Association of Copayments with Healthcare Utilization and Expenditures among Medicaid Enrollees with a Substance Use Disorder

Abstract

The purpose of this study was to examine the association between copayments and healthcare utilization and expenditures among enrollees with substance use disorders. This study used claims data (2020–2021) from a private insurer participating in Arkansas’s Medicaid expansion. We compared service utilization and expenditures for enrollees in different Medicaid program structures with varying copayments. Enrollees with incomes above 100% FPL (N=10,240) had copayments for substance use treatment services while enrollees below 100% FPL (N=2,478) did not. Demographic, diagnostic, utilization, and cost information came from claims and enrollment information. Substance use and clinical comorbidities were identified using claims from July through December 2020, and utilization and costs were evaluated in 2021. Generalized linear models (GLM) were used to estimate outcomes using single equation and two-part modeling. A gamma distribution and log link were used to model expenditures, and negative binomial models were used to model utilization. A falsification test comparing behavioral health telemedicine utilization, which had no cost sharing in either group, was used to assess whether differences in the groups may be responsible for observed findings. Substance use enrollees with copayments were less likely to have a substance use or behavioral health outpatient (−0.04 PP adjusted; p=0.001) or inpatient visit (−0.04 PP; p=0.001) relative to their counterparts without copayments, equal to a 17% reduction in substance use or behavioral health outpatient services and a nearly 50% reduction in inpatient visits. The reduced utilization among enrollees with a copayment was associated with a significant reduction in total expenses ($954; p=0.001) and expenses related to substance use or behavioral health services ($532; p=0.001). For enrollees with at least one behavioral health visit, there were no differences in outpatient or inpatient utilization or expenditures between enrollees with and without copayments. Copayments had no association with non-behavioral health or telemedicine services where neither group had cost sharing. Copayments serve as an initial barrier to substance use treatment, but are not associated with the amount of healthcare utilization conditional on using services. Policy makers and insurers should consider the role of copayments for treatment services among enrollees with substance use disorders in Medicaid programs.

1. Introduction

More than 1 in 5 Americans suffer from some form of substance use, and over 80% of individuals that have substance use disorders do not receive treatment (Dave & Mukerjee, 2011; Olfson et al., 2019). One of the most commonly cited barriers to treatment for substance use disorders is financial barriers (Ali et al., 2017; Browne et al., 2016; Han et al., 2017; Priester et al., 2016; Tai & Volkow, 2013). This issue is exacerbated by the fact that many individuals with a substance use disorder are unable to obtain insurance through an employer given their frequent state of cognitive impairment (McCabe & Wahler, 2016) and that individuals with substance use disorders are more likely to have a mental illness comorbidity (Harris et al., 2019; Minkoff, 2013).

Medicaid expansion efforts under the Affordable Care Act aimed to decrease the number of uninsured individuals, and the effort is estimated to have reduced the number of uninsured individuals with substance use disorders by at least 13.7% (Andrews et al., 2019). Despite this gain in coverage, barriers continue to exist for individuals with a substance use disorder (Creedon & Cook, 2016), and individuals still face considerable barriers in accessing behavioral health treatment (Dickson-Gomez et al., 2022; Olfson et al., 2021). Additionally, some evidence suggests that enrollees with substance use disorder in the post Affordable Care Act (ACA) period had greater rates of comorbidities compared to their pre-ACA counterparts (Campbell et al., 2018) likely reflecting greater access to care, but also underscoring the need for treatment.

Arkansas uniquely expanded Medicaid by using Medicaid funding to purchase qualified health plans on the Marketplace, with individuals having varying copayment structures based on the federal poverty level of their household income (Holder, 2018). Newly enrolled individuals below 100% the federal poverty level retain no cost sharing such as those enrolled in traditional Medicaid, but individuals above this income threshold can be subject to cost sharing that does not exceed 5% of their family income (Guyer et al., 2015). There were more than 300,000 individuals in Medicaid private health plans within Arkansas in 2021 (Arkansas Health and Opportunity for Me (ARHOME) Section 1115 Demonstration Waiver, 2021). One of the two private insurers in Arkansas granted our research team access to their claims data through a contractor agreement at no cost to the insurer. Through this agreement, we were provided a unique opportunity to study the impact of copayments on access to substance use treatment for Medicaid enrollees.

