Spatiotemporal Analysis of the Association Between Pain Management Clinic Laws and Opioid Prescribing and Overdose Deaths

Abstract

Pain management clinic (PMC) laws were enacted by 12 states to promote appropriate opioid prescribing, but their impact is inadequately understood. We analyzed county-level opioid overdose deaths (National Vital Statistics System) and patients filling long-duration (≥30 day) or high-dose (≥90 morphine milligram equivalents per day) opioid prescriptions (IQVIA, Inc.) in the United States in 2010–2018. We fitted Besag-York-Mollié spatiotemporal models to estimate annual relative rates (RRs) of overdose and prevalence ratios (PRs) of high-risk prescribing associated with any PMC law and 3 provisions: payment restrictions, site inspections, and criminal penalties. Laws with criminal penalties were significantly associated with reduced PRs of long-duration and high-dose opioid prescriptions (adjusted PR = 0.82, 95% credible interval (CrI): 0.82, 0.82, and adjusted PR = 0.73, 95% CI: 0.73, 0.74 respectively) and reduced RRs of total and natural/semisynthetic opioid overdoses (adjusted RR = 0.86, 95% CrI: 0.80, 0.92, and adjusted RR = 0.84, and 95% CrI: 0.77, 0.92, respectively). Conversely, PMC laws were associated with increased relative rates of synthetic opioid and heroin overdose deaths, especially criminal penalties (adjusted RR = 1.83, 95% CrI: 1.59, 2.11, and adjusted RR = 2.59, 95% CrI: 2.22, 3.02, respectively). Findings suggest that laws with criminal penalties were associated with intended reductions in high-risk opioid prescribing and some opioid overdoses but raise concerns regarding unintended consequences on heroin/synthetic overdoses.

Abbreviations:

 
• CrI

  credible interval

 
• MME

  morphine milligram equivalents

 
• PMC

  pain management clinic

 
• PR

  prevalence ratio

 
• RR

  relative rate

The United States is in the midst of a drug overdose epidemic. The drug overdose death rate in 2018 was 3.6 times higher than in 1999 (1) and almost 70% of drug overdose deaths in 2018 involved an opioid (2). While the bulk of opioid overdoses in 2018 involved nonmethadone synthetic opioids (e.g., fentanyl) or heroin, almost a third involved prescription opioids (2). Further, individuals often use prescription opioids before they initiate heroin or illegally manufactured synthetic opioids, making prevention of prescription opioid misuse a priority to reduce all opioid overdoses (3, 4).

One response to reducing prescription opioid misuse has focused on reducing inappropriate, high-volume opioid prescribing without clear medical need in certain pain management clinics (PMCs), which were a major source of opioids diverted into the illegal drug supply (5–7). PMC laws were enacted in 12 states (8), imposing operational, personnel, inspection, and other requirements on facilities that specialize in pain treatment (9, 10). Laws vary between states, with different provisions including certification, training, physician ownership, on-site medical directors, inspections, payment restrictions, and civil and criminal penalties. While greater oversight of PMCs may reduce inappropriate prescribing and diversion of prescription opioids, it is also postulated that by restricting the availability of prescription opioids in the secondary market, these policies may inadvertently lead some people who use prescription opioids to transition to illicit opioids, resulting in more overdoses from heroin and illegally manufactured synthetic opioids (4). To date, evidence on the impact of PMC laws on opioid overdose deaths is mixed. Two national studies reported a reduction in prescription opioid overdoses but not in heroin overdose deaths following enactment of PMC laws (11, 12). PMC enactment was associated with a reduction in both prescription opioid and heroin overdose deaths in Florida (13), but no change was found in Tennessee and Ohio (14).

Three factors may contribute to inconsistent findings. First, treatment of PMC laws as binary variables without considering variation in their provisions may explain differences in their impact across states. Second, previous studies did not account for within-state variation in opioid overdose over time and in the distribution of demographic characteristics that could affect opioid overdose, raising concerns about aggregation bias (15). Third, prior work considered the immediate timeframe after PMC law enactment, potentially missing lagged cumulative estimates of effect that may emerge over time. Notably, most prior work considered PMCs through 2012, missing potential impacts in the current phase of the overdose epidemic where synthetic opioid overdoses predominate.

