- Journal List
- Am J Epidemiol
- PMC7443765
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsem*nt of, or agreement with, the contents by NLM or the National Institutes of Health.
Learn more: PMC Disclaimer | PMC Copyright Notice
Am J Epidemiol. 2020 Sep; 189(9): 885–893.
Published online 2020 Feb 20. doi:10.1093/aje/kwaa015
PMCID: PMC7443765
PMID: 32077469
Kenneth A Feder, Ramin Mojtabai, Elizabeth A Stuart, Rashelle Musci, and Elizabeth J Letourneau
Author information Article notes Copyright and License information PMC Disclaimer
Abstract
In 2011, Florida established a prescription drug monitoring program and adopted new regulations for independent pain-management clinics. We examined the association of those reforms with drug overdose deaths and other injury fatalities. Florida’s postreform monthly mortality rates—for drug-involved deaths, motor vehicle crashes, and suicide by means other than poisoning—were compared with a counterfactual estimate of what those rates would have been absent reform. The counterfactual was estimated using a Bayesian structural time-series model based on mortality trends in similar states. By December 2013, drug overdose deaths were down 17% (95% credible interval: −21, −12), motor vehicle crash deaths were down 9% (95% credible interval: −14, −4), and suicide deaths were unchanged compared with what would be expected in the absence of reform. Florida’s opioid prescribing reform substantially reduced drug overdose deaths. Reforms may also have reduced motor vehicle crash deaths but were not associated with a change in suicides. More research is needed to understand these patterns. Bayesian structural time-series modeling is a promising new approach to interrupted time-series studies.
Keywords: drug policy, interrupted time series, mortality, opioids
Abbreviations
- BSTS
- Bayesian structural time series
- CDC
- Centers for Disease Control and Prevention
- CrI
- credible interval
- PDMP
- prescription drug monitoring program
Editor’s note:
An invited commentary on this article appears on page 894, and the authors’ response appears on page 898.
In 2011, the state of Florida adopted a number of reforms to reduce irresponsible prescribing of opioid pain relievers. We examined whether those reforms were associated with reductions in 3 sources of mortality—drug overdose, motor vehicle crash, and suicide—that could be linked to opioid access. This question has important implications for other states seeking to combat opioid-related mortality through policy.
The opioid epidemic
Opioid-related problems have reached epidemic proportions in the United States (1). Annual numbers of drug overdose deaths now exceed 70,000 (2)—more than deaths due to motor vehicle accidents and firearms (3).
The roots of the present crisis began in the 1990s, when pharmaceutical companies and professional societies began to advocate for more aggressive long-term management of chronic, noncancer pain with opioid pain relievers (1). As prescriptions increased, overdose deaths increased nearly in parallel (1). Two subsequent upticks in mortality were linked to this initial prescription opioid problem: the first when many adults who were misusing prescription opioids switched to heroin in the early 2010s (4) and the second in recent years when extremely potent synthetic opioids like fentanyl entered the illicit drug supply (5).
Addressing opioid overdose through policy—the example of Florida
Opioid overdose deaths in Florida consistently exceeded the national average from the 1990s to the late 2000s (6). Identifying problematic opioid prescribing as a possible driver of these high overdose rates, in 2010–2011, Florida adopted a number of reforms to try to reduce prescription drug-related mortality.
First, Florida’s legislature authorized the creation of a prescription drug monitoring program (PDMP). All dispensers of controlled substances were required to check the PDMP to review each patient’s prescription history before dispensing a controlled substance and to log each prescription made in the PDMP. The law also allowed certain investigators from law enforcement and health agencies to access the PDMP (7).
Second, Florida’s legislature officially defined “pain-management clinics” to be programs that either advertised themselves as such or had a majority of their patients receiving pain medication. Florida required these programs to register with the state. Then, beginning July 2011, Florida adopted a “pill mill” law. This law required physician ownership of pain-management clinics, prohibited those clinics from operating onsite pharmacies, and permitted opioids to be prescribed only if the prescription was accompanied by a medical examination and follow-up care (7).
Primary impacts of Florida policy on opioid outcomes.
Following the adoption of these reforms:
More than 500 of Florida’s 900 independent pain-management clinics closed (7).
Numbers of opioid prescriptions fell in comparison with other, similar states, with the largest reductions being seen among physicians making the highest volume of prescriptions and patients receiving the highest volume of prescriptions (8).
