The impact of expanded Medicaid eligibility on access to naloxone

Abstract

Background and Aims

Federal, state and local US governments have sought interventions to reduce deaths due to opioid overdoses by increasing the availability of naloxone. The Affordable Care Act (ACA) expanded Medicaid coverage to low-income, childless adults, potentially giving this group financial access to naloxone. The aims of this paper are: (1) to describe the changes in the amount of Medicaid-covered naloxone used between 2009 and 2016 and (2) to quantify the differential change in the amount of dispensed naloxone between states that expanded their Medicaid programs and states that did not.

Design

A quasi-experimental approach based on states’ ongoing choice to expand their Medicaid program to all adults with incomes between 100 and 138% of the federal poverty line (FPL), starting in 2014. As of 2018, 37 states had expanded and 14 states had not. Estimation of the policy impact relies on a difference-in-difference method.

Setting

US state Medicaid programs.

Participants and measurements

Data are from the Medicaid Drug Rebate Program and include all dispensed prescriptions of naloxone through the Medicaid program. State/quarters with fewer than 10 prescriptions are suppressed; n = 1632.

Findings

Prior to Medicaid expansion, the number of Medicaid-covered naloxone prescriptions was very similar in expansion and non-expansion states. On average, states that expanded Medicaid had 78.2 (95% confidence interval = 16.0–140.3, P = 0.02) more prescriptions per year for naloxone compared with states that did not expand Medicaid coverage, a nearly 10 increase over the pre-expansion years. Medicaid expansion contributed to this growth in Medicaid-covered naloxone more than other state-level naloxone policies.

Conclusions

Medicaid accounts for approximately a quarter of naloxone sales. Medicaid expansion generated 8.3% of the growth in naloxone units from 2009 to 2016, holding other factors constant.

INTRODUCTION

The Centers for Disease Control and Prevention reported that there were 63 632 deaths from drug poisonings in 2016 [1]. Of those, 42249 (66%) deaths were estimated to be due to fatal opioid poisonings. These deaths are just the latest in an epidemic that has been under way since the late 1970s. Since then, the growth rate in drug poisonings has increased at an average rate of 9% per year [2]. However, a more recent year-over-year change (2015–16) represents a 21% increase in deaths. The increase in mortality has been accompanied by rapid growth in emergency department visits and hospital admissions for drug-related poisonings [3,4].

One harm reduction response to the expanded role of fentanyl in the epidemic is to increase the availability of naloxone. It is the short-acting opioid antagonist that reverses the effects of an opioid-related overdose by displacing opioid agonists from their receptors in the nervous system [5]. Naloxone has been available on the US market since 1971, originally marketed as an injectable product. The active ingredient (naloxone hydrochloride) in its injectable form has been available as a generic since 1985. Recently, the Food and Drug Administration approved two ‘user-friendly’ formulations of naloxone. One is naloxone administered by an auto-injector, similar to the one used for epinephrine, which launched in 2015. The other version is a nasal spray formulation that has been available since 2016.

Two recent reviews, based largely on quasi-experiments and observational studies, demonstrate that lay people can be trained to use and will make use of naloxone to reverse overdoses. These reviews suggest that naloxone is associated with one life saved for every 14.5 prescriptions of naloxone distributed [6,7]. Other work has demonstrated that there are approximately 22 opioid overdose reversals for every 100 people trained in naloxone administration. [8]

Public policy responses to the opioid epidemic and its increasing lethality has involved providing targeted grants to states and localities for purchasing naloxone [3,9]. For 2017 and 2018, almost $1 billion was distributed to all 50 states to address the opioid epidemic under the 21st Century Cures Act [10,11]. Additionally, the Substance Abuse and Mental Health Services Administration (SAMHSA) has given $11 million to 12 states to help purchase naloxone and train first responders in its use. The 2016 Comprehensive Addiction Recovery Act provided approximately $46 million over 5 years to fund grants in 22 states to help first responders and providers to address the needs of people at high risk of overdose. In addition, many states passed laws permitting the following: third-party prescription of naloxone, prescription via standing orders, possession of naloxone without a prescription, civil (and in some states criminal) immunity for prescribers and dispensers of naloxone and Good Samaritan laws [12].