Our study addressed these gaps. First, we examined the impact of PMC laws on annual long-duration (≥30 day) and high-dose (≥90 milligram equivalents (MME) per day) opioid prescriptions and opioid overdose deaths, characterizing PMC laws 3 ways: 1) enactment of any PMC law; 2) enactment of 3 specific provisions in PMC laws: restrictions on payment, authorization of inspections, and criminal penalties for violations of the law; and 3) number of provisions adopted in the PMC law (range 0–3). Second, we adopted a hierarchical Bayesian spatiotemporal approach (16–19) to examine the impact of state-level PMC laws on county-level opioid prescribing and fatal opioid overdoses, to account for within-state variation in the level and growth rate and spatial autocorrelation in opioid prescribing and overdose deaths across counties and states. Third, we examined the association between PMC laws and opioid outcomes with 0- to 3-year cumulative lags from implementation, using the most up-to-date prescription, mortality, and legislative data available through 2018.

METHODS

Outcome measures

We used the National Vital Statistics System’s restricted-use detailed multiple cause-of-death (MCOD) mortality “All Counties” files to tabulate opioid-related overdose deaths in 2010–2018 (20). This timeframe was selected to focus on the most recent phase of the epidemic and because only 1 state (Louisiana) implemented a PMC law before 2010. Codes from the International Classification of Diseases, Tenth Revision, were used to identify drug overdose deaths, including X40–44, X60–64, X85, and Y10–14 as underlying causes. Opioid overdose deaths were defined using multiple cause-of-death codes T40.0–T40.4 and T40.6, based on national poisoning surveillance standards for opioid overdose (21). Annual counts of overdose deaths were aggregated by county of death (n = 3,109, excluding Alaska due to boundary changes) and year of death.

Prescribing data were obtained from the 2010–2018 IQVIA (IQVIA Inc., Durham, North Carolina) longitudinal prescription database, a repository containing over 85% of US retail pharmacy sales (22). Data were restricted to opioid prescriptions, excluding cough and cold formulations and medications for opioid use disorder. Transactions were limited to retail pharmacies only. We aggregated annual counts of unique patients by county of the dispensing pharmacy for 2 measures of high-risk opioid prescriptions: patients receiving prescriptions for ≥30 days (long-duration prescriptions) and patients receiving doses of ≥90 morphine MME per day (high-dose prescriptions). Counties with 3 or fewer pharmacies were suppressed in the IQVIA data, resulting in a subset of 2,609 counties included in the prescribing analyses (see Web Table 1, available at https://doi.org/10.1093/aje/kwab192).

Pain management clinic law measures

Data on the characteristics of PMC laws from January 1, 2010, through June 30, 2018, were obtained from the Prescription Drug Abuse Policy System (PDAPS) (8). Three provisions were selected a priori because: 1) they reflected the important domains of increased professionalism (restrictions on payments) and clinic accountability (authorization of inspections; criminal penalties for violations of PMC laws); and 2) heterogeneity in their implementation across different states over time allowed for analyses of specific provisions. For each law and provision, we calculated the proportion of each year in which it was in effect (e.g., if a state enacted a PMC law effective July 2014, it was coded as 0 for all years prior to 2014, 1 for all years after 2014, and 0.5 for 2014). We measured the proportion of the year a state had: 1) any PMC law; 2) a PMC law with payment-type restrictions; 3) a PMC law authorizing inspection of facilities; or 4) a PMC law authorizing criminal penalties for violations of the law. We also modeled the combination of these 3 provisions as an index, with variables representing the proportion of the year with 0, 1, 2, or all 3 characteristics in effect, versus no PMC law.

