Transit and treatment: Aligning systems to address substance use in Connecticut

Abstract

Objective

Test hypotheses that proximity to new transit improves substance use disorder treatment provider cost efficiency (i.e., economies of scale and scope).

Data Sources and Study Setting

Connecticut substance uses disorder treatment providers/programs. A 2015 rapid transit line opening with 10 stations, near some providers/programs. Providers' annual operating costs from publicly available federal tax forms (2013–2018). Annual client counts, service‐type (including substance use disorder and/or mental health, among others), and location data, for 50 providers and their programs, from Department of Mental Health and Addiction Services, with an unbalanced panel of 285 provider‐years.

Study Design

Economies of scale occur when the percent change in operating costs is less than the percentage change in clients. Economies of scope occur when operating costs fall as providers treat clients with multiple service needs. With our quasi‐experimental, multivariate regressions approach, we test hypotheses that proximity to a new transit line enhances economies of scale and scope (i.e., lowers unit operating costs).

Data Collection/Extraction Methods

Annual provider‐level operating costs merged with new transit station locations and Department of Mental Health and Addiction Services program/provider‐level secondary data (locations, client counts/completions/dates, service types, and average demographics).

Principal Findings

For providers with programs within 1‐mile of new transit (compared with a “control” sample beyond 1‐mile of new transit), (i) a 10% increase in clients leads to a 0.12% lower operating costs per client; (ii) a 10% increase in clients completing treatment results in a 1.5% decrease in operating costs per client; (iii) a 10% increase in clients receiving treatment for multiple services causes a 0.81% lower operating costs per client; (iv) offering multiple services leads to 6.3% lower operating costs.

Conclusions

New transit proximity causes operating cost savings for substance use disorder/mental health treatment providers. System alignment may benefit transit and health care sectors.

What is known on this topic

• Substance use disorder treatment in the US is costly and with scarce resources, not everyone in need can receive treatment.

• Transportation is often a barrier to substance use disorder treatment that can also lead to missed/late appointments and idle clinician time, increasing provider operating costs.

• Sometimes, budgetary constraints may imply higher operating costs could lead to a need to cut back on clients served. Prior research has not adequately addressed this potential reverse causality.

What this study adds

• A quasi‐experimental approach develops causal estimates of the impact of proximity to new transit on provider operating costs.

• There are statistically significant lower per‐client operating costs as client counts rise (i.e., enhanced economies of scale) for providers with some programs near the new transit, compared with providers farther from the new transit lines.

• Similarly, proximity to a new transit line leads to lower operating costs (i.e., improved economies of scope) for providers offering multiple (e.g., substance use disorder and mental health) services.

• System alignment between transit and substance use disorder/mental health treatment can be beneficial by leading to lower provider operating costs.

1. INTRODUCTION

Illicit substance use and prescription misuse cost our economy more than $600 billion each year. ¹ With roughly 2.7 million people, 12+ years of age receiving treatment in 2020 at a substance use disorder specialty treatment facility annually, ² providers' operating cost efficiency is an important aspect of providing services.

One measure of operating cost efficiency is economies of scale, where the percentage change in operating costs is less than the percentage increase in number of clients. A classic example of economies of scale in health care sectors is a hospital with lots of vacant beds. Since the hospital needs to heat/cool the facility and pay nurses/staff regardless of whether these beds are filled, the operating costs per patient can fall if a small number of additional beds are filled. Applied to substance use disorder treatment, changes that reduce client absences and reduce staff downtime due to “no‐shows” can lead to economies of scale. As another example, when a provider rents space that is not used to full capacity due to inability to fill appointment slots, the percentage change in operating costs associated with a percentage increase in patients is higher than when the facility is fully staffed and filled to capacity.

Another cost efficiency measure is economies of scope, where operating costs are less as providers offer multiple treatment services, opposed to only one of these treatments. ³ The classic example of economies of scope in health care is also a hospital with vacant beds. Here, filling some of the vacant beds can be accomplished by offering additional services, which in turn can reduce operating costs.

We hypothesize operating costs (and in turn, economies of scale and scope) improved along both these metrics in the presence of new bus rapid transit service, for providers “close” to the new transit station(s) relative to “far”. Improved transit could enhance client engagement with scheduled services, thereby impacting provider operating cost. Smaller operating cost increases can enable more clients to receive treatment. With evidence that better transit can support this, intentional systems alignment could be beneficial.