Beyond increasing supply of naloxone via targeted grants and laws that seek to improve access to naloxone, state Medicaid programs play an increasingly important role in the opioid epidemic by providing financial access to treatment for opioid use disorder (OUD) and naloxone for individuals at risk of opioid-related overdose and their friends and family. Medicaid programs in the United States provide health insurance coverage to low-income children, parents and disabled adults. Although income eligibility limits for these groups vary by state, most states had eligibility limits below 100% FPL. Additionally, most states did not provide coverage to single, childless adults without a disability, leaving few health insurance coverage options for this population. Indeed, more than half of adults with an OUD had incomes below 250% FPL, and adults with an OUD were disproportionately uninsured (authors’ tabulations from National Survey on Drug Use and Health).

The Affordable Care Act (ACA) passed in 2010 included a provision that required states to expand Medicaid income eligibility limits for adults to 138% of the FPL and remove other ‘categorical’ requirements for eligibility (i.e. any adult who met the income eligibility limit would now be eligible for Medicaid). The US Supreme Court ruling in Sebelius v. NFIB in 2012, however, found this requirement to be unconstitutional, and gave states the option to expand their Medicaid programs. To promote expansion, the federal government paid for the bulk of the cost of covering the newly eligible people under the ACA.

States that chose to expand eligibility for their Medicaid programs could do so beginning in 2014. Since then, 36 states and the District of Columbia have expanded their Medicaid program to all adults with incomes under 138% of the FPL.

In the following analysis of the impact of Medicaid expansion on the use of naloxone, we focus on the changes in the use of naloxone as measured by dispensed naloxone prescriptions. Specifically, the aims of our paper are (1) to describe the changes in the amount of Medicaid-covered naloxone between 2009 and 2016 and (2) to quantify the differential change in the amount of dispensed naloxone between states that expanded their Medicaid programs and states that did not during this study period.

DATA AND METHODS

Our approach is to take advantage of the natural experiment created by the state option to expand Medicaid based on the Supreme Court’s ruling in NFIB v. Sebelius. Using data on the amount of Medicaid-covered dispensed naloxone, we make use of a quasi-experiment that makes comparisons before and after Medicaid expansion for those states that expanded and those that did not. Our impact estimates rely on a difference-in-differences analysis. Our unit of analysis is the state-quarter-year, and our study period is from 2009 to 2016. A more granular study period allows us to capture more accurately the timing of the Medicaid expansions. That is because they often occurred in months other than January, and the study period encompasses the beginning of the opioid epidemic’s exponential increase in 2009–10 and provides 3 years of post-expansion data for most states.

Description of data

Our outcome measure of interest, Medicaid-covered dispensed out-patient naloxone prescriptions, was obtained from the Medicaid Drug Rebate Program (MDRP), maintained by the Centers for Medicare and Medicaid Services (CMS). CMS provides state-quarterly reports of out-patient prescription drugs from drug manufacturers participating in the MDRP. In accordance with the National Drug Rebate Agreement and Section 1927(b) [3](A) of the Social Security Act, a participating pharmaceutical company must provide an average manufacturing price (AMP) report for each drug in the MDRP during that quarter-year. Pharmaceutical companies that do not provide the quarterly AMP report for each drug are subject to removal from the MDRP or civil monetary penalties from the Office of the Inspector General of the Department of Health and Human Services. These reports provide information on each prescription drug’s National Drug Code (NDC), market date, product name, unit type and units per package size. State-quarter-drug code fields with fewer than 11 [11] prescriptions are suppressed to protect the privacy of individuals. Only state-quarter-year combinations where all state-quarter-NDC code observations are labeled as suppressed have a value of 1 for the suppression dummy variable.

Similar to other reports using these data for naloxone [13], we classified a drug as naloxone for opioid reversal if the drug was:

1. Labeled as naloxone, naloxone hydrochloride, Narcan (the brand name nasal injector) or Evzio (the brand name auto-injector); or

2. Had an NDC code that corresponded to naloxone but was not combined with buprenorphine (which is used to inhibit opioid-related cravings or withdrawal symptoms) or pentazocine (for management of severe pain).