Covariates

We accounted for the implementation of other opioid laws that could have affected opioid overdose rates. These included operationalization status of electronic prescription drug monitoring programs (PDMPs) and, specifically, provisions requiring providers to check PDMPs before prescribing opioids (23), mandatory limits on initial prescriptions of opioids for acute pain (24), medical cannabis laws (25–27), overdose Good Samaritan laws, and naloxone access laws (28, 29). Based on prior studies of opioid prescribing laws (30–32), we also selected the following county-level annual estimates of demographic characteristics, obtained from GeoLytics Inc. (Branchburg, New Jersey) (33), as covariates: population density (thousands of people/square mile); age composition (% of the population aged 0–19, 20–44, 45–64, and ≥65 years); racial/ethnic composition (% non-Hispanic White, non-Hispanic Black, Hispanic); % male; and socioeconomic conditions (% of families in poverty, median household income, % unemployed). We also accounted for the overall mortality rate per 1,000 residents in the county.

Spatiotemporal models

Utilizing an extended Besag-York-Mollié model specification (see Web Appendix 1) (16–19), we estimated county-by-year prevalence ratios of patients filling: 1) long-duration and 2) high-dose opioid prescriptions, and relative rates of 3) any opioid overdose, 4) natural and semisynthetic opioid overdose other than heroin, 5) synthetic opioid overdose other than methadone, and 6) heroin overdose. Using “no PMC laws” as the reference group, the exposures were: 1) any PMC laws, 2) PMC laws with payment restrictions, 3) PMC laws authorizing facility inspections, 4) PMC laws with criminal penalties, and 5) the index of PMC law provisions, controlling for demographic characteristics, overall mortality rates, and co-occurring prescription opioid and harm reduction laws. We used linear distributed lags to estimate concurrent policy impacts as well as impacts over the 3 subsequent years after implementation of PMC laws to establish temporal order between the law and outcomes.

Our analytical approach was similar to a difference-in-difference approach. It used state-level fixed effects to account for stable differences between states that did and did not enact PMC laws; it assumed (by fitting a county-level random intercept) that each county could start out at a different level of overdose; and it accounted for nonlinear secular trends common to all states by fitting year dummy variables. However, our approach also provides a separate time trend (i.e., random slope) for each county, thus allowing us to obtain unbiased estimates in the context of heterogeneous growth between counties over time, and avoiding biases due to over- and underdifferencing (34). Further, by incorporating a conditional autoregressive (CAR) spatial random effect, our models accounted for the lack of independence of overdose counts in spatially contiguous counties and avoided biases due to small area effects (17). The Poisson-specified CAR model also included corrections for overdispersion, similar to a negative binomial specification (35). All analyses were performed using R (R Foundation for Statistical Computing, Vienna, Austria) (36) with the R-INLA (R-INLA Project) package (37). We estimated posterior means and 95% credible intervals (CrIs) for the PMC law predictor variables and covariates (38). Exponentiated estimates are interpreted as prevalence ratios (PRs) for prescribing outcomes and relative rates (RRs) for mortality outcomes.

Sensitivity analyses

We conducted 3 types of sensitivity analyses. First, to assess the magnitude of the associations of outcomes with PMC laws before and after adjustment for other laws and policies, we removed control variables for other opioid laws in the total opioid mortality model. Second, we replicated our main analyses excluding states that a previous study had found to substantially underestimate opioid mortality rates (Alabama, Indiana, Louisiana, and Pennsylvania) (39, 40). Third, we modeled total overdose death rates due to all substances as an outcome measure, to address concerns about variation over time and across states in death certificate coding of specific drugs implicated in drug overdose deaths.

The study protocol was reviewed and approved by the New York University Langone Health Institutional Review Board.

RESULTS

Descriptive statistics

Twelve states had a PMC law in effect as of June 1, 2018 (Table 1). Eleven states enacted at least 1 of the 3 provisions of interest to our study (restriction of source or form of payment, authorization of inspections, and/or authorization of criminal penalties). Three states enacted 2 of the 3 provisions (Florida, Georgia, Texas), and 2 states had all 3 provisions in effect as of June 1, 2018 (Kentucky, Tennessee).