Previous studies evaluated the impact of different insurance coverage on healthcare service utilization among enrollees with substance use disorders. Cost sharing in the form of higher deductibles among substance use enrollees was shown to result in lower healthcare services utilization (Satre et al., 2020). Type of insurance was also associated with the odds of receiving substance use treatment, with evidence suggesting that private enrollees are only half as likely to receive substance use treatment compared to Medicaid enrollees (Bouchery et al., 2012). Despite this important work, literature that examines the impact of varying copayments on different types of behavioral health and substance use healthcare utilization among substance use enrollees is lacking. Furthermore, the impact of cost sharing on the utilization of substance use services is largely unknown among the Medicaid population which historically has had higher prevalence of substance use disorders relative to individuals with other types of insurance (Wells et al., 2002). Therefore, this study evaluated the impact of cost sharing (copayments) on the utilization of behavioral health and substance use services among a cohort of Medicaid enrollees with private coverage through a qualified health plan.

2. Material and methods

2.1. Participants

To study the impact of copayments on healthcare utilization among enrollees with a substance use disorder, we used claims data from a private insurer, years 2020–2021, for Medicaid enrollees aged 19–64 with continuous coverage in the qualified health plan. Enrollees with a substance use disorder were identified using claims data in the 6 months before 2021 (July 2020–December 2020) based on ICD-10 codes for substance use disorders (F10–F19) (Baillargeon et al., 2021; Peterson et al., 2021; Smelson et al., 2012). The individual was identified as having had a substance use disorder if they had any of the specified F codes as a primary or secondary diagnosis, regardless of claim type.

2.2. Outcomes of Interest

The primary goal was to compare utilization and expenditures for both substance use services and services that were not substance use related. The study measured utilization and expenditures in 2021 (January 1st to December 31st) to reduce endogeneity between substance use disorder diagnosis and utilization. Expenditure outcomes included total expenses and behavioral health and/or substance use (BH/SU) expenses. Expenditures were defined as the sum of the amount the insurer paid and all out-of-pocket expenditures (if applicable) by the enrollee which included copayments, deductibles, and coinsurance. The study identified outcomes related to BH/SU services using a series of internal flags provided by the insurer. Utilization outcomes included BH/SU outpatient visits, telehealth visits, BH/SU inpatient visits, and non-BH/SU inpatient visits, measured as counts.

Falsification test

There was a period in 2021 (January 1st to October 1st) during which the private insurer provided telehealth visits at $0 cost sharing for all enrollees to mitigate the access burdens associated with COVID-19. Thus, telehealth visits for BH/SU visits served as a falsification test in our study. Specifically, if we observed no differences in telehealth utilization between the cost-sharing and no cost-sharing groups, we would have greater confidence that the differences observed in other outcomes (e.g., BH/SU visits) could be attributed to the difference in the cost-sharing structure for the other services, rather than demographic differences and other unobserved differences between groups.

2.3. Independent Variables

Our primary independent variable of interest was the cost-sharing structure among Medicaid participants. There are two types of cost structures: the first cost structure has $0 cost-sharing for all services, and the second structure has limited cost sharing for some services, including a $4 copayment for outpatient services and a $60 copayment for inpatient services. Enrollees were assigned to one of the two cost sharing structures based on their household income. Individuals that had an income below 100% of the federal poverty level were placed into the $0 cost-sharing group. Individuals earning between 100% and 138% of the federal poverty level were assigned to the group with the limited cost-sharing structure.