Measures

Our outcome of interest is the total number of dispensed naloxone prescriptions paid for by Medicaid. We treat the outcome variable as normally distributed in our regression models. We consider two transformations of the outcome—population-adjusted prescriptions per 100 000 Medicaid enrollees and the natural log of total prescriptions to account for potential skewedness in the data. We conduct ordinary least-squares regressions on these transformed outcomes.

Obtaining consistent estimates relies on our ability to account for a complex policy and illness environment. For instance, if state laws were passed that expanded access to naloxone, we would want to be able to account for the separate effects of these laws in estimating the impacts of the Medicaid expansion. Therefore, our regression models control for state policy initiatives that aim to expand the use of naloxone, especially by drug users, their friends and families. These initiatives include laws that permit standing orders for naloxone, laws that permit pharmacists to dispense naloxone and laws that allow third parties to carry naloxone without a prescription. Our measurement of these policies is based on data assembled by the Prescription Drug Abuse Policy System Network for Public Health Law [12,14].¹ Additionally, if states chose to expand their Medicaid programs because of the burden of opioid epidemic, we also want to be able to account for these selection differences in the decision to expand. Thus, we included rates of non-medical use of opioid pain relievers and substance use disorder from the National Survey on Drug Use and Health (NSDUH) from the SAMHSA [15]. In order to control for baseline differences in state demographics, we also obtained data on each state’s unemployment rate, gender and age composition, as well as racial and ethnic make-up for each state and year [16]. Additionally, new formulations of naloxone enabled out-patient prescription of naloxone products to at-risk patients as part of routine clinical care. Prior to approval of these user-friendly formulations, naloxone was generally only provided to at-risk patients through public health programs that assembled kits including naloxone, along with the equipment and training needed to administer it (e.g. syringes plus needles or an atomizer). We include dummy variables for the state-quarter-year introductions of both Evzio and Narcan to account for the increases in naloxone attributable to introduction of these novel products.

Statistical analysis

We begin by describing the annual trends in Medicaid purchases of naloxone by state Medicaid expansion status. In addition to visual inspection of parallel pre-trends, we also conduct a formal test of parallel pre-trends by regressing an interaction between a dummy variable that indicates whether the state expanded its Medicaid program during the study period and a pre-period linear time trend on the outcome of interest.

As is standard in difference-in-differences, our preferred regression specification includes a dummy variable for treatment (i.e. states that expanded Medicaid) a dummy for post (i.e. quarter-years after Medicaid expansion for each state) and an interaction between these dummy variables. Because some states expanded their Medicaid program after January 2014, we have multiple treatment periods. This does not impact our estimation or computation strategy, but does affect the interpretation of results. We include the state demographic characteristics described above, indicator variables for the enactment of other naloxone policies and state fixed-effects to account for state-specific, time-invariant unobservable factors that may affect the demand for and availability of naloxone. We check the robustness of our results by systematically removing covariates included in our main regression specification.

We also estimate a difference-in-differences regression model that allows for a separate effect for each post-period year, rather than averaging the 3 years together. Relaxing the linear assumption in the post-period may capture more accurately the relationship between Medicaid expansion and Medicaid-covered naloxone prescriptions. All regressions have clustered standard errors at the state level to address potential serial autocorrelation.

Sensitivity analyses

To test the sensitivity of our results to suppression of small cell counts, we assume that state-quarter-NDC code observations that are not present in the data set are zero and code them as such. If the state-quarter-NDC code observation is in the data set but labeled as suppressed, we include a dummy variable indicating suppression. For suppressed state-quarter-NDC values, we use three imputations: one, five or 10, to test the sensitivity to suppression. After imputation, we conduct regression analyses similar to those discussed above. We explore variants on the model specifications, including functional form differences and the exclusion of different covariates.

During our observation period, there was a significant rise in the presence and lethality of fentanyl in New England. It is possible that the fentanyl crisis in New England (with three states expanding in 2014, one state expanding in 2015, and one state not expanding) is driving the results obtained in our difference-in-differences analysis. In order to test for this possibility, we exclude the five New England states from our analysis and compare those estimates to those of the full sample.

