Current Research

“The Impact of Physcian Networks on Provider Opioid Prescribing Behavior ” (with T. Flottenmesch, M. Ali, J. Jones, R. Mutter, A. Hohlbauch).

Drug overdose is now the most common cause of injury death in the United States, and opioid analgesics are involved in almost half of these deaths. Curbing the opioid overdose epidemic is a national public health priority. Efforts to reduce over-prescribing of opioid analgesics among health care providers are recognized as an important step in improving the health of populations with opioid use disorders (OUDs). In 2016, the CDC issued guidelines regarding the appropriate use of opioid analgesics. To examine the relationship between patterns of opioid prescribing across empirically identified social networks and examine the relationship between incidence of four risky prescribing behaviors by individual providers and their integration within a care community. We applied social network analysis (SNA) to the IBM Watson Health Medicaid MarketScan® Research Database for the years 2010-2015. Our SNA first identified provider care communities and then the level of each provider’s integration (i.e. centrality) within his/her identified community. A nested multivariable logistic regression considering the likelihood of any and the risk of repeated risky opioid prescribing controlling for year, and patient mix examined the relationship between provider integration within an identified care community and risky prescribing behaviors. We focused upon four risky prescribing behaviors mentioned in the 2016 CDC guideline. These five behaviors were: 1) Greater than 90-days continuous supply of high-dose opioid analgesic for chronic pain, 2) Multiple providers prescribing opioid analgesics to the same patient, 3) Overlapping opioid prescriptions, 4) Co-occurring opioid and benzodiazepine prescriptions, and 5) Prescribing an extended release form for an acute pain diagnosis. Analysis was at the provider-year level. For each opioid prescriptions its prescribing provider was identified by examining all encounters occurring within seven days of the prescription fill. For multivariable models, we adjusted for patient mix, provider specialty, can care community. Prescribing provider centrality ranged from .008 to .621. Multivariable models indicate greater provider integration is associated with a lower risk of certain risky prescribing behaviors, but has indeterminate associations with others. Greater provider integration does appear to lower the risk of a provider repeatedly engaging in risky opioid prescribing; however, it does not eliminate the chance of the provider ever engaging in such a behavior.


“Initial Effects of Legalizing Recreational Marijuana on Traffic Fatalities” (with Aalap Trivedi).

The recent introduction of state laws legalizing recreational marijuana, coupled with literature on the relationship between marijuana and driving impairment, motivate an examination of the effect of these laws on traffic fatalities. This paper uses data from the National Highway Traffic Safety Administration in a Synthetic Control framework to analyze the association between Colorado and Washington's implementation of recreational marijuana use and a) the number of motor vehicle fatalities per 100 million miles traveled and b) the number of alcohol related motor vehicle fatalities per 100 million miles. Using a synthetic control approach, the analysis finds evidence to suggest these laws are associated with an increase in traffic fatalities, but no change in drunk driving fatalities. These results suggest policymakers in states with legalized marijuana should implement policies to mitigate the adverse effects of recreational marijuana on traffic fatalities.


“County Disparities in Opioid Related Mortality”

The general increase in opioid related deaths is well documented, and disparities in geographies and demographics have been documented. However, the distribution of opioid related deaths among U.S. Counties and the trends in that distribution have not been explored. This study examines the inequality in opioid death rates to assess convergence or divergence in opioid related mortality between counties. Using mortality data from the NVSS for 2000-2014, this study examines the Gini coefficient of county opioid mortality distribution. The distribution of opioid mortality became more equal, with the Gini coefficient falling from 0.81 in 2000 to 0.61 in 2014. Counties with lower initial opioid mortality rates experienced faster growth in mortality than counties with high initial mortality. Counties have experienced a convergence in opioid mortality rates. This poses potential challenges for addressing the crisis as measures must become much broader in scope and be implemented in areas in which the dangers of the opioid crisis are not as apparent.

“Suicides and Mental Health Treatment: An Instrumental Variable Approach” (with R. Mutter and M. Ali).

In 2015, suicide was the tenth leading cause of preventable death, with 44,193 individuals committing suicide. Additionally, while preventable mortality from other causes has been declining in recent years, suicide deaths have been increasing. This study examines the impact of mental health treatment on suicide attempts by employing an instrumental variable strategy that uses the exogenous variation in access to mental health services to identify the impact of mental health treatment. Greater access to mental health treatment, as measured by both proximity to mental health providers and mental health provider density, is associated with more mental health treatment, which in turn significantly reduces suicide attempt risks. Policies designed to expand availability of and access to mental health treatment should be considered as part of the response to reduce suicides.

“The Impact of Abuse-Deterrent Formulation of Extended-Release OxyContin on Non-medical Use of Prescription Pain Relievers and Heroin Use” (with M. Ali, W. Dowd, C. Wolff, A. Meinhofer, L. Glos, A. Schick, M. Rosenberg, and L. Sherman).


“Price Elasticity of Long-Acting Opioids” (with R. Mutter, M. Ali, R. Kaestner, K. Fingar, T. Gibson, and W. Olesiuk).