Covariates included health status and demographic characteristics. The study controlled for health status by the number of Elixhauser comorbidities identified for each participant in the 6 months prior (i.e., July 2020– December 2020) to the outcome measurement period of 2021 (Simard et al., 2018). Demographic covariates included gender and geographic region using seven defined regions in Arkansas. Additionally, we linked the CDC 2018 Social Vulnerability Index at the county-level to control for county-level sociodemographic information such as poverty, unemployment, income, education (percent without a high school diploma), minority composition, English proficiency, and access to a vehicle (CDC/ATSDR Social Vulnerability Index (SVI), 2022). Finally, the study controlled for access to healthcare providers by using the number of psychiatric physicians per 1,000 residents in each county, health provider shortage area (HPSA) designation provided by the Area Health Resource File, and the median distance within a county to the nearest hospital with alcohol and drug abuse inpatient from the Agency for Healthcare Research and Quality Social Determinants of Health Database (Area Health Resources Files, 2022; Social Determinants of Health Database (Beta Version), 2023). All measures other than gender, HPSA designation, and geographical region were continuous variables.

2.4. Statistical Analysis

Descriptive statistics included calculations of differences in covariates among individuals in each of the two cost-sharing groups. Two sample t-tests and chi square tests assessed unadjusted differences in the outcome measures between the two cost-sharing groups.

Health care expenditures and utilization outcomes are typically right-skewed and prone to heteroscedasticity. Thus, a generalized linear model (GLM) was used with a gamma distribution and log link to model expenditures (Manning et al., 2005), and negative binomial models were used for utilization measures to address over-dispersed counts (variances greater than the means) (Deb & Norton, 2018).

Additionally, healthcare expenditures and utilization variables tend to have a high distribution of 0 or null values. Thus, it becomes difficult to separate whether differences are observed because of the number of individuals using a resource (extensive margin) or due to the individuals already using resources increasing or decreasing their utilization (intensive margin) (Mullahy, 1998). The study evaluated the proportion of non-zero values for each outcome to assess whether the third model (estimating the intensive margin) was necessary given if only a small proportion of zeros were present there would be no need to derive the intensive margin (Belotti et al., 2015; Mullahy, 1998).

After assessing group differences and the proportion of zeros, we ran three separate models for each of the outcomes to assess differences in the extensive and intensive margin for each outcome. The first model used the appropriate generalized linear model (GLM) with the entire population (i.e., the model contained individuals with values equal to or greater than 0). The second model followed the two-part logic with a logit regression that predicted the probability of having a value greater than 0 (or probability of having some level of utilization or expenses) to estimate the extensive margin followed by a conditional expenditure or utilization model to estimate the intensive margin. We report differences between the two cost-sharing groups as the percentage point difference. These first two models evaluated differences within the extensive margin for each of the outcomes. The third model used the appropriate GLM for each of the outcomes among those that had at least some utilization or expenses defined by the outcome measure being greater than 0, which measures differences at the intensive margin. After the regression analyses, we computed the adjusted means for each of the outcomes so that percent change among those with and without copays could be computed based on the average marginal effects.

Each of the modeling approaches used the following model specification:

$$\text{Y}={\beta }_0+{\beta }_1{X}_{\text{copayment}}+{\beta }_2{X}_{\text{gender}}+{\beta }_3{X}_{\text{age}}+{\beta }_{4\text{Elixhauser\_number}}+{\beta }_5{X}_{\text{Psychiatrist\_number}}+{\beta }_6{X}_{\text{HPSA}}+{\beta }_7{X}_{\text{Distance\_Inpatient\_Substance}}+{\beta }_8{X}_{\text{Region}}+{\beta }_{9-15}{X}_{\text{CountySDOH\_Measures}}+\epsilon$$

with the primary coefficient of interest being β₁, which indicates the difference in the specified outcome associated with having a copayment.