Numbers of oxycodone overdose deaths fell sharply (9), even as they continued to increase in nearby North Carolina (10).
Taken together, these studies suggest that Florida’s policy changes succeeded in achieving their primary objective—preventing unsafe opioid prescribing and overdose.
Secondary impacts of Florida policy on other injury deaths.
While overdose deaths have been the subject of past research, the impacts of Florida’s opioid prescribing reforms may not end at overdose deaths.
Motor vehicle crash deaths. First, intoxication, primarily with alcohol, is a major cause of motor vehicle crash deaths in the United States (11). Opioid use can induce drowsiness and impair cognitive function (12), which could increase risk for car crashes. However, while opioid use is associated with unsafe driving behavior (13), at least 1 literature review found no evidence that opioid use was associated with motor vehicle crashes (14). Findings about whether motor vehicle crash deaths involving opioids have increased during the present epidemic are also mixed (15, 16).
Suicide deaths. Further, some critics of restrictions on opioid prescribing argue that crackdowns on opioid prescribing may lead to poor management of chronic pain and, in some cases, increased risk of suicide (17). Chronic pain is associated with increased risks of suicide attempts and suicide completion (18), and much higher rates of suicidal ideation and attempt have been found among veterans whose physicians terminated their prescription opioid use (19).
Study question and hypotheses
In summary, current research on the effects of Florida’s opioid prescribing reforms on drug overdose are limited by the challenge of selecting adequate controls. Further, no studies (to our knowledge) have examined the possible influence of Florida’s opioid prescribing reforms on important secondary sources of mortality—motor vehicle crashes and suicides. This study addresses these important gaps by examining whether the introduction of Florida’s PDMP and pill mill laws were associated with changes in these 3 causes of mortality—drug overdose, motor vehicle crash, and suicide—using a novel machine learning approach that combines information from 17 different control states.
METHODS
Data
Monthly mortality counts were extracted from publicly available data for Florida and all comparison states through the Centers for Disease Control and Prevention’s (CDC’s) online WONDER database (https://wonder.cdc.gov/) (20). Drug overdose death included all deaths for which the underlying cause of death was determined to be drug-related. Note that this could include accidental overdoses, suicides, or homicides. Suicide death included all injury-related deaths for which the injury intent was determined to be suicide, excluding deaths where the mechanism of injury was poisoning. We excluded poisoning deaths in order to distinguish this outcome from the drug overdose deaths outcome and to isolate suicide deaths that were not caused by drug overdose (however, see the “Sensitivity analysis” section below). Motor vehicle crash, as well as the 2 control outcomes (see the “Model-checking and inference” subsection), deaths from major cardiovascular disease and malignant neoplasm, each included all deaths in the corresponding “113 Causes of Death” type (21). To compute monthly mortality rates, we divided these counts by annual average population totals taken from US Census intercensal estimates of the population of each state (22).
Study sample
The time period for this study was January 2005 through December 2013. Twenty-three states were selected as possible comparison states because they already had some type of PDMP law in place as early as 2005. These states were identified using the LawAtlas Project of the Policy Surveillance Program at Temple University Beasley School of Law (23). Restricting comparison states to states with a PDMP in 2005 was necessary because, in order to serve as comparisons, states could not have made a policy change similar to that of Florida’s. (We considered using states that did not have a PDMP for the entire observation period as comparison states; it turned out that there were no states that met this criterion.)
The CDC suppresses the monthly mortality count for each state reporting fewer than 10 deaths in that month. For this reason, in the analysis of each cause of death, some of the 23 eligible comparison states were not actually included as comparisons because at least 1 of their mortality totals was suppressed. The exact comparison states included in each analysis are shown in Table 1. Every analysis had at least 17 comparison states.
