Are Rural Opioid Treatment Program (OTP) Facilities Associated with Lower Deaths?

Abstract

Objectives: Rural areas have been particularly affected by the opioid epidemic in the U.S., with treatment options being scarce. This study focuses on opioid treatment programs (OTPs), which combine counseling services and opioid-related medication provision. In the South census region (comprised of 16 states and D.C.), 156 urban counties and 27 rural counties contained OTPs as of 2013. We examine whether their presence is associated with lower opioid-related death rates during 2014–2016. Methods: Coarsened exact matching (CEM) is used to match treated and untreated counties on demographic characteristics and opioid-related deaths from 2011 to 2013. Two treatments are considered: (1) if a county had an OTP in 2013; and (2) if an OTP existed in a neighboring county in 2013. The matched samples are then used in weighted least square regression models, with propensity score matching serving as a robustness check. Rural and urban counties are analyzed separately to determine if the impact of OTPs differs between these areas. Results: Results show that the presence of an OTP mostly has no statistical association with the rate of (or change in) future opioid deaths, in either rural or urban counties. Proximity to a neighboring county OTP displays a similar lack of association. Conclusions: The findings suggest that OTPs are not associated with fewer opioid-related deaths in the South over the near term, regardless of rural or urban location. These results could be attributed to outside factors that hinder this relationship. Continued assessment of varied approaches to the rural opioid crisis is encouraged.

Keywords:

• Opioid epidemic
• opioid treatment
• medication assisted treatment (MAT)
• coarsened exact matching

Declaration of interest

The authors report no conflict of interest.

Notes

1 The states included in the South census region are: Alabama, Arkansas, Delaware, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia

2 We use the terms metro / non-metro and urban / rural interchangeably, although we note that since our data is at the county-level metro / non-metro may be more appropriate.

3 The National Institute on Drug Abuse suggests that methadone, a prescription medication used in certain MAT programs, should be given to patients for a minimum of 12 months in order to prevent relapse (National Institute on Drug Abuse, 2018). Patients are able to take home doses of medication at a certain point during their treatment.

4 The codes represent the following drug-related deaths, T40.0: opium, T40.1: heroin, T40.2: other opioids, T40.3: methadone, T40.4: other synthetic narcotics

5 Age-adjusted death rates account for the counties’ differences in age distributions, so that the counties in the sample can be compared to one another (Anderson & Rosenberg, 1988). When death rates are age adjusted, the CDC provides age adjusted death rates for counties with death counts greater than 20. Age adjusted death rates were calculated manually for counties with 10 to 20 deaths using the average change (from crude death rate to age adjusted rate) for counties with a similar population size that do have CDC-provided age-adjusted death rates. The final age-adjusted death rates are compared to CDC values (Centers for Disease Control and Prevention, 2018a) to verify calculation accuracy.

6 CEM differs from a synthetic control approach because it does not generate a singular control group from units not receiving the treatment (Abadie, Diamond, & Hainmueller, 2010).

7 Ideally, we would be most interested in OTPs that came in existence after the initial spike, and measure their resulting impact. Data limitations prevent this analysis.

8 A neighbor is considered to be any county that shares a border with the county of interest (queen contiguity matrix).

9 The L1 values are calculated by determining the difference in the empirical distributions between the treated and control groups (Iacus et al., 2011).

10 In nearest neighbor matching, treated and control observations are matched based on similarities in the covariates. A commonly used approach is to include kernel matching as a robustness check for nearest neighbor matching, because it corrects for any large differences in covariates between the treated and control units (Whitacre, Gallardo, & Strover, 2014). Kernel matching determines the distance between one treatment observation and many control observations, and gives the closer units a higher weight in the matching procedure.