Lastly, we perform the difference-in-differences analysis with sublingual buprenorphine prescriptions as the outcome, rather than naloxone. Efforts to address opioid overdose in the United States have encouraged expanded use of both naloxone for overdose reversal and sublingual buprenorphine for opioid use disorder in clinical practice. Formulation changes to naloxone may have fueled its growth during the study period, but there were no formulation changes to sublingual buprenorphine. If we see similar estimates for Medicaid-covered, dispensed buprenorphine prescriptions and naloxone, then we can be more confident that the change in the rate of prescribing is driven by Medicaid expansion than naloxone reformulation.

All analyses were conducted in Stata version 15.0 (College Station, TX, USA).

RESULTS

Characteristics of expansion versus non-expansion states

Twenty-four (75%) of the expansion states implemented a standing order for the dispensing of naloxone (‘standing order’), compared to 13 (75%) of the non-expansion states before Medicaid expansion began in 2014 (Table 1). Similarly, 27 (84%) expansion states implemented a policy allowing pharmacists to dispense naloxone without a prescription [‘pharmacist dispensing’ compared to 15 (83%) of the non-expansion states]. Finally, 25 (78%) of the expansion states implemented a policy from 2009 to 2013 allowing for prescribing to third parties (‘third-party naloxone’) compared to 16 (89%) of the non-expansion states.

Table 1.

Demographic characteristics of states by Medicaid expansion status, 2009–13.

Covariate	Non-expansion states (n= 18)	Expansion states (n = 32)
Naloxone policies
Standing order	13 (72%)	24 (75%)
Pharmacist dispensing	15 (83%)	27 (84%)
Third-party naloxone	16 (89%)	25 (78%)
Demographic characteristics
% with SUD	2.3 (2.2–2.3)	2.6 (2.6–2.7)
% non-medical use of opioids	4.4 (4.3–4.5)	4.8 (4.7–4.8)
% in poverty	15.7 (15.4–16.0)	14.3 (14.1–14.6)
% unemployed	4.4 (4.3–4.5)	4.7 (4.7–4.8)
% male	49.3 (49.2–49.4)	49.3 (49.2–49.4)
% White, non-Latino	78.9 (77.8–79.9)	76.5 (75.3–77.7)
% Black, non-Latino	13.0 (11.8–14.1)	10.0 (9.1–10.8)
% Latino	8.9 (8.1–9.8)	11.1 (10.3–11.9)
%18–24 years of age	10.9 (10.1–11.8)	10.0 (10.0–10.1)
%25–44 years of age	25.8 (25.7–25.9)	26.3 (26.2–26.4)
%45–64 years of age	25.2 (24.3–26.1)	27.0 (26.9–27.1)
% 65+ years of age	13.5 (13.3–13.7)	13.7 (13.6–13.8)

95% confidence intervals (CIs) displayed in parentheses. 95% CIs may not be symmetrical due to rounding. Authors’ analyses of data from the Prescription Drug Abuse Policy System (PDAPS; naloxone policies), the National Survey on Drug Use and Health (NSDUH) produced by the Substance Abuse and Mental Health Services Agency [SAMHSA; rates of substance use disorder (SUD) and non-medical use of prescription pain relievers], and the American Community Survey produced by the US Census Bureau (poverty and unemployment rate; age, gender, race/ethnic composition).

Expansion states, on average, have populations with a higher proportion of younger adults than non-expansion states: specifically, expansion states have a greater proportion of their population aged between 25 and 44 years compared to non-expansion states (26.3 versus 25.8) (Table 2). These are ages when the prevalence of opioid use disorder is especially high. Additionally, expansion states have higher rates of substance use disorder (2.6% versus 2.3), non-medical use of opioid pain relievers (4.8% versus 4.4) and unemployment (4.7% versus 4.4) compared to non-expansion states prior to Medicaid expansion. Taken together, these factors suggest that the burden of the opioid epidemic may be higher in expansion states. Indeed, Ohio’s Governor Kasich and others have suggested that the opioid epidemic was part of the reason that many Republican-led states (e.g. Ohio, Michigan and West Virginia) expanded their Medicaid program [17].

Table 2.

Medicaid-covered dispensed naloxone prescriptions per 100 000 Medicaid enrollees by expansion status, 2009–16.