The study conducted data pre-processing in SAS Enterprise Guide 8.3, and statistical analysis in R 4.2.2 (R: The R Project for Statistical Computing, 2023). Outcomes measurements are reported as average marginal effects and calculated using the margins package (Leeper, 2021). Standard errors are clustered at the county level using the lmtest package (Zeileis & Hothorn, 2002). The University of Arkansas for Medical Sciences Institutional Review Board deemed this study non-human subjects research (#274731).

3. Results

The insurance company covered over 300,000 Medicaid beneficiaries in 2020–2021. We identified 12,733 adults with continual insurance in 2020 and 2021 and substance use disorders in 2020. Among the final sample 10,240 individuals had no cost sharing and 2,478 had some degree of cost-sharing (Table 1). The Medicaid group with some cost sharing had a greater proportion of females than the $0-cost sharing group, with no other significant differences between the two groups (Table 1).

Table 1.

Descriptive statistics for Medicaid Enrollees, by Cost Sharing Type

Socio-Economic Demographics	Insurance Plan
	Medicaid no Cost Sharing	Medicaid Cost Sharing
Total (N)	10,240	2,478
Gender
Female	54.4%	63.4%
Age	42.6	42.4
Number of Elixhauser comorbidities	2.9	2.7
Number of psychiatric physicians per 1000a	0.067	0.061
Health Provider Shortage Areaa
Underserved	16.0%	13.4%
Partially	78.7%	81.3%
Distance in miles to the nearest hospital with alcohol and drug use inpatient carea	17.3	17.8
Percent povertya	18.5%	18.3%
Percent unemployeda	6.0%	5.8%
Per capita incomea	$24,869	$24,745
Percent no high school diplomaa	14.3%	14.4%
Percent minority race/ethnicitya	27.3%	25.5%
Percent limited Englisha	1.4%	1.5%
Percent no vehiclea	6.6%	6.4%
Region
Central North	27.3%	25.5%
North East	17.6%	17.4%
North West	15.2%	17.8%
South Central	7.8%	8.9%
South East	9.3%	8.2%
South West	10.3%	8.8%
West Central	12.5%	13.5%

^aCounty-level variable

When comparing the means for each of the outcome measures, the group with $0 cost-sharing had higher means for all outcomes except for the falsification tests for telemedicine visits (Table 2). Notably, the greatest difference was in inpatient visits related to BH/SU, which were more than double in the group with no copayment compared to the Medicaid group with a $60 copayment for the service (0.61 visits vs. 0.29 visits; p=0.001). In addition, the majority of individuals did not have any healthcare utilization in the form of BH/SU outpatient visits (less than 25%), telehealth visits (12% or less), or inpatient visits (less than 9%). Approximately a quarter of participants had either an inpatient or outpatient expense associated with BH/SU. However, virtually all participants (99.7% for both groups) had greater than $0 total expenses.

Table 2.

Unadjusted means of each study outcome and percent with each study outcome, by insurance plan.

Outcome Variables	Insurance Plan		pa	Insurance Plan		pb
	No Cost Sharing	Cost Sharing		No Cost Sharing Medicaid no Cost Sharing	Medicaid Cost Sharing
	Mean (SD)	Mean (SD)		%	%
BH/SU Outpatient Visits	2.08 (6.02)	1.81 (5.64)	0.033	23.6%	19.9%	<0.001
BH/SU Tele-Health Visits	0.64 (2.42)	0.66 (2.66)	0.774	12.0%	11.0%	<0.001
BH/SU Inpatient Visits	0.61 (2.74)	0.29 (1.88)	0.001	8.2%	4.0%	0.020
Non-BH/SU Inpatient Visits	0.17 (0.53)	0.14 (0.53)	0.016	11.5%	9.9%	0.868
Total Expenses	$9,033 (15,069)	$7,756 (13,207)	0.001	99.7%	99.7%	<0.001
BH/SU Expenses	$1,286 (4,358)	$812 (3,247)	0.001	27.8%	22.5%	0.173