Counties in the prescribing (n = 2,609) and the mortality (n = 3,109) analyses represent 83% and 99% of United States counties, respectively. Table 2 shows mean county demographic characteristics for covariates used in our models, as well as average prescribing and mortality outcomes, stratified by PMC law status. Counties in states that implemented PMC laws tended to have lower median incomes, higher rates of poverty, lower population density, more residents who identified as Black or Hispanic, and higher rates of high-risk prescribing and overdose compared with counties in states that did not implement PMC laws (Table 2). Counties suppressed in the prescribing data (n = 500) tended to have lower rates of overdose, lower population densities, younger age composition, and more poverty (Web Table 1).

An annual average of 72 patients per 1,000 population filled high-dose prescriptions (≥90 MME per day), with a net decrease of 32 patients per 1,000 from 2010–2018, while 37 patients per 1,000 population filled long-duration opioid prescriptions (≥30 days), with little net change (Figure 1A). On average, 13.1 total overdose deaths per 100,000 county population occurred annually in 2010–2018, with 7.6 per 100,000 involving any opioids, 3.8 per 100,000 involving natural and semisynthetic opioids such as oxycodone and hydrocodone, 2.5 per 100,000 involving synthetic opioids such as fentanyl, and 1.5 per 100,000 involving heroin. Opioid overdose mortality rates generally increased over the time period, with some decline in deaths coded as natural/semisynthetic opioid overdoses (e.g., hydrocodone, oxycodone) after 2016, while heroin and especially synthetic opioid (e.g., fentanyl) deaths continued to increase (Figure 1B).

Prescription models

Adoption of any state PMC law was associated with reduced PRs of long-duration opioid prescribing, from 4% in the year of implementation (adjusted PR = 0.96, 95% CrI: 0.96, 0.96) to 8% 3 years after implementation (adjusted PR = 0.92, 95% CrI: 0.91, 0.92) (Figure 2A, Web Table 2). Similar trends were found for the association between any state PMC law and high-dose opioid prescribing (year of implementation, adjusted PR = 0.97, 95% CrI: 0.97, 0.98; 3 years after, adjusted PR = 0.93, 95% CrI: 0.92, 0.93) (Figure 2E). Of the 3 PMC provisions examined, PMC laws with criminal penalties had the strongest association with opioid prescribing. Controlling for other opioid laws and county demographic characteristics, PMC laws with criminal penalties were associated with reductions in the prevalence ratios of long-duration opioid prescribing ranging from 4% (adjusted PR = 0.96, 95% CrI: 0.96, 0.96) in the same year to 18% in the third year of implementation (adjusted PR = 0.82, 95% CrI: 0.82, 0.82) (Figure 2D). Prevalence ratios of high-MME opioid prescribing were likewise reduced by 5% in the year of implementation (adjusted PR = 0.95, 95% CrI: 0.95, 0.95) to 27% 3 years later (adjusted PR = 0.73, 95% CrI: 0.73, 0.74) (Figure 2H).

We observed a dose-response relationship in the number of provisions implemented and reduced PRs of patients filling long-duration prescriptions, which by the third year of implementation ranged from reductions of 5% (adjusted PR = 0.95, 95% CrI: 0.94, 0.95) for PMC laws with none of the 3 provisions to 25% (adjusted PR = 0.75, 95% CrI: 0.75, 0.76) for laws with all 3 key provisions (Figure 3A–D). The same dose response was observed for patients receiving high-MME opioid prescriptions, reaching a maximum reduction of 37% in year 3 for all 3 provisions (Figure 3E–H).