Table 1
Comparison States Included in Analyses of Florida’s Opioid Prescribing Crackdown, by Cause of Death, January 2005–December 2013
| State | Cause of Death | ||||
|---|---|---|---|---|---|
| DrugOverdose | Motor VehicleCrash | Suicide | Major CVD | MalignantNeoplasm | |
| Alabama | Xa | X | X | X | X |
| California | X | X | X | X | X |
| Colorado | X | X | X | X | X |
| Florida | X | X | X | X | X |
| Hawaii | X | X | |||
| Idaho | X | X | |||
| Illinois | X | X | X | X | X |
| Indiana | X | X | X | X | X |
| Kentucky | X | X | X | X | X |
| Maine | X | X | |||
| Massachusetts | X | X | X | X | X |
| Michigan | X | X | X | X | X |
| Nevada | X | X | X | X | X |
| New Mexico | X | X | X | X | |
| New York | X | X | X | X | X |
| Ohio | X | X | X | X | X |
| Oklahoma | X | X | X | X | X |
| Pennsylvania | X | X | X | X | X |
| Rhode Island | X | X | |||
| Texas | X | X | X | X | X |
| Utah | X | X | X | X | |
| Virginia | X | X | X | X | X |
| West Virginia | X | X | X | X | |
| Wyoming | X | X | |||
Open in a separate window
Abbreviation: CVD, cardiovascular disease.
a “X” indicates that the state was included in the analysis.
Analytical approach
Motivation.
The goal of the analysis was to determine how different Florida’s mortality rate for each cause of injury mortality would have been had it not adopted the opioid prescribing reforms described above (hereafter called the “intervention”). The analytical approach was the same for all mortality measures, so from here on we refer generally to the “mortality rate.” We could only observe Florida’s behavior in the presence of the intervention, so the goal was to estimate Florida’s behavior in the absence of intervention. By taking the difference between our estimate of the counterfactual and the observed value, we estimated the impact of the intervention.
Bayesian structural time-series models.
We adopted a novel approach described by Brodersen et al. (24)—Bayesian structural time-series (BSTS) models—to forecast Florida’s behavior in the absence of the intervention. BSTS models use the flexibility of Bayesian model averaging to combine a number of different time-series models into a single forecast. In this analysis, we averaged 2 simple models for Florida’s behavior in the postintervention period. The first was a seasonal model, where Florida’s mortality rate was modeled using a dummy variable for 3-month periods (e.g., January–March, April–June, etc.). The second was a “spike-and-slab” linear regression model (25), where Florida’s mortality rate in each month of the preintervention period was regressed on the mortality rate in the comparison states. Spike-and-slab regression is a machine learning approach that uses “shrinkage” to down-weight covariates that do less to improve predictive accuracy, reducing model variance and improving out-of-sample predictions as compared with a simple linear regression (25).
To all regression coefficients, seasonal dummy variables, and residual variances, we assign so-called “noninformative” prior distributions—a common default choice in Bayesian analysis, and the default in the modeling package used (see “Estimation and software”). The spike-and-slab model expected model size was set at 1; this prior functioned as a form of “shrinkage” to reduce model variance and improve accuracy. This was a conservative prior choice, because it did not presume that any state was more or less predictive, and it is the default for the modeling package used (see “Estimation and software”).
BSTS models are described elsewhere in detail (24, 26, 27).
Estimation of postreform change.
Using the BSTS model, we forecasted Florida’s mortality rate in the postintervention period in the absence of the reform. We determined the preintervention observed mortality trend and then compared the model-estimated mortality rate trend with the observed mortality rate trend in the postintervention period (Figure 1). We then estimated 2 measures of the association of the intervention with changes in mortality.
Figure 1
Annual rates of observed and model-estimated mortality per 100,000 persons, by cause of death, Florida, 2005–2014.
First, for each month in the postintervention period, we converted predicted rates into predicted counts and took the absolute difference between the observed death count and the model-estimated death count. This was an estimate of the change in mortality that followed the introduction of reforms in each month (“monthly estimate”).
Second, for each month in the postintervention period, we then added the model-estimated monthly count to the model-estimated count in all prior months and took the absolute difference and relative difference (percent change) between this model-estimated value and the observed value. This was an estimate of the cumulative change, up to that month, in mortality that followed the introduction of reforms (“cumulative estimate”).
Presentation of results.
Because the analysis produced estimates for each month following the intervention, results were best summarized using a figure. Specifically, Figure 2 shows monthly and cumulative estimates of the postreform change in number of deaths, for each month, for each cause of death.
Figure 2
Estimated monthly and cumulative changes in numbers of deaths following opioid prescribing reform, by cause of death, Florida, July 2011–December 2014. A) Monthly change in drug overdose deaths; B) cumulative change in drug overdose deaths; C) monthly change in motor vehicle crash deaths; D) cumulative change in motor vehicle crash deaths; E) monthly change in suicide deaths; F) cumulative change in suicide deaths.