Increasing access and retention in treatment services is a critical step in this process of minimizing treatment provider operating costs. Transportation barriers are an impediment to successful treatment completion, ⁴ , ⁵ , ⁶ , ⁷ and fewer treatment completions can be a source of higher operating costs. ⁸ , ⁹ , ¹⁰ While expanding transit can be beneficial, these social services are costly, and their added value should be considered.

Understanding how enhanced transit impacts operating costs via economies of scale/scope may have implications for alignment of systems across medical and social services sectors. An example of this type of situation is a densely populated area where large numbers of individuals need treatment. Providers might treat a greater number of clients while experiencing a relatively smaller increase in the percent of their operating costs, so that they can spread out their fixed costs over greater patient numbers. In a city, a steady flow of potential clients can lead to economies of scale. Opportunities for system alignment with transit may exist to yield similarly lowered operating costs. Transit may offer providers a steady client flow that can help ensure full utilization of space and clinician time, and, therefore, improve their economies of scale. Better transit access may also facilitate treatment of clients who otherwise might need to rely on more costly transportation modes to reach their appointments (such as ride share services).

Some substance use disorder treatment providers also offer a variety of other (e.g., mental health) services, so operating cost savings of treating multiple health issues in one facility (economies of scope) are possible. In other words, two separate providers—for example, one specializing in substance use disorder treatment and another specializing in mental health treatment—may both be operating at less than full capacity. But a single clinic offering multiple services may enhance clinical and billing capacity. In this case, offering multiple services may result in a less than proportionate increase in operating costs, or cost less than providing each service in a separate facility, implying economies of scope. When patients have access to multiple services at the same provider, better transit access may also lower the provider's operating costs by reducing the travel time of clients and in turn making it more likely that they will attend their appointments—again reducing clinician underutilization. We hypothesize that substance use disorder treatment services more closely aligned with transit lead to improvements in providers' economies of scope (that is, lower their operating costs). These estimates have a direct bearing on the ability of society to offer treatment to more patients in need.

Relatively few studies have examined economies of scale and scope for substance use disorder treatment using a causal framework. ³ , ¹¹ , ¹² , ¹³ Duffy et al. ¹² base their analysis on a cross section of U.S. outpatient treatment facilities in 1997. They find statistically significant evidence of economies of scale for outpatient admissions. Beaston‐Blaakman et al. ¹¹ use the count of active clients on a specific day of the year as a proxy for size, and find evidence of economies of scale. Dunlap et al. ¹³ estimate two models for methadone treatment, one based on patient “average daily census,” and the other based on “primary services” at methadone treatment programs, and find some evidence of economies of scale. Cohen and Morrison Paul ³ find economies of scale at hospitals providing outpatient substance use disorder treatment in Washington State. But it is also possible that the causality works in both directions (in other words, budgetary constraints may imply higher operating costs could lead to a need to cut back on clients served). None of this previous research uses a strategy that overcomes this potential simultaneity of client counts.

The introduction of an expanded transit system in 4 municipalities in Connecticut in 2015 (called CTfastrak) provides an opportunity to use a quasi‐experimental empirical estimation approach to assess the causality of new transit proximity on economies of scale and scope. Our approach (as in Autor ¹⁴ ) enables us to overcome the simultaneity between client count variables and operating costs. We explore how economies of scale and scope are different for providers with programs near the new transit line.

2. METHODS

The scientific approach for testing the two hypotheses relies on a set of quasi‐experiments. These experiments include a statistical technique to determine causality, as in Autor, ¹⁴ and an operating cost analysis approach as in Cohen and Morrison Paul. ³ The latter is a technique originating from the Industrial Organization literature in economics that has also been applied widely to issues in health care economics, focusing on the production process. This particular operating cost analysis approach has been applied to many different industry studies (ranging from hospitals to manufacturing) to aid in decisions of how many firms, how much of each input each firm should use, and what size firms are optimal in a given industry. Some studies (e.g., Cohen and Morrison Paul ³ ) have considered substance use disorder treatment provider operating costs using this approach. This type of operating cost analysis can describe economies of scale, which can help with decisions about whether it is more cost efficient for many small firms to produce small amounts, or fewer large firms to produce large amounts, of a product or service in an industry. This approach has also been widely used to estimate economies of scope.

