IntroductionOpioid misuse and associated harms have grown to epidemic proportions. From
1999–2015, opioid prescriptions in the U.S. tripled,1 paralleled by rates of opioid-related
overdose deaths, emergency department visits, and substance use treatment
admissions.2,3 In 2015, prescription opioids were involved
in 15,000 deaths.4 Most people
using opioid analgesics non-medically obtained their drugs from a prescription
written either for them or for an acquaintance.5 Thus, efforts to reduce opioid-related harm often focus on
moderating prescribing.
State prescription drug monitoring programs (PDMPs) electronically store
information about prescriptions filled by patients for drugs with misuse potential,
including opioids. Because they offer a wealth of information to prescribers, the
Centers for Disease Control and Prevention describe PDMPs as “among the most
promising state-level interventions to improve painkiller prescribing, inform
clinical practice, and protect patients at risk,” and recommend regular PDMP
queries in their opioid prescribing guidelines for chronic pain.6,7
PDMP implementation has proliferated since 1990, and now all states except
Missouri have operational programs. Many early programs had technical deficiencies,
lacked comprehensive data,8,9 and were little-used. Their
effectiveness in reducing opioid prescriptions and opioid-related overdoses was
limited.10–18 More recently, states have
attempted to increase PDMP participation and improve the clinical utility of PDMP
data. Today’s most robust PDMPs require prescribers to register
(“registration mandate”) and to query a database prior to prescribing
opioids (“use mandate”), and allow prescribers’ delegates to
check the PDMP.19,20 Online Appendix A details the features of
robust PDMP policies.21
Recent literature finds that PDMPs—particularly those with use
mandates or that are implemented in combination with pain clinic laws that regulate
the ownership, operation, and management of facilities primarily engaged in the
treatment of pain—are associated with decreases in opioid prescribing, doctor
shopping, and overdoses.22–26 However, one study found that
registration mandates, not use mandates, reduce opioid prescribing among Medicaid
patients.27 As well,
previous research does not evaluate effects of a package of robust PDMP features on
opioid prescribing26 or on PDMP
effects among adults with commercial insurance, who account for over half of opioid
prescriptions in many states.28
Using a rigorous longitudinal design, we compare prescribing outcomes in
states with “robust” PDMPs (defined below) to outcomes in states with
weak or no PDMPs. We hypothesized that states introducing robust PDMPs would
experience reductions in several population-level measures of opioid prescribing,
including the proportion of patients filling opioid prescriptions and the total
dosage of opioids prescribed per patient. We also hypothesized that enrollees
receiving opioids in states with robust PDMPs would experience reductions, relative
to comparison states and to a pre-implementation period, in high-dose opioid
prescribing and receipt from multiple prescribers and pharmacies.
MethodsOur quasi-experimental analyses of individual-level health insurance claims
data (2010–2014) for commercially-insured adults compared states implementing
robust PDMPs to matched states without robust PDMPs.
States with robust PDMPs and comparison states.Based on careful review of state PDMP laws, we devised a comprehensive,
novel coding scheme to identify states that implemented robust PDMPs before
January 1, 2014. Robust PDMPs exhibited by the end of 2013 at least 8 of 10
features that facilitate prescriber access to comprehensive, timely data and/or
have been established in prior PDMP evaluation literature as important to
increasing prescriber use and utility of the data. These were prescriber access
to the PDMP; a use mandate of any kind (including those that require prescribers
to check the database only if they suspect misuse or diversion, rather than
based on objective prescribing criteria); a comprehensive use
mandate that requires prescribers to check the PDMP regularly (including for any
initial prescription of Schedule II-III drugs to a patient); operation by a
health agency; at least weekly dispensing data updates; monitoring of all
controlled substances on the federal schedules II-IV; a registration mandate (or
automatic registration); delegate access; proactive reporting of suspicious
prescribing or dispensing; and no prescriber immunity for failure to check the
PDMP.11,19–20,22–27 Online Appendix A and Appendix Exhibits A1-A2 describe if
and when robust PDMP features were implemented across all 50 states.21
We identified Kentucky, New Mexico, Tennessee, and New York as
“intervention states” with robust PDMPs. Each state implemented a
robust PDMP as of the quarter in which it exceeded the 8-feature threshold by
unveiling a set of reforms on top of baseline program features. Kentucky
implemented 4 new features (a use mandate, a comprehensive use mandate, a
registration mandate, and delegate access) and New Mexico implemented 3 new
features (a use mandate, a comprehensive use mandate, and a registration
mandate) in the third quarter of 2012 to qualify as having a robust PDMP,
thereby providing 9 post-implementation quarters for observation in each state.