The main results highlight only the estimated cumulative percent change in mortality at the end of each calendar year (i.e., the estimated cumulative change in number of cases by the end of each year divided by the model-estimated total number of cases by the end of that year). The text of the Results section is even more restricted. Specifically, in Results we describe the cumulative postreform change in deaths by December 2013 and the average monthly change in deaths (calculated by dividing cumulative change by 30 months).
Model-checking and inference.
The analysis relied on the assumption that any deviation between our forecast of Florida’s postintervention mortality rate and its true postintervention rate was attributable to the intervention and not model misspecification. We tested the plausibility of this assumption in 2 ways.
First, we used a simplified version of the test proposed by Abadie et al. (28). We repeated the same analysis for each of the comparison states. Since comparison states did not adopt any intervention during the observation period, estimated “intervention associations” should have been much smaller than those observed in Florida. If intervention associations estimated in Florida were comparable in magnitude to those observed in the comparison states, we would be concerned that these associations were merely the result of random error or model misspecification.
Second, we repeated the entire analysis for 2 causes of mortality that should not have been affected by Florida’s prescribing reform—major cardiovascular diseases and malignant neoplasms. If the modeling approach could not accurately predict postintervention trends in these unaffected causes of death, we would be concerned that any observed associations with the outcomes of interest were merely the result of the poor predictive accuracy of the model.
Sensitivity analysis.
As we noted above, the estimated suicide death rate excluded deaths caused by poisoning, which might be affected differently than other methods of suicide. As a sensitivity analysis, we repeated our analysis of suicides including all suicides, including those involving poisoning.
Because of the relatively novel methodological approach, we show in Web Table 1 (available at https://academic.oup.com/aje) the results of a reanalysis of the data using a simple “difference-in-differences” approach. Briefly, for each type of mortality, we calculated for Florida and each control state the average mortality rate in the preintervention and postintervention periods, as well as the difference in mortality rates between these periods. We then calculated the difference between the prereform/postreform change in Florida and the prereform/postreform change in each of the control states.
Estimation and software.
All parameter estimates are accompanied by 95% credible intervals, defined as the 0.025th and 0.975th quantiles of the estimated Bayesian posterior for that parameter. All models were fitted using the “bsts” package in R (26, 29).
RESULTS
Drug overdose
Numbers of drug overdose deaths in Florida increased from 2005 to 2011, before declining by the end of 2013 (Table 2).
Table 2
Monthly Numbers and Ratesa of Mortality From Drug Overdose and Other Causes Before and After the July 2011 Opioid Prescribing Crackdown, Florida, January 2005–December 2013
| Cause of Death | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Monthand Year | Drug Overdose | Motor Vehicle Crash | Suicide (Nonpoisoning) | Major CVD | Malignant Neoplasm | |||||
| No. ofDeaths | Mortality Rate | No. of Deaths | Mortality Rate | No. of Deaths | Mortality Rate | No. of Deaths | Mortality Rate | No. of Deaths | Mortality Rate | |
| January 2005 | 197 | 1.1 | 310 | 1.7 | 156 | 0.9 | 5,522 | 30.9 | 3,417 | 19.2 |
| July 2011 | 274 | 1.4 | 202 | 1.1 | 204 | 1.1 | 4,254 | 22.3 | 3,520 | 18.5 |
| December 2013 | 233 | 1.2 | 232 | 1.2 | 182 | 0.9 | 4,898 | 25.1 | 3,618 | 18.5 |
Open in a separate window
Abbreviation: CVD, cardiovascular disease.
a Number of deaths per 100,000 residents.
As compared with the BSTS-estimated counterfactual estimates, by December 2013, cumulative numbers of drug overdose deaths over the full observation period were down by about one-sixth (−16.8%, 95% credible interval(CrI): −21.2, −11.8) (Table 3, Figure 2). This corresponds to a reduction of 1,377 deaths, or an average of 86 deaths per month.