Substance use disorder treatment is costly. A crucial point about the operating cost analysis approach used here is that it aids in determining how much of a product firms should make, and how the firms should produce the products, in order to operate “cost efficiently” (that is, to ensure operating costs increase less than proportionately with the firm's “output” or “product”). When substance use disorder treatment providers are not using the optimal approach, financial resources can be wasted, and some people may not get the care they need. In practical terms, this inefficiency might reflect unbillable time and/or underutilized space. Also, integrated care models, which treat mental health and substance use disorders concurrently, are effective ¹⁵ but are rarely available in community clinics. ¹⁶ However, decisions to offer such services are complex given multiple other variables affecting a firm's decision of how to produce its product(s), such as client demographics (race/ethnicity and age). We can control for these factors with our operating cost analysis approach. Regression analysis is used for estimating the parameters in this operating cost analysis model and testing hypotheses on economies of scale/scope. While some studies have attempted to calculate the aggregate costs savings of some policies, ²¹ , ²² the focus here is on the operating costs effects of incremental changes in client counts due to proximity to new transit.

There are a number of different approaches to empirically implementing the operating cost analysis. One of the common approaches was used by Li and Rosenman ¹⁷ and Cohen and Morrison Paul ³ for hospital costs, and Morrison and Schwartz ¹⁸ and Cohen and Morrison Paul ¹⁹ for public infrastructure impacts on manufacturing. This relatively tractable approach is to use regression analysis in the context of our problem, comparing how more clients affect operating costs for providers with programs near versus far from the transit line, after versus before the transit line opening:

$$\ln \left(\mathit{\text{OpCos}}{t}_{\mathit{it}}\right)=\begin{array}{l}{\beta }_0+{\beta }_1\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)+{\beta }_3\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\\ \times \mathit{\text{Distanc}}{e}_i+{\beta }_4\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\times {\mathit{\text{Post}}}_i\\ +{\beta }_5\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\times \mathit{\text{Distanc}}{e}_i\times {\mathit{\text{Post}}}_i+{\beta }_6{\mathit{\text{Post}}}_i\\ +{\beta }_7\mathit{\text{Distanc}}{e}_i+{\beta }_8\mathit{Yea}{r}_i+{\beta }_{9}Z+{ϵ}_{\mathit{it}}\end{array}$$ (1)

In the context of research on substance use disorder treatment providers and social services, $\mathit{\text{OpCos}}{t}_{\mathit{it}}$ is provider i operating costs in year t; $\mathit{\text{NClient}}{s}_{\mathit{it}}$ is the service produced, or “output”, which in one specification here represents the number of clients treated by provider i in year t; and in another specification here represents the number of clients completing treatment with provider i in year t. $Z$ represents a vector of other covariates, including the demographic variables for measuring characteristics of clients that may be associated with different operating costs. For this study, $\mathit{\text{Distanc}}{e}_i$ is a distance cutoff indicator variable for proximity to transit (defined here as taking a value of 1 if the provider has any programs within 1‐mile of the new transit, and 0 otherwise); ${\mathit{\text{Post}}}_i$ is a dummy variable that equals 1 if observation i is after the opening of the transit line (i.e., post‐2015), and 0 otherwise; and ${ϵ}_{\mathit{it}}$ is a random error. A $\mathit{\text{Distanc}}{e}_i$ cutoff of less than 1 mile offers too small of a sample of affected providers, while more than 1 mile is too far to consider transit accessible to clients. The effect of our transit quasi‐experiment is the regression coefficient estimate, ${\beta }_5$. If ${\beta }_5$ is negative and statistically significant, this indicates economies of scale are improved for providers with programs in proximity to transit after the transit opening.

Similarly, when considering multiple treatment services offered by the same provider, the economies of scope estimate can be derived by estimating the regression equation (2) below:

$$\ln \left(\mathit{\text{OpCos}}{t}_{\mathit{it}}\right)=\begin{array}{l}{\beta }_0+{\beta }_1\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)+{\beta }_2{\mathit{\text{ShareBoth}}}_{\mathit{it}}\\ +{\beta }_3\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\times \mathit{\text{Distanc}}{e}_i+{\beta }_4{\mathit{\text{ShareBoth}}}_{\mathit{it}}\\ \times \mathit{\text{Distanc}}{e}_i+{\beta }_5\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\times {\mathit{\text{Post}}}_i\\ +{\beta }_6{\mathit{\text{ShareBoth}}}_{\mathit{it}}\times {\mathit{\text{Post}}}_i+{\beta }_7\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)\\ \times \mathit{\text{Distanc}}{e}_i\times {\mathit{\text{Post}}}_i+{\beta }_8{\mathit{\text{ShareBoth}}}_{\mathit{it}}\times \mathit{\text{Distanc}}{e}_i\\ \times {\mathit{\text{Post}}}_i+{\beta }_9{\mathit{\text{Post}}}_i+{\beta }_{10}\mathit{\text{Distanc}}{e}_i+{\beta }_{11}\mathit{Yea}{r}_i+{\beta }_{12}Z+{ϵ}_{\mathit{it}}\end{array}$$ (2)