	Non-expansion states (n= 18)		Expansion states (n = 32)
	Mean	95% CI	Mean	95% CI
2009	18.5	9.5–27.4	7.2	0.9–13.5
2010	25.0	11.1–38.9	16.6	7.31–25.8
2011	31.5	18.9–44.2	15.8	7.7–23.9
2012	23.2	15.2–31.2	21.1	13.9–28.3
2013	31.2	16.4–46.0	24.5	16.8–32.
Pre-expansion	25.7	20.4–31.0	17.4	13.9–20.8
2014	35.4	21.1–49.7	56.9	41.1–72.7
2015	70.0	25.1–115.0	139.8	100.7–178.8
2016	133.3	65.2–201.3	381.9	229.1–475.0
Post-expansion	83.1	53.4–112.8	215.6	172.1–259.1
Pre-post difference	57.4		195.2
Difference-in-differences			140.8

Units are not population-standardized. Standardized analyses look similar. Suppressed values are not included as ‘0’ in these means. However, true ‘0’s are included these means. Means represent mean number of dispensed prescriptions per state-quarter in given year. The number of state-quarter-year observations included varies by rate of suppression. Authors’ analysis of Medicaid Drug Rebate Program Data (MDRP). Naloxone for opioid overdose reversal is defined using NDCs and excludes naloxone combined with buprenorphine (e.g. Suboxone, used for treatment of opioid use disorder) and naloxone combined penatzocine (used for treatment of pain); t-tests use un-pooled variance. Expansion states include all states that expanded during the study period, which are fewer than have expanded as of 31 December 2018. CI = confidence interval; NDC = National Drug Code.

Medicaid expansion and naloxone prescriptions

Prior to Medicaid expansion, the number of naloxone prescriptions paid for by Medicaid was essentially identical in expansion states compared to non-expansion states (Fig. 1); that changed beginning in 2012–13. This pre-dates the bulk of the Medicaid expansion that started in 2014. However, it is important to recognize that five states and the District of Columbia initiated at least partial coverage expansion via Medicaid between 2010 and 2012. Some of the states that expanded their Medicaid programs before 2014 also have high rates of opioid misuse and overdose (e.g. Connecticut, New Jersey and District of Columbia).

Figure 1.

Prescriptions of Medicaid-covered naloxone per 100 000 Medicaid enrollees by state expansion status, 2009–16. Naloxone prescriptions include Evzio and Narcan, but exclude naloxone hydrochloride combined with buprenorphine or buprenorphine hydrochloride and pentazocine. Authors’ analysis of Medicaid Drug Rebate Program data, which are produced by the Centers for Medicaid and Medicare Services. Data are provided quarterly at the National Drug Code (NDC)-state level, and aggregated to the state-year level for all naloxone hydrochloride NDCs. [Colour figure can be viewed at wileyonlinelibrary.com]

Between the years 2009–13, non-expansion states had a total of 3800 Medicaid-covered naloxone prescriptions and expansion states had total of 4025 naloxone prescriptions. Examining per-capita use, non-expansion states had a higher number per 100 000 population of Medicaid-covered naloxone prescriptions in each of the pre-expansion years compared to expansion states (Table 2 reports state-quarter-year average). This pattern changed with Medicaid expansion. The difference is clear in 2016, when expansion states had more than 38000 Medicaid-covered prescriptions for naloxone, while non-expansion states had fewer than 7000. The unadjusted difference-in-differences estimate suggests that Medicaid expansion was associated with an increase of more than 31000 naloxone prescriptions in expansion states compared to non-expansion states after Medicaid expansion. Per-capita comparisons are similar: non-expansion states averaged 25.7 prescriptions per 100 000 Medicaid enrollees before expansion, while expansion states averaged 17.4 prescriptions per 100 000 Medicaid enrollees. Post-expansion, states that expanded Medicaid had a greater number of naloxone prescriptions than states that did not expand (215.6 versus 83.1 per 100 000 Medicaid enrollees).

