Telehealth and Solving the Rural Opioid Crisis

A key issue plaguing the United States is the opioid epidemic, one of the most devastating public health crises in the nation’s history. In accordance with the principles outlined by the previous article in the series, our first policy initiative can help alleviate the opioid crisis in a swift way that has bipartisan support. The following proposal will first introduce the crisis, discuss an overlooked solution to that crisis, and conclude with the potential impact and political implications of the policy solution.

Introduction

Across the United States, drug overdose deaths continue to rise across the country driven largely by the opioid epidemic. In 2020, the country suffered the largest increase in opioid overdoses with the United States seeing a 37% increase in total fatalities from 51,333 fatalities in 2019 to 70,168 in 2020 ^[1]. The sudden spike in opioid deaths can be attributed to the introduction of new drugs, such as fentanyl. Conditions continued to worsen throughout the COVID-19 pandemic, with opioid deaths reaching their peak in 2022. While opioid overdose rates have slightly decreased since 2023, the number of opioid seizures declined at a much lower rate. This indicates that interventions like methadone have decreased opioid deaths but the underlying rate of opioid use has hardly moved^[2]. 

During the pandemic, social distancing and quarantine mandates forced the Drug Enforcement Administration (DEA) to relax their restrictions on medications for opioid use disorder (MOUD). During this state of emergency, providers were allowed to prescribe MOUD through telehealth, otherwise known as telemedicine, without an in-person evaluation^[3]. However, telehealth’s efficacy hinges on the assumption that patients have consistent access to the internet. For many vulnerable populations, this is often not the case and a majority of healthcare providers cite unreliable internet access as a major obstacle for patients who wish to receive it^[4]. This disproportionately affects Americans living in rural areas, as broadband infrastructure and internet usage is not as developed as in urban areas.

This growing inequity in internet access is coined as the “digital divide”, the gap between communities that do and do not have reliable access to the internet ^[5]. The first stage of the digital divide is in the physical inequality characterized by a lack of reliable physical access to the internet. This can be seen in internet connectivity speeds, with only 51.6% of rural U.S. residents having internet speeds over 250 megabits per second compared to 94% of urban residents in 2020^[6]. The COVID-19 pandemic further exacerbated the digital divide, prompting the launch of the Emergency Broadband Benefit (EBB) in 2021, a federal program created to maintain broadband connectivity in low-income households. In 2022, the program became the Affordable Connectivity Program (ACP), which was mostly the same but rebranded to establish that it would remain beyond the duration of the pandemic. Enrolled households received a 30$ per month discount on internet service (up to 75$ on tribal lands) and were eligible for a $100 discount on an electronic device such as a laptop or desktop computer^[7].

The ACP was discontinued on June 1^st, 2024 due to a lack of congressional funding^[8]. Since its cancellation, the inequities between rural and urban communities persist with drug overdose rates rising much faster in rural areas with less infrastructure to support OUD treatment ^[9]. Although the digital divide encompasses more than just affordability, the cost of an internet plan is still the primary hurdle for those in need of telehealth. According to an Federal Communication Commission (FCC) survey, 72% of ACP recipients used ACP devices or internet access to schedule or attend appointments physically or virtually ^[10]. Combined with the lack of funding and shutdown of rural clinics across the country ^[11], increasing telehealth access through the ACP could be key to combating the opioid crisis in rural communities. This paper investigates the ACP’s impact on drug overdoses through increasing access to MOUD telehealth prescriptions and discusses its role in a future policy platform.

Data

To examine the effects of the ACP on rural drug overdose deaths, we used annual health measures, program enrollment, and drug overdoses for all rural counties across the United States. ZIP code areas and census tract geographical boundaries were unstable on an annual basis for these measures and aggregating into states would lead to insufficient statistical power. Annual county level observations were chosen since the boundaries for counties remain stable across years, and there are enough observations to potentially observe a moderate treatment effect.