Table 3
Estimated Cumulative Percent Change in Mortality Following the July 2011 Opioid Prescribing Crackdown, Florida, July 2011–July 2014
| Cause of Death | Time Period | |||||
|---|---|---|---|---|---|---|
| July 2011–December 2011 | July 2011–December 2012 | July 2011–December 2013 | ||||
| Point Estimate | 95% CrI | Point Estimate | 95% CrI | Point Estimate | 95% CrI | |
| Drug overdose | −5.9 | −12.9, 2.3 | −13.6 | −18.6, −7.8 | −16.8 | −21.2, −11.8 |
| Motor vehicle crash | −5.6 | −15.5, 5.7 | −6.2 | −12.5, 0.6 | −9.1 | −14.5, −3.6 |
| Suicide (nonpoisoning) | −0.3 | −9.8, 10.7 | 0.2 | −8.5,9.4 | 0.4 | −7.0, 8.3 |
| Major CVD | −1.2 | −4.7, 2.6 | −0.4 | −2.7, 2.2 | −0.7 | −2.7, 1.7 |
| Malignant neoplasm | 0.3 | −1.8, 2.4 | −0.2 | −1.5, 1.1 | 0.1 | −1.1, 1.3 |
Open in a separate window
Abbreviations: CrI, credible interval; CVD, cardiovascular disease.
Motor vehicle crash
Numbers of motor vehicle crash deaths declined from 2005 to 2011, before increasing slightly by the end of 2013 (Table 2).
As compared with the BSTS-estimated counterfactual estimates, cumulative numbers of motor vehicle crash deaths over the full observation period were down by about one-tenth (−9.1%, 95% CrI: −14.5, −3.6) (Table 3, Figure 2). This corresponds to a reduction of 615 deaths, or an average of 38 deaths per month.
Suicide
Numbers of nonpoisoning suicide deaths fluctuated over the course of the study period (Table 2).
As compared with the BSTS counterfactual estimates, numbers of suicides were essentially unchanged (0.4%, 95% CrI: −7.0, 8.3) (Table 3, Figure 2).
Model-checking
After repeating the analysis in all comparison states, the magnitude of the percent change in cause-specific mortality at 30 months in Florida was the third (out of 18) largest for drug overdose, third (out of 18) largest for motor vehicle crashes, and 17th (out of 18) for suicide (Web Figures 1–3).
Repeating the analysis for mortality due to major cardiovascular disease and malignant neoplasms showed no association of prescribing reforms with these control outcomes (Web Figures 4 and 5).
Sensitivity analysis
Results of sensitivity analysis that included poisoning suicides were the same (not shown). Associations in the difference-in-differences sensitivity analysis were not qualitatively different from those in our main analysis (Web Table 1).
DISCUSSION
Drug overdose deaths
Our analysis provides strong evidence that policies that reduce high-volume or insufficiently supervised opioid prescribing prevent drug overdose deaths. We found that drug overdose mortality declined sharply following the introduction of Florida’s opioid prescribing reforms, preventing 1,377 drug overdose deaths during the 30 months following the introduction of the pill mill law. This is similar to the conclusions of Kennedy-Hendricks et al. (10), who analyzed a slightly different 34-month period and found that reforms prevented 1,029 prescription opioid overdose deaths. We find it encouraging that these 2 studies—which used different data sources and different analytical approaches—reached very similar conclusions.
These findings are also important in light of forthcoming research showing mixed outcomes following similar “pill mill” laws—specifically that pill mill laws reduced drug overdose deaths in Ohio but not in Tennessee (30). It is notable that, as we mentioned, Florida’s pill mill law was strongly enforced, and more than half of the state’s pain-management clinics closed. This suggests that strong enforcement is essential to effectiveness.
It is important to note that we (and Kennedy-Hendricks et al.) ended our observation period in 2013. Beginning in 2013, deaths from synthetic opioids such as fentanyl increased dramatically. Synthetics quickly became the leading cause of opioid overdose death, both in Florida and nationally, and annual numbers of drug overdose deaths more than doubled (2). We cannot be sure that the benefits of Florida’s prescribing reforms were sustained in this new era of synthetic opioids. However, we find it encouraging that our monthly estimates (Figure 2) suggested that numbers of all drug overdose deaths were consistently down over our 2.5-year study period. This suggests that the prescribing reforms did not lead opioid users to immediately substitute other illegal drugs.
Prescribing reforms could prevent mortality in a number of ways: by preventing accidental overdoses among people receiving an opioid prescription who have no addiction; by preventing accidental overdoses among people receiving an opioid prescription who have an addiction; by preventing the formation of new opioid addictions; or by preventing diversion of drugs onto the black market. We could not determine who was affected by the reform—older or younger opioid users, men or women, or people living in urban areas or rural areas. Understanding who is affected by such reforms and mechanisms of action are important topics for future research.
Other sources of mortality
In addition to drug overdose death, we were interested in how Florida’s prescribing reforms might have affected other causes of death that may be linked to opioid use. Our findings were mixed.