Difference-in-differences analyses

We augmented the visual inspection of parallel pre-trends (Fig. 1) with a formal test of parallel pre-trends that supports the assumption as reasonable. The formal test performed by regressing the treatment variable on the pre-trend period resulted in a small and insignificant estimate [0.14, 95% confidence interval (CI) = −0.44, 0.72]. Our preferred difference-in-differences specification yielded estimates showing that the implementation of the ACA’s Medicaid expansion is associated with an increase of 78.2 (95% CI = 16.0, 140.3) Medicaid-covered dispensed naloxone prescriptions (Table 3).²

Table 3.

Changes in Medicaid-covered naloxone prescriptions between expansion and non-expansion states, 2009–16.

	Estimate (SE)
	Pooled post-period		Unpooled post-period
Variable	Without state FEs	With state FEs	Without state FEs	With state FEs
Expansion state	−50.5 (82.3)	−33.2 (104.9)	–	–
Post-expansion	−11.7 (17.6)	−8.2 (16.0)	–	–
DD estimate (2014–16)	79.2*(32.0)	78.2*(31.0)	–	–
Expansion state	–	–	−68.5 (87.0)	52.7 (105.9)
2014	–	–	16.4 (14.0)	24.1 (12.0)
2015	–	–	47.7 (26.5)	51.5 (27.7)
2016	–	–	−75.6 (166.8)	−57.3 (177.3)
2014 DD estimate	–	–	16.4 (15.3)	7.4 (12.3)
2015 DD estimate	–	–	43.7 (35.6)	45.9 (33.3)
2016 DD estimate	–	–	203.1**(68.4)	205.5**(67.0)
n	1632	1632	1632	1632
R2	0.33	0.05	0.05	0.05

Naloxone prescriptions include Evzio and Narcan, but exclude naloxone hydrochloride combined with buprenorphine or buprenorphine hydrochloride and pentazocine. Regressions without state fixed effects adjusted for state-level naloxone access policies and opioid use disorder burden, socio-economic, gender, age and racial/ethnic composition of the state in each year. State-level naloxone policies include standing order laws, third-party prescribing laws and Good Samaritan laws. Opioid use disorder burden includes proportion of population with substance use disorder and proportion of population that misused prescription opioid pain relievers in the past 12 months. Regressions with state fixed effects are adjusted for socio-economic, gender, age and racial/ethnic composition of the state in each year. All regressions account for serial autocorrelation by calculating clustered standard errors. FE = fixed effect; DD = difference-in-differences; SE = standard error.

^*P < 0.05

^**P < 0.01

^***P < 0.001.

Authors’ analysis of Medicaid Drug Rebate Program (MDRP) data. These data are produced quarterly for each National Drug Code (NDC) code–state combinations from 2009 to 2016 and aggregated up to the state-year level for all naloxone hydrochloride NDC codes. State-level naloxone laws are compiled from the Prescription Drug Abuse Policy System (pdaps.org). Demographic regression covariates are 1-year estimates from the American Community Survey, produced by the US Census Bureau. Opioid use disorder burden data are from the National Survey on Drug Use and Health, produced by the Substance Abuse and Mental Health Services Agency (SAMHSA).

Similar to the unadjusted trends seen in Fig. 1, the difference between expansion and non-expansion states post-expansion seems to be driven by an increasing impact of Medicaid expansion over time, as indicated by the year-specific difference-in-difference estimates. Specifically, the difference-in-differences estimate for 2016 (205.5; 95% CI = 70.8, 340.2) is 27.8 (7.4; 95% CI = −17.3, 32.0) and 4.5 (45.9; 95% CI = −21.1112.9) times larger than the estimate for 2014 and 2015, respectively (Table 3).

Sensitivity analysis

The difference-in-differences estimate for our preferred specification is quite stable throughout a number of different specifications, including adding naloxone state regulations and removing state fixed effects (Supporting information, Table S1 and Table 2, respectively). This suggests that the changes in dispensed naloxone prescriptions are probably the result of Medicaid expansion and not some other policy shock, including the introduction of user-friendly formulations of naloxone. The removal of other demographic and other state characteristics does not change the difference-in-differences estimate substantially (Supporting information, Table S2). Again, this suggests that Medicaid expansion was responsible for the increase in naloxone prescriptions.

