Systematic Evaluation of State Policy Interventions Targeting the US Opioid Epidemic, 2007-2018

This cross-sectional study assesses state drug policies implemented between 2007 and 2018 to ascertain whether they are associated with variations in opioid misuse, opioid use disorder, and drug overdose mortality.


Introduction
The current opioid epidemic in the US has its historical roots in the movement during the 1990s to address undertreated chronic pain. In response, opioid-producing pharmaceutical companies engaged in aggressive marketing and prescribers overcorrected, relying on powerful opioid analgesics to treat acute and minor pain in addition to chronic and severe pain. Subsequently, the widespread use of opioid analgesic agents created demand for long-term and non-medical use of prescription and illicit opioids. 1,2 To address the growing opioid epidemic, policy makers have focused largely on controlling the prescription and use of opioid analgesics through the implementation of supply-side drug policies. These include prescription drug monitoring programs (PDMPs), pain clinic laws, and prescription limit laws to reduce inappropriate prescribing behavior. In tandem, policy measures to reduce harms and barriers associated with treating and reporting drug overdose, including naloxone access laws and Good Samaritan laws, have been introduced.
Previous studies have provided mixed evidence on the impact that these state policies have had on opioid misuse, nonfatal overdose, and opioid mortality. For example, some research indicates that access to a PDMP, which allows prescribers to review patients' prescription histories, substantially reduces prescription of opioids by 6 percentage points and oxycodone distribution by 8 percentage points. 3,4 In contrast, other studies have found that PDMP access is ineffective [5][6][7] or is only effective if a review of patient records is mandatory. 8,9 Other types of policies such as naloxone access and pain clinic laws have reported contradictory evidence. [10][11][12] For example, naloxone access laws were estimated by 1 study to reduce opioid-related fatal overdose by 0.387 per 100 000 people in 3 or more years after adoption, 10 whereas another study reported that there were no significant changes in mean opioid-related mortality but a 14% increase in the Midwest after the implementation of naloxone access laws. 11 A likely explanation for the conflicting evidence on opioid policy outcomes is methodological limitations of existing work. First, discrepant findings may be attributable to differences in data coverage, such as variation in the states and time periods included in analyses. 13,14 This sample selection issue makes it difficult to compare and synthesize existing evidence. Second, there is disagreement about how to operationalize and model the timing of policy implementation. 15 Third, the most common modeling approach, the 2-way fixed-effects model in difference-in-differences analysis (ie, controlling for both state and period indicators) has an important limitation: likely violation of the parallel trends assumption. 16,17 Because many states enacted new policies after 2013, it is urgent to conduct an updated assessment of the impact of state opioid policy using more recent data.
Herein, we present the most comprehensive study to date on state policies that target the US opioid epidemic, focusing on the consequences of policies for both prescription opioid misuse and overdose mortality. Are state drug policies significantly associated with variations in opioid misuse, opioid use disorder, and drug overdose mortality? To answer this question, we use panel matching to implement a rigorous difference-in-differences approach in conjunction with extensive data coverage that includes observations through 2018 (2007-2018; across 50 states). We also assemble and refine the policy timing data across the 6 most widely studied policies; to our knowledge, these data have never before been investigated in tandem.

Methods
This analysis draws on medical and pharmacy claims data from the Optum Clinformatics Data Mart Database (2007-2018) and a publicly available mortality data set from the National Center for Health Statistics (NCHS). The Optum database is a large deidentified database from a national private insurance provider 18-20 that includes medical and prescription claims for the full population of patients ever prescribed any controlled substance between 2007 and 2018 (approximately 23 million). eAppendix 1 and eTable 1 in the Supplement include a description of the Optum database and the population coverage by state. Excluded from the study were patients with cancer and those receiving palliative care because they are expected to be outliers with respect to (medically necessary) opioid consumption. We measured all indicators at the patient level and aggregated them to the level of state of residence. We obtained state-level overdose mortality data from the NCHS Multiple Cause of Death file from 1999 to 2018. In addition, we obtained data on several confounders (ie, the proportion of female individuals; those aged <40 years, 40-60 years, and >60 years; White, Black, Asian, and Hispanic individuals, and individuals of other races [American Indian, Pacific Islander, and multiple racial categories except for Hispanic]; those who were unemployed; those living below the poverty line; the state population; and state implementation of Medicaid expansion) from current population surveys. We aggregated individual-level data to quarterly state-level data using the provided survey weights. Herein we report results across quarterly units because we find that these units generate more precise estimates. This study was approved by the institutional review board at Indiana University, and the requirement for informed consent was waived because deidentified data were used. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Prescription Opioid Indicators
We used an extensive set of prescription opioid outcomes. First, we measured the proportion of patients who received any opioid during a given period (ie, quarter). Second, using the subset of In addition, we measured the traditional doctor-shopping indicator (ie, the proportion of patients who visit Ն4 unique doctors and Ն4 unique pharmacies for opioids within 90 days) 22 and the proportion of patients with overlapping opioid prescriptions. We also examined opioid treatment, which was defined as the proportion of patients who were prescribed any medicationassisted treatment (MAT) drug that includes buprenorphine, buprenorphine/naloxone, buprenorphine hydrochloride and naltrexone. 23 We excluded methadone from the MAT drug list because it is often used as an opioid analgesic when prescribed by primary care physicians and other

