
Probability of a High Dose
Opioid by Law

Probability of Opioid Abuse or
Overdose by Law
Probability of Prescriber
Shopping by Law
Probability of
Doctor/Pharmacy Shopping by
Law

Count of Opioid Fills
Prescribed by Law
Average MME per Day
Prescribed by Law
Share of Prescriptions that are
High Dose by Law

19.1%
21.3%
19.4%
18.4%
18.1%
19.7%
16.0%
17.0%
18.0%
19.0%
20.0%
21.0%
22.0%
C T C T C T
3.1
3.0
3.2
3.0
3.2
3.1
2.8
2.9
3.0
3.1
3.2
3.3
C T C T C T
15.7%
9.7%
16.0%
14.1%
17.4%
14.5%
0.0%
5.0%
10.0%
15.0%
20.0%
C T C T C T
2.9%
3.1%
2.9%
3.0%
2.8%
3.1%
2.5%
2.6%
2.7%
2.8%
2.9%
3.0%
3.1%
3.2%
C T C T C T
16.6%
15.1%
17.0%
15.1%
16.6%
16.5%
14.0%
14.5%
15.0%
15.5%
16.0%
16.5%
17.0%
17.5%
C T C T C T
3.6%
3.5%
3.7%
3.3%
3.5%
3.6%
3.1%
3.2%
3.3%
3.4%
3.5%
3.6%
3.7%
3.8%
C T C T C T
24.2
26.9
24.7
22.9
23.4
24.8
20.0
21.0
22.0
23.0
24.0
25.0
26.0
27.0
28.0
C T C T C T
52.3
43.3
52.8
49.8
54.6
50.6
0.0
10.0
20.0
30.0
40.0
50.0
60.0
C T C T C T
10.2%
5.5%
10.4%
9.2%
11.1%
9.5%
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
12.0%
C T C T C T
Tamper-Resistant
Prescriptions
Pain Management Clinic
Regulation
Prescription Drug Monitoring
Program

66

Table 2. Difference-in-Differences Estimates of Opioid Utilization Outcomes
Specification 1 2 3
Panel 1: Probability of being an Opioid User estimated by linear probability model
Sample Mean 19.1%
Pain Clinic Law
-0.0015 -0.0009 -0.0038*
(0.002) (0.002) (0.002)
Tamper Resistance
-0.0061*** -0.0054*** -0.0035**
(0.002) (0.002) (0.001)
PDMP
-0.0028 -0.0028 -0.0025*
(0.002) (0.002) (0.001)
Observations 64,208,328 64,208,328 64,208,328
Panel 2: Count of Opioid Fills estimated by log-linear regression
Sample Mean 3.1
Pain Clinic Law
0.0101 0.0069 0.0009
(0.006) (0.007) (0.006)
Tamper Resistance
-0.0159** -0.0196*** -0.0135**
(0.007) (0.007) (0.006)
PDMP
-0.0005 -0.0011 -0.0010
(0.004) (0.004) (0.005)
Observations 12,284,478 12,284,478 12,284,478
Panel 3: Probability of High Dose (>100MME) Rx estimated by linear probability model
Sample Mean 15.5%
Pain Clinic Law -0.0312** -0.0329*** -0.0342***

(0.013) (0.011) (0.011)
Tamper Resistance 0.0134** 0.0055 0.0066

(0.005) (0.005) (0.005)
PDMP -0.0021 -0.0017 -0.0017

(0.006) (0.006) (0.006)
Observations 12,284,478 12,284,478 12,284,478
Demographics x
Eligibility Status x
Specialty x
Control Laws x x
Quarter-Year FE x x x
State FE x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses. Fills and Prob(>100 MME per day) is conditional on
being an opioid user.

67

Table 3. Difference-in Differences Estimates of Opioid Abuse or Misuse Outcomes
Specification 1 2 3
Panel 1: Abuse or Overdose Diagnosis estimated by linear probability model
Sample Mean 2.9%
Pain Clinic Law
-0.0021 -0.0018 -0.0031**
(0.001) (0.001) (0.001)
Tamper Resistance
0.0004 -0.0001 0.0011
(0.001) (0.001) (0.001)
PDMP
0.0007 0.0009 0.0009
(0.001) (0.001) (0.001)
Panel 2: Prescriber Shopping estimated by linear probability model
Sample Mean 16.5%
Pain Clinic Law
0.0043 0.0050 0.0005
(0.004) (0.003) (0.003)
Tamper Resistance
-0.0066 -0.0062 -0.0024
(0.003) (0.004) (0.003)
PDMP
0.0008 0.0010 0.0015
(0.003) (0.003) (0.002)
Panel 3: Prescriber and Pharmacy Shopping estimated by linear probability model
Sample Mean 3.6%
Pain Clinic Law 0.0003 0.0005 -0.0011
(0.002) (0.002) (0.002)
Tamper Resistance -0.0041*** -0.0048*** -0.0034**
(0.001) (0.001) (0.001)
PDMP -0.0002 -0.0001 0.0000
(0.001) (0.001) (0.001)
Demographics x
Eligibility Status x
Specialty x
Control Laws x x
Quarter-Year FE x x x
State FE x x x
Observations 12,284,478 12,284,478 12,284,478
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

68

Table 4. Difference-in-Difference Estimates of Opioid Prescribing Outcomes
Specification 1 2 3
Panel 1: Count of Fills estimated by log-linear regression
Sample Mean 24.2
Pain Clinic Law
-0.0284*** -0.0225*** -0.0141**
(0.005) (0.006) (0.007)
Tamper Resistance
-0.0258 -0.0221 -0.0201
(0.015) (0.016) (0.016)
PDMP
-0.0085 -0.0069 -0.0055
(0.009) (0.009) (0.010)
Panel 2: MME per Day estimated by log-linear regression
Sample Mean 52.1
Pain Clinic Law
-0.0915** -0.0886** -0.0918***
(0.036) (0.035) (0.034)
Tamper Resistance
0.0236 0.0079 0.0094
(0.024) (0.022) (0.023)
PDMP
-0.0049 -0.0015 -0.0009
(0.015) (0.015) (0.015)
Panel 3: Share of Prescriptions >100 MME per day estimated by LPM
Sample Mean 10.1%
Pain Clinic Law -0.0336** -0.0341*** -0.0342***
(0.013) (0.012) (0.012)
Tamper Resistance 0.0095 0.0030 0.0031
(0.006) (0.005) (0.005)
PDMP 0.0000 0.0008 0.0009
(0.005) (0.005) (0.005)
Demographics x
Eligibility Status x
Specialty x
Control Laws x x
Quarter-Year FE x x x
State FE x x x
Observations 5,980,677 5,980,677 5,980,677
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

69

Table 5. Summary of Difference-in-Differences Estimates of Mean Law Impact on Patient
Outcomes by Patient Diagnostic Categories
Prob
(Opioid
User)
Opioid Fills Prob
(>100 MME
per day)
Opioid
Abuse or
Overdose
Physician &
Pharmacy
Shopping
Pain Clinic Regulation
All Diagnoses -2.0% -22.1% -10.4%
Cancer/Fracture -20.7% -16.5%
Chronic Pain -18.2% -7.5%
Tamper Resistant Prescriptions
All Diagnoses -1.8% -1.3% -9.1%
Cancer/Fracture -1.1% -9.5%
Chronic Pain -1.8% -7.3%
Prescription Drug Monitoring Program
All Diagnoses -1.3%
Cancer/Fracture 7.8%
Chronic Pain
Notes: All estimates are based off of specification 3 from the regression results. Values in bold are significant at
the p<0.05 level and values in regular type are significant at the p<0.1 level. Blank cells represent results with no
significant findings.

70

Table 6. Summary of Difference-in-Differences Estimates of Law Impact on Patient
Outcomes by Patient Medicare Eligibility Status
Prob
(Opioid User)
Opioid Fills Prob
(>100 MME
per day)
Opioid Abuse
or Overdose
Diagnosis
Physician &
Pharmacy
Shopping
Pain Clinic Regulation
All Users -2.0% -22.1% -10.4%
Disabled -1.4% -15.5% -7.4%
Non-Disabled -2.6% -24.9% -8.8%
Tamper Resistant Prescriptions
All Users -1.8% -1.3% -9.1%
Disabled -2.7% -1.7% -9.1%
Non-Disabled
Prescription Drug Monitoring Program
All Users -1.3%
Disabled -1.1%
Non-Disabled
Notes: All estimates are based off of specification 3 from the regression results. Values in bold are significant at
the p<0.05 level and values in regular type are significant at the p<0.1 level. Blank cells represent results with no
significant findings.

71

Table 7. Summary of Sensitivity Analyses

Panel 1: Opioid Use

Prob
(Opioid User)
Fills
Prob
(>100 MME per day)
Preferred Result (Specification 3)
Pain Clinic
Law
-0.0038* -0.0342***
Tamper
Resistance
-0.0035** -0.0135**
PDMP -0.0025*
Inclusion of State-Specific Linear Time Trends (Specification 4)
Pain Clinic
Law
Increases significance of
finding
Lowers effect size, still
negative and significant
Tamper
Resistance
Finding not significant Lowers effect size, still
negative and significant

PDMP Finding not significant Significant, negative effect
Estimation of Count Models by Fixed Effect Poisson (Specification 3 Controls)
Tamper
Resistance
Increases effect size
Estimation at Annual Rather than Quarterly Level (Specification 3 Controls)
Pain Clinic
Law
Increases effect size and
significance of finding
Lowers effect size, still
negative and significant
Tamper
Resistance
Increases effect size Finding not significant
PDMP Finding not significant

Panel 2: Opioid Misuse
Abuse or Overdose Prescriber Shopping
Physician & Pharmacy
Shopping
Preferred Result (Specification 3)
Pain Clinic
Law
-0.0031**
Tamper
Resistance
-0.0034**
Inclusion of State-Specific Linear Time Trends (Specification 4)
Pain Clinic
Law
Finding not significant Positive, significant effect Positive, significant effect
Tamper
Resistance
Negative, significant effect Lowers effect size, still
negative and significant
Estimation at Annual Rather than Quarterly Level (Specification 3 Controls)
Pain Clinic
Law
Finding not significant
Tamper
Resistance
Negative, significant effect Lowers effect size, still
negative and significant

72

Panel 3: Opioid Prescribing
Fills MME per day
Share of Fills >100 MME
per day
Preferred Result (Specification 3)
Pain Clinic
Law
-0.0141** -0.0918*** -0.0342***
Inclusion of State-Specific Linear Time Trends (Specification 4)
Pain Clinic
Law
Increases effect size Lowers effect size, still
negative and significant
Lowers effect size, still
negative and significant
Estimation of Count Models by Fixed Effect Poisson (Specification 3 Controls)
Pain Clinic
Law
Finding not significant Increases effect size
Estimation at Annual Rather than Quarterly Level (Specification 3 Controls)
Pain Clinic
Law
Finding not significant Lowers effect size, still
negative and significant
Lowers effect size, still
negative and significant
Tamper
Resistance
Negative, significant effect

Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Green shading
indicates a new finding where the previous effect size was estimated to be zero; orange shading indicates a finding
of no effect where previously there was an estimated effect; and, no shading indicates qualitatively similar findings.
Missing rows also indicate qualitatively similar findings. Grey shading indicates that the sensitivity analysis was not
run on those outcomes. At the annual level, probability of being an opioid user is defined as multiple fills in a
calendar year.