BH/SU=Behavioral Health/Substance Use

^aP-values come from two-sample t-tests

^bP-values come from chi square tests

To examine whether differences seen between the two cost-sharing groups were due to differences in the extensive or intensive margin we conducted three separate models for each outcome (Table 3). Please see the appendix for all coefficients from each of the models. The logistic regression for BH/SU outpatient visits indicated that the probability of having an outpatient visit is significantly less (4 percentage points less likely; p = 0.001) for the group with the $4 copayment compared to the group with no copayment (Table 4). In the GLM with all participants (individuals with 0 visits included) the model indicated that on average the group with the $4 copayment has almost half (−0.46; p= 0.002) an outpatient visit less per individual. However, when we conducted the same GLM on only individuals that had at least one outpatient visit, we saw no differences between the two cost-sharing groups, indicating the differences observed between the two groups are due to differences in the extensive margin and not the intensive margin. There was a period of 9 months (January 1^st, 2021 – October 1^st, 2021) in which $0 cost-sharing occurred for any enrollee who had a telemedicine visit. Thus, we used telehealth service as a falsification test to determine if differences in healthcare utilization between the two groups could be attributed to other factors. The falsification test produced null results for the logistic regression (having telemedicine visits related to BH/SU) and for both GLM models (among all individuals and among users of telemedicine services). The logistic regression for inpatient visits associated with BH/SU indicated the probability of having an inpatient BH/SU visit was also 4 percentage points lower for the group with a $60 copayment vs. the group with no cost sharing. Additionally, the GLM that included all participants indicated that those with the $60 copayment had a reduced number of BH/SU inpatient visits (−0.36; p = 0.001). However, in the GLM among those who had at least one BH/SU inpatient visit, we saw no differences between the groups, indicating the differences observed in inpatient usage for BH/SU services are due to differences in the extensive margin rather than the intensive margin. For non-BH/SU inpatient services we observe no significant differences between the two cost-sharing groups in either the intensive or extensive margin.

Table 3.

Adjusted associations^a of having a copayment for outpatient services and selected outpatient and inpatient utilization.

	Logistic			GLM			GLM among those with >=1
	AME	95%CI	p	AME	95%CI	p	AME	95%CI	p
BH/SU outpatient	−0.04	(−0.06, −0.02)	0.001	−0.46	(−0.74, −0.18)	0.001	0.13	(−0.76, 1.03)	0.772
BH/SU telehealth	−0.01	(−0.03, 0.01)	0.119	−0.09	(−0.26, 0.08)	0.313	0.42	(−0.27, 1.11)	0.231
BH/SU inpatient	−0.04	(−0.05, −0.03)	0.001	−0.36	(−0.48, −0.24)	0.001	−0.06	(−1.36, 1.24)	0.927
Non-BH/SU inpatient	−0.01	(−0.02, 0.01)	0.079	−0.15	(−0.04, 0.01)	0.196	0.03	(−0.08, 0.14)	0.619
Total expenses ($)	N/A	N/A	N/A	−954	(−1519, −390)	0.001	N/A	N/A	N/A
BH/SU expenses ($)	−0.05	(−0.07, −0.03)	0.001	−532	(−716, −350)	0.001	−798	(−1278, −318)	0.001

AME = Average Marginal Effects

BH/SU=Behavioral Health/Substance Use

GLM = Generalized Linear Model

^aFor all models, having a copayment for outpatient BH/SU visits was the “treatment” level, with the reference category being no copayment. There were virtually no patients with total expenses of $0, so no Logistic regression or GLM>1 was necessary for that outcome. All models were conducted among the full sample and were adjusted for patient gender, age, number of Elixhauser comorbidities, Region, and county-level variables. All coefficients are reported as average marginal effects.

Table 4.

Adjusted^a means for the outcomes.