Our model estimated an approximate 9% reduction in motor vehicle crash fatalities attributable to Florida’s prescribing reforms. This is consistent with our hypothesis that fewer opioid prescriptions would lead to less opioid-impaired driving. However, unlike numbers of drug overdose deaths—which fell almost immediately in the second half of 2011—most of this reduction came from lower-than-expected rates of motor vehicle crash death in 2013. We did not anticipate this delayed impact a priori. Moreover, the fact that the decline in crash deaths accelerated in 2013 is particularly telling because, in that same year, Florida banned texting while driving (31). Our analysis could not distinguish between the influence of prescribing reforms and the new ban on texting while driving. Given this limitation, we think it is essential that this finding be replicated in another jurisdiction before concluding that opioid prescribing reforms reduced motor vehicle crash deaths.
By contrast, we found no evidence that suicide mortality changed following Florida’s opioid prescribing reforms. This was true both of all suicides and of all suicides other than poisonings. This is encouraging, because both medical organizations and the popular press have raised concerns that prescribing reforms may lead to increased numbers of suicide deaths among people whose pain was previously treated by opioids (17, 32). Opioid prescribing has declined across the country since 2011 (33), a process that has probably accelerated because of new CDC guidelines on opioid prescribing (34). While careful monitoring of these broader declines and of the appropriateness of the CDC guidelines is needed, it is encouraging that even after Florida’s dramatic reforms to opioid prescribing, suicide rates did not budge relative to rates in other, similar states.
Limitations
This analysis had a number of limitations. First, the main assumption of our analysis was that Florida’s PDMP and pill mill laws were the only state interventions that might have affected the mortality rates analyzed here. We cannot know whether some other policy change made in Florida or in one of the comparison states highly predictive of Florida’s preintervention trends was responsible for the associations described here; in fact, as we noted above, the observed decline in motor vehicle crash deaths could be explained by a ban on texting while driving that was introduced around the same time. Second, mortality rates were estimated by dividing CDC-reported monthly death counts by census mean annual population estimates. Since the population changed over the course of the year, there was some error in these estimated rates, although likely very small. Third, this was an ecological study, and the outcome analyzed was a rate calculated at the state level. Research on individuals is needed to determine the effects of reducing or eliminating opioid use on individual risk for the types of mortality examined here.
Conclusion
The results presented here offer strong evidence that Florida’s opioid prescribing reforms sharply reduced numbers of drug overdose deaths over the 30-month period following their introduction. They also offer comforting evidence that Florida’s suicide mortality rate did not change following these reforms. Finally, they offer preliminary evidence that prescribing reforms may have reduced motor vehicle crash fatalities, but this finding deserves further investigation given the possibility that results were confounded by a contemporaneous ban on texting while driving.
These findings can provide at least 2 important lessons for other states that have recently enacted or are considering similar opioid prescribing reforms (23, 30). First, reforms that reduce irresponsible or unnecessary opioid prescribing prevent drug overdose deaths. States and the federal government should continue to promulgate regulations that reduce prescribing of opioids and promote alternative treatment of chronic pain. Second, while these reforms may have unanticipated consequences, we did not identify any negative unintended consequences like an increase in suicide deaths. However, other reforms that were structured differently might have different associations with different outcomes. It will be important to carefully monitor pain patients in locations instituting opioid prescribing reforms with regard to multiple outcomes, to ensure that any reduced risk of overdose death is accompanied by overall reductions in morbidity and mortality risk.
ACKNOWLEDGMENTS
Author affiliations: Department of Mental Health, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland (Kenneth A. Feder, Ramin Mojtabai, Elizabeth A. Stuart, Rashelle Musci, and Elizabeth J. Letourneau).
This research was conducted as part of K.A.F.’s doctoral dissertation, which was supported by a training grant from the National Institute on Drug Abuse (grant F31DA044699).
K.A.F. thanks Drs. Becky Genberg and Michael Fingerhood for serving on his dissertation committee.
Conflict of interest: none declared.
REFERENCES
1. Kolodny A, Courtwright DT, Hwang CS, et al.. The prescription opioid and heroin crisis: a public health approach to an epidemic of addiction. Annu Rev Public Health. 2015;36:559–574. [PubMed] [Google Scholar]
2. National Institute on Drug Abuse Overdose death rates. https://www.drugabuse.gov/related-topics/trends-statistics/overdose-death-rates. Published August 9, 2018. Revised January 2019Accessed December 6, 2018.