Overdose models

Adoption of any state PMC law was not associated with changes in county-level relative rates of total opioid overdose deaths (3 years after implementation, adjusted RR = 1.02, 95% CrI: 0.97, 1.07) (Figure 4A, Web Table 3) or overdose deaths involving natural and semisynthetic opioids (adjusted RR = 0.94, 95% CrI: 0.88, 1.01) (Figure 4E). Natural and semisynthetic opioids (e.g., oxycodone, hydrocodone, oxymorphone, hydromorphone) are the types of opioids typically prescribed by PMCs. However, PMC laws with criminal penalties were associated with 14% reductions in the relative rate of total opioid overdose mortality (adjusted RR = 0.86, 95% CrI: 0.80, 0.92) (Figure 4D) and 16% for natural and semisynthetic opioids (adjusted RR = 0.84, 95% CrI: 0.77, 0.92) in the third year of implementation (Figure 4H). Conversely, the relative rate of overdose involving synthetic opioids other than methadone (e.g., fentanyl, meperidine) increased significantly after implementation of all categories of PMC laws, doubling by year 3 of implementation for all PMC models (Figure 4I–L, Web Table 3). Similarly, the relative rate of heroin overdose increased following implementation of any PMC laws, especially for those with criminal penalties, ranging from an adjusted RR of 1.41 (95% CrI: 1.24, 1.59) in the year of implementation to 2.59 (95% CrI: 2.22, 3.02) in year 3 (Figure 4M–P).

The strength of the association was greatest for PMC laws with all 3 provisions, reaching a 19% (adjusted RR = 0.81, 95% CrI: 0.72, 0.92) and 42% (adjusted RR = 0.58, 95% CrI: 0.50, 0.68) reduction, respectively, in the relative rates of total opioid (Figure 5A–D) and natural/semisynthetic opioid (Figure 5E–H) overdose mortality in the third year of implementation. There was inconsistent evidence of a dose-response relationship with increasing number of PMC law provisions on synthetic (Figure 5I–L) and heroin deaths (Figure 5M–P). Results were robust in the 3 sensitivity analyses (Web Table 4), with similar reductions in the relative rates of opioid overdoses associated with criminal penalties in models that held out other policy covariables (adjusted RR = 0.77, 95% CrI: 0.72, 0.82), when excluding states found in prior studies to have differences in reported and imputed opioid overdose rates (37, 38) (adjusted RR = 0.80, 95% CrI: 0.77, 0.84), and for overdose deaths from all substances (adjusted RR = 0.90, 95% CrI: 0.86, 0.94).

DISCUSSION

Our study found that the type of PMC law enacted matters. In the simple binary metric of any PMC laws, we observed small declines in the prevalence ratios of high-dose and long-duration opioid prescribing but no overall association with total opioid overdoses. However, we found substantial declines reaching 16%–27% by the third year of implementation in the prevalence ratios of county-level long-duration and high-dose opioid prescribing and in natural/semisynthetic opioid overdose deaths in the 6 states that enacted PMC laws with criminal penalties for noncompliance. Larger declines in PRs of long-duration and high-dose prescribing and RRs of natural/semisynthetic overdose, ranging from 24% to 42%, respectively, by year 3, were found in states that enacted all 3 of the PMC provisions (Kentucky and Tennessee). Such findings suggest that criminal penalties for PMC law violations may have been necessary to reduce harmful prescribing practices and high-volume diversion of prescription opioids (10), consistent with prior evaluations of PMC laws finding decreased opioid prescribing (41–43), diversion of prescription opioids (44), and prescription opioid overdoses (11–13). The mechanisms through which such laws affect opioid prescribing and overdose rate deserve further investigation, as PMC laws both influence the management of pain clinics and provide a deterrent signal to prescribers across clinical settings.

Meanwhile PMC laws were associated with increased relative rates of heroin and synthetic opioid overdose deaths, raising concerns about unintended consequences. Targeting PMCs that served as sources of prescription opioid diversion to the illegal market and spillover effects on broader opioid prescribing patterns may have contributed to the demand for heroin and illegally manufactured synthetics (10). Our findings differ from studies that found no association between PMC laws on heroin overdose (12, 13, 15) or a potential decrease in heroin overdose (13). These studies focused on an earlier phase of the opioid overdose epidemic when the rates of overdoses from heroin and illegally manufactured synthetics were lower and when many states had not yet enacted PMC laws. The net impact of PMC provisions on overdoses may vary over time and with characteristics of the opioid epidemic in a particular state.