In equation (2) above, two separate specifications are estimated. In the first specification, ${\mathit{\text{ShareBoth}}}_{\mathit{it}}$ is the share of clients who are receiving multiple services at provider i in year t. In the second specification, ${\mathit{\text{ShareBoth}}}_{\mathit{it}}$ is an indicator variable taking the value of 1 if provider i in year t has any clients receiving multiple types of services, and 0 otherwise. The “average treatment effect” in equation (2) is ${\beta }_8$. In other words, ${\beta }_8$ is the estimate of how economies of scope change for the “treatment group” of providers that are “near” the new transit line, relative to the “control group” of providers that (i) offers only one type of services and (ii) all providers with no programs “near” the new transit and (iii) all programs before the opening of the new transit line. If ${\beta }_8$ is statistically significant and negative, this implies economies of scope are enhanced by proximity to new transit.

3. DATA

One reason for few substance use disorder treatment operating cost analysis studies is the lack of sufficient publicly available data on annual provider‐level client counts, demographics, and operating costs. To address this issue, collaborative data agreements were formed with a state agency in Connecticut. The annual data coverage is for 2013–2018, at the provider level. These data include client counts (admissions), overall client demographics, primary treatment(s), and treatment completions. Duffy et al. ¹² explain the importance of controlling for client mix. Others, such as Yeom and Shepard ²⁰ note that gender differences can have impacts on costs. Therefore, client mix variables were included as “shift” covariates in the model. The set of case‐mix‐related variables in our analysis include the percent of provider's clients who are African American, who are Hispanic, who are female, who are in age groups 18–25, in age groups 26–34, and in age groups 35–44. This study was determined to not meet the criteria for human subjects research due to the use of publicly available datasets and provider‐level data.

Cost data (measured as annual total operating costs) at the provider‐level were obtained from publicly available Internal Revenue Service 990 forms for nonprofit organizations. These operating costs data were merged with secondary annual client count and treatment type data at the provider level, which was obtained from the Connecticut Department of Mental Health and Addiction Services from years 2013 to 2018. The Internal Revenue Service 990 forms were obtained annually during the same time period for all providers included in this study. These records for providers contain annual detailed information on total wages and salaries paid; value of owned physical capital (buildings and equipment); and total operating costs. Most providers have multiple program sites throughout Connecticut. Due to some provider and program openings and closings during the study time period, the data are an unbalanced panel, which includes 50 providers across 6 years, resulting in 285 provider‐years. The 1692 program‐years observations were used in calculating whether each provider had at least one program within 1 mile of the nearest CTfastrak station.

Table 1 provides summary statistics for key variables. The average provider operating costs in a given year of the sample, among all 50 providers in Connecticut that operated during the 2013–2018 period, were approximately $23.8 million. Approximately 40% of these operating costs were comprised of salaries, on average. Among the total of 285 provider‐year observations in Connecticut, there were 66 provider‐year observations with at least one program within 1 mile of one of the new CTfastrak stations, with average program distance from the station of 0.59 miles among that group. Approximately 17% of all programs in the state were within one mile of a CTfastrak station. On average, 35% of patients completed treatment. Females comprised 43% of the total patient population. Twenty percent of all clients were between the ages of 25 and 34, while 19% of the patient population were African American and 18% identified as Hispanic.

TABLE 1.

Summary statistics for programs and providers.