Outcome Variables	Insurance Plan
	Medicaid no Cost Sharing	Medicaid Cost Sharing
BH/SU Outpatient Visits	23.6%	19.7%
BH/SU Tele-Health Visits	12.1%	10.7%
BH/SU Inpatient Visits	8.1%	4.3%
Non-BH/SU Inpatient Visits	11.4%	10.3%
BH/SU Expenses	27.8%	22.4%

AME = Average Marginal Effects

BH/SU=Behavioral Health/Substance Use

^aAll models were conducted among the full sample and were adjusted for patient gender, age, number of Elixhauser comorbidities, Region, and county-level variables.

A single equation model was used for total expenses because virtually all participants had some expenses (99.7%; Table 2). The GLM among all individuals indicated that on average those in the group with some copayment generated $954 (p = 0.001) fewer expenses annually compared to individuals in the group with no copayment. Additionally, we looked at expenses specifically related to BH/SU (Table 3). The logistic regression indicated that the probability of having a BH/SU expense among those with some copayment is significantly less (5 percentage points less likely; p = 0.001) likely than for those with no copayment. The GLM among all participants suggests those with some copayment have $532 (p = 0.001) lower BH/SU-related expenses annually than those with no copayment. In addition, BH/SU expenses are the only outcome variable that indicated an effect in the intensive margin as well. Those with a copayment had $798 (p = 0.001) fewer BH/SU expenses annually compared to those with no copayment among those who had at least some BH/SU expenses (Table 3).

The adjusted means for the percent of individuals utilizing each service is expressed in Table 4. Using the adjusted mean from Table 4 and the average marginal effect in Table 3, we computed the percent change in the utilization of each service given the individual’s insurance plan (i.e., cost sharing or not). If individuals with 0 cost sharing had to pay the $4 copayment for BH/SU outpatient services, we would expect an approximate reduction of 4.1 percentage points. Thus, we expect a 17% reduction (−4.1/23.6) in the utilization of BH/SU outpatient services with the introduction of a $4 copayment. The percent change is even greater for BH/SU inpatient services with an almost 50% percent reduction (−4.1/8.1) in the utilization of inpatient services with the introduction of a $60 copayment.

To investigate whether differences seen in expenditures among the two cost-sharing groups were due to the number of services used or due to reimbursement differences between the two cost-sharing groups, we compared the reimbursement amount (allowed amount) for the most common CPT procedure associated with outpatient visits (90837) [results not shown]. Both groups reported the same median value of $148 for the service. Given Psychotherapy 1 (or CPT 90837) constitutes more than 40% of all outpatient services related to BHA/SU, this provides confidence that the significant difference seen in expenses is more likely due to the number of services used rather than differences in reimbursement.

4. Discussion

This study used insurance claims data from a private insurer in Arkansas to assess whether cost sharing in the form of copayments affected the utilization of outpatient visits, inpatient visits, and healthcare expenditures within a Medicaid population with plans through the Marketplace. We found that copayments were associated with a decrease in the utilization of behavioral health and substance use related services. Substance use disorders affect millions of Americans, and most of these individuals never seek treatment (Dave & Mukerjee, 2011; Lipari, 2014; Olfson et al., 2021). Previous studies have shown that utilization of health services among substance use populations differs based on payer type (Bouchery et al., 2012), and additional cost sharing can negatively impact healthcare utilization among enrollees with substance use disorders (Dickson-Gomez et al., 2022; Satre et al., 2020). The findings from our study further highlight the impact financial barriers (i.e., copayments) on access to care for BH treatment among low-income individuals (Medicaid enrollees).