Study findings are also consistent with research on impacts of other types of policy approaches to regulate opioid prescribing. For example, prior research found that more robust prescription drug monitoring programs were consistently associated with reduced relative rates of overdoses from natural/semisynthetic opioids but increased relative rates of overdoses from synthetic opioids and sometimes heroin (45). Taken together, the findings point to potential unintended consequences of laws and policies that regulate the prescription opioid supply and strongly suggest that attempts to reduce inappropriate opioid prescribing be coupled with targeted outreach to connect individuals receiving prescription opioids with pain medicine specialists, opioid-use-disorder treatment, or risk-reduction resources as appropriate for that individual.

Limitations are noted. First, use of International Classification of Diseases, Tenth Revision, codes may misclassify overdose deaths by specific drugs (46, 47). In particular, prior studies have found that the quality of ascertainment of specific drug involvement in overdose deaths varies across states and over time, raising concerns about differential misclassification by state PMC law enactment (40). However, findings were robust to sensitivity analyses conducted to address this concern. Future work incorporating contributing causes of death into overdose estimation is important. Second, we focus on the impact of PMC laws on fatal opioid overdose. Future studies should consider the impact of PMC laws on a range of outcomes related to opioid-related harm. Third, future research should consider patterns of use of prescribed and street opioids at the time of enactment. Interventions in states where rogue pain clinics proliferated in a less-regulated era may not have the same impact in different periods and contexts. Fourth, we cannot rule out the possibility of interference, or spillover effects across states after the implementation of opioid laws in other states. Fifth, concerns about type I error arise when the number of states enacting a policy is small; however, Poisson models with corrections for overdispersion perform well in terms of producing type I error rates near 0.05 (48). Finally, as in any observational study, we could not control completely for all potential confounders that could explain observed changes.

In conclusion, state adoption of stricter PMC laws was associated with greater reductions in the prevalence ratios of patients filling high-dose and long-duration opioid prescriptions. Decreases in the relative rates of deaths related to natural/semisynthetic opioids (e.g., oxycodone, hydrocodone), the specific targets of these laws, were also observed after implementation of PMC laws with critical provisions, especially criminal penalties for failure to comply with the law. However, increases in relative rates of heroin and synthetic opioid overdose deaths following implementation of PMC laws raise substantial concerns about negative secondary consequences and raise questions about the net benefits of such laws. These increases also highlight the need to couple measures to restrict the prescription opioid supply with policies that reduce demand for opioids, including by increasing access to evidence-based treatment for opioid use disorder and nonopioid therapy for individuals with pain, as well as by addressing structural drivers of opioid use, such as socioeconomic deprivation and income inequality.

ACKNOWLEDGMENTS

Author affiliations: Center for Opioid Epidemiology and Policy, Department of Population Health, Grossman School of Medicine, New York University, New York, New York, United States (Magdalena Cerdá, Katherine Wheeler-Martin); Department of Epidemiology, Columbia University Mailman School of Public Health, New York, New York, United States (Emilie Bruzelius, Katherine Keyes, Deborah Hasin, Kara E. Rudolph, Silvia S. Martins); Prevention Research Center, Pacific Institute of Research and Evaluation, Berkeley, California, United States (William Ponicki, Paul Gruenewald); Department of Biostatistics, Columbia University Mailman School of Public Health, New York, New York, United States (Christine Mauro); Institute for Health, Health Care Policy, and Aging Research, Rutgers University, New Brunswick, New Jersey, United States (Stephen Crystal); and Network for Public Health Law, Los Angeles, California, United States (Corey S. Davis).

This work was supported by the National Institute on Drug Abuse (grants R01DA048572 to M.C. and S.S.M., R01DA048860 to D.H., and R01DA047347 to S.C.), the National Center for Advancing Translational Science (grant UL1TR003017 to S.C.), the Agency for Healthcare Research and Quality (grant R18HS023258 to S.C.), and the New Jersey Health Foundation and the Foundation for Opioid Response Efforts, which also supported S.C.

Data availability: Restricted-use mortality data (National Vital Statistics System) and opioid prescribing data (IQVIA, Inc.) may not be re-released.

K.K. has testified as an epidemiology expert witness in opioid litigation. D.H. receives funding from Syneos Health to support measurement and quality control for a Food and Drug Administration–mandated study of the risks of opioid addiction among patients prescribed opioids for pain.

REFERENCES