Tennessee implemented 3 new features (a comprehensive use mandate, a
registration mandate, and weekly data updates) in the first two quarters of 2013
to qualify, thereby providing 6 post-implementation quarters for observation.
New York implemented 6 new features (a use mandate, a comprehensive use mandate,
proactive reporting, no prescriber immunity, at least weekly data updates, and
delegate access) in the third quarter of 2013 to qualify, thereby providing 5
post-implementation quarters for observation. Importantly, no other relevant
program feature changes closely preceded robust PDMP implementation in any
intervention state.21
We selected neighboring comparison states without robust PDMPs that had
similar primary outcome trends during the pre-implementation period and 5 or
fewer of the 10 PDMP features. We compared Missouri to Kentucky, Texas to New
Mexico, Georgia to Tennessee, and New Jersey to New York, as shown on a map in
online Appendix Exhibit
A3.21
Opioid prescribing.We used deidentified Optum data (OptumInsight, Eden Prairie, MN) to
quantify prescribed opioids dispensed to enrollees in plans offered by a large
national health insurer.29,30 From the study states, we
included prescription claims for adults aged 18 to 64 years enrolled between
January 1, 2010 and December 31, 2014. The study start date provided adequate
time to evaluate baseline trends, including effects of two national
interventions associated with decreases in opioid-related overdoses and
prescribing that occurred in the fourth quarter of 2010: reformulation of
OxyContin to a tamper-resistant extended-release form and withdrawal of
propoxyphene from the market.29 Because rates in our outcomes of interest declined
similarly among comparator states following these national interventions and
well before our state PDMP implementation periods, we included 2010 data in our
analysis (Exhibits 1–2).
Statistical analyses.We used two quasi-experimental designs to assess robust PDMP effects
independent of other changes to opioid prescribing: interrupted time series with
comparison series and difference-in-difference analysis.
The first design, population-level comparative interrupted time series
analyses, captured quarterly event rates contemporaneously in 4
intervention/comparison states pairs. The primary outcomes modeled were
percentage of enrollees filling opioid prescriptions and mean morphine
equivalent dosage (MED) dispensed per enrollee, a measure that standardizes
opioid dosages prescribed. We adjusted quarterly primary outcomes for changes in
this open cohort’s characteristics during the study period and between
state comparator sets.31 We
then used aggregate-level segmented linear regression32 to model the differenced outcomes
between intervention and comparison states for each of the 4 pairs separately.
We detail the interrupted time series statistical analyses in online Appendix A.21
To ensure that patients entering or leaving the study population were
not biasing results and to generate interpretable relative change estimates, we
conducted sensitivity difference-in-differences analyses on cohorts of adults
with any opioid receipt (i.e., at least 1 opioid prescription fill during the
study period) and adults with chronic non-cancer-related opioid receipt who were
continuously enrolled from 1 year before to 1 year after the robust PDMP
implementation quarter. We defined adults with “chronic
non-cancer-related opioid receipt” as those with opioid fills in each
quarter of the year prior to the PDMP implementation quarter(s) for reasons not
related to a non-benign cancer diagnosis. Adjusting for individual covariates,
we modeled mean MED dispensed, the number of opioid fills, and the following
“high-risk opioid prescribing measures”: average daily MED of
≥ 100 mg, and mean number of quarters in which enrollees used ≥ 3
doctors or ≥ 3 pharmacies to fill opioid prescriptions. We repeated
analyses using an alternative comparison state for each intervention state. We
detail sensitivity difference-in-differences analyses in online Appendix B.21
We performed analyses using SAS 9.3 (Cary, NC) and Stata 12 (College
Station, Texas).
Study limitations.We used administrative data and therefore cannot observe opioid
dispensing not billed to insurance (e.g., cash purchases). Nevertheless, our
data represent a substantial market share; approximately two-thirds of
controlled substances are paid for by commercial insurance.28,33 Second, our cohort was limited to commercially-insured
adults ages 18 to 64—a population with high opioid analgesic
use34—and our
results may not be generalizable to other populations. Third, we focused
primarily on opioid prescribing outcomes, rather than opioid-related injuries.