3. National Center for Injury Prevention and Control 10 Leading Causes of Injury Deaths by Age Group Highlighting Unintentional Injury Deaths, United States—2014Atlanta, GA:Centers for Disease Control and Prevention; 2016. https://www.cdc.gov/injury/wisqars/pdf/leading_causes_of_injury_deaths_highlighting_unintentional_injury_2014-a.pdf. Accessed June 28, 2018. [Google Scholar]
4. Muhuri PK, Gfroerer JC, Davies MC. Associations of nonmedical pain reliever use and initiation of heroin use in the United States. (CBHSQ Data Review, vol. 17)Rockville, MD:Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental Health Services Administration, US Department of Health and Human Services; 2013. [Google Scholar]
5. Miller JM, Stogner JM, Miller BL, et al.. Exploring synthetic heroin: accounts of acetyl fentanyl use from a sample of dually diagnosed drug offenders. Drug Alcohol Rev. 2017;37(1):121–127. [PubMed] [Google Scholar]
6. National Institute on Drug Abuse Florida opioid summary. https://www.drugabuse.gov/drugs-abuse/opioids/opioid-summaries-by-state/florida-opioid-summary. Published February 28, 2018. Revised May 2019Accessed June 28, 2018.
7. Gau JM, Moreto WD, Potter RH, et al.. Non-Medical Use of Prescription Drugs: Policy Change, Law Enforcement Activity, and Diversion Tactics. Orlando, FL:Department of Criminal Justice, University of Central Florida; 2017. https://www.ncjrs.gov/pdffiles1/nij/grants/250603.pdf. Accessed August 18, 2019. [Google Scholar]
8. Rutkow L, Chang H-Y, Daubresse M, et al.. Effect of Florida’s prescription drug monitoring program and pill mill laws on opioid prescribing and use. JAMA Intern Med. 2015;175(10):1642–1649. [PubMed] [Google Scholar]
9. Delcher C, Wagenaar AC, Goldberger BA, et al.. Abrupt decline in oxycodone-caused mortality after implementation of Florida’s prescription drug monitoring program. Drug Alcohol Depend. 2015;150:63–68. [PubMed] [Google Scholar]
10. Kennedy-Hendricks A, Richey M, McGinty EE, et al.. Opioid overdose deaths and Florida’s crackdown on pill mills. Am J Public Health. 2016;106(2):291–297. [PMC free article] [PubMed] [Google Scholar]
11. National Center for Statistics and Analysis, National Highway Traffic Safety Administration Alcohol-Impaired Driving. (Traffic Safety Facts, 2017 data). (DOT publication HS 812 630). Washington, DC:National Highway Traffic Safety Administration; 2018. https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812630. Accessed August 18, 2019. [Google Scholar]
12. Altilio T, Brennan M, Dahl J, et al.. Treatment Options: A Guide for People Living With Pain. Baltimore, MD:American Pain Foundation; 2007. https://ce4less.com/Tests/Materials/E019Materials.pdf. Accessed August 18, 2019. [Google Scholar]
13. Dubois S, Bédard M, Weaver B. The association between opioid analgesics and unsafe driving actions preceding fatal crashes. Accid Anal Prev. 2010;42(1):30–37. [PubMed] [Google Scholar]
14. Fishbain DA, Cutler RB, Rosomoff HL, et al.. Can patients taking opioids drive safely? A structured evidence-based review. J Pain Palliat Care Pharmacother. 2002;16(1):9–28. [PubMed] [Google Scholar]
15. Governors Highway Safety Association Drug-Impaired Driving: Marijuana and Opioids Raise Critical Issues for States. Washington, DC:Governors Highway Safety Association; 2018. https://www.insurancejournal.com/app/uploads/2018/05/GHSA_DrugImpairedDriving_FINAL-1.pdf. Accessed September 28, 2018. [Google Scholar]
16. Duren M, Ehsani J, Grant B, et al.. Geographic variation and trends in opioid-involved crash deaths in Maryland: 2006–2017. Accid Anal Prev. 2019;125:1–6. [PubMed] [Google Scholar]
17. Levine A.The government’s solution to the opioid crisis feels like a war to pain patients. Huffington Post. July 31, 2018. https://www.huffingtonpost.com/entry/government-crackdown-opioid-prescriptions-pain-patients_us_5b51ec57e4b0fd5c73c4a42e. Accessed August 18, 2019. [Google Scholar]
18. Racine M.Chronic pain and suicide risk: a comprehensive review. Prog Neuropsychopharmacol Biol Psychiatry. 2018;87(B):269–280. [PubMed] [Google Scholar]
19. Demidenko MI, Dobscha SK, Morasco BJ, et al.. Suicidal ideation and suicidal self-directed violence following clinician-initiated prescription opioid discontinuation among long-term opioid users. Gen Hosp Psychiatry. 2017;47:29–35. [PubMed] [Google Scholar]
20. National Center for Health Statistics, Centers for Disease Control and Prevention About underlying cause of death, 1999–2018. (Underlying cause of death data for 1999–2016 from the WONDER database)http://wonder.cdc.gov/ucd-icd10.html. Published 2017Accessed July 30, 2018.