73

CHAPTER 4
EMERGENCY DEPARTMENT CONTRIBUTION TO THE
PRESCRIPTION OPIOID EPIDEMIC
1

I. Introduction
Propelled by pharmaceutical marketing, a purported low-addiction potential, and
regulatory pressure to “ensure all patients the right to appropriate assessment and management of
their pain,” prescriptions for opioid analgesics have risen dramatically over the past 15 years. By
some estimates, the total amount of milligrams of morphine equivalents (MME) has increased by
300% from 180 MME per capita in 1997 to 710 MME per capita in 2010. (Cantrill et al., 2012)
(Betses & Brennan, 2013) Deaths and hospitalizations due to prescription opioid use have
paralleled this dramatic rise in prescriptions. In 2010, 15,000 people died of prescription drug
overdose in the United States (up from 4,030 in 1999). (Betses & Brennan, 2013; Paulozzi,
2012) (Paulozzi, Weisler, & Patkar, 2011) Death from opioid analgesics is now the most
common cause of injury death among patients aged 35-54. (C. M. Jones, Mack, & Paulozzi,
2013) Recent evidence shows that individual risk factors associated with death from opioid
prescriptions include filling opioid prescriptions from four or more prescribers of opioids within
1 year, using more than 4 pharmacies, most significantly, history of mental health disorders, and
receiving greater than 100 MME of opioid prescriptions daily. (Baumblatt et al., 2014)
In response to these trends, recent prescription guidelines aimed at reducing opioid
misuse have emerged focusing on Emergency Department (ED) prescribing. (Cai, Crane,
Poneleit, & Paulozzi, 2010) In New York City, then Mayor Michael Bloomberg held a large
press conference to announce the health authority’s ED prescribing guidelines that included a 3-

1
This paper was co-authored with Michael Menchine and Seth Seabury.

74

day limit to the quantity of opioids that could be prescribed from New York City public hospital
EDs. In the state of Washington, Medicaid required EDs to adopt opioid diversion guidelines to
ensure re-imbursement for care they provide. (Juurlink, Dhalla, & Nelson, 2013) Other states and
many health plans are working on similar opioid diversion plans centering on emergency care. In
general, these and similar plans have been supported by state level chapters of the American
College of Emergency Physicians. Central to the success of these policies in impacting the
opioid epidemic in the United States is the belief that EDs significantly contribute to the national
drug problem. However, a recent clinical policy review notes that the contribution of EDs to the
national epidemic of prescription opioid misuse almost entirely unknown.
The goal of this investigation is to evaluate the relative contribution of EDs to the
national prescription opioid epidemic. Specifically, we aim to 1) estimate the temporal trend in
total quantity of opioids prescribed by site of care (e.g. ED vs. Office visit) 2) evaluate average
quantities and strength-of-dose of opioids prescribed 3) determine the likelihood of receiving a
high dose opioid prescription by site of care and 4) identify the relative importance of single or
refilled prescriptions in opioid prescribing. By analyzing a variety of outcomes we aim to better
characterize the contribution of the EDs to opioid prescription availability and misuse.

II. Data and Sample
To answer our research questions, we rely on a retrospective analysis of the Medical
Expenditure Panel Survey (MEPS) from 1996 to 2012.
2
The MEPS is a nationally representative
sub-survey of the annual National Health Interview Survey administered by the Agency for
Healthcare Research and Quality. It is administered to approximately 15,000 individuals

2
We exclude data from 2013 as opioid prescriptions are not completely characterized by active ingredient and thus
cannot be reliably converted to milligrams of morphine equivalence.

75

annually. In MEPS, non-institutionalized individuals within a selected household give detailed
description of health care events over the preceding year and then are followed with five in-
person visits throughout the year to record current health care use. As such, any given individual
may provide health care use data for a period of up to two years. Health care events are grouped
according to the following categories: 1) Dental visits, 2) visits to outpatient settings 3) office
visits 4) ER visits 5) Inpatient visits 6) home health visits 7) miscellaneous and 8) prescribed
medications. MEPS personnel then obtain an authorization from the survey respondent to contact
service providers (including pharmacies) and obtain detailed records for each event. Each event
file contains the ICD-9 codes, services provided and out-of-pocket expense associated with that
event. The MEPS utilizes a complex survey design that oversamples certain populations and
requires using sampling weights to obtain national estimates. Details of the survey strategy are
available at http://meps.ahrq.gov/mepsweb/. Subjects were included in the analysis if they had at
least 1 prescription opioid event during the study period and were over age 18. This investigation
was certified exempt from review by the local Institutional Review Board.
II.I Linking Procedures
AHRQ creates various analysis files according to the grouping of care described earlier.
To analyze the role of source of care in opioid prescribing, we have to link various files together
following the most conservative methodology. Each of the prescribed medicines files was linked,
using the appendix file provided through MEPS, to each of the medical visits files. We
performed a one-to-many merge as a given prescribed medicine was in certain cases linked with
more than one medical event. For example, if a patient entered the ED, was given a pain killer,
and then admitted to the hospital, that pain killer could be linked to both the ED event and the
hospital inpatient event.

76

Because the primary goal of this analysis is to examine the relationship between opioid
prescriptions and emergency room visits, it was important to be able to ascribe each prescription
event to a single source of care. Therefore, for all prescription drug events with multiple
associated sources of care, we followed the following set of rules. First, if any prescription event
was associated with more than one medical event of the same type (e.g. two outpatient visits),
that single source was associated with the prescription. Second, if a prescription was associated
with multiple types of medical events, the most common one was associated with the
prescription (e.g. if 2 ED events and 1 dental event was associated with the prescription, it was
recorded as an ED prescription). Finally, for the remaining prescription events, in order to
provide conservative estimates of the difference between ED and office based prescribing they
were labeled first as ED events if they were ever associated with the ED, then as dental events if
ever associated with dental visits, then as office based, and finally as inpatient prescriptions.
Prescriptions that were not associated with any medical event were labeled as “other.”
II.II Identifying Opioids and Morphine Equivalence
For the prescribed medications file, the event file contains a National Drug Code (NDC)
for each prescribed agent. The NDC is a unique identifier that describes the compound,
formulation, dose and quantity of a given medication. The MEPS then uses the Multum Lexicon
to group NDC codes into larger clinically meaningful categories. For example, the NDC for
hydrocodone would be captured under the Multum code 060 or 191 for opioid analgesics.
Misclassification was a significant concern. As such, we evaluated the actual compound names
for the following Multum codes and eliminated those observations associated with codes that
were not opioid analgesics or combination opioid analgesics. We examined the compound names
within Multum codes that could be reasonably be expected to contain opioid analgesics. We did

77

not review compound names in Multum codes that would not likely contain an opioid analgesic
(e.g. anti-hypertensives).
Once we narrowed the analysis to opioids, we determined the total potency of these
opioids so that they can be directly compared to one-another. To do so, we converted these drugs
to their milligrams of morphine equivalence (MME) using the conversion factors in the table in
the appendix which come from published morphine equivalence charts. We then calculated the
total morphine equivalence of the prescription event which was equal to the prescription’s
strength in milligrams multiplied by the MME conversion factor and the quantity of the
prescription. Prescriptions lacking data on both the active ingredient and the NDC were dropped
from the sample. For prescriptions lacking valid NDCs or data on strength or quantity, those
values were replaced by the median value recorded for the drug’s active ingredient.
II.III Identifying One-Time or Refill Prescriptions
In addition to categorizing prescriptions by the source of medical care, the analysis also
breaks out prescriptions by whether they are one-time (e.g. no refills) prescriptions or
prescriptions that contain refills. Our categorization of refills relies on the MEPS architecture in
the prescribed drug event files. Each record is given a unique prescription identifier as well as an
identifier for linking prescription events to medical events. We classify a prescription as
containing refills if there are multiple observations of the unique prescription identifier within
the event identifier. That is, if the individual obtains the same drug (the same NDC) in the same
quantity, linked to the same medical encounter more than once in a calendar year, that is
classified as a refill prescription. If they obtain that drug only once in a calendar year, or it is
associated with a different medical encounter, it is classified as a one-time prescription.
Following this procedure, we specify the first of these identical fills or a one-time prescription as

78

the “incident” prescription and apply the identified refills to determine the quantity and strength
of the entire prescription.

III. Empirical Strategy
The goal of this paper is to estimate the relative contribution of various sources of care to
the opioid epidemic in the United States. We do so by employing a variety of outcomes of
interest and using multivariate regression analysis to control for differences by patient
demographics, diagnostic history, and insurance status. We then report regression-adjusted
estimates of the relative contribution of a variety of sources of care.
III.I Regression Adjustment
We implement a multivariate regression analysis of the form in equation 1. In this
equation, r indexes the prescription and t the year and δ
t
is a year fixed effect. Y is one of
the outcomes described in detail below, X is a vector of patient demographics associated
with the patient receiving the prescription including age, gender, race, region, marital
status, poverty status, urban status, or indicators for a variety of diagnoses; or, X is a
vector of prescription characteristics such as whether it has refills and the type of
pharmacy used. We include dummy variables for patients reporting clinical conditions
that are associated with high opioid use--these include painful joint conditions, acute
injury, back pain, headache and major mental health diagnoses. Source indicates which
source of care the prescription is tied to—office based, emergency department, inpatient,
outpatient, dental or other and α is the key coefficient of interest.

(1) y
rt
= α*source
rt
+ βX
rt
+ δ
t
+ ε
rt

79

Where the outcomes of interest are counts, equation 1 is estimated via log-linear
regressions and the reported coefficients can be interpreted as the percent change in the outcome
associated with a given characteristic. When reporting regression adjusted outcomes, we employ
the smearing methodology from Duan (1983) to account for differences in the underlying
distribution functions when transforming the results from log scale.(Duan, 1983) Where
outcomes are binary, equation 1 is estimated using logit regressions and odds ratios and marginal
effects are reported.
III.II Outcomes of Interest
To more completely characterize opioid prescribing by site of care, we employ five key
outcomes: (1) the total milligrams of morphine equivalence per prescription (2) the strength per
dose per prescription (3) the quantity in milligrams of the prescription (4) the rate of receipt of
high dose prescriptions and (5) the rate of receipt of “chronic” prescriptions. Both the strength
and quantity outcomes come directly from the data reported to MEPS by individual patients. The
total milligrams of morphine equivalence is calculated from that original data and the MME
conversion factor explained earlier. A prescription is determined to be “high dose” if its strength
per dose (dosage*MME conversion factor) equals or exceeds 100MME. Since days supply is not
well-populated, this is our best estimate of the clinical guideline identified as an overdose risk
factor. It is likely a conservative estimate as patients often consume multiple doses per day. We
alternatively identify “high dose” prescriptions as those with an active ingredient that is more
potent than morphine (e.g. has an MME conversion factor greater than 1).
3
And finally, chronic
prescriptions are estimated as either a prescription with more than 90 doses per fill or as a
prescription with more than 6 refills per year.

3
These active ingredients include Butorphanol, Fentanyl, Fentanyl Citrate, Hydromorphone, Levorphanol,
Oxymorphone, and Oxycodone.

80

IV. Results
We identified 47,551 unique individuals receiving 83,987 opioid prescriptions during the
study period. The mean age of individuals included in the study was 49 years (std dev. 17.2
years) and 63% were female. More complete characteristics of the sample are given in Table 1.
IV.I Unadjusted Trends in Opioid Prescribing
Overall, from 1996 to 2012, there was a 700% increase in the total MME of opioids
prescribed in the United States, before regression adjustment. The contribution of office based
opioid prescribing to the total amount of MMEs increased from 34 percent of the total in 1996 to
49 percent of the total in 2012. In contrast, the share of total MMEs prescribed originating from
the emergency department declined from 4.8 percent of the total in 1996 to 2.3 percent of the
total in 2012. Figure 1 shows the relative unadjusted increase in milligrams of morphine
equivalent opioids prescribed from 1996 to 2012 in the office-based, inpatient, and emergency
department settings. There is a greater than ten-fold increase in total MMEs prescribed in an
office setting, about a five-fold increase in MMEs prescribed in inpatient settings and a near-
quadrupling of MMEs prescribed in the Emergency Department.
IV.II Regression-Adjusted Trends in Prescribing
After controlling for the covariates described earlier, total, regression-adjusted MMEs
prescribed by the Emergency department more than doubled from about 1.2 billion MMEs to 3.2
billion MMEs while the regression-adjusted MMEs prescribed by office-based prescribers
quadrupled from about 10 to 47 billion MMEs from 1996 to 2012. Total opioid prescribing
increased by about 400 percent. The share attributed to office-based prescribers increased by
about 5 percentage points from 38 to 43 percent of the total. The share attributed to both
inpatient and emergency department physicians decreased over the observation period by 1.1 and

81

1.6 percentage points, respectively. While the magnitude of the increase in total and office-based
prescribing is dampened after regression-adjustment, the vast majority of the remaining increase
is still attributable to office based prescriptions.
Notably, much of that increase in opioid prescribing can be attributed the increase of
prescriptions that contain refills. In Figure 2, each line reflects regression-adjusted estimates of
the total MMEs attributable to office based prescriptions with and without refills as well as
emergency department prescriptions with and without refills. The share of milligrams of
morphine equivalence prescribed by the emergency department for prescriptions with and
without refills has been fairly equal from 1996 to 2012 with a slight decrease in the share
attributable to prescriptions with refills. In office based-settings, on the other hand, the share of
office-based MMEs attributable to prescriptions with refills has increased from about 70 percent
in 1996 to 82 percent in 2012. The growth in billions of MMEs, as shown in Figure 2, is even
more striking.
IV.III Strength and Quantity of Prescriptions by Source of Care
While our findings point to a fairly small contribution to the opioid epidemic on the part
of Emergency Departments in total MME, as MME is composed of both the quantity of the
prescription as well as its strength, it is important to determine whether EDs are dispensing
disproportionately large or strong prescriptions. After adjusting for patient and prescription
characteristics, it appears that EDs prescribe both smaller quantities and weaker opioids than do
office-based prescribers. As shown in Table 2, all else equal, ED prescriptions are about 27%
smaller and 17% less potent than office-based prescriptions (p<0.01). In general, older, female,
and publicly insured patients were associated with larger quantity prescriptions, while minority
patients were associated with shorter, less potent prescriptions than their White counterparts.