These intermediate outcomes are more proximal to PDMP implementation, but do not
directly measure the health impact of PDMPs. Fourth, by defining one cohort for
difference-in-difference analyses as persons with opioid fills at any time
during the study period, we may have introduced bias if those receiving opioids
when robust PDMPs were and were not in place differ. To mitigate this risk, we
required this cohort to be continuously enrolled and controlled for individual
characteristics.
Fifth, in our comparator pairs, baseline levels were different and
baseline trends for some outcomes were not parallel between the states compared
in each set.32 However, effect
estimates from comparative interrupted time series designs are robust even with
such differences, particularly when effects observed at the time of an
intervention are immediate and dramatic. Finally, our analysis did not account
for other state policy interventions that may have occurred at the same time,
although we researched opioid policies in the study states and identified none
that could have explained our findings except perhaps for Kentucky’s pain
clinic law.35
ResultsAll 4 states with robust PDMPs exhibited reductions in the opioid dosages
prescribed to commercially-insured individuals versus their comparisons, and only
Kentucky also exhibited a sustained reduction in the percentage of persons filling
opioid prescriptions. The number of opioid prescription fills per person receiving
opioids in a year also declined following robust PDMP implementation in all 4 states
examined versus comparison states. High-risk opioid prescribing measures only
declined among individuals receiving opioids in Kentucky relative to Missouri.
Sample characteristics.In the interrupted time series analyses of an open cohort of all adult
enrollees in commercial insurance plans, enrollment and demographic composition
within each state remained quite consistent over time. Sample sizes ranged from
36,100–867,500 enrollees in the month before robust PDMP implementation.
Average age was 39–42, half were male, and mean enrollment was
34–52 months. Most of the sample was white, with substantial proportions
of persons classified as Hispanic and African American in some states. Online
Appendix A and
Exhibits A4-A7
detail open cohort characteristics within each state comparator set.21
Changes in population-level opioid prescribing.By the end of 2014, opioid dosages prescribed declined significantly and
in clinically meaningful quantities in all states with robust PDMPs relative to
their comparisons. Also, the proportion of enrollees filling opioid
prescriptions declined in Kentucky versus its comparison state.
Pre-implementation trends in the percentage of enrollees filling opioid
prescriptions per quarter were parallel among each state comparator set. Among 2
of the 4 state comparator sets, pre-implementation trends in mean MED dispensed
per enrollee were parallel (Exhibits
1–4). Pre-implementation
levels were generally higher in intervention than comparison states, albeit
still usually parallel. Online Appendix Exhibits A8 and A9 display the trends in percentage of
enrollees filling opioid prescriptions per quarter and quantify the differences
in pre- and post-robust PDMP implementation levels and trends between state
comparator sets, respectively.21
In the quarter after robust PDMP implementation, the percentage of
sampled persons filling opioid prescriptions declined significantly in 3 states.
In Kentucky, where the rate of opioid fills was approximately 10.2% in the
quarter before robust PDMP implementation, the absolute percentage fell by 1.3%
compared to Missouri in the quarter immediately after PDMP implementation.
Tennessee exhibited similar reductions to Kentucky along this outcome, while New
York’s reductions were smaller in magnitude. However, these initial
reductions were not sustained in Tennessee and New York over time. By the end of
the study period, or the fourth quarter of 2014, Kentucky exhibited the most
sustained decline, with 1.6% fewer persons filling opioid prescriptions compared
to Missouri.
In the quarter after robust PDMP implementation, the mean MED level
dispensed per person declined significantly in 3 states. In Kentucky, where mean
MED was approximately 192.4 mg per quarter pre-implementation, the absolute
level fell by 9.2 mg within Kentucky and by 31.6 mg relative to Missouri (Exhibit 1). Just after PDMP implementation,
mean MED also fell in Tennessee and New York relative to their comparators, but
by smaller amounts than in Kentucky (Exhibits
3, 4). Enrollees in Kentucky and
New Mexico also experienced downward post-implementation trends versus
comparisons (Exhibits 1, 2). By the fourth quarter of 2014, Kentucky exhibited
the greatest decline in mean MED dispensed per enrollee, an absolute reduction
of 77.1 mg relative to Missouri. However, absolute reductions were also
significant by the end of the study period – albeit in smaller magnitudes
– for the other 3 states relative to their comparators (Exhibits 2–4). In Kentucky, New Mexico, and Tennessee, these reductions
approximately equate to each commercially-insured person receiving 10, 8, and 5
fewer 5-mg oxycodone tablets in a quarter, respectively.