Overdose Mortality
We used ICD-10 codes to identify overdose mortality and to differentiate the cause of death. Deaths with drug overdose as the underlying cause were first identified using codes X40-X44

State Policies
We compiled a data set on 6 opioid-related policies with 2 broad objectives: to control the supply of prescription opioids or reduce harms and barriers to medical assistance for overdose. eTables 6 and 7 in the Supplement show the exact year and month of all policy implementation dates by state. The opioid-related policies include (1) PDMP access laws that provide access to the PDMP, an electronic database that tracks controlled substance prescriptions in a state; (2) mandatory PDMPs that require prescribers under certain circumstances to access the PDMP database prior to prescribing opioids; (3) prescription limit laws that impose limitations on the number of days that medical professionals dispense opioids for acute pain; (4) pain clinic laws that regulate the operation of pain clinics; (5) Good Samaritan laws that provide immunity or other legal protection for those who call for help during overdose events; and (6) naloxone access laws that provide civil or criminal immunity to licensed health care clinicians or lay responders for administration of opioid antagonists, such as naloxone hydrochloride, to reverse overdose. eAppendix 2 in the Supplement describes the information collection process for these state policies. Based on the dates of these policies, we defined treatment indicators, which were assigned a value of 1 if the law was active in a given quarter, otherwise a value of 0 was assigned. In addition, we created a separate indicator for state implementation of Medicaid expansion that we obtained from the Kaiser Family Foundation to control for its potential impact on the statistical inferences.

Statistical Analysis
To ascertain whether state policy altered the prevalence of opioid abuse and misuse indicators, we used a difference-in-differences approach. A state was considered to be a treated case if the policy had been changed at a specified time, otherwise it was considered to be an untreated case. The goal was to compare the outcomes of interest for a state under the new policy regime at a specific time with the outcome under the old policy regime if the policy was not enacted at the same time. The key challenge was to find a suitable control case for the treated case. The difference-in-differences approach imputed the change of outcomes in the control case as a comparison case for the change of outcomes in the treatment case under the parallel trends assumption (ie, potential outcomes have