82

Diagnoses of injury and headache were associated with lower quantities and potencies of opioids
while other diagnoses were associated with higher quantities and potencies.
IV.IV High Dose and Chronic Prescriptions
Our findings tend to indicate that Emergency Departments are only partially responsible
for the large increase in opioid prescribing in the United States. However, it might be the case
that patients using high-dose opioids or patients receiving particularly long or chronic
prescriptions are more likely to do so in the Emergency Department setting. As shown in Table
3, the odds of receiving a high-dose opioid in the ED are 0.09 that of receiving one in an office-
based setting (p<0.01) and the odds of receiving an opioid with a potent active ingredient are
about 3 percent lower than in an office based setting (p<0.01). Similarly, the odds of receiving a
large or chronic prescription were 0.46 and 0.42 that of an office based visit (p<0.01). As odds
ratios can be difficult to interpret, Table 4 shows the marginal effect of the source of care on the
same outcomes of high dose or chronic prescriptions. As shown, on average, the probability of
receiving an opioid in the ED was about 0.2% as compared to 1.8% in an office based setting, all
else equal. The probability of receiving a chronic prescription—either as measured by doses or
fills—was also much larger in the office based as compared to the ED setting. Interestingly, the
probability of receiving a relatively more potent prescription was highest in the inpatient setting.
IV.V Limitations
Our study has several important limitations. The Medical Expenditure Panel Survey from
which the data are derived relies on a self-report of medical encounters to trigger a pharmacy
review. Subjects wishing to conceal opioid misuse may not disclose some prescriptions resulting
in an underestimate of the total amount of opioids dispensed. We believe this does not
significantly undermine our findings for several reasons. Foremost is that our estimates of the

83

growth in opioid use are consistent with other reports that do not rely on self-report but market
sales data. In addition, there is no reason to suspect that survey respondents would differentially
disclose prescriptions that originated from an office visits compared with one from an ED visit.
Therefore, even if the total magnitude of opioid prescriptions were underestimated, the ED
contribution would likely be unbiased.
Another limitation is that certain patient groups have been traditionally under-represented
in the MEPS--particularly patients with unstable addresses (e.g. homeless) or undocumented
immigration status. If these patients have higher ED use than the MEPS survey respondents in
general, ED estimates may be slightly affected. Additionally, the MEPS only follows patients
for one year prospectively, and therefore, there is insufficient power to link high dose
prescription use with meaningful clinical outcomes such as overdose or death.
Finally, about half of all opioid prescriptions reported in the MEPS are not attributed to a
particular source of care. We include these prescriptions in our analysis when calculating the
total increase in MMEs of opioids in the sample. If we excluded these opioids, the general trends
still hold: office based prescriptions increased as a share of the total while emergency based
MMEs remained constant. To try to determine whether this lack of a source of care might affect
our results, we analyzed the share of patients reporting a site of care who did and did not report
any emergency department visits. We found that patients with at least one ED visit were more
likely to be able to identify a point of care for their prescriptions than those with no ED visits.
While not a perfect solution, it provides some evidence that the propensity to not report a source
of care is not higher among those patients with ED visits.

84

V. Discussion
We observed a massive 400% increase in opioid prescribing for non-cancer pain in the
United States between 1996 and 2012. However, the proportion of prescription opioids
originating from US emergency departments during this study period is both modest and
declining, even after controlling for patient and prescription characteristics. The value and
impact of policies directed at curbing ED prescriptions of opioids should be re-evaluated in light
of these findings.
In the 1990’s, multiple scientific and policy reports emerged documenting poor
assessment and management of acute painful conditions, particularly for ethnic and racial
minorities. US EDs were specifically cited as offenders in the ‘oligoanalgesia’ epidemic.
(Albrecht et al., 2013) (Fosnocht, Swanson, & Barton, 2005; Goldfrank & Knopp, 2000) (Todd
et al., 2007; Todd, Sloan, Chen, Eder, & Wamstad, 2002) (Pines & Perron, 2005) In 2001 the
Joint Commission responded to this concern by labeling pain as the ‘5
th
vital sign’ and required
hospital settings to adopt policies to guarantee pain assessment if not pain medication provision.
(Lanser & Gesell, 2001) More recently, excessive opioid prescriptions have become a major
public health issue. Deaths from prescription opioids have increased from 4,030 in 1999 to
15,000 in 2010 and are now the leading cause of accidental death in the US for persons aged 35-
54. (C. M. Jones et al., 2013) (Paulozzi, 2012) (Paulozzi & Ryan, 2006) Not surprisingly, health
officials are scrambling to enact policies that can curb this epidemic and interestingly EDs have
been a focal point of these efforts. In Washington, the state Medicaid office tied reimbursement
to enacting ED-based opioid prescribing guidelines. In New York, the Mayor announced the
implementation of the City’s ED prescribing guidelines through a massive media release.
(Juurlink et al., 2013; Neven, Sabel, Howell, & Carlisle, 2012) Other states and health plans are

85

following and state-level ACEP charters have generally helped develop and/or endorsed the
guidelines. Various justifications for focusing on ED opioid prescriptions have included that 1)
EDs treat a large number of painful conditions 2) EDs dispense a higher proportion of opioid
prescriptions per visits than other venues of care 3) EDs are the 3
rd
leading prescriber for opioids
among patients between the ages of 10-40 and 4) frequent ED users have disproportionately high
numbers of opioid prescriptions. (Neven et al., 2012; N. D. Volkow, McLellan, Cotto,
Karithanom, & Weiss, 2011)
Taken together, these justifications suggest that EDs are a major source of opioids in the
US. However, our findings suggest the opposite. The proportion of opioids originating from ED
in 2012 was a mere 3%. This, despite over 10% of all outpatient visits occurring in US EDs.
Perhaps more importantly, this proportion is in decline and this decline pre-dates any specific,
large-scale ED focused opioid diversion program. It is true that we observed a doubling in the
absolute amount of opioids prescribed through the study period. However, given that the
beginning of the study period encompassed a widely publicized era of oligoanalgesia, this
increase may be reasonable and even desirable.
Though the overall contribution to the opioid epidemic by EDs is small, it is possible that
EDs disproportionately supply opioids to high-risk individuals and therefore merit special
attention. Though limited, our findings do not support this contention. A recent report examining
deaths from opioids in Tennessee identified high dose prescriptions (>100 MME daily) as being
the single greatest independent predictor of subsequent opioid death (adjusted Odds Ratio =
11.2). (Baumblatt et al., 2014) We found that a mere 0.2% of ED prescriptions fell into that
category while 2.3% of office-based prescriptions were for more than 100 MME daily. Similarly
we note that ED-derived prescriptions are for markedly lower quantities of pills and lower

86

potency opioids. Since EDs are not a major source of the total amount of opioids or high-risk
opioid prescriptions (≥100 MME daily), ED guidelines are unlikely to prevent opioid deaths.
The question that emerges is ‘could there be harm in dramatically restricting prescriptions
from EDs’? We offer a theoretical reason to be suspicious. First, patients addicted to opioids are
unlikely to be deterred from procuring opioids because an ED refuses to prescribe them. Rather,
they may find unscrupulous medical providers or worse yet turn to the illicit drug markets.
(Betses & Brennan, 2013; Werb et al., 2011) Illicit opioid users are much more likely to
experience overdose, violence, incarceration, unemployment and other destructive social
consequences. (Darke & Farrell, 2014; Grund, Latypov, & Harris, 2013; Khan, Khazaal,
Thorens, Zullino, & Uchtenhagen, 2014; Lipton & Johnson, 1998; Nurco, Ball, Shaffer, &
Hanlon, 1985; Werb et al., 2011) Moreover, the emphasis on restricting opioid use in an
environment that has been generally found to undertreat pain is likely to have a profound impact
on patients suffering from acute painful conditions while having minimal if any impact on opioid
misuse. Indeed, regulators from the Centers for Medicaid and Medicare Services recently
expressed concern that posting restrictive pain medication guidelines in ED waiting rooms or
treatment areas might violate hospitals’ Emergency Medical Treatment and Active Labor Act
(EMTALA) obligations and recommended removal of all public signage on the matter. This
highlights that policies and advertisements emphasizing denial of desired medications may
adversely affect the therapeutic relationship between providers and patients.
In lieu of restrictive opioid practices in the ED, we believe that emphasis should be on
developing and disseminating tools to help providers identify high-risk individuals and refer
them to proven treatment programs.

87

VI. Figures and Tables
Figure 1. Increase in Total Opioids Prescribed Relative to 1996 Levels by Source of Care

Note: totals exclude prescriptions for patients with cancer diagnoses

0
2
4
6
8
10
12
1996 1998 2000 2002 2004 2006 2008 2010 2012
Growth in Prescribing of total MMEs Relative to 1996
Emergency Department Inpatient Office Based

88

Figure 2. Total Adjusted Milligrams of Morphine Equivalence by Source of Care and Type
of Prescription, 1996 – 2012

Note: totals exclude prescriptions for patients with cancer diagnoses

0
5
10
15
20
25
30
35
40
45
1996 1998 2000 2002 2004 2006 2008 2010 2012
Billions of Milligrams of Morphine Equivalents
Office Based, Single Rx Emergency Department, Single Rx
Office Based, Refill Rx Emergency Department, Refill Rx

89

Table 1. Characteristics of the Study Population, 1996 - 2012

Individual
Observations 47,551
Average Age 48.7 (17.2)
Average Years of Education 12.8 (2.5)
Male 37%
Married 53%
Urban 78%
Region

Northeast 13%

Midwest 23%

South 41%

West 23%
Race or Ethnicity

White 63%

Black 17%

Hispanic 15%

Asian 2%
Insurance Status

Private 61%

Public 29%

Uninsured 10%
Diagnostic History

Joint Pain 17%

Injury 16%

Back Pain 12%

Headache 3%

Cancer Pain 2%

Mental Health 0%

Notes: Values in parentheses are standard deviations.

90

Table 2. Log-Linear Regression Results
Log(Total MME) Log(Quantity) Log(Strength)
Coefficient
(S.E.)
Coefficient
(S.E.)
Coefficient
(S.E.)
Age 0.029***
(0.0016)
0.015***
(0.001)
0.014***
(0.001)
Male -0.025**
(0.010)
-0.035***
(0.008)
0.010
(0.007)
Race/Ethnicity
Reference = White
Black -0.214***
(0.013)
-0.063***
(0.010)
-0.151***
(0.009)
Hispanic -0.233***
(0.015)
-0.081***
(0.012)
-0.152***
(0.011)
Asian -0.259***
(0.036)
-0.081***
(0.013)
-0.179***
(0.022)
Insurance
Reference = Private
Public 0.082***
(0.013)
0.050***
(0.010)
0.032***
(0.010)
Uninsured -0.006
(0.018)
0.029**
(0.014)
-0.036***
(0.012)
Source of Care
Reference = Office Based
Inpatient -0.098***
(0.017)
-0.051***
(0.013)
-0.047***
(0.013)
ED -0.442***
(0.016)
-0.274***
(0.012)
-0.168***
(0.010)
Diagnoses
Joint Pain 0.155***
(0.013)
0.149***
(0.010)
0.005
(0.010)
Cancer Pain 0.175***
(0.033)
0.063**
(0.028)
0.112***
(0.028)
Injury -0.089***
(0.013)
-0.051***
(0.010)
-0.039***
(0.009)
Back Pain 0.253***
(0.014)
0.115***
(0.011)
0.137***
(0.011)
Headache -0.182***
(0.027)
-0.124***
(0.024)
-0.059***
(0.019)
Mental Health 0.324***
(0.063)
0.155***
(0.048)
0.169***
(0.048)
Observations 82,613 82,628 82,613
R-squared 0.43 0.44 0.07
Notes: Regressions also control for age
2
, year dummies, poverty status, pharmacy source, region, marital status,
education, urban status, and whether the prescription carries refills.