Differences in functional forms for models of naloxone prescriptions do not affect the difference-in-differences results significantly. Standardizing the number of prescriptions per 100 000 Medicaid enrollee results in a difference-in-differences estimate of 5.3 prescriptions (95% CI = 1.0, 9.3). Similarly, adjusted difference-indifferences estimation using the natural logarithm of prescriptions (as the distribution of prescriptions is probably right-skewed) suggests that Medicaid expansion is associated with an increase in the number of naloxone prescriptions of 109.6% (95% CI = 77.1, 148).

When we exclude the five New England states from our analysis, we find that the results are largely unchanged (78.8; 95% CI = 12.5, 145.1) for all transformations of the outcome (Supporting information, Table S3). This suggests that the fentanyl crisis in New England was probably not the driver of the increased supply of naloxone in expansion states compared to non-expansion states after Medicaid expansion.³ To further explore this point we estimated a reverse regression where we put Medicaid expansion status on the right-hand side of our models and pre-period fentanyl mortality on the left-hand side. The estimated coefficient for the expansion variable was not significantly different from zero. Lastly, when we conduct the same difference-in-differences analysis with buprenorphine prescriptions rather than naloxone prescriptions, the same qualitative pattern emerges (Supporting information, Table S4). However, the relationship between Medicaid expansion and changes in buprenorphine prescriptions (44 124.3; 95% CI = 13 506.7, 74741.9) is orders of magnitude larger than the relationship between Medicaid expansion and changes in naloxone prescriptions (78.2; 95% CI = 16.0,140.3).

DISCUSSION

Our results underscore that Medicaid expansion has been and will probably continue to be an important tool for states in the opioid epidemic. Expansion has probably put naloxone into the hands of some of the people best positioned to prevent death from an overdose—individuals without opioid use disorder and their friends and members. Medicaid spending, overall and as a share of national naloxone spending [18], increased during our study period (2009–16). In 2009, Medicaid sales of naloxone represented fewer than 1% of total naloxone sales and remained between 1 and 2% of all national spending until 2015, when Medicaid spending represented 9% of total naloxone spending. By 2016, Medicaid spending grew to a quarter of the naloxone spending nation-wide. Our difference-in-differences result implies that Medicaid expansion alone accounted for approximately 8.3% of the growth in naloxone units from 2009 to 2016. The $19 million spent by state Medicaid programs in 2016 on naloxone represents a sum considerably larger than most of the federal grant programs aimed at expanding naloxone use.

Our estimate of a 78-unit increase of naloxone per state-year associated with Medicaid expansion is smaller than unadjusted individual state reports. For example, states such as Arizona, Ohio and Maryland all report substantial increases in the use of naloxone in their Medicaid programs, despite significant differences among them. The Maryland Medicaid program reported an increase in naloxone prescriptions dispensed of nearly 1000 from March—September of 2014 to October 2014-December 2015 and 275 from December 2015-June 2016 for a 3-year average increase (since 2013) of 427 prescriptions [19]. Additionally, our estimated increase in naloxone prescriptions attributable to Medicaid expansion implies that expansion saves an estimated 22.7 lives per state-year if the historical ratio of lives saved per prescription continued.It is also notable that state legislation promoting the use of naloxone appears to have little impact on the levels of naloxone dispensing within Medicaid. This may indicate that financial access was a greater barrier to obtaining naloxone compared to concerns about liability or the ability to obtain a prescription for naloxone.

Finally, it is important to emphasize that Medicaid and its expansion are but one part of a state’s strategy to promote the availability and use of naloxone, particularly in an environment where overdoses are increasingly probable and lethal. Naloxone-specific grants to states, reduced prices for user-friendly versions of naloxone and improved first-responder awareness of and education about naloxone are also critically important. Policymakers and naloxone manufacturers and distributors must also rethink what constitutes an appropriate dose of naloxone and to make this dosage available in standardized kits as the role of fentanyl in the opioid epidemic continues to grow.

Supplementary Material

Supplemental Material

Table S1 Full regression results for the changes in Medicaid-covered naloxone prescriptions between expansion and non-expansion states, 2009–2016.

Table S2 Robustness of difference-in-differences estimates to the removal of covariates.

Table S3 Robustness of difference-in-difference estimates to the exclusion on New England (NE) states and their fentanyl crisis.

Table S4 Comparison of difference-in-differences estimates for naloxone and buprenorphine.