Our falsification test comparing utilization of no cost sharing telehealth services strengthened our assumption that differences in utilization for all other outcomes were associated with differences in copayments for services rather than observed and unobserved group differences. Otherwise, we would have expected to see a significant difference in telemedicine utilization between the two groups, similar to the difference we observed with outpatient visits related to behavioral health or substance use among the group with a $4 copayment. This difference occurred in the extensive margin or that individuals with no copayment were more likely to have initiated BH/SU treatment (or to have had at least 1 outpatient visit) than those with the $4 copayment. However, we did not observe any significant difference in the intensive margin, where individuals who have had at least one outpatient visit (related to behavioral health or substance use) did not differ in the number of outpatient visits. We see a similar effect when looking at inpatient visits related to behavioral health or substance use. Having a $60 copayment was associated with having fewer inpatient visits related to behavioral health or substance use relative to having no copayment. Again, we did not see any difference in association within the intensive margin for inpatient visits related to behavioral health or substance use. These findings are consistent with previous literature demonstrating that the vast majority of individuals with substance use may not initially perceive a need for treatment (Ali et al., 2015; Andrade et al., 2014). On the other hand, we found no difference in the utilization of non-BH/SU inpatient services among the two groups suggesting that non-BH/SU inpatient services may be less reliant on cost-sharing structures. Some previous evidence among low-income enrollees suggested that inpatient utilization was unaffected by the cost-sharing structure (Roberts et al., 2020). The overall findings suggest that any form of copayment may serve as an initial barrier to treatment among substance use enrollees, but once this barrier is overcome differences in utilization of services is negligible. Additionally, the results suggest treatments related to substance use are more sensitive to financial barriers than other services. This is evident by the fact we observe no differences in utilization of inpatient services not related to substance use among the two populations.

The argument that substance use services may be more sensitive to copayments may be further strengthened by the fact that virtually all study participants have at least some expenses suggesting they received some form of healthcare services. Total health care expenditures were higher among the group with no copayment. Expenses related to behavioral health or substance use treatment were also higher among the group with no copayment both within the extensive and intensive margins. The difference seen in the intensive margin may partially be explained by the greater proportion of individuals with inpatient-related services in the group with no copayment. Thus, among individuals with some BH/SU expenses, having no copayments was associated with higher average expenses than those with positive BH/SU expenses that had copayments.

4.1. Limitations

The study has limitations that need to be discussed. First, while our falsification test held, there is still the possibility that we have not accounted for unobserved factors (e.g., social determinants of health) that can explain differences in utilization of health services among the two cost sharing groups. Cross-sectional observational data does not allow for true randomization, thus it is possible for endogeneity between cost sharing and group outcomes (Athey & Imbens, 2017). However, the falsification test did hold, and group differences that are typically expected to increase the use of health services did not outweigh the effect of the difference in copayments. For example, the group with some copayment for each service were 100% to 138% of the federal poverty level, while those in the no-cost sharing group were below 100% of the federal poverty level. We would expect those with higher incomes (copayment group) to have higher utilization of health services, given higher income levels; however, we found that the copayments outweighed the potential effect of higher income. Furthermore, our identification of patients with substance use disorders did not consider whether the diagnosis of substance use was specifically documented for the purpose of treating the substance use itself. Also, we did not control for the severity of an individual’s level of substance use (Johnson et al., 2020). Finally, differences in total expenses may be due to the use of services unrelated to behavioral health or substance use. However, we aimed to mitigate this concern by first showing that differences in expenses related to behavioral health or substance use exist. Second, we showed that the difference in expenses is unlikely explained by the difference in reimbursement rates given both cost-sharing groups had the same expense related to the most used service (Psychotherapy).

4.2. Conclusion

Copayments may have served as an initial barrier to substance use treatment, but once this barrier was overcome, copayments were not associated with reduced healthcare utilization. The reduction in healthcare utilization due to the financial barrier (copayments) was also associated with a significant reduction in expenditures. The main implication for the observed findings involves the role of copayments for enrollees with substance use disorder. Imposing cost sharing on vulnerable populations can result in unintended consequences necessitating policy makers to ensure that the benefits of cost sharing for substance use services outweigh any unintended costs.

Highlights.

• Copays decreased healthcare utilization among substance use enrollees.

• Copays serve as an initial barrier to behavioral health service use.

• Behavioral health services were more sensitive to copays than other services.