91

Table 3. Logit Regression Results
Prob
(High Dose)
Prob
(High Potency)
Prob
(>90 Doses)
Prob
(6 Refills)
Odds Ratio
(S.E.)
Odds Ratio
(S.E.)
Odds Ratio
(S.E.)
Odds Ratio
(S.E.)
Age 1.086***
(0.017)
1.038***
(0.004)
1.0381
(0.005)
1.014
(0.009)
Male 0.963
(0.064)
0.807***
(0.018)
0.979
(0.025)
0.921*
(0.039)
Race/Ethnicity
Reference = White
Black 0.467***
(0.050)
0.732***
(0.023)
0.760***
(0.028)
1.036
(0.057)
Hispanic 0.627***
(0.086)
0.582***
(0.022)
0.744***
(0.031)
0.836**
(0.061)
Asian 0.193***
(0.078)
0.695***
(0.058)
0.752***
(0.078)
0.743
(0.171)
Insurance
Reference = Private
Public 1.073
(0.089)
1.130***
(0.032)
1.167***
(0.036)
1.184***
(0.061)
Uninsured 0.767
(0.122)
0.866***
(0.037)
1.061
(0.051)
1.341***
(0.106)
Source of Care
Reference = Office Based
Inpatient 0.450***
(0.092)
2.359***
(0.088)
0.609***
(0.035)
0.548***
(0.065)
ED 0.088***
(0.030)
0.974***
(0.040)
0.467***
(0.028)
0.419***
(0.055)
Diagnoses
Joint Pain 0.894
(0.070)
0.780***
(0.022)
1.371***
(0.037)
1.010
(0.045)
Cancer Pain 1.821***
(0.296)
1.342***
(0.092)
1.201**
(0.094)
1.208
(0.153)
Injury 0.752**
(0.095)
0.827***
(0.027)
0.754***
(0.031)
0.849**
(0.061)
Back Pain 1.602***
(0.121)
1.094***
(0.031)
1.364***
(0.040)
1.068
(0.051)
Headache 0.474***
(0.117)
0.792***
(0.053)
0.898
(0.067)
1.098
(0.111)
Mental Health 1.300
(0.289)
1.608***
(0.188)
1.400***
(0.160)
1.516**
(0.261)
Observations 82,613 82,628 82,628 31,518
+

Pseudo R
2
0.13 0.06 0.12 0.02
Notes: Regressions also control for age
2
, year dummies, poverty status, pharmacy source, region, marital status,
education, urban status, and whether the prescription carries refills.
+
There are far fewer observations as we only consider prescriptions with refills.

92

Table 4. Marginal Effect of Source of Care on High Dose and Chronic Opioid Prescriptions

1996 - 2000 2001 - 2004 2005 - 2008 2009 - 2012 Mean
High Dose Prescriptions (>100 MME per dose)
All 0.4% 1.7% 2.2% 2.2% 1.8%
Office Based 0.5% 2.2% 2.9% 2.8% 2.3%
Inpatient 0.2% 1.0% 1.3% 1.3% 1.0%
ED 0.0% 0.2% 0.3% 0.3% 0.2%
Prescription with Active Ingredients More Potent than Morphine
All 16.4% 22.0% 23.6% 27.9% 23.1%
Office Based 16.0% 21.6% 23.3% 27.7% 22.8%
Inpatient 30.3% 38.6% 40.8% 46.5% 39.9%
ED 15.6% 21.2% 22.8% 27.1% 22.3%
Chronic Prescriptions (>90 Doses)
All 11.7% 11.8% 16.8% 23.8% 17.1%
Office Based 12.5% 12.6% 18.1% 25.6% 18.3%
Inpatient 8.2% 8.3% 12.1% 17.8% 12.4%
ED 6.4% 6.5% 9.6% 14.3% 9.8%
Chronic Prescriptions (>6 Fills Per Year)
All 12.3% 13.9% 14.4% 12.4% 13.3%
Office Based 14.0% 15.8% 16.4% 14.2% 15.1%
Inpatient 8.2% 9.3% 9.7% 8.3% 8.9%
ED 6.4% 7.4% 7.7% 6.5% 7.0%

Notes: Active ingredients stronger than morphine include butorphanol, fentanyl, fentanyl citrate,
hydormorphone, levorphanol, oxymorphone, and oxycodone.

93

REFERENCES
Albrecht, E., Taffe, P., Yersin, B., Schoettker, P., Decosterd, I., & Hugli, O. (2013).
Undertreatment of acute pain (oligoanalgesia) and medical practice variation in
prehospital analgesia of adult trauma patients: a 10 yr retrospective study. Br J Anaesth,
110(1), 96-106. doi: 10.1093/bja/aes355
Anderson, Z. (2015, March 18, 2015). Pain clinic rules cut overdose deaths. Herald-Tribue.
Baehren, D. F., Marco, C. A., Droz, D. E., Sinha, S., Callan, E. M., & Akpunonu, P. (2010). A
statewide prescription monitoring program affects emergency department prescribing
behaviors. Ann Emerg Med, 56(1), 19-23 e11-13. doi:
10.1016/j.annemergmed.2009.12.011
Baumblatt, J. A., Wiedeman, C., Dunn, J. R., Schaffner, W., Paulozzi, L. J., & Jones, T. F.
(2014). High-Risk Use by Patients Prescribed Opioids for Pain and Its Role in Overdose
Deaths. JAMA Intern Med. doi: 10.1001/jamainternmed.2013.12711
Becker, G. S., Grossman, M., & Murphy, K. M. (1991). Rational Addiction and the Effect of
Price on Consumption. The American Economic Review, 81(2), 237-241. doi:
10.2307/2006861
Becker, G. S., Grossman, M., & Murphy, K. M. (1994). An Empirical Analysis of Cigarette
Addiction. The American Economic Review, 84(3), 396-418. doi: 10.2307/2118059
Bernheim, D., & Rangel, A. (2005). From neuroscience to public policy: a new economic view
of addiction. Swedish Economic Policy Review, 12(2), 99.
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How Much Should We Trust Differences-In-
Differences Estimates? The Quarterly Journal of Economics, 119(1), 249-275. doi:
10.1162/003355304772839588
Besley, T., & Burgess, R. (2004). Can Labor Regulation Hinder Economic Performance?
Evidence from India. The Quarterly Journal of Economics, 119(1), 91-134. doi:
10.1162/003355304772839533
Besley, T., & Case, A. (2000). Unnatural Experiments? Estimating the Incidence of Endogenous
Policies. The Economic Journal, 110(467), F672-F694. doi: 10.2307/2667771
Betses, M., & Brennan, T. (2013). Abusive prescribing of controlled substances--a pharmacy
view. N Engl J Med, 369(11), 989-991. doi: 10.1056/NEJMp1308222
Bickel, W. K., Jarmolowicz, D. P., Mueller, E. T., & Gatchalian, K. M. (2011). The behavioral
economics and neuroeconomics of reinforcer pathologies: implications for etiology and
treatment of addiction. Current psychiatry reports, 13(5), 406-415.

94

Birnbaum, H. G., White, A. G., Schiller, M., Waldman, T., Cleveland, J. M., & Roland, C. L.
(2011). Societal costs of prescription opioid abuse, dependence, and misuse in the United
States. Pain Med, 12(4), 657-667.
Bohnert, A. S., Valenstein, M., Bair, M. J., Ganoczy, D., McCarthy, J. F., Ilgen, M. A., & Blow,
F. C. (2011). Association between opioid prescribing patterns and opioid overdose-
related deaths. JAMA: The Journal of the American Medical Association, 305(13), 1315-
1321.
Braden, J. B., Russo, J., Fan, M. Y., Edlund, M. J., Martin, B. C., DeVries, A., & Sullivan, M. D.
(2010). Emergency department visits among recipients of chronic opioid therapy. Arch
Intern Med, 170(16), 1425-1432. doi: 10.1001/archinternmed.2010.273
Brocas, I., & Carrillo, J. D. (2008). The brain as a hierarchical organization. The American
Economic Review, 1312-1346.
Buurma, H., Bouvy, M. L., De Smet, P. A. G. M., Floor-Schreudering, A., Leufkens, H. G. M.,
& Egberts, A. C. G. (2008). Prevalence and determinants of pharmacy shopping
behaviour. Journal of Clinical Pharmacy and Therapeutics, 33(1), 17-23. doi:
10.1111/j.1365-2710.2008.00878.x
Cai, R., Crane, E., Poneleit, K., & Paulozzi, L. (2010). Emergency department visits involving
nonmedical use of selected prescription drugs in the United States, 2004-2008. J Pain
Palliat Care Pharmacother, 24(3), 293-297. doi: 10.3109/15360288.2010.503730
Cantrill, S. V., Brown, M. D., Carlisle, R. J., Delaney, K. A., Hays, D. P., Nelson, L. S., . . .
American College of Emergency Physicians Opioid Guideline Writing, P. (2012).
Clinical policy: critical issues in the prescribing of opioids for adult patients in the
emergency department. Ann Emerg Med, 60(4), 499-525. doi:
10.1016/j.annemergmed.2012.06.013
Cepeda, M. S., Fife, D., Chow, W., Mastrogiovanni, G., & Henderson, S. C. (2013). Opioid
Shopping Behavior: How Often, How Soon, Which Drugs, and What Payment Method.
The Journal of Clinical Pharmacology, 53(1), 112-117. doi: 10.1177/0091270012436561
Cerda, M., Ransome, Y., Keyes, K. M., Koenen, K. C., Tracy, M., Tardiff, K. J., . . . Galea, S.
(2013). Prescription opioid mortality trends in New York City, 1990-2006: examining the
emergence of an epidemic. Drug Alcohol Depend, 132(1-2), 53-62. doi:
10.1016/j.drugalcdep.2012.12.027
Chen, J. H., Humphreys, K., Shah, N. H., & Lembke, A. (2016). DIstribution of opioids by
different types of medicare prescribers. JAMA internal medicine, 176(2), 259-261. doi:
10.1001/jamainternmed.2015.6662
Cicero, T. J., Surratt, H., Inciardi, J. A., & Munoz, A. (2007). Relationship between therapeutic
use and abuse of opioid analgesics in rural, suburban, and urban locations in the United
States. Pharmacoepidemiol Drug Saf, 16(8), 827-840. doi: 10.1002/pds.1452

95

Control, C. f. D., & Prevention. (2013). Vital Signs: Overdoses of Prescription Opioid Pain
Relievers and Other Drugs Among Women-United States, 1999-2010. MMWR. Morbidity
and mortality weekly report, 62(26), 537.
Darke, S., & Farrell, M. (2014). Would legalizing illicit opioids reduce overdose fatalities?
Implications from a natural experiment. Addiction. doi: 10.1111/add.12456
Dart, R. C., Surratt, H. L., Cicero, T. J., Parrino, M. W., Severtson, S. G., Bucher-Bartelson, B.,
& Green, J. L. (2015). Trends in Opioid Analgesic Abuse and Mortality in the United
States. New England Journal of Medicine, 372(3), 241-248. doi:
doi:10.1056/NEJMsa1406143
Deshpande, A., Furlan, A. D., Mailis ‐Gagnon, A., Atlas, S., & Turk, D. (2007). Opioids for
chronic low ‐back pain. The Cochrane Library.
Duan, N. (1983). Smearing estimate: a nonparametric retransformation method. Journal of the
American Statistical Association, 78(383), 605-610.
Dunn, K. M., Saunders, K. W., Rutter, C. M., Banta-Green, C. J., Merrill, J. O., Sullivan, M. D.,
. . . Von Korff, M. (2010). Opioid prescriptions for chronic pain and overdose: a cohort
study. Ann Intern Med, 152(2), 85-92. doi: 10.7326/0003-4819-152-2-201001190-00006
Edlund, M. J., Martin, B. C., Fan, M. Y., Devries, A., Braden, J. B., & Sullivan, M. D. (2010).
Risks for opioid abuse and dependence among recipients of chronic opioid therapy:
results from the TROUP study. Drug Alcohol Depend, 112(1-2), 90-98. doi:
10.1016/j.drugalcdep.2010.05.017
Edlund, M. J., Steffick, D., Hudson, T., Harris, K. M., & Sullivan, M. (2007). Risk factors for
clinically recognized opioid abuse and dependence among veterans using opioids for
chronic non-cancer pain. Pain, 129(3), 355-362. doi: 10.1016/j.pain.2007.02.014
Fishman, S. M., Papazian, J. S., Gonzalez, S., Riches, P. S., & Gilson, A. (2004). Regulating
Opioid Prescribing Through Prescription Monitoring Programs: Balancing Drug
Diversion and Treatment of Pain. Pain Med, 5(3), 309-324. doi: 10.1111/j.1526-
4637.2004.04049.x
Fosnocht, D. E., Swanson, E. R., & Barton, E. D. (2005). Changing attitudes about pain and pain
control in emergency medicine. Emerg Med Clin North Am, 23(2), 297-306. doi:
10.1016/j.emc.2004.12.003
Frakt, A. B., & Bagley, N. (2015). Protection or Harm? Suppressing Substance-Use Data. New
England Journal of Medicine, 372(20), 1879-1881. doi: doi:10.1056/NEJMp1501362
Goldfrank, L. R., & Knopp, R. K. (2000). Racially and ethnically selective oligoanalgesia: is this
racism? Ann Emerg Med, 35(1), 79-82.