	Mean	SD
Number of clients per program per year	86.4	248
Average values for providers (N = 285 provider‐years)
Operating costs	23,835,567	19,993,850
Total salaries	10,356,922	11,025,185
Total assets	12,808,726	11,324,051
Percentages of clients by provider
Age 18–25	14%	8%
Age 26–34	20%	8%
Age 35–44	18%	5%
Female	43%	11%
African American	19%	13%
Hispanic	18%	9%
Percentages of programs (N = 1692 program‐years)
Client treatment completed %	35%	37%
Programs within 0.5 miles of a new station	6%	24%
Programs within 1 miles of a new station	17%	22%
Distance from programs to nearest station
Full sample (N = 285 provider‐years)	25.90 miles	15.84 miles
Affected providers (having programs within one mile of a station) (N = 66 provider‐years)	0.59 miles	0.56 miles

Note: This table provides summary statistics for key variables. Data from Connecticut Department of Mental Health and Addiction Services Dashboard data from 2013 to 2018, combined with financial data from the Internal Revenue Service (IRS). Coverage includes programs and providers in operation throughout state of Connecticut that are recorded in the Dashboard data. The average provider operating costs in a given year of the sample, among all 50 providers in the state of Connecticut that operated during the 2013–2018 period, were approximately $23.8 million. Approximately 40% of these operating costs were comprised of salaries, on average. There were 66 provider‐year observations with at least one program within 1 mile of one of the new CTfastrak stations, with average program distance from the station of 0.59 miles. Approximately 17% of all programs (among 1692 program‐year observations throughout the entire state of Connecticut) were within one mile of a CTfastrak station. On average, 35% of patients completed treatment. Females comprised 43% of the total patient population. Twenty percent of all clients were between the ages of 25 and 34, while 19% of the patient population was African American and 18% identified as Hispanic.

4. RESULTS

Figure 1 plots a map of the locations of programs and CTfastrak stations. The greatest concentration of programs is in the Hartford area, which is also a large population center and the state capital. The green dots are the CTfastrak station locations. The blue and red dots are program locations. The red dots are program locations within 1 mile of a CTfastrak station location (denoted as the “affected programs”). There are 11 providers with at least one program within 1 mile of a CTfastrak station, for each year over the 6 years of the sample, leading to an N = 66 for the affected group of providers (those within 1 mile, for the years after the opening of the transit line). Data on all providers in the databases throughout the entire state of Connecticut are included in the sample, which consist of an unbalanced panel of 285 provider‐year observations and 1692 program‐year observations.

FIGURE 1.

Locations of CTfastrak transit stations and substance use disorder treatment programs near Hartford, Connecticut. Figure shows a map of Hartford, Connecticut, and surrounding areas with programs and CTfastrak stations labeled. Note that all programs throughout Connecticut were included in the regression analyses of this paper, while those programs within one mile of a CTfastrak station were all in the Hartford region. The additional programs included in the study in regions beyond the map boundary are not shown in this figure. Affected programs are defined as those programs operating within one mile of a CTfastrak Station after the bus rapid transit line opened in 2015. CTfastrak stations shown in green with a solid line are an estimate of the bus rapid transit route connecting the stations. (Sources: Connecticut Department of Transportation, Connecticut Department of Mental Health and Addiction Services, and Authors' Calculations).

4.1. Economies of scale

We first report estimates of economies of scale. Our model in (1) has operating cost as the dependent variable and total volume of clients as an independent variable. The effect is given by the regression coefficient ${\beta }_5$. Our models consider programs within 1 mile of the nearest new station as the “treatment” group, while providers throughout the entire state are included in the regression analysis. Reducing this 1 mile cutoff results in too few providers/programs, while beyond 1 mile is not walkable.

Table 2 presents two versions of our main results. In columns 1 and 2, the “average treatment effects” are based on performing a regression with equation (1), of the $\ln \left(\mathit{\text{OpCos}}{t}_{\mathit{it}}\right)$ on log of client volume, $\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)$, and the other covariates. The effect is ${\beta }_5$, which is the regression coefficient of the product of the $\mathit{\text{Distanc}}{e}_i$ (within 1 mile) indicator (taking a value of 1 if the provider had at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (post‐2015) indicator, ${\mathit{\text{Post}}}_i$, and the logarithm of the provider's “client volume”, $\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)$.

TABLE 2.

Economies of scale “average treatment effect” estimates for generalized‐difference‐in‐difference‐in‐differences specification.