Sources

County level enrollment data for the EBB and ACP was obtained from the FCC’s Enrollment and Claims tracker. Since the enrollment tracker uses monthly data, the monthly estimates for enrollment had to be aggregated into yearly estimates. To find ZIP code level eligibility, estimates were scraped from the RURAL LISC Dashboard ^[12]. Their methodology for estimating the eligible population is listed in the Appendix. These ZIP code eligibility numbers were aggregated into county wide estimates using HUD USPS ZIP Code Crosswalk files. 

Annual opioid use deaths for each county were sourced from the National Center for Health Statistics. Other covariates, such as percentage of rural population, non-Hispanic white population, unemployment rate, children in poverty rate, and uninsured adult rate, were taken from the County Health Rankings and Roadmaps datasets. The original sources for each covariate can be found in the Appendix. All data and code used for this paper is attached below and the original data sources are also all publicly available on their respective websites. 

Measures

County-level panel data was used for three years, 2019, 2022, and 2023 with a total of 10029 observations. 2020 was removed as during this year, lockdowns and quarantines were most intense across jurisdictions in the United States. 2021 was omitted as it was the year when the EBB launched and while the ACP and EBB were similar programs, their eligibility requirements and level of stimulus were quite different^[13]. Additionally, 2024 and 2025 were also removed as those were years after the ACP was discontinued. Years before 2017 and 2018 were not used in the main specification as drug overdose deaths were structurally different by that point (opiates like heroin were less common while opioids like fentanyl were becoming more common). Counties that were first treated in 2021 were also removed in the sample, as they would have experienced a transition between the EBB and ACP which could have potentially caused confusion for that cohort of program recipients. Counties that had less than 10% of their population eligible for the program were omitted from the sample. Only non-core (rural) counties were kept in the sample as defined by the NCHS rural-urban classification scheme. The classification scheme has 6 categorical variables ranking how metropolitan a county is and is designed for assessing health risks.

For privacy reasons, many counties had suppressed drug overdose deaths when their counties had between 1-9 drug overdose deaths in a given year. For our analysis, we implement the Dean & Kimmel imputation scheme for drug overdose deaths ^[14]. If during any year between 2019 and 2023 the county had a year where drug overdose deaths were observed and not suppressed, their suppressed drug overdose deaths were imputed as 9. Otherwise, the county’s drug deaths were imputed as 5. 

Methods

To estimate the effect of ACP enrollment on county-level drug overdose mortality (per 100,000), we use the doubly-robust staggered difference-in-differences estimator from Callaway & Sant’Anna ^[15]. For each treatment cohort g and post-treatment year t, the estimator compares changes in overdose death rates relative to a common pre-treatment base among counties baseline among counties first treated in year g against never-treated counties, after reweighting controls to be comparable on observable characteristics:

\($$ ATT(g,t) = E[Y_t(g) - Y_t (\infty ) | G_i = g] $$\)

 The advantage of the doubly-robust method is if our model is specified correctly for either a propensity score or linear regression model, a causal effect can still be identified even if the specification is wrong for one of these models. For the sake of reproducibility, a consistent seed of 123 in the R programming language is set before running the model and all robustness checks. 

Treated vs. control group. Counties that had at least 40% of their eligible households enrolled to the ACP were considered treated. The subscription rate was defined as the percentage of eligible households in the county that were subscribed to the ACP. The 40% threshold was chosen as the main model specification as it was the most robust. The control group was any county that had never reached this 40% enrollment threshold.

Parallel Trends. Our model assumes parallel trends hold while conditioning on the following covariates: uninsured rate, unemployment rate, child poverty rate, baseline overdose mortality in 2019, and the rural percentage of the county. When pretrends tests were done in these years, there was no statistically significant difference between the treated and untreated cohorts. To further support the parallel trends assumption, a pretrends test was done with additional years 2017 and 2018. Despite 2017 and 2018 being structurally different years, there was no statistically significant difference found between the treated and untreated cohorts even when these additional years were included. A county’s ACP enrollment rate was also not inherently tied to its overdose deaths. Rather, ACP enrollment was based on the program’s local outreach efforts, the broadband infrastructure, and cultural attitudes around internet use in the county. To help determine covariate selection, a standardized differences table was used. Of particular note, there was no imbalance in COVID death rate between treated and untreated populations.