“Macroeconomic Effects of VAT Evasion and Enforcement” Under Review, Journal of International Tax and Public Finance

Lower tax revenues have a theoretically mixed effect on growth as they create more disposable income for investment, but simultaneously reduce funds for public goods. This study combines firm level data on tax evasion and enforcement from 79 countries with macroeconomic data to examine the effects of tax enforcement measures and tax revenue shortfall on economic growth. This study finds that while increased enforcement measures reduce growth, high tax revenue collection serve to increase growth. These results suggest that reforms focusing on increasing revenue without resorting to greater enforcement measures are desirable.

“The ACA and Risky Behavior” (with Benjamin Ukert). Under Review, American Journal of Health Economics.

This paper estimates the impact of the private marketplace insurance market expansion as part of the Affordable Care Act on legal and illegal substance use. We utilize detailed drug use data from the National Survey on Drug Use and Health (NSDUH) from 2011 to 2015 and employ a difference-in-difference estimation strategy that takes advantage of variation across time and state uninsured rate. First, we confirm that our methodological approach identifies similarly large increases in insurance coverage as documented in the literature. Second, our estimates on the impact of insurance coverage on legal and illegal drug use shows statistically significant decreases in marijuana, stimulant, and inhalant use but does not impact legal substance use, such as drinking and smoking, and any other illegal substance use. Our results provide new evidence that health insurance expansions reduce illegal substance use and that the extension of health insurance does not lead to moral hazard in the form of increases in risky health behaviors.


“Losing Public Insurance and Mental Healthcare and Mental Illness: Evidence from a Large Scale Medicaid Disenrollment” (with Catherine Maclean and Sebastian Tello-Trillo).

In this study we explore the effects of losing public insurance on mental healthcare utilization and mental illness. We leverage plausibly exogenous variation in insurance coverage offered by a large-scale and unexpected Medicaid disenrollment in Tennessee that occurred between August 2005 and May 2006. Disenrollees were predominately childless and non-disabled adults. We apply differences-in-differences models, comparing changes in Tennessee to changes in geographically similar U.S. states. Our findings suggest that, post-disenrollment, mental healthcare use declined and shifted from ambulatory care to hospitalizations. We further show that Medicaid coverage and use of Medicaid to pay for treatment declined, and we provide suggestive evidence that mental illness increased in Tennessee relative to geographically similar southern states post-disenrollment. These findings are immediately important given current policy debates within the United States on whether or not to maintain the Affordable Care Act Medicaid expansions to non-disabled low-income childless adults.

“Behavioral Health’s Integration Within a Care Network and Healthcare Utilization” (with T. Flottenmesch, M. Ali, J. Jones, R. Mutter, A. Hohlbauch, D. Whalen, N. Nordstrom). Revise and Resubmit, Health Services Research.

Greater integration of behavioral health providers within a provider care network will be significantly associated with lower total medical costs, fewer emergency department (ED) visits, and less frequent inpatient admissions. This study estimates the relationship between behavioral health integration and the patient-level outcomes of total health care costs, ED visits, and inpatient admissions in care communities. Social network analysis (SNA) using data from six Medicaid plans in the Truven Health MarketScan® Research Databases was used to identify care communities and estimate how integrated, or central, behavioral health providers are within each identified care community across the years 2011-13. Enrollees in these same plans then were attributed to a community where they had the greatest numbers of encounters, and their total health care costs, number of ED visits, and inpatient admissions for each year were tabulated. The relationship between behavioral health centrality and patient outcomes was estimated using multivariable generalized linear models adjusting for patient age, sex, number of prescriptions and Charlson Comorbidity Score. Increased behavioral centrality was significantly associated with a reduced number of ED visits, less frequent inpatient admissions, and lower overall health care costs.

“Opioid-Overdose Laws and Naloxone Programs’ Association with Opioid Use and Mortality” (with B. Lambdin, M. Ali, R. Mutter, C Davis, E Wheeler, M. Pemberton, A Kral). Under Review, Addictive Behaviors.

Since the 1990’s, governmental and non-governmental organizations have adopted several measures to increase access to the overdose reversal medication naloxone. These include the implementation of programs that distribute naloxone to laypeople, laws that increase layperson naloxone access, and overdose-specific Good Samaritan laws that protect those reporting overdoses from criminal sanction. The association of these initiatives with overdose mortality and opioid use is unknown. We assess the relationship of (1) naloxone access laws, (2) overdose Good Samaritan laws and (3) naloxone programs with opioid-overdose mortality and non-medical opioid use in the United States. We used 2000–2014 National Vital Statistics System data, 2002–2014 National Survey on Drug Use and Health data, and primary datasets of the location and timing of naloxone access laws, overdose Good Samaritan laws, and naloxone programs. By 2014, 30 states had a naloxone access and/or Good Samaritan law, and 259 counties had a naloxone program. States with naloxone access laws or Good Samaritan laws had a 14% (p=0.033) and 15% (p=0.05) lower incidence of opioid-overdose mortality, respectively. Both law types exhibited differential association with opioid-overdose mortality by race and age. Naloxone programs are associated with lower opioid-overdose mortality among those 45 to 54 years of age. No significant relationships were observed between the examined policies and opioid use. Generally, laws designed to increase layperson engagement in opioid-overdose reversal were associated with reduced opioid-overdose mortality. We found no evidence that these measures were associated with increased non-medical opioid use.