Dependent variable: log(OpCost)	(1)	(2)
Average treatment effect	−0.012***	−0.150**
	(0.003)	(0.062)
Observations	285	285
R 2	0.989	0.988

Note: In columns 1 and 2, the “average treatment effects” are based on performing a difference‐in‐difference‐in‐differences regression of the log of operating cost on log of client volume. The “average treatment effect” is the regression coefficient of the product of a “near transit” (within 1 mile) indicator (taking a value of 1 if the provider has at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (post‐2015) indicator, and the provider's “client volume”. The “average treatment effect” estimate in column 1 represents the difference‐in‐difference‐in‐differences coefficient where “client volume” is the (log of) provider's client counts. The “average treatment effect” estimate of −0.012 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, experienced 0.12% lower operating costs in response to a 10% increase in clients receiving treatment, relative to the “control” data observations. The “average treatment effect” in column 2 represents the coefficient from a difference‐in‐difference‐in‐differences regression where the “client volume” variable is the proportion of each provider's clients completing treatment. The “average treatment effect” estimate of −0.150 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, experienced 1.5% lower operating costs in response to a 10% increase in clients completing treatment, relative to the “control” data observations. Both of these regressions include year fixed effects. Other regressors (not shown) are included for percent of provider's clients who are African American, who are Hispanic, who are female, who are in age groups 18–25, in age groups 26–34, and in age groups 35–44. Standard errors in parenthesis. Significantly different than zero at 90 (*), 95 (**), 99 (***) percent confidence levels. The regression model uses data from Connecticut Department of Mental Health and Addiction Services from 2013 to 2018 combined with provider‐level financial data (operating costs) from the Internal Revenue Service.

In Column 1 of Table 2, the “output” covariate corresponding to ${\beta }_5$ is the (logarithm of) provider's client counts. This estimate of −0.012 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, experienced 0.12% lower operating costs in response to a 10% increase in clients receiving treatment, relative to the “control” group of data observations (that is, economies of scale were improved). The p‐value on this estimate is less than 0.01.

In Column 2 of Table 2, the “output” covariate corresponding to ${\beta }_5$ is the proportion of each provider's clients completing treatment. The estimate, ${\beta }_5$, of −0.150 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, experienced 1.5% lower operating costs in response to a 10% increase in clients completing their treatment, relative to the “control” data observations. The p‐value on this estimate is less than 0.05. Once again, the economies of scale estimate is enhanced by having programs within 1 mile of a CTfastrak station, after the new transit line opened.

4.2. Economies of scope

To estimate economies of scope, we generate two scope variables, a scope percentage and a scope indicator. The scope indicator takes a value of 1 if a provider offers two or more services. The scope percentage consists of the percentage of clients who receive two or more services for each provider. On average, before the CTfastrak opening, 36% of providers have zero clients who received two or more services (that is, 36% of providers have all of their clients only receiving one service). In other words, before the new transit stations were opened, only 32 of the 50 providers in Connecticut have patients receiving multiple services. After the transit opening, this number increases to 36 of the 50 providers. No providers reduced their number of treatment types after the new transit opened. In all of these cases, providers added new service types and no providers switched from one type of service to another type of service. Our measure of increasing scope is much broader than merely the categories of both mental health and substance use disorder. Each provider offers a number of mental health and/or substance use disorder services (our data set defines 14 possible types of services: Case Management, Medication Assisted Treatment, Residential Services, Outpatient, Intensive Outpatient Programs, Social Rehabilitation, Inpatient Services, Employment Services, Acceptance and Commitment Therapy, Community Support, Forensics Community‐based, Recovery Support, Crisis Services, Education Support). A provider increases their scope when moving from offering one type to at least one additional type of service.

The Table 3 regression results are based on the regression coefficient ${\beta }_8$ in equation (2) above. In Columns 1 and 2 of Table 3, the effects are based on performing a regression of $\ln \left(\mathit{\text{OpCos}}{t}_{\mathit{it}}\right)$ on $\ln \left(\mathit{\text{NClient}}{s}_{\mathit{it}}\right)$, along with other covariates, $Z$, as described below.

TABLE 3.

Economies of scope “average treatment effect” estimates for generalized‐difference‐in‐difference‐in‐differences specification.