Changes in opioid prescribing among persons who received opioids.Difference-in-difference analyses among continuously enrolled
individuals who received any opioids were consistent with the population-level
analyses of all adult enrollees, which revealed reductions in opioid dosages
dispensed to individuals after robust PDMP implementation. Also among this
continuous cohort who received opioids, a robust PDMP was associated with
significant relative reductions in the number of opioid prescriptions filled
across all states studied. In Kentucky, the robust PDMP was also associated with
significant reductions in high-risk opioid prescribing measures, such as those
receiving opioids in high dosages and from multiple prescribers or
pharmacies.
Among individuals with any opioid receipt in Kentucky, average opioid
prescription fills were 2.4 and 2.1 in the years pre- and post-robust PDMP
implementation, respectively, as compared to 2.1 and 2.3 in Missouri. Enrollees
in Kentucky received mean MED of 4177.1 mg and 4024.6 mg pre- and post-robust
PDMP implementation, as compared to 3922.4 mg and 4627.6 mg in Missouri. Online
Appendix A10
provides estimates for pre- and post-robust PDMP implementation outcome levels
among cohorts with opioid receipt.21
According to the difference-in-difference analyses, the mean MED
dispensed in the year after robust PDMP implementation declined in Kentucky by
857.6 (relative: −18.3%) compared to Missouri. Relative reductions in
this measure were also significant, albeit smaller in magnitude than in
Kentucky, in New Mexico, Tennessee, and New York (Exhibit 5). We also detected significant reductions
in the number of opioid fills in Kentucky of 0.4 (relative: −16.2%), and,
to a lesser degree, in the other 3 states in the year after implementation
relative to their comparators (Exhibit
5).
In analyses of potentially high-risk opioid prescribing among persons
who received any opioids, only Kentucky experienced significant
post-implementation reductions in relation to its comparator, Missouri. The
percentage of persons in Kentucky with daily MED ≥ 100 mg declined by
0.2% (relative: −20.4%). Also in Kentucky, the number of quarters when
opioid prescriptions were filled with ≥ 3 doctors per enrollee or at
≥ 3 pharmacies per enrollee fell by 0.02 (relative: −40.4%) and
0.01 (relative: −38.1%), respectively (Exhibit 5).
Our findings were generally consistent with results generated when
examining opioid prescribing outcomes among continuous enrollees with baseline
chronic non-cancer-related opioid receipt. The magnitude of pre-and
post-measurements along all outcomes was higher for this chronic cohort than for
the cohort with any opioid receipt. For example, enrollees in Kentucky were
filling 13.5 and 11.0 opioid prescriptions for non-cancer-related conditions on
average pre- and post-robust PDMP implementation, respectively, as compared to
14.1 and 12.8 in Missouri.
As compared to the cohort with any opioid receipt,
difference-in-difference changes for the chronic non-cancer-related opioid
receipt cohort were larger in absolute terms, smaller in relative terms, and
somewhat less significant. As in the main analyses, the most dramatic and
consistent reductions along all outcomes among this chronic cohort were observed
in Kentucky relative to Missouri. For example, post-implementation absolute and
relative reductions in the following outcomes were observed in Kentucky as
compared to Missouri: enrollees with daily MED ≥ 100 mg declined by 1.43%
(relative: −13.3%); quarters when opioid prescriptions were filled with
≥ 3 doctors per enrollee fell by 0.06 (relative: −37.8%); and
quarters when opioid prescriptions were filled at ≥ 3 pharmacies per
enrollee dropped by 0.04 (relative: −32.7%). Online Appendix B and Exhibits A10-A13 present complete
difference-in-differences results for the chronic non-cancer-related opioid
receipt cohort.21
Our findings were also generally consistent with results generated when
using alternative comparison states. Notably, our Kentucky findings relative to
Missouri, a state with no PDMP, were qualitatively similar to those relative to
Indiana, a state with a PDMP only lacking in a comprehensive PDMP mandate and
registration mandate during the study period. Online Appendix B and Exhibits A10-A13 provide complete
difference-in-differences results for alternative comparator state
sets.21
DiscussionWe evaluated changes in opioid prescribing after robust PDMP implementation
in 4 states. In all states studied, robust PDMP implementation was associated with
sustained declines in the total opioid dosage prescribed and number of opioid fills.