Discussion
Recent trends in the US opioid epidemic present a paradox: opioid overdose mortality has continued to increase despite declines in opioid prescriptions since 2012. 28, 29 The opioid paradox may arise from the success-not failure-of state interventions to control opioid prescriptions. This finding is supported by a comprehensive assessment of multiple opioid policies on a range of outcomes, including opioid misuse and overdose mortality with extensive data coverage. We found that supplycontrolling policies were associated with a reduction in the amount of prescription opioid misuse and the number of overdose deaths attributable to natural opioids as well as an increase in the number of patients receiving MAT drugs. In tandem, the significant increase of overdose deaths from synthetic opioids, heroin, and cocaine after the enactment of PDMP access, pain clinic laws, and naloxone access laws suggests that current drug policies may have the unintended consequence of motivating opioid users to switch to illicit drugs. An important implication of our findings is that there is no easy policy solution to reverse the epidemic of opioid dependence and mortality in the US.
Hence, to resolve the opioid paradox, it is imperative to design policies to address the fundamental causes of overdose deaths (eg, lack of economic opportunity, persistent physical, and mental pain) and enhance treatment for drug dependence and overdose rather than focusing on opioid analgesic agents as the cause of harm. 2,30 Prescription drug monitoring programs are the most widely studied policy responses to the opioid epidemic. Previous research on their impact indicates that providing access to PDMPs is not associated with significant improvement, but PDMPs have reduced prescription opioid misuse when accessing the databases was required for physicians. 8,22,31 The present study found that mandatory PDMPs also reduced opioid misuse in a commercially insured population. In addition, we found that prescription limit laws and pain clinic laws were associated with a reduction in opioid abuse and an increase in the proportion of patients receiving MAT drugs. A study found that these laws as designed significantly reduce the length of initial prescription, although they also increase the likelihood of new (ie, first time) opioid use and the strength of initial prescription. 32 In addition, pain clinic laws have been associated with modest decreases in opioid prescribing in Florida and Texas. 33,34 The results of the present study are broadly consistent with those of other studies.
This extensive analysis may settle some of the contradictory findings in the literature and contributes to the previous research on opioid policy outcomes. Previous research on naloxone access and Good Samaritan laws has yielded inconsistent results. Namely, the enactment of naloxone access laws has been associated with substantial reductions in a fatal overdose but increased nonfatal overdoses. 10,12 These results, however, have been contradicted by other studies that suggested that expansion of naloxone access laws leads to more opioid-related emergency department visits and thefts without any substantial reductions in opioid mortality. 11,35 Likewise, previous research has suggested that Good Samaritan laws are not associated with heroin-related mortality, but substantially reduce mortality from other opioids. 12 The situation is similar to studies examining the association between PDMPs and overdose mortality. 6,7,13,14 Differences between our results and those of previous studies may be explained by our rigorous and extensive analytic approach, which uses panel matching for difference-in-differences analysis to mitigate the violation of the parallel trends assumption and the use of more recent data while simultaneously examining temporal associations. Given that most state policies were enacted after 2013, the present study used data through 2018 to provide the most up-to-date evidence on the opioid policy landscape.

JAMA Network Open | Health Policy
We believe that our findings on the role of harm reduction policies in accelerating drug overdose deaths, especially those attributable to synthetic opioids and heroin, have important implications.
Good Samaritan laws are designed to remove the threat of liability for people who call for emergency assistance in the event of a drug overdose. It is theoretically possible that provision of immunity may lead to greater reporting of overdose events in the absence of actual increases, although it is less likely to explain the increase of overdose deaths after the enactment of a Good Samaritan law.
Naloxone laws provide civil immunity to licensed health care professionals or lay responders for opioid antagonist administration. Although expanded access to naloxone can reverse an opioid overdose and save lives, we found that naloxone access laws were associated with a substantial increase rather than a decrease in overdose deaths, especially deaths from illicit drugs. It is possible that the prospect of getting access to overdose-reversing treatment may instead induce moral hazards by encouraging people to use opioids and other drugs in riskier ways than they would have without the safety net of naloxone. 11 Although naloxone access laws were estimated to increase naloxone dispensing, 36 a recent study found that only 2135 of 138 108 high-risk patients (1.5%) in the US were prescribed naloxone in 2016. 37 This finding suggests that policies designed to dramatically improve treatment for overdose are needed.

Limitations
Several important limitations of this study are worth noting. First, a difference-in-differences approach through panel matching cannot account for spillover effects between states and between policies. 26 Although we believe that considering multiple types of drug policies simultaneously is important, teasing out the association between a single policy and outcome from those of other policies is difficult because of correlated policy responses. Likewise, because prior research suggests that states' adoption of Medicaid expansion has led to an increase in opioid overdose-related mortality 38 but a decrease in opioid-related hospital use, 39 we accounted for this factor by adjusting for Medicaid expansion in the modeling framework. However, identifying policy consequences on outcomes while controlling for the adoption of Medicaid expansion may produce underestimates or overestimates given the policy response lag and overlaps among different policy responses. Future research may consider, for example, sequence analysis or other clustering methods combined with the difference-in-differences approach to examine the impact of correlated policy responses.
Second, because we draw on data from a commercially insured population, these findings may not extend to other populations or to those individuals with insurance who pay cash for opioid prescriptions. Opioid misuse and overdose rates are higher among Medicare beneficiaries in our patient population with a low-income subsidy (eTable 4 in the Supplement), which suggests that this analysis may be missing a substantial proportion of the population at risk for opioid problems.
Although the results on opioid misuse are consistent with those of an earlier report on mandatory PDMPs up to 2013 using a Medicare part D sample, 8