21. National Center for Health Statistics Underlying and Multiple Cause of Death Codes. Hyattsville, MD:National Center for Health Statistics; 2018. https://www.cdc.gov/nchs/data/datalinkage/underlying_and_multiple_cause_of_death_codes.pdf. Accessed March 27, 2020. [Google Scholar]
22. Bureau of the Census, US Department of Commerce Revised 2010 demographic analysis middle series estimates. https://www2.census.gov/programs-surveys/popest/. Published2017. Accessed September 10, 2018.
23. The Policy Surveillance Program, A LawAtlas Project State laws and other regulatory policies related to pain care. http://lawatlas.org/datasets/state-laws-and-other-regulatory-policies-related-to-pain-care-final. Published2018. Accessed March 27, 2020.
24. Brodersen KH, Gallusser F, Koehler J, et al.. Inferring causal impact using Bayesian structural time-series models. Ann Appl Stat. 2015;9(1):247–274. [Google Scholar]
25. Ishwaran H, Rao JS. Spike and slab variable selection: frequentist and Bayesian strategies. Ann Stat. 2005;33(2):730–773. [Google Scholar]
26. Scott SL.Fitting Bayesian structural time series with the bsts R package. In: The Unofficial Google Data Science Blog [Web log]. http://www.unofficialgoogledatascience.com/2017/07/fitting-bayesian-structural-time-series.html. Published July 11, 2017. Accessed March 22, 2020.
27. Scott SL, Varian HR. Predicting the present with Bayesian structural time series. Int J Math Model Numer Optim. 2014;5(1-2):4–23. [Google Scholar]
28. Abadie A, Diamond A, Hainmueller J. Synthetic control methods for comparative case studies: estimating the effect of California’s tobacco control program. J Am Stat Assoc. 2010;105(490):493–505. [Google Scholar]
29. R Core Team R: A Language and Environment for Statistical Computing .Vienna, Austria:R Foundation for Statistical Computing; 2017. https://www.R-project.org/. Accessed March 22, 2020. [Google Scholar]
30. McGinty B, Brighthaupt S-C. Evaluating state pill mill laws’ effects on opioid overdose deathsPresented at the APHA 2018 Annual Meeting and Expo, San Diego, California, November 10–14, 2018https://apha.confex.com/apha/2018/meetingapp.cgi/Paper/410643. Accessed April 24, 2019. [Google Scholar]
31. Florida Ban on Texting While Driving LawFla Stat ch 316, §305. 2018.
32. Kertesz S, Manhapra A, Oliva E, et al.. VA data about prescription opioids and overdose and suicide: clinical implications [PowerPoint presentation]. Presented at the National Prescription Drug Abuse and Heroin Summit, Atlanta, Georgia, April 5–8, 2018https://www.scribd.com/document/375496421/VA-April-4th-Data-Opioids-Suicides. Accessed August 18, 2019. [Google Scholar]
33. The IQVIA Institute Medicine use and spending in the U.S.: a review of 2017 and outlook to 2022. Durham, NC:IQVIA Institute for Human Data Science; 2018. https://www.iqvia.com/institute/reports/medicine-use-and-spending-in-the-us- review-of-2017-outlook-to-2022. Accessed March 22, 2020. [Google Scholar]
34. Dowell D, Haegerich TM, Chou R. CDC guideline for prescribing opioids for chronic pain—United States, 2016. MMWR Recomm Rep. 2016;65(1):1–49. [PubMed] [Google Scholar]
Articles from American Journal of Epidemiology are provided here courtesy of Oxford University Press