96

Gomes, T., Mamdani, M. M., Dhalla, I. A., Paterson, J. M., & Juurlink, D. N. (2011). Opioid
dose and drug-related mortality in patients with nonmalignant pain. Arch Intern Med,
171(7), 686.
Grady, D., Berkowitz, S. A., & Katz, M. H. (2011). OPioids for chronic pain. Arch Intern Med,
171(16), 1426-1427. doi: 10.1001/archinternmed.2011.213
Green, T. C., Grau, L. E., Carver, H. W., Kinzly, M., & Heimer, R. (2011). Epidemiologic trends
and geographic patterns of fatal opioid intoxications in Connecticut, USA: 1997-2007.
Drug Alcohol Depend, 115(3), 221-228. doi: 10.1016/j.drugalcdep.2010.11.007
Grossman, M., & Chaloupka, F. J. (1998). The demand for cocaine by young adults: a rational
addiction approach. Journal of Health Economics, 17(4), 427-474.
Grund, J. P., Latypov, A., & Harris, M. (2013). Breaking worse: the emergence of krokodil and
excessive injuries among people who inject drugs in Eurasia. Int J Drug Policy, 24(4),
265-274. doi: 10.1016/j.drugpo.2013.04.007
Gugelmann, H. M., & Perrone, J. (2011). Can prescription drug monitoring programs help limit
opioid abuse? JAMA, 306(20), 2258-2259. doi: 10.1001/jama.2011.1712
Haffajee, R. L., Jena, A. B., & Weiner, S. G. (2015). MAndatory use of prescription drug
monitoring programs. JAMA, 313(9), 891-892. doi: 10.1001/jama.2014.18514
Ives, T. J., Chelminski, P. R., Hammett-Stabler, C. A., Malone, R. M., Perhac, J. S., Potisek, N.
M., . . . Pignone, M. P. (2006). Predictors of opioid misuse in patients with chronic pain:
a prospective cohort study. BMC Health Serv Res, 6, 46. doi: 10.1186/1472-6963-6-46
Jayawant, S. S., & Balkrishnan, R. (2005). The controversy surrounding OxyContin abuse:
issues and solutions. Therapeutics and Clinical Risk Management, 1(2), 77-82.
Jena, A. B., Goldman, D., Weaver, L., & Karaca-Mandic, P. (2014). Opioid prescribing by
multiple providers in Medicare: retrospective observational study of insurance claims.
BMJ: British Medical Journal, 348.
Jena, A. B., & Goldman, D. P. (2011). Growing Internet use may help explain the rise in
prescription drug abuse in the United States. Health affairs, 30(6), 1192-1199.
Jones, C. M., Mack, K. A., & Paulozzi, L. J. (2013). Pharmaceutical overdose deaths, United
States, 2010. JAMA, 309(7), 657-659. doi: 10.1001/jama.2013.272
Jones, C. M., Paulozzi, L. J., & Mack, K. A. (2014). Sources of Prescription Opioid Pain
Relievers by Frequency of Past-Year Nonmedical Use: United States, 2008-2011. JAMA
internal medicine, 174(5), 802-803.
Juurlink, D. N., Dhalla, I. A., & Nelson, L. S. (2013). Improving opioid prescribing: the New
York City recommendations. JAMA, 309(9), 879-880. doi: 10.1001/jama.2013.1139

97

Khan, R., Khazaal, Y., Thorens, G., Zullino, D., & Uchtenhagen, A. (2014). Understanding
Swiss Drug Policy Change and the Introduction of Heroin Maintenance Treatment. Eur
Addict Res, 20(4), 200-207. doi: 10.1159/000357234
Laibson, D. (2001). A cue-theory of consumption. The Quarterly Journal of Economics, 116(1),
81-119.
Lanser, P., & Gesell, S. (2001). Pain management: the fifth vital sign. Healthc Benchmarks, 8(6),
68-70, 62.
Larochelle, M. R., Zhang, F., Ross-Degnan, D., & Wharam, J. (2015). RAtes of opioid
dispensing and overdose after introduction of abuse-deterrent extended-release
oxycodone and withdrawal of propoxyphene. JAMA Internal Medicine, 175(6), 978-987.
doi: 10.1001/jamainternmed.2015.0914
Lipton, D. S., & Johnson, B. D. (1998). Smack, crack, and score: two decades of NIDA-funded
drugs and crime research at NDRI 1974-1994. Subst Use Misuse, 33(9), 1779-1815.
Loewenstein, G. (2000). Emotions in economic theory and economic behavior. American
Economic Review, 426-432.
Loewenstein, G., & O’Donoghue, T. (2004). Animal spirits: Affective and deliberative processes
in economic behavior: mimeo.
Manchikanti, L., Giordano, J., Boswell, M. V., Fellows, B., Manchukonda, R., & Pampati, V.
(2007). Psychological factors as predictors of opioid abuse and illicit drug use in chronic
pain patients. J Opioid Manag, 3(2), 89-100.
Martell, B. A., O'Connor, P. G., Kerns, R. D., Becker, W. C., Morales, K. H., Kosten, T. R., &
Fiellin, D. A. (2007). Systematic review: opioid treatment for chronic back pain:
prevalence, efficacy, and association with addiction. Ann Intern Med, 146(2), 116-127.
McCabe, S. E., Teter, C. J., Boyd, C. J., Knight, J. R., & Wechsler, H. (2005). Nonmedical use
of prescription opioids among US college students: Prevalence and correlates from a
national survey. Addictive behaviors, 30(4), 789-805.
McDonald, D. C., & Carlson, K. E. (2013). Estimating the prevalence of opioid diversion by
“doctor shoppers” in the United States. PLoS One, 8(7), e69241.
Morden, N. E., Munson, J. C., Colla, C. H., Skinner, J. S., Bynum, J. P., Zhou, W., & Meara, E.
R. (2014). Prescription Opioid Use among Disabled Medicare Beneficiaries: Intensity,
Trends and Regional Variation. Medical care, 52(9), 852.
Neven, D. E., Sabel, J. C., Howell, D. N., & Carlisle, R. J. (2012). The development of the
Washington State emergency department opioid prescribing guidelines. J Med Toxicol,
8(4), 353-359. doi: 10.1007/s13181-012-0267-6

98

Nurco, D. N., Ball, J. C., Shaffer, J. W., & Hanlon, T. E. (1985). The criminality of narcotic
addicts. J Nerv Ment Dis, 173(2), 94-102.
Office, G. A. (2011). Medicare Part D: Instances of Questionable Access to Prescription Drugs.
Report to Congressional Requesters, 1-27.
Office of the Attorney General, S. o. F. (2014). Statewide Task Force on Prescription Drug
Abuse & Newborns: 2014 Progress Report: Office of the Attorney General.
Pacula, R. L., Powell, D., & Taylor, E. (2015). Does Prescription Drug Coverage Increase
Opioid Abuse? Evidence from Medicare Part D: National Bureau of Economic Research.
Paulozzi, L. J. (2012). Prescription drug overdoses: a review. J Safety Res, 43(4), 283-289. doi:
10.1016/j.jsr.2012.08.009
Paulozzi, L. J., Kilbourne, E. M., & Desai, H. A. (2011). Prescription drug monitoring programs
and death rates from drug overdose. Pain Med, 12(5), 747-754. doi: 10.1111/j.1526-
4637.2011.01062.x
Paulozzi, L. J., & Ryan, G. W. (2006). Opioid analgesics and rates of fatal drug poisoning in the
United States. Am J Prev Med, 31(6), 506-511. doi: 10.1016/j.amepre.2006.08.017
Paulozzi, L. J., & Stier, D. D. (2010). Prescription drug laws, drug overdoses, and drug sales in
New York and Pennsylvania. J Public Health Policy, 31(4), 422-432. doi:
10.1057/jphp.2010.27
Paulozzi, L. J., Weisler, R. H., & Patkar, A. A. (2011). A national epidemic of unintentional
prescription opioid overdose deaths: how physicians can help control it. J Clin
Psychiatry, 72(5), 589-592. doi: 10.4088/JCP.10com06560
Peirce, G. L., Smith, M. J., Abate, M. A., & Halverson, J. (2012). Doctor and pharmacy
shopping for controlled substances. Medical care, 50(6), 494-500.
Pines, J. M., & Perron, A. D. (2005). Oligoanalgesia in ED patients with isolated extremity
injury without documented fracture. Am J Emerg Med, 23(4), 580.
Pletcher, M. J., Kertesz, S. G., Kohn, M. A., & Gonzales, R. (2008). Trends in opioid prescribing
by race/ethnicity for patients seeking care in US emergency departments. JAMA, 299(1),
70-78. doi: 10.1001/jama.2007.64
Policy, O. o. N. D. C., & President, U. E. O. o. t. (2011). Epidemic: Responding to America's
Prescription Drug Abuse Crisis.
Pradel, V., Delga, C., Rouby, F., Micallef, J., & Lapeyre-Mestre, M. (2010). Assessment of
abuse potential of benzodiazepines from a prescription database using 'doctor shopping'
as an indicator. CNS Drugs, 24(7), 611-620. doi: 10.2165/11531570-000000000-00000

99

Prevention, C. f. D. C. a. (July 11, 2012). Prescription Drug Overdose: State Laws. Retrieved
December 15, 2014, from
www.cdc.gov/HomeandRecreationalSafety/Poisoning/laws/index.html
Prevention, C. f. D. C. a. (2013). Vital Signs: Overdoses of Prescription Opioid Pain Relievers
and Other Drugs Among Women-United States, 1999-2010. (1545-861X).
Reisman, R. M., Shenoy, P. J., Atherly, A. J., & Flowers, C. R. (2009). Prescription opioid usage
and abuse relationships: An evaluation of state prescription drug monitoring program
efficacy. Substance Abuse: Research and Treatment, 2009(3), 41-51.
Robert Dufour, J. M., Margaret K. Pasquale, David Schaaf, George A. Andrews, Nick C. Patel.
(2014). Understanding Predictors of Opioid Abuse: Predictive Model Development and
Validation. Am J Pharm Benefits, 6(5), 208-216.
Robst, J., Levy, J. M., & Ingber, M. J. (2007). Diagnosis-based risk adjustment for medicare
prescription drug plan payments. Health Care Financing Review, 28(4).
Saunders, K. W., Dunn, K. M., Merrill, J. O., Sullivan, M., Weisner, C., Braden, J. B., . . . Von
Korff, M. (2010). Relationship of opioid use and dosage levels to fractures in older
chronic pain patients. Journal of General Internal Medicine, 25(4), 310-315.
Schloff, L. A., Rector, T. S., Seifeldin, R., & Haddox, J. D. (2004). Identifying controlled
substance patterns of utilization requiring evaluation using administrative claims data.
American Journal of Managed Care, 10, 783-790.
Sehgal, N., Colson, J., & Smith, H. S. (2013). Chronic pain treatment with opioid analgesics:
benefits versus harms of long-term therapy.
Simeone, R., & Holland, L. (2006). An evaluation of prescription drug monitoring programs.
Simeone Associates, Inc. Albany, NY. Retrieved on September, 11, 2012.
Solomon, D. H., Rassen, J. A., Glynn, R. J., Garneau, K., Levin, R., Lee, J., & Schneeweiss, S.
(2010). The comparative safety of opioids for nonmalignant pain in older adults. Arch
Intern Med, 170(22), 1979-1986.
Strassels, S. A. (2009). Economic burden of prescription opioid misuse and abuse. Journal of
managed care pharmacy: JMCP, 15(7), 556-562.
Sung, H.-E., Richter, L., Vaughan, R., Johnson, P. B., & Thom, B. (2005). Nonmedical use of
prescription opioids among teenagers in the United States: Trends and correlates. Journal
of Adolescent Health, 37(1), 44-51.
Surratt, H. L., O'Grady, C., Kurtz, S. P., Stivers, Y., Cicero, T. J., Dart, R. C., & Chen, M.
(2014). Reductions in prescription opioid diversion following recent legislative
interventions in Florida. Pharmacoepidemiol Drug Saf, 23(3), 314-320. doi:
10.1002/pds.3553