Implementation was less consistently associated with reduced percentages of patients
prescribed opioids. The magnitude and statistical significance of effect varied
across intervention states, with Kentucky exhibiting the most dramatic and
consistent decreases along all outcomes, followed by Tennessee and New Mexico. We
also found that Kentucky’s robust PDMP was associated with reductions in
potentially high-risk opioid prescribing measures (i.e., high-dose prescriptions and
individuals receiving prescriptions from multiple providers).
Our results are generally consistent with recent studies in other
populations that have found reductions in opioid prescribing following
implementation of PDMP mandates.22–24,27 Specifically, other analyses have found that
registration mandates, rather than use mandates, reduce the number of Schedule II
opioid prescriptions among Medicaid enrollees;27 use mandates are associated with reductions in quantity
prescribed, as well as doctor and pharmacy shopping in the Medicare
population;22,24 and use mandates paired with pain clinic
laws are associated with lower opioid dosages prescribed.23
Our study benefits from a more detailed legal and policy analysis and
classification of PDMP robustness than previous studies. We also employed a
longitudinal, controlled design and compared multiple states along many opioid
prescribing outcomes. We assessed PDMP effects among commercially-insured adults, a
population not previously investigated but among whom opioid prescribing is highly
prevalent.28,34 We also focused on prescribing outcomes more
proximal to PDMP use, rather than more distal outcomes like overdoses.25,26 We found differential effects even among states with robust
PDMPs, which may explain mixed or conflicting findings in other studies.10–18,22–24,27
We observed the smallest effects in New York, where rates of opioid
prescribing were lower at baseline, so PDMP checks might have been less likely to
yield salient information and influence prescribers’ decisions. Additionally,
New York’s PDMP had fewer robust features than the other intervention states.
Most notably, it lacked a registration mandate.
Kentucky’s program might serve as a model for other jurisdictions
given the consistent and dramatic effects following robust PDMP implementation that
we detected. In July 2012, Kentucky implemented PDMP use and
registration mandates and allowed delegates to use the registry. Delegate PDMP
checks have helped prescribers to satisfy the use mandate requirement.20 Kentucky’s PDMP also
benefited from increased administrative staffing to support its operations, more
frequent (daily) updates to the data, and input from prescribers that made the PDMP
user-friendly.20,35 From 2011 to 2014, PDMP queries increased
over 500%, and from 2012 to 2013, PDMP registrants increased by almost 70%. By July
2013, 95% of in-state practitioners with the authority to prescribe controlled
substances were PDMP-registered.20,35–36
Of note, Kentucky implemented a pain clinic law concurrent with PDMP
upgrades in 2012, which led to closure of many pain clinics and perhaps may have
contributed to some effects observed.24,35Our data do not
reflect changes in prescription opioids paid for with cash, a common practice in
pain clinics.37 However, to the
extent our data do capture some prescription opioids dispensed at pain clinics, any
of the prescribing outcomes measured in Kentucky could be affected by the pain
clinic law in addition to the robust PDMP.
Both Tennessee and New Mexico, which have PDMPs with features similar to
Kentucky’s and also benefited from implementation support and increased PDMP
usership, experienced significant effects after implementation. Concurrent with
robust PDMP implementation, Tennessee upgraded hardware configuration to handle
increased query volume and educated prescribers about the PDMP.38 From 2012–2013, Tennessee PDMP
registrants increased 56% while PDMP queries more than doubled from 1.86 to 4.50
million.38 Although New
Mexico does not make available pre-implementation registration and use rates, a
gradual monthly increase in PDMP queries after robust implementation from 34,000
queries in January 2013 to 100,000 queries in December 2014 suggests that PDMP
participation increased.39 Because
these states, like Kentucky, allowed delegates to use PDMPS in addition to
implementing comprehensive use and registration mandates for prescribers, this trio
of policy features may be important to include in a PDMP for significant and
clinically meaningful reductions in opioid prescribing.
Unintended consequences of increased PDMP use, such as reduced prescribing
when medically indicated or patient substitution to illicit opioid sources, should
be monitored closely, although a qualitative assessment of Kentucky’s robust
PDMP implementation did not detect adverse unintended consequences.35 Newer features not studied here
also deserve future attention, including automated interstate sharing of data and
interoperability of PDMP data and medical records.20 Despite these areas of uncertainty, the
evidence is compelling that robust PDMPs can make a significant difference in
curbing opioid prescribing.