Results

Treated counties that had at least one previous year of ACP treatment had a dynamic average treatment term of -15.0935 drug overdose deaths per 100,000 people. The general average treatment term was -16.0689 per 100,000 people. In this case, the average treatment term is the average effect of the ACP on drug overdose deaths. This suggests that among the cohort that was first treated in 2022, there was a 37.11% reduction in mean drug overdose deaths due to the ACP. Based on this figure, we estimate that between 2022 and 2023, 7169 drug overdose deaths were prevented and that 20058 more overdoses could have been prevented if the untreated counties had more ACP enrollment. 

The lagged benefits of the ACP for drug overdoses highlights that increasing access to telehealth could significantly decrease opioid deaths greater than even our difference-in-difference analysis suggests. The ACP, compared to other federal programs like SNAP or SSA, was not as well known and was greatly overshadowed by the context of the pandemic. A survey conducted near the end of the program's lifespan found that 71% of ACP eligible households either did not know the ACP existed, heard of the program and didn't know much about it, or knew of it and didn't apply ^[16]. This indicates that if the program was not discontinued and reached higher penetration levels, it could have further decreased rates of drug overdoses in rural counties through telehealth flexibilities.

Limitations

These results are not sensitive to a specific imputation assumption but are sensitive to whether suppressed values are imputed or dropped. Due to the significant loss in statistical power from dropping suppressed values, the confidence intervals become too wide to detect even a large treatment effect. Several factors caused the ACP to be adopted nonrandomly. Inequities in broadband infrastructure likely constrained potential demand for the ACP, as there are areas where even if the internet becomes affordable due to the program, the internet is still too slow. Enrollment was also constrained by both physical infrastructure and "network externalities" or a lack of others demonstrating the utility of a technology^[17]. 

Our results remain robust when specifying different thresholds for defining a treated county, from 40% to 70%. Critically, these ATT coefficients also remain stable when using two different imputation schemes - replacing all suppressed values with any integer value 2 to 9, and imputing according to the Dean & Kimmel scheme with different values. The only specification that did not survive robustness checks is the imputation that replaced all suppressed values with 1. This value is implausible as it drastically underestimates the amount of opioid deaths in these counties. Notably, the Dean & Kimmel scheme that replaced observed counties with 2 deaths per 100k and replaced unobserved counties with 1 death per 100k did remain robust. The results also remain robust when specifying estimation methods besides doubly robust methods. Using a propensity score or regression model on its own gives similar results to the doubly robust model used as our main specification. 

Placebo outcome tests were done, replacing drug overdose deaths with alcohol-impaired driving deaths and COVID cases per 100k, and the placebo coefficients were found to be insignificant. To check for structural stability, different covariates were included and excluded. The model was not sensitive to the inclusion of unemployment, children in poverty, uninsured rate, and rurality as covariates. The model was only sensitive to the removal of the 2019 baseline as a control, which was necessary for establishing conditional parallel trends. The model also remained robust when we included counties beyond non-rural counties. 

An important distinction is that while these results are suggestive for opioid related deaths, the response variable contains all drug overdose deaths. Part of the ATT captures the effect of telehealth on treating substance use disorder, not just opioid use disorder. As a result, our results are most likely conservative and underestimate the proportional impact on decreasing opioid overdoses. Further studies should explore differences in telehealth use for patients that use different kinds of substances. 

Policy Implications

Reviving the Affordable Connectivity Program

Literature surrounding telehealth flexibilities primarily focus on interventions for providers, while neglecting pain points the patient faces when getting to their appointment. Although telecommunication policy research addresses the potential benefits of the ACP for telehealth, rural health literature does not discuss the ACP as a public health intervention. In order for policymakers to fully take advantage of telehealth flexibilities, they must design policies that address the obstacles to both providers and patients.