100

Tamayo-Sarver, J. H., Dawson, N. V., Cydulka, R. K., Wigton, R. S., & Baker, D. W. (2004).
Variability in emergency physician decisionmaking about prescribing opioid analgesics.
Ann Emerg Med, 43(4), 483-493. doi:
http://dx.doi.org/10.1016/j.annemergmed.2003.10.043
Tamayo-Sarver, J. H., Dawson, N. V., Hinze, S. W., Cydulka, R. K., Wigton, R. S., Albert, J.
M., . . . Baker, D. W. (2003). The effect of race/ethnicity and desirable social
characteristics on physicians' decisions to prescribe opioid analgesics. Acad Emerg Med,
10(11), 1239-1248.
Tamayo-Sarver, J. H., Hinze, S. W., Cydulka, R. K., & Baker, D. W. (2003). Racial and ethnic
disparities in emergency department analgesic prescription. American Journal of Public
Health, 93(12), 2067-2073.
Todd, K. H., Ducharme, J., Choiniere, M., Crandall, C. S., Fosnocht, D. E., Homel, P., . . .
Group, P. S. (2007). Pain in the emergency department: results of the pain and emergency
medicine initiative (PEMI) multicenter study. J Pain, 8(6), 460-466. doi:
10.1016/j.jpain.2006.12.005
Todd, K. H., Sloan, E. P., Chen, C., Eder, S., & Wamstad, K. (2002). Survey of pain etiology,
management practices and patient satisfaction in two urban emergency departments.
CJEM, 4(4), 252-256.
Tudor, C. G. (2012). Supplemental Guidance Related to Improving Drug Utilization Review
Controls in Part D [Press release]
Turk, D. C., Brody, M. C., & Okifuji, E. A. (1994). Physicians' attitudes and practices regarding
the long-term prescribing of opioids for non-cancer pain. Pain, 59(2), 201-208. doi:
http://dx.doi.org/10.1016/0304-3959(94)90072-8
Turk, D. C., & Okifuji, A. (1997). What Factors Affect Physicians' Decisions to Prescribe
Opioids for Chronic Noncancer Pain Patients? Clin J Pain, 13(4), 330-336.
Volkow, N. D., Frieden, T. R., Hyde, P. S., & Cha, S. S. (2014). Medication-Assisted Therapies
— Tackling the Opioid-Overdose Epidemic. New England Journal of Medicine, 370(22),
2063-2066. doi: doi:10.1056/NEJMp1402780
Volkow, N. D., McLellan, T. A., Cotto, J. H., Karithanom, M., & Weiss, S. R. (2011).
Characteristics of opioid prescriptions in 2009. JAMA, 305(13), 1299-1301. doi:
10.1001/jama.2011.401
Webster, L. R., & Webster, R. M. (2005). Predicting Aberrant Behaviors in Opioid ‐Treated
Patients: Preliminary Validation of the Opioid Risk Tool. Pain Med, 6(6), 432-442.
Werb, D., Rowell, G., Guyatt, G., Kerr, T., Montaner, J., & Wood, E. (2011). Effect of drug law
enforcement on drug market violence: a systematic review. Int J Drug Policy, 22(2), 87-
94. doi: 10.1016/j.drugpo.2011.02.002

101

White, A. G., Birnbaum, H. G., Schiller, M., Tang, J., & Katz, N. P. (2009). Analytic models to
identify patients at risk for prescription opioid abuse. Am J Manag Care, 15(12), 897-
906.
Wilsey, B. L., Fishman, S. M., Gilson, A. M., Casamalhuapa, C., Baxi, H., Lin, T. C., & Li, C. S.
(2011). An analysis of the number of multiple prescribers for opioids utilizing data from
the California Prescription Monitoring Program. Pharmacoepidemiol Drug Saf, 20(12),
1262-1268. doi: 10.1002/pds.2129
Wilsey, B. L., Fishman, S. M., Tsodikov, A., Ogden, C., Symreng, I., & Ernst, A. (2008).
Psychological Comorbidities Predicting Prescription Opioid Abuse among Patients in
Chronic Pain Presenting to the Emergency Department. Pain Med, 9(8), 1107-1117. doi:
10.1111/j.1526-4637.2007.00401.x

102

APPENDIX TO CHAPTER 2
Appendix Table 1. ICD-9 Codes for Diagnosis Definitions
Diagnosis Name ICD-9 Codes
Opioid Abuse 304.0x, 305.5x, 304.7x
Other Drug Abuse 292.xx, 304.1x - 304.6x, 304.8x – 304.9x, 305.1x – 305.4x, 305.6x –
305.9x
Alcohol Abuse 305.0x, 303.xx, 291.xx
Opioid Related
Overdose
965.02, 965.09, 965.80, 965.90
Headache 346.xx, 784.0, 307.81, 339.xx
Hepatitis C 070.41, 070.44, 070.51, 070.54, 070.70, 070.71
Chronic Pain 338.2x, 338.0, 338.4
Myalgia 729.1x – 729.9x
Cancer 140.xx – 209.xx
Arthritis 714.xx, 711.xx, 730.xx, 715.xx
Fracture 800.xx – 829.xx
Sprain or Strain 830.xx – 848.xx
Mental Health Diagnosis 295.xx – 301.xx
Back Pain 847.2x, 847.3x, 847.4x, 847.9x, 839.30, 839.20, 805.90, 805.70,
895.69, 805.50, 805.40, 756.12, 756.11, 739.40, 739.30,724.03,
724.02, 722.93, 722.83, 722.73, 722.52, 722.32 722.10, 355.50,
353.40, 353.10,738.50, 738.40, 737.42, 737.40, 733.13, 720.xx,
721.3x-721.91, 724.2x-724.9x, 737.2x, 806.4x – 806.9x, 839.4x
839.5x, 952.2x – 952.9x, 953.2x – 953.3x, 953.5x – 953.9x, 846.xx

103

Appendix Table 2. Morphine Equivalence by Active Ingredient
Active Ingredient Morphine
Equivalence
Conversion
Propoxyphene 0.23
Codeine 0.15
Hydrocodone 1
Butalbital & Codeine 0.15
Dihydrocodeine 0.25
Pentazocine 0.37
Morphine Sulfate 1
Codeine Sulfate 0.15
Oxycodone 1.5
Hydromorphone 4
Meperidine HCL 0.1
Fentanyl Citrate (transmucosal) 125
Fentanyl (transdermal) 100
Oxymorphone 3.0
Levorphanol 11.0
Methadone 3.0
Nalbuphine 1
Tapentadol 0.4
Butorphanol 7
Buprenorphine 10
Sufentanil* 125

Source: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3286630/
Notes: *There is no established morphine equivalence for Sufentanil, so I assume equivalence with Fentanyl, its
parent drug. This is a conservative estimate as it is said to be between 5 and 10 times as potent as Fentanyl.

104

Appendix Table 3. Regression Results, Opioid Utilization Measures

Opioid Use Log(Fills) Log(Days) Log(MME) High Dose

Coeff SE Coeff SE Coeff SE Coeff SE Coeff SE
Age -0.006 (0.000) 0.002 (0.000) 0.008 (0.000) -0.002 (0.000) -0.014 (0.000)
Female 0.113 (0.001) 0.011 (0.001) 0.036 (0.001) -0.012 (0.001) 0.026 (0.002)
Race/Ethnicity: Reference White
Black -0.122 (0.002) -0.140 (0.001) -0.216 (0.002) -0.307 (0.002) -0.263 (0.004)
Asian -0.850 (0.004) -0.360 (0.003) -0.439 (0.004) -0.672 (0.005) -0.598 (0.012)
Hispanic -0.452 (0.002) -0.247 (0.001) -0.273 (0.002) -0.349 (0.003) -0.290 (0.005)
Other -0.069 (0.004) -0.060 (0.002) -0.065 (0.004) -0.148 (0.004) -0.167 (0.007)
Beneficiary Eligibility
Health status 0.449 (0.002) -0.017 (0.001) -0.061 (0.001) -0.072 (0.002) -0.014 (0.003)
Dual eligibility 0.373 (0.001) 0.202 (0.001) 0.289 (0.001) 0.160 (0.002) -0.177 (0.003)
LIS 0.388 (0.002) 0.142 (0.001) 0.280 (0.002) 0.287 (0.003) 0.033 (0.005)
Died 0.404 (0.003) -0.166 (0.001) -0.132 (0.002) -0.175 (0.003) 0.544 (0.004)
Disabled 0.604 (0.002) 0.309 (0.001) 0.607 (0.002) 0.614 (0.002) 0.114 (0.003)
ESRD 0.544 (0.004) -0.036 (0.002) -0.156 (0.004) -0.195 (0.004) -0.095 (0.007)
Beneficiary Diagnoses
Myalgia 0.387 (0.001) 0.067 (0.001) 0.058 (0.001) 0.063 (0.001) 0.128 (0.002)
Cancer 0.284 (0.002) -0.037 (0.001) -0.107 (0.002) -0.015 (0.002) 0.222 (0.003)
Opioid abuse 0.991 (0.010) 0.323 (0.003) 0.467 (0.004) 0.835 (0.005) 0.966 (0.008)
Other abuse 0.545 (0.002) 0.102 (0.001) 0.095 (0.002) 0.146 (0.002) 0.141 (0.003)
Alcohol -0.363 (0.005) -0.135 (0.002) -0.299 (0.004) -0.412 (0.004) -0.312 (0.007)
Overdose 1.177 (0.025) 0.187 (0.005) 0.296 (0.008) 0.421 (0.011) 0.475 (0.017)
Back Pain 0.848 (0.001) 0.155 (0.001) 0.358 (0.001) 0.418 (0.001) 0.200 (0.003)
Headache 0.324 (0.002) 0.014 (0.001) -0.014 (0.002) -0.034 (0.002) 0.008 (0.003)
Fracture 0.674 (0.002) -0.035 (0.001) -0.193 (0.002) -0.158 (0.002) 0.131 (0.003)
Sprain 0.333 (0.002) -0.082 (0.001) -0.266 (0.002) -0.248 (0.002) 0.043 (0.003)
Mental -0.025 (0.001) 0.097 (0.001) 0.086 (0.001) 0.036 (0.002) 0.004 (0.003)
Hep C 0.287 (0.005) 0.095 (0.002) 0.149 (0.004) 0.265 (0.005) 0.325 (0.007)
Chronic pain 1.845 (0.003) 0.486 (0.001) 0.817 (0.002) 0.957 (0.002) 0.654 (0.003)
Arthritis 0.708 (0.001) 0.106 (0.001) 0.220 (0.001) 0.250 (0.001) 0.095 (0.002)

Obs.
21,120,682 5,137,581 5,137,565 5,137,581 5,137,581
R-squared
0.193 0.173 0.161 0.172 0.059*

Notes: *r-squared value is a pseudo r-squared value.