Dependent variable: log(OpCost)	(1)	(2)
Average treatment effect	−0.081**	−0.063*
	(0.040)	(0.025)
Observations	285	285
R 2	0.989	0.988

Note: In Columns 1 and 2, the “average treatment effects” are based on performing a difference‐in‐difference‐in‐differences regression of log of provider operating cost on log of client volume. No providers reduced their number of treatment type offerings after the opening of the new transit line in 2015. After the transit opening, 36 providers served clients with multiple services, which rose from 32 prior to the new transit opening. All cases were additions of new service types and no provider switched from one type of service to another type of service. Our measure of increasing economies of scope is not restricted from only servicing one of mental health and substance use disorder to both. Rather, each provider offers a number of mental health and substance use disorder services (our data set defines 14 types of services: Case Management, Medication Assisted Treatment, Residential Services, Outpatient, Intensive Outpatient Programs, Social Rehabilitation, Inpatient Services, Employment Services, Acceptance and Commitment Thearapy, Community Support, Forensics Community‐based, Recovery Support, Crisis Services, Education Support). A provider increases their scope when moving from offering one type to at least one additional type of service. In Column 1, the “average treatment effect” is the regression coefficient of the product of a “near transit” (within 1 mile) indicator (taking a value of 1 if the provider has at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (post‐2015) indicator, and the proportion of clients receiving mental health and addiction treatment. The “average treatment effect” estimate of −0.081 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, experienced 0.81% lower operating costs in response to a 10% increase in clients receiving multiple services, relative to the “control” data observations. Economies of scope were greater for these “treatment group” of providers, in other words, they experienced greater cost efficiency. In Column 2, the “average treatment effect” is the regression coefficient of the product of a “near transit” (within 1 mile) indicator (taking a value of 1 if the provider has at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (post‐2015) indicator, and an indicator variable for whether the program has any clients receiving multiple forms of treatment. The “average treatment effect” estimate of −0.063 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, and having any clients receiving multiple treatment types, experienced 6.3% lower operating costs, relative to the “control” data observations. Cost efficiency improved for these “treatment group” of providers; their economies of scope estimates were enhanced. These regressions include fixed effects for years. Other regressors (not shown) are included for percent of provider's clients who are African American, who are Hispanic, who are female, who are in age groups 18–25, in age groups 26–34, and in age groups 35–44. Standard errors in parenthesis. Significantly different than zero at 90 (*), 95 (**), 99 (***) percent confidence. The regression model uses Connecticut Department of Mental Health and Addiction Services data from 2013 to 2018 combined with provider operating costs data from the Internal Revenue Service.

In Column 1 of Table 3, the regression coefficient ${\beta }_8$ is the coefficient on the product of the $\mathit{\text{Distanc}}{e}_i$ indicator (taking a value of 1 if the provider i has at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (${\mathit{\text{Post}}}_i$) indicator, and the proportion of clients receiving multiple services at the same provider, here denoted as ${\mathit{\text{ShareBoth}}}_{\mathit{it}}$. The estimate of −0.081 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 transit line opening, experience 0.81% lower operating costs in response to a 10% increase in clients receiving multiple services at the same provider, relative to the “control” group of data observations. The p‐value for this estimate is less than 0.05. In other words, these providers' economies of scope improve; it is more cost efficient to treat clients receiving multiple services at providers with at least one program within 1‐mile of the new transit line.

In Column 2 of Table 3, the regression coefficient ${\beta }_8$ is the coefficient on the product of a $\mathit{\text{Distanc}}{e}_i$ indicator (taking a value of 1 if the provider has at least one program within 1 mile of the nearest CTfastrak station), an after transit opening (${\mathit{\text{Post}}}_i$) indicator, and ${\mathit{\text{ShareBoth}}}_{\mathit{it}}$. Here, ${\mathit{\text{ShareBoth}}}_{\mathit{it}}$ is an indicator variable for whether the program has any clients receiving multiple forms of treatment. The estimate of −0.063 implies providers with at least one program within 1 mile of a CTfastrak station, after the 2015 opening, and having any clients receiving multiple services, experience 6.3% lower operating costs, relative to the “control” data observations. The p‐value for this estimate is less than 0.10. Once again, operating cost efficiency improves at providers who were serve patients with multiple services; economies of scope are enhanced for providers with at least one program near the new transit line.

5. DISCUSSION

We uncover new causal empirical evidence, using an operating cost analysis framework and a quasi‐experimental strategy, related to economies of scale and scope for substance use disorder treatment providers in Connecticut. These findings may have some implications for the benefits of system alignment between substance use disorder treatment, mental health treatment, and transit services. Specifically, we find that proximity to a new transit line has led to greater economies of scale from treating additional clients with substance use disorder, for providers with programs within 1 mile of the nearest new transit station after the opening of the line, relative to those providers farther away. We also find evidence that treating clients for multiple services by the same provider, where that provider has at least one program within 1 mile of a new transit station, has led to greater economies of scope, relative to other providers.