The success of the ACP as an incidental public health intervention must be contextualized by its program implementation. A key factor in the ACP's success hinged on effective outreach investment. While the ACP began during the COVID-19 pandemic which distracted potential participants from the program's launch, the majority of eligible participants were still not aware of the program after the end of the pandemic. From 2021-2022, the ACP mostly utilized the interim outreach strategy from the EBB and delayed development of a comprehensive outreach strategy until 2023. In an official report regarding the ACP, the Government Accountability Office concluded that the FCC’s outreach plan was underdeveloped and did not follow leading practices for consumer outreach ^[18].

Any comprehensive plan to overcome the opioid crisis must include the revival and expansion of the Affordable Connectivity Program. Despite a flawed consumer outreach program and launching in the midst of the COVID-19 pandemic, the program still succeeded in delivering positive public health outcomes even though it was not designed as a public health intervention. A second iteration of the ACP should immediately continue its outreach grant program but prioritize giving grants to clinics that service populations most afflicted by the opioid crisis. ACP consumer outreach has primarily focused on getting eligible users to enroll, with tertiary consideration for developing internet usage patterns in communities with lower internet use. The FCC should significantly invest in internet literacy training as part of a future ACP outreach strategy. Within these trainings, community members should be encouraged to train their friends and family on internet literacy. 

The political capital needed to reimplement the ACP would be minimal since both iterations of the policy, EBB and ACP, were passed as part of bipartisan policy initiatives (2021 Consolidated Appropriations Act and Infrastructure Investment and Jobs Act). The program is estimated to generate at least twice the economic savings than it costs to subsidize the program, due to increasing employment and decreasing healthcare costs^[19]. There is still institutional knowledge for how to run and improve the ACP for its next iteration since the program was only shut down recently. The program was only “shut down” due to a lapse of funding, not a legal termination. Unlike most policy proposals, which create entirely new policies, reviving the ACP only requires a new round of funding from congress in its next session. 

To maximize the impact of the ACP, further investment into broadband infrastructure for rural areas is necessary. As the ACP is a consumer-side intervention, its impact does not address the underlying inequities in broadband availability and quality of internet access between counties. Nearly 23% of Americans in rural counties lack fixed broadband services^[20]. Many of the eligible counties analyzed likely did not reach their peak ACP enrollment rate because low income, rural communities are underserved by broadband infrastructure. Even amongst rural counties, there is heterogeneity in treatment due to different regions having ISPs who offer plans with different internet speeds and device limits. Increasing broadband infrastructure in rural communities is not only vital for economic investment but also can improve the public health outcomes of rural areas. 

For the ACP to succeed as a health intervention, legal barriers to telemedicine providers must be streamlined or removed. The telehealth flexibilities for MOUD currently exist in a state of legal precarity, where they are extended yearly by the DEA ^[21]. Enshrining these flexibilities into permanent legislation would give providers the confidence they need to invest in providing telehealth for their communities. This legislation should also make it easier for providers to obtain telehealth licenses across multiple states ^[22]. Incentives should be provided for states to adopt a universal quality standard for telehealth licensure and standardize reciprocity agreements across states. Minimizing the time between providers applying for licenses and providing telehealth would increase the number of providers across the country– especially in rural counties most afflicted by the opioid crisis and with the lowest proportion of healthcare providers. The extensive regulations around MOUD prescriptions such as the Harrison Act have created a stigma surrounding receiving MOUD treatment ^[23]. By increasing the amount of telehealth providers for MOUD, the stigma surrounding MOUD prescriptions decreases and telehealth utilization will increase, saving lives in the process.

Appendix

ACP ZIP Code Eligible Population Estimation Formula

Total Eligible Households by ZIP Code = $27,180 + $9,440(HS-1)

• HS = average household size in a ZIP code.
• $27,180 = 200% FPL for a 1-person household
• $9,440(HS - 1) = the rate at which the total number of individuals living at 200% FPL changes within a ZIP code for each person added to a household

Source: 

https://www.lisc.org/media/filer_public/13/4d/134d8f53-fe33-4ab9-b54b-5523c0193c80/acp_methodology_faqs_1_20_2023.pdf

Data Sources Table

Sources

Final dataset and code: https://github.com/Nathaniel-AW/Research-on-Internet-Connectivity