105

Appendix Table 4. Regression Results, Opioid Persistence Measures

Persistent Use Log(Fills) Log(Days) Log(MME) High Dose

Coeff SE Coeff SE Coeff SE Coeff SE Coeff SE
Age 0.004 (0.000) 0.000 (0.000) 0.004 (0.000) -0.007 (0.000) -0.019 (0.000)
Female 0.189 (0.002) -0.027 (0.001) -0.046 (0.002) -0.118 (0.002) -0.062 (0.003)
Race/Ethnicity: Reference White
Black -0.208 (0.002) -0.123 (0.001) -0.201 (0.002) -0.296 (0.003) -0.256 (0.005)
Asian -1.057 (0.006) -0.322 (0.004) -0.377 (0.007) -0.612 (0.008) -0.506 (0.019)
Hispanic -0.555 (0.003) -0.213 (0.002) -0.233 (0.003) -0.308 (0.004) -0.267 (0.007)
Other -0.156 (0.005) -0.046 (0.003) -0.061 (0.004) -0.134 (0.006) -0.134 (0.010)
Beneficiary Eligibility
Health status 0.346 (0.002) -0.038 (0.001) -0.100 (0.002) -0.122 (0.002) -0.049 (0.004)
Dual eligibility 0.593 (0.002) 0.170 (0.001) 0.180 (0.002) 0.075 (0.002) -0.175 (0.004)
LIS 0.411 (0.003) 0.120 (0.002) 0.207 (0.003) 0.237 (0.003) 0.039 (0.006)
Died 0.319 (0.003) -0.323 (0.002) -0.376 (0.003) -0.405 (0.003) 0.492 (0.006)
Disabled 0.918 (0.002) 0.206 (0.001) 0.391 (0.002) 0.446 (0.002) 0.149 (0.005)
ESRD 0.464 (0.005) -0.083 (0.003) -0.244 (0.004) -0.303 (0.005) -0.173 (0.010)
Beneficiary Diagnoses
Myalgia 0.292 (0.002) 0.063 (0.001) 0.031 (0.001) 0.044 (0.002) 0.147 (0.003)
Cancer 0.109 (0.002) -0.026 (0.001) -0.075 (0.002) 0.019 (0.002) 0.257 (0.004)
Opioid abuse 0.723 (0.009) 0.236 (0.003) 0.309 (0.004) 0.671 (0.006) 0.927 (0.009)
Other abuse 0.388 (0.002) 0.088 (0.001) 0.078 (0.002) 0.132 (0.002) 0.138 (0.004)
Alcohol -0.470 (0.006) -0.127 (0.003) -0.283 (0.005) -0.419 (0.006) -0.331 (0.010)
Overdose 0.587 (0.021) 0.126 (0.006) 0.195 (0.008) 0.324 (0.012) 0.470 (0.020)
Back Pain 0.751 (0.002) 0.110 (0.001) 0.251 (0.001) 0.353 (0.002) 0.245 (0.003)
Headache 0.274 (0.002) -0.012 (0.001) -0.068 (0.002) -0.093 (0.002) -0.016 (0.004)
Fracture 0.228 (0.003) 0.015 (0.001) -0.107 (0.002) -0.071 (0.003) 0.180 (0.005)
Sprain 0.093 (0.002) -0.053 (0.001) -0.208 (0.002) -0.218 (0.002) 0.001 (0.004)
Mental 0.023 (0.002) 0.082 (0.001) 0.050 (0.002) 0.008 (0.002) 0.008 (0.004)
Hep C 0.241 (0.006) 0.073 (0.003) 0.105 (0.004) 0.250 (0.005) 0.381 (0.008)
Chronic pain 1.571 (0.003) 0.362 (0.001) 0.569 (0.002) 0.731 (0.002) 0.625 (0.004)
Arthritis 0.659 (0.002) 0.064 (0.001) 0.123 (0.001) 0.155 (0.002) 0.080 (0.003)

Obs.
19,245,288 2,693,598 2,693,598 2,693,598 2,693,598
R-squared
0.194 0.146 0.125 0.162 0.068*

Notes: *r-squared value is a pseudo r-squared value.

106

Appendix Table 5. Regression Results, Opioid Prescribing Measures

Physician Fills MME per Day Days Share High Dose

Coeff SE Coeff SE Coeff SE Coeff SE
Provider Specialty, Reference: General Medicine
Dental -1.411 (0.002) -0.144 (0.001) -1.386 (0.001) -0.040 (0.000)
Emergency Medicine -0.634 (0.003) 0.040 (0.001) -1.321 (0.001) -0.034 (0.000)
Nurse -0.677 (0.003) -0.003 (0.002) -0.238 (0.002) -0.002 (0.000)
Surgery -0.461 (0.002) 0.239 (0.001) -0.803 (0.001) 0.005 (0.000)
Anesthesia -0.352 (0.013) 0.191 (0.006) 0.001 (0.006) 0.032 (0.002)
Palliative or Hospice -0.552 (0.035) 0.279 (0.024) -0.356 (0.022) 0.073 (0.007)
Pain Medicine 1.295 (0.011) 0.315 (0.005) 0.397 (0.003) 0.051 (0.001)
Addiction Medicine -0.054 (0.033) 0.709 (0.020) 0.170 (0.015) 0.293 (0.008)
Other/Unknown -0.933 (0.002) 0.033 (0.001) -0.557 (0.001) -0.003 (0.000)
Patient Characteristics
Age 0.014 (0.000) -0.007 (0.000) 0.009 (0.000) -0.001 (0.000)
Share Male 0.015 (0.002) 0.029 (0.001) 0.002 (0.001) -0.001 (0.000)
Share White 0.363 (0.002) 0.136 (0.001) 0.033 (0.001) 0.019 (0.000)
Health status -0.197 (0.002) -0.012 (0.001) -0.047 (0.001) -0.008 (0.000)
Dual eligibility 0.332 (0.002) -0.119 (0.001) 0.117 (0.001) -0.014 (0.000)
LIS 0.203 (0.004) 0.000 (0.002) 0.138 (0.002) 0.005 (0.001)
Died -0.121 (0.004) 0.055 (0.003) 0.144 (0.003) 0.024 (0.001)
Disabled 0.516 (0.003) 0.036 (0.002) 0.330 (0.002) 0.009 (0.000)
ESRD -0.153 (0.004) -0.074 (0.003) -0.080 (0.003) -0.021 (0.001)
Patient Diagnoses
Myalgia -0.015 (0.002) 0.003 (0.001) 0.014 (0.001) 0.000 (0.000)
Cancer -0.224 (0.002) 0.093 (0.001) -0.046 (0.001) 0.006 (0.000)
Other abuse -0.039 (0.003) 0.047 (0.002) -0.074 (0.002) 0.005 (0.000)
Alcohol -0.258 (0.005) -0.083 (0.003) -0.164 (0.003) -0.016 (0.001)
Overdose -0.097 (0.012) 0.064 (0.009) 0.001 (0.008) 0.025 (0.002)
Back Pain 0.109 (0.002) 0.022 (0.001) 0.180 (0.001) 0.006 (0.000)
Headache -0.071 (0.002) -0.037 (0.001) -0.073 (0.002) -0.007 (0.000)
Fracture -0.109 (0.003) 0.009 (0.002) -0.084 (0.002) -0.003 (0.000)
Sprain -0.147 (0.002) -0.034 (0.001) -0.181 (0.001) -0.012 (0.000)
Mental 0.080 (0.002) -0.039 (0.001) -0.002 (0.001) -0.002 (0.000)
Hep C -0.092 (0.005) 0.093 (0.003) 0.038 (0.003) 0.029 (0.001)
Chronic Pain 0.255 (0.003) 0.162 (0.002) 0.250 (0.002) 0.034 (0.000)
Arthritis 0.206 (0.002) 0.016 (0.001) 0.151 (0.001) 0.000 (0.000)
Opioid Abuse -0.067 (0.006) 0.249 (0.004) 0.043 (0.004) 0.097 (0.001)

Obs. 3,158,639 3,158,639 3,158,639 3,158,639
R-squared 0.185 0.065 0.353 0.176

107

Appendix Table 6. Regression Results, Opioid Misuse Measures

Opioid Abuse Opioid Overdose

Coeff SE Coeff SE
Age -0.033 (0.000) -0.003 (0.001)
Female -0.185 (0.008) 0.048 (0.016)
Race/Ethnicity: Reference White
Black -0.100 (0.011) -0.458 (0.027)
Asian -0.133 (0.049) -0.135 (0.098)
Hispanic 0.034 (0.015) -0.193 (0.035)
Other 0.017 (0.020) -0.023 (0.043)
Beneficiary Eligibility
Health status 0.027 (0.007) 0.422 (0.012)
Dual eligibility -0.037 (0.010) 0.001 (0.020)
LIS 0.039 (0.015) 0.087 (0.031)
Died 0.045 (0.019) 0.664 (0.028)
Disabled 0.322 (0.014) 0.254 (0.026)
ESRD -0.331 (0.024) 0.356 (0.041)
Beneficiary Diagnoses
Myalgia 0.131 (0.008) 0.188 (0.017)
Cancer -0.110 (0.012) 0.017 (0.022)
Opioid abuse 1.265 (0.022)
Other abuse 1.288 (0.009) 0.872 (0.020)
Alcohol 0.759 (0.010) 0.574 (0.024)
Overdose 1.356 (0.023)
Back Pain 0.744 (0.008) 0.422 (0.017)
Headache 0.104 (0.008) 0.259 (0.017)
Fracture 0.064 (0.010) 0.262 (0.018)
Sprain -0.106 (0.008) -0.031 (0.017)
Mental 0.682 (0.009) 0.907 (0.020)
Hep C 0.985 (0.011) 0.141 (0.027)
Chronic pain 1.050 (0.008) 1.151 (0.018)
Arthritis 0.050 (0.008) 0.085 (0.016)
0.131 (0.008) 0.188 (0.017)
Obs.
4,317,955 5,137,581
Pseudo R-squared
0.284 0.192

108

APPENDIX TO CHAPTER 3
Appendix Table 1. Specification 4, Results with State-Specific Linear Time Trends
DV
Prob (Opioid User) Fills Prob (Fill>100 MME per day)
Estimation Method LPM log-linear LPM
Sample Mean 19.1% 3.1 15.5%
Pain Clinic Law
-0.0037** 0.0005 -0.0149*
(0.002) (0.002) (0.008)
Tamper Resistance
-0.0011 -0.0102** 0.0004
(0.002) (0.004) (0.004)
PDMP
-0.0003 -0.0045* -0.0024
(0.002) (0.003) (0.003)
Observations 64,208,328 12,284,478 12,284,478
DV Abuse or Overdose Prescriber Shopping
Physician & Pharmacy
Shopping
Estimation Method LPM LPM LPM
Sample Mean 2.9% 16.5% 3.6%
Pain Clinic Law
-0.0005 0.0014**
0.0010**
(0.001) (0.001)
(0.000)
Tamper Resistance
-0.0011** -0.0017
-0.0026**
(0.000) (0.002)
(0.001)
PDMP
0.0004 -0.0001
0.0002
(0.001) (0.002)
(0.001)
Observations 12,284,478 12,284,478 12,284,478
DV Fills Prescribed
MME per Day
Prescribed
Share of Fills Prescribed >100
MME per day
Estimation Method log-linear log-linear LPM
Sample Mean 24.2 52.1 10.1%
Pain Clinic Law -0.0163** -0.0477* -0.0186*
(0.008) (0.028) (0.010)
Tamper Resistance -0.0021 0.0055 0.0006
(0.021) (0.018) (0.004)
PDMP -0.0038 -0.0061 -0.0008
(0.009) (0.011) (0.002)
Observations 5,980,677 5,980,677 5,980,677
Demographics
x x X
Eligibility Status
x x X
Control Laws
x x X
Quarter-Year FE
x x X
State FE
x x X
State Time Trends
x x X

109

Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level.
Robust standard errors clustered at the state level are in parentheses.