These general findings are consistent with the findings of some other operating cost analysis studies, such as Cohen and Morrison Paul, ³ although our quasi‐experimental approach leads to more precise and valid estimates because we identify a causal relationship, opposed to the correlations that others have found. Also, prior studies have not considered the implications of economies of scale/scope for system alignment across the sectors targeted for this analysis, specifically transit with various types of mental health treatment and substance use disorder treatment.

There are some potentially confounding factors associated with the results from this study. These pertain to concurrent developments in substance use disorder treatment, and other confounding factors related to transportation, in the Greater Hartford area compared with other parts of Connecticut. First, this type of bus rapid transit service is unique to the Hartford area. Regulations for development of new public transit are different by municipality, including Hartford and the surrounding towns where CTfastrak operates. ²³

Second, Connecticut, like many other states, experienced increases in persons needing and seeking treatment for substance use disorders in the time period analyzed. The introduction of the Affordable Care Act, Connecticut's expansion of patients eligible for Medicaid, the “CT Health Exchange,” and the ongoing Opioid Crisis all served to increase the number of patients needing and seeking substance use disorder treatment. The impacts of these policies are large. The state's uninsured population is believed to have fallen from 13.2% in fiscal year 2013 to 4% in fiscal year 2015, and the Department of Mental Health and Addiction Services estimates a 10% increase in clients served in substance use disorder services. ²⁴ With these overlapping influences, data suggest the number of Connecticut clients receiving substance use disorder treatment increased, with 33,267 patients receiving treatment on a single day in March 2013, while 41,873 patients in treatment on that same day in March 2017. ²⁵ These changes were concentrated in urban areas, as are the transit changes under study, leading to a potential confound in that increases in clients served are likely to be steeper in urban settings (Hartford) near the transit lines. Thus, our findings related to transit changes are against a backdrop of larger influences yielding increased treatment.

In a robustness test of how these potential confounding factors may be impacting some of our results, we re‐estimate equation (1) above for the model based on total clients served, but now only including data from providers in the largest urban areas in the estimations. These urban areas are Hartford, New Haven, Bridgeport, and Waterbury, with an unbalanced panel of 26 providers over 6 years (135 total observations). With this sub‐sample, we find that the coefficient estimate, and the p‐value, on the “Average Treatment Effect” are −0.021 and 0.059, respectively. Since these results using the urban sample are similar to the results using the full sample, we believe it is unlikely that confounding factors are driving our results.

Thus, the introduction of the new transit line in Connecticut permits a unique opportunity to examine proximity as a potential influencer on provider operating costs. We argue that the results of this past (unintentional) systems alignment have important future implications for the further of alignment of health care and transit planning. To explore these issues, we have convened an advisory panel of experts and stakeholders in the relevant sectors, which has been meeting monthly from 2020 to 2023, to discuss system alignment strategies, past experiences, and challenges and limitations to system alignment approaches. Panelists represent the Connecticut Department of Mental Health and Addiction Services, the Connecticut Department of Public Health, the Connecticut Department of Transportation, the Capitol Region Council of Governments, 5 local treatment providers, and University of Connecticut and UConn Health Medical School researchers.

Our approach of exploring for empirical evidence of past system alignment success, and projecting that forward to future system alignment efforts, could be generalizable to other health care settings beyond substance use disorder and mental health treatment. For instance, the operating cost efficiency of other health care sectors could benefit from proximity to new transit. There are other types of system alignment issues that could be examined. These include promoting ride sharing services, encouraging transit operators to shift route schedules to align arrival times with appointment starting times at providers, adding more daily bus service, offering transit subsidies for clients to encourage them to use transit, and encouraging providers to open new program sites near transit stations.

FUNDING INFORMATION

Support for this work was provided (for Cohen, Rash, Huleatt, and Murphy) by the Robert Wood Johnson Foundation through the Systems for Action National Coordinating Center, grant #S4A‐78117. Support (for Rash) was provided by NIH grant P50‐AA027055.

Cohen JP, Huleatt S, Murphy S, Rash CJ. Transit and treatment: Aligning systems to address substance use in Connecticut. Health Serv Res. 2024;59(Suppl. 1):e14268. doi: 10.1111/1475-6773.14268