110

Appendix Table 2. Difference in Differences Estimates Employing Alternative Definitions
of Opioid User
Specification
1 2 3
Panel 1: 1 Fill per Quarter, Multiple Fills per Year
Sample Mean
19.1%
Pain Clinic Law
-0.0015 -0.0009 -0.0038*
(0.002) (0.002) (0.002)
Tamper Resistance
-0.0061*** -0.0054*** -0.0035**
(0.002) (0.002) (0.001)
PDMP
-0.0028 -0.0028 -0.0025*
(0.002) (0.002) (0.001)
Observations
64,208,328 64,208,328 64,208,328
Panel 2: 1 Fill per Quarter
Sample Mean
22.6%
Pain Clinic Law
-0.0020 -0.0015 -0.0044**
(0.003) (0.002) (0.002)
Tamper Resistance
-0.0064*** -0.0060*** -0.0040**
(0.002) (0.002) (0.002)
PDMP
-0.0026 -0.0025 -0.0023*
(0.002) (0.002) (0.001)
Observations
12,284,478 12,284,478 12,284,478
Panel 3: 2 Fills per Quarter
Sample Mean 14.2%
Pain Clinic Law
0.0005 0.0006 -0.0019

(0.002) (0.002) (0.001)
Tamper Resistance
-0.0059*** -0.0055*** -0.0038***
(0.002) (0.002) (0.001)
PDMP
-0.0025 -0.0026 -0.0025*

(0.002) (0.002) (0.001)
Observations
12,284,478 12,284,478 12,284,478
Demographics
x
Eligibility Status
x
Control Laws
x x
Quarter-Year FE
x x x
State FE
x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

111

Appendix Table 3. Count Outcomes Using Different Estimation Methods

Specification
2 3 4
DV
Count of Fills (Opioid Users)
Estimation Method Fixed Effect Poisson
Pain Clinic Law
0.0053 -0.0012 0.0015
(0.011) (0.010) (0.002)
Tamper Resistance
-0.0223*** -0.0153** -0.0117**
(0.008) (0.008) (0.005)
PDMP
0.0015 0.0013 -0.0034
(0.005) (0.006) (0.003)
Observations 12,284,478 12,284,478 12,284,478
DV Count of Fills Prescribed (Opioid Prescribers)
Estimation Method Fixed Effect Poisson
Pain Clinic Law
-0.0035 0.0027 -0.0171
(0.010) (0.009) (0.011)
Tamper Resistance
-0.0314* -0.0257 0.0032
(0.018) (0.017) (0.018)
PDMP
-0.0144 -0.0137 -0.0024
(0.011) (0.013) (0.012)
Observations 5,980,677 5,980,677 5,980,677
DV MME Prescribed per Day (Opioid Prescribers)
Estimation Method Fixed Effect Poisson
Pain Clinic Law
-0.1230*** -0.1256*** -0.0704*

(0.041) (0.041) (0.036)
Tamper Resistance
0.0112 0.0130 0.0047

(0.021) (0.022) (0.018)
PDMP
0.0074 0.0075 -0.0020

(0.015) (0.016) (0.011)
Observations 5,980,677 5,980,677 5,980,677
Demographics
x x
Eligibility Status
x x
Control Laws
x x x
Quarter-Year FE
x x x
State FE
x x x
State Time Trends
x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

112

Appendix Table 4. Annual Level Analysis, Opioid Utilization

Specification
1 2 3
DV
Probability (Opioid User)
Estimation Method LPM
Pain Clinic Law
-0.0031* -0.0018 -0.0041***
(0.002) (0.002) (0.001)
Tamper Resistance
-0.0075*** -0.0080*** -0.0055***
(0.002) (0.002) (0.001)
PDMP
-0.0009 -0.0012 -0.0010
(0.002) (0.002) (0.002)
Observations 17,586,169
DV Count of Fills
Estimation Method Log-linear
Pain Clinic Law
0.0062 0.0044 -0.0041
(0.007) (0.007) (0.005)
Tamper Resistance
-0.0122** -0.0127** -0.0049
(0.006) (0.006) (0.004)
PDMP
-0.0040 -0.0032 -0.0019
(0.006) (0.006) (0.005)
Observations 4,317,871
DV Probability (>100 MME per day)
Estimation Method LPM
Pain Clinic Law
-0.0329*** -0.0245*** -0.0259***

(0.010) (0.008) (0.008)
Tamper Resistance
0.0123 0.0125 0.0137

(0.010) (0.013) (0.013)
PDMP
0.0061 0.0026 0.0023

(0.010) (0.009) (0.009)
Observations 4,317,871
Demographics
x
Eligibility Status
x
Control Laws
x x
Quarter-Year FE
x x x
State FE
x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

113

Appendix Table 5. Annual Level Analysis, Opioid Abuse or Misuse

Specification
1 2 3
DV
Probability (Abuse or Overdose)
Estimation Method LPM
Pain Clinic Law
-0.0005 0.0004 -0.0007
(0.002) (0.001) (0.001)
Tamper Resistance
0.0019* 0.0018 0.0026
(0.001) (0.001) (0.002)
PDMP
0.0003 0.0000 0.0001
(0.001) (0.001) (0.001)
Observations 4,317,871
DV Prescriber Shopping
Estimation Method LPM
Pain Clinic Law
0.0022 0.0023 -0.0012
(0.002) (0.002) (0.002)
Tamper Resistance
-0.0059*** -0.0079*** -0.0055**
(0.002) (0.002) (0.002)
PDMP
0.0009 0.0012 0.0019
(0.002) (0.002) (0.002)
Observations 4,317,871
DV Prescriber & Pharmacy Shopping
Estimation Method LPM
Pain Clinic Law
0.0001 0.0007 -0.0012

(0.001) (0.001) (0.001)
Tamper Resistance
-0.0026 -0.0042*** -0.0029***

(0.002) (0.001) (0.001)
PDMP
0.0010 0.0011 0.0014

(0.001) (0.001) (0.001)
Observations 4,317,871
Demographics
x
Eligibility Status
x
Control Laws
x x
Quarter-Year FE
x x x
State FE
x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

114

Appendix Table 6. Annual Level Analysis, Opioid Prescribing

Specification
1 2 3
DV
Fills
Estimation Method Log-linear
Pain Clinic Law
-0.0214** -0.0124 -0.0082
(0.010) (0.010) (0.008)
Tamper Resistance
-0.0204 -0.0174* -0.0244**
(0.016) (0.010) (0.010)
PDMP
0.0056 0.0025 0.0004
(0.009) (0.008) (0.010)
Observations 2,360,336
DV MME per Day
Estimation Method Log-linear
Pain Clinic Law
-0.0951*** -0.0750*** -0.0769***
(0.019) (0.018) (0.018)
Tamper Resistance
-0.0041 -0.0131 -0.0122
(0.024) (0.025) (0.025)
PDMP
0.0093 0.0061 0.0060
(0.016) (0.015) (0.015)
Observations 2,360,336
DV Share of Fills >100 MME per Day
Estimation Method LPM
Pain Clinic Law
-0.02538*** -0.0200*** -0.0199***

(0.006) (0.005) (0.005)
Tamper Resistance
0.0044 0.0032 0.0032

(0.005) (0.007) (0.007)
PDMP
0.0043 0.0032 0.0031

(0.004) (0.004) (0.004)
Observations 2,360,336
Demographics
x
Eligibility Status
x
Specialty
x
Control Laws
x x
Quarter-Year FE
x x x
State FE
x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

115

Appendix Table 7. Diagnosis-Level Results, Patient Use, Abuse, or Misuse Measures
Specification 3 3 3 3
DV Fills Prob(>100 MME)
Prob(Abuse or
Overdose)
Prescriber &
Pharmacy
Shopping
Estimation Method Log-linear LPM LPM LPM
Diagnosis Cancer or Fracture
Sample Mean 3.1 16.3% 2.9% 4.4%
Pain Clinic Law
0.0028 -0.0341*** -0.0048*** -0.0007
(0.006) (0.012) (0.001) (0.001)
Tamper Resistance
-0.0108** 0.0051 0.0005 -0.0043***
(0.005) (0.006) (0.001) (0.001)
PDMP
0.0023 -0.0031 0.0021* 0.0015
(0.004) (0.005) (0.001) (0.001)
Observations 3,378,153
Diagnosis Chronic or Back Pain
Sample Mean
3.6 19.3% 5.5%
6.6%
Pain Clinic Law
0.0000 -0.0357*** -0.0041*** -0.0006

(0.007) (0.010) (0.002) (0.003)
Tamper Resistance
-0.0176** 0.0062 0.0002 -0.0050**

(0.007) (0.005) (0.001) (0.002)
PDMP
-0.0053 -0.0028 0.0018 -0.0005

(0.004) (0.006) (0.002) (0.002)
Observations 4,784,075
Demographics
x x x x
Eligibility Status
x x x x
Control Laws
x x x x
Quarter-Year FE
x x x x
State FE
x x x x
Notes: *** indicates significance at the p<0.01 level, ** at the p<0.05 level, and * at the p<0.1 level. Robust
standard errors clustered at the state level are in parentheses.

116

APPENDIX TO CHAPTER 4
Appendix Table 1. Milligrams of Morphine Equivalence Conversions
Active Ingredient MME Conversion Factor
Butorphanol 7
Codeine 0.15
Dihydrocodeine 0.25
Fentanyl 120*
Fentanyl Citrate 130
Hydrocodone 1
Hydromorphone 4
Levorphanol 11
Meperidine 0.1
Methadone 3.0
Morphine 1.0
Nalbuphine 1.0
Opium 1
Oxycodone 1.5
Oxymorphone 3
Pentazocine 0.37
Propoxyphene 0.2
Tapentadol 0.4
Tramadol 0.1
Notes: *We assume transdermal Fentanyl patches last for 6 hours so the effective conversion factor is 720 or 120* 6
hours

Abstract (if available)

Abstract Despite growing attention to the problem of prescription opioid use, misuse, and abuse in the United States, there are many unanswered questions. Further, many policies aimed at curbing opioid use have been enacted ahead of reliable evidence on the likely efficacy of such interventions. The goal of my dissertation is to fill in these major gaps in our understanding. Each paper adds critical knowledge that deepens our understanding of the scope and trajectory of opioid utilization in the United States, the effect of existing policies on both appropriate and inappropriate opioid utilization, and the role of physicians in prescribing these drugs. Each paper relies on the analysis of large datasets—both survey-based and claims-based—and statistical methods to add detail to our understanding of this “epidemic.” ❧ The first paper focuses on quantifying opioid utilization in Medicare. The majority of research on opioid utilization and abuse has focused on the under-65 population. The few studies that tackle the question of opioid use in Medicare provide single-year snapshots or report on a limited number of outcomes. My results fill that gap by chronicling multiple measures of opioid utilization in a 20 percent sample of Medicare beneficiaries from 2006 to 2012. I estimate multivariate regression models to show patterns of use, misuse, and prescribing that control for demographics, patient health, socio-economic status, and eligibility status. Results indicate that in contrast to the general population, opioid utilization in Medicare—while high—is fairly stable. Despite stable use, opioid abuse in the Medicare population is on the rise

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Essays on access to prescription drugs

PDF

A multi-system evaluation of medication for opioid use disorder for people who use opioids

PDF

Sportiod and teen athletes

PDF

Health care utilization and spending of the U.S. aging population

PDF

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

How it hurts: culture, markets, and pain in the U.S. opioid epidemic

PDF

Opioid withdrawal symptoms, opioid use, and injection risk behaviors among people who inject drugs (PWID)

PDF

Addressing federal pain research priorities: drug policy, pain mechanisms, and integrative treatment

PDF

Thesis on Medicare Part D

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Essays on the use of microsimulation for health and economic policy analysis

PDF

The Linkage to Hope Project

PDF

Essays on the U.S. market for substance use treatment and the impact of Medicaid policy reform

PDF

A qualitative study of street fentanyl in Dayton, Ohio: drug markets, trajectories, and overdose risk reduction

PDF

Three essays on health, aging, and retirement

PDF

The context of leadership in the development of California’s innovation hubs

PDF

US biosimilar market entry and uptake: US commercial payer preferences in access decision making

PDF

Essays in health economics: evidence from Medicare

PDF

Essays on the economics of cities

Asset Metadata

Creator Axeen, Sarah Anne (author)

Core Title Essays in opioid use and abuse

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Publication Date 04/20/2016

Defense Date 03/22/2016

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics of addiction,Health Economics,Health policy,OAI-PMH Harvest,opioid,prescription drugs,substance abuse

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Goldman, Dana (committee chair), Joyce, Geoffrey (committee member), Lakdawalla, Darius (committee member)

Creator Email axeen@usc.edu,sarah.axeen@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-235958

Unique identifier UC11279358

Identifier etd-AxeenSarah-4316.pdf (filename),usctheses-c40-235958 (legacy record id)

Legacy Identifier etd-AxeenSarah-4316.pdf

Dmrecord 235958

Document Type Dissertation

Format application/pdf (imt)

Rights Axeen, Sarah Anne

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA