The adoption and diffusion of evidence-based pharmaceutical technologies in addiction treatment. Carolyn J. Heinrich University of Wisconsin-Madison

Transcription

1 The adoption and diffusion of evidence-based pharmaceutical technologies in addiction treatment Carolyn J. Heinrich University of Wisconsin-Madison Grant R. Cummings University of Wisconsin-Madison January 2011 This research was funded by an Institute for Clinical and Translational Research (ICTR) Type 2 Translational Research Health Policy Assistantship Award and Regina Laughlin Scholar funds at the University of Wisconsin-Madison. Page 1 of 79

2 After more than three decades of research in drug and alcohol treatment, there is a consensus that the pace of adoption of research findings has been too slow and that a disconnect between science and practice and between public beliefs and research findings persists (Brown and Flynn, 2002; Thomas and McCarty, 2004). At the same time, treatment practices are being strongly influenced by managed care and related economic pressures to limit services, which further diminish interest in adopting new evidence- based practices that are viewed as increasing pressure to do more with less (Simpson, 2002:172). In fact, researchers suggest that barriers or supports to transferring research and evidence- based technologies to practice increasingly operate at the policy, environmental, and organizational levels (e.g., regulatory and financial pressures, insurance coverage, inter- organizational linkages, organizational climate, staffing and resources, management structures and philosophies, etc.) (Backer et al., 1995; Fuller et al., 2005; Kovas et al., 2007; Simpson et al., 2007). There is considerable potential to advance both the conceptual and empirical modeling of the diffusion of innovations in substance abuse treatment. Research that explores the adoption and diffusion of evidence-based innovations in addiction treatment is at an early stage of development, largely focusing on the application of alternative methodologies of technology transfer (e.g., staff training) and other program and staff characteristics associated with adoption and diffusion (Simpson, 2002; Roman et al., 2006). In this study, we examine the adoption and diffusion of evidence-based pharmaceutical technologies in substance abuse treatment facilities over time, measuring and estimating the influence of state policies and other Page 2 of 79

3 external factors, as well as internal treatment facility characteristics, on the adoption and diffusion of three pharmaceutical agents: disulfiram, naltrexone and buprenorphine. 1 Each of these three pharmaceutical agents has different attributes and applications in addiction treatment: disulfiram, used for decades to treat alcohol abuse, produces immediate hangover-like symptoms when alcohol is consumed; naltrexone, approved by the Food and Drug Administration (FDA) in 1994 as an adjunct to the treatment of alcohol dependence, blocks alcohol-induced stimulation of endogenous opioids and dulls the high feeling produced by alcohol; and buprenorphine, approved by the FDA for treatment of opioid addiction in 2002, reduces cravings and has relatively lower risks of abuse and side effects than other opioid agonists such as methadone. Clinical trials for these pharmaceutical agents have generally confirmed their potential and have also examined the treatment conditions under which they are more or less effective. For example, O Malley et al. (2003) found that naltrexone may be as effective when prescribed by physicians in primary care settings as in specialized treatment settings, and Anton et al. (2006) found that naltrexone was equally effective alone (in the presence of medical management) as combined with behavioral therapies. While each is used in substance abuse treatment, they vary in terms of the particular condition they are intended to treat, their physical effects, and their stage of adoption. Thus, examining whether and how correlates of their adoption and diffusion differ is also potentially important. To date, studies of naltrexone, buprenorphine, and related pharmaceutical agents (or pharmacotherapies) have focused almost exclusively on treatment centers or programs as the unit of analysis, exploring the relationships of internal organization characteristics (size, age, staff education and training, ownership status, affiliations, services, etc.) to the adoption or use of 1 The brand names are, respectively, Antabuse, ReVia and Suboxone/Subutex. In this paper, we interchangeably use the words pharmaceutical agents/technologies and pharmacotherapies to describe these medical treatments. Page 3 of 79

4 these medications. Although this research confirms that treatment facility attributes are important predictors of the use of evidence-based pharmaceutical technologies (Ducharme et al., 2007; Fuller et al., 2005; Knudsen et al., 2005, 2006; Roman et al., 2006), few studies have explicitly considered the role of inter-organizational and other external professional and environmental factors (including state policies and administrative factors) on adoption and diffusion. In addition, recent findings suggest that different constellations of factors may influence adoption decisions of different pharmaceutical agents (Knudsen et al., 2006), and that in analyzing diffusion, it is important to model it as a process, not a discrete event. A majority of existing studies use a single cross-section of data to predict adoption (with some retrospective in design), and limited or narrow samples of treatment organizations have contributed to some initial conflicting findings in the literature. A few studies use multiple waves of survey data to look at adoption over time (Fuller et al., 2005, Oser et al., 2007). For example, Oser et al. (2007) use data from four waves of the National Treatment Center Study (collected between 1994 and 2003) to examine factors that influence treatment centers adoption of naltrexone over time (165 centers and 398 center-wave observations). Focusing on these three pharmaceutical agents disulfiram, naltrexone and buprenorphine we examine treatment facilities decisions to adopt (and/or discontinue) these medications over a four-year period when each was at a different stage of adoption or diffusion. We assembled data on state policies intended to promote (or that may unintentionally impede) diffusion of pharmaceutical technologies, as well as other external/environmental factors that theory, prior empirical research, and practice suggest should influence adoption and diffusion. We merged these data with nationally representative substance abuse treatment facility-level data that are linked across years, allowing us to observe the offering/availability of these Page 4 of 79

5 pharmaceutical agents by specific treatment facilities over time. A key objective of this study is to produce findings that will be directly relevant to policymakers and practitioners who are looking for policy levers or strategies to increase the adoption and diffusion of evidence-based innovations in addiction treatment and improve treatment access and quality. In the next section, we review the broader literature on the adoption and diffusion of innovations and discuss its implications for substance abuse treatment research, and we also review studies and their findings on the adoption and diffusion of pharmacological innovations in addiction treatment. In subsequent sections, we describe our data, measures and methodological approach to this study, followed by a discussion of our empirical findings. As we discuss in greater detail in this latter section and in the concluding section, we confirm some of the implications of theory and identify policy and administrative levers that policymakers and treatment facility staff can use to increase the adoption of evidence-based pharmaceutical technologies and their continued use by treatment facilities. Literature Review The broader theoretical literature on the adoption and diffusion of innovations, accumulating over four decades, identifies important roles for external pressures and environmental influences (in addition to organizational characteristics) (Berry and Berry, 1990; Borins, 1998; Damanpour, 1991; Rogers, 2003). Summarizing key theoretical tenets, we know that awareness ( early knowing ) and salience of innovations is an important first step, and that the dissemination of evidence showing that the use of an innovation is feasible or accessible and effective is critical. Research across substantive areas suggests that exposure to this type of information and the adoption behavior of others through direct (relational) ties and indirect, structural correspondence (i.e., environmental scanning ) Page 5 of 79

6 influences the adoption of new practices, with the threshold for knowledge transfer lower in more collaborative settings. In other words, the social and communication structures in which treatment organizations operate and the norms (or patterns of behavior they establish) will affect the rate of adoption and diffusion (Rogers, 2003). Both the number of adopters, as well as the proximity of adopting entities, also influences the likelihood that a particular organization will decide to implement an innovation (Walker, 1969; Berry and Berry, 1990, 1992). As more organizations adopt over time, social pressures to conform (theoretically) increase, along with the perceived legitimacy of adopting (which in turn reduces costs associated with evaluating evidence on the technology). In addition, as innovations and related policies gain legitimacy in the eyes of policy makers and are adopted in more neighboring jurisdictions, they are more likely to be viewed as viable solutions to problems, even if politically or socially contentious (Jensen, 2003). In the adoption of pharmaceutical agents, states thus have the potential to play important roles in allocating resources for disseminating research evidence and in encouraging and legitimizing adoption of these medications, through, for example, their identification on state Medicaid drug formularies and other Medicaid/managed care policies that promote their use. Licensing, accreditation and oversight bodies likewise have a role in encouraging innovation by requiring evidence of quality improvement, reinforcing professional norms that support adoption of innovations, and influencing the capacity- building efforts of potential adopters (Oser et al., 2007). In addition, we know that learning takes place over time and can reduce the level of uncertainty surrounding an innovation (both external and internal), improving conditions necessary for adoption (Mooney, 2001). In this regard, new third- party initiatives that aim to Page 6 of 79

7 increase inter- organizational connections and treatment facilities familiarity with these medications, such as the Clinical Trials Network, Drug Free Coalitions, Community Treatment Programs supported by the National Institute on Drug Abuse, and Screening and Brief Intervention efforts, may hold promise for increasing adoption and diffusion of beneficial pharmaceutical technologies. Timing and rate, of course, are other elemental variables in the modeling of the diffusion of innovations. Among the four pharmaceutical agents we are examining, it appears that none has reached a majority adopters stage. In other words, among the medications that have been in use longer (disulfiram and naltrexone), adoption rates have leveled off at relatively low levels. Buprenorphine, a more recent innovation, is still in an early adoption phase, with approximately 10 percent of treatment facilities offering it in However, the low or stable year-to-year levels of adoption of pharmaceutical agents obscure variation in facilities continuing use (or their discontinuation) of the medication over time. For example, although we observe (over a four-year period) a statistically and substantively strong association between pharmaceutical adoption in adjacent years, up to 38 percent of facilities that adopted naltrexone in one year did not adopt it the following year, and up to 42 percent of facilities that adopted buprenorphine in one year did not adopt it the following year. Once an innovation is adopted or implemented, what factors contribute to its continued use or availability and possible diffusion to other organizations over time? Research suggests that the compatibility of innovations the degree to which they are believed to be consistent with values, objectives, capabilities, and other technologies (e.g., treatment approaches) and client needs and the complexity or difficulty and costs of using them are important determinants of both their adoption and continued use (Rogers, 2003). Prior research also indicates a role for state health care and regulatory policies and the Page 7 of 79

8 general health care environment in the adoption of pharmaceutical agents in substance abuse treatment (Etheridge et al., 1997, Heinrich and Hill, 2008), relationships which we explore over time in this study. Recent findings on innovations in substance abuse treatment In recent years, new research findings on the adoption of innovations in substance abuse treatment, and particularly the adoption and diffusion of naltrexone, buprenorphine, and related pharmaceutical agents, have emerged. Researchers engaged in The National Treatment Center Study (NTCS) at the University of Georgia are using a nationally representative, stratified random sample of substance abuse treatment facilities, including private- and publicly-funded centers, therapeutic communities, and treatment programs affiliated with the Clinical Trials Network. Their unit of analysis is the treatment center (full sample=1,386 centers), and data were collected from program administrators, clinical directors, and counselors in these organizations, with the measures used in the published empirical analyses primarily constructed at the treatment center level (Durcharme et al., 2007; Knudsen and Roman, 2004; Knudsen et al., 2005a; Knudsen et al., 2005b; Knudsen et al., 2006; Roman et al., 2006). The dependent variables of interest in their multivariate analyses have included the adoption of pharmacological treatments, including disulfiram, buprenorphine, naltrexone, LAAM, and others, and psychosocial therapies. The NTCS researchers report that treatment center size, for-profit ownership, and accreditation are positively associated with the odds of buprenorphine adoption, as were programs offering inpatient and outpatient detoxification services and those that were already using naltrexone in treatment (Knudsen et al., 2006; Roman et al., 2006). Access to a physician, the proportion of opiate-dependent clients, and direct exposure to a clinical trial protocol were Page 8 of 79

9 also positively associated with buprenorphine adoption (Ducharme et al., 2007). Alternatively, the utilization of disulfiram was not related to treatment center ownership status, but it was more likely to be used in hospital settings, in-patient programs, and in centers with more professional counseling staff and those that did not have a spiritual emphasis in treatment (Knudsen et al., 2005b). Of particular note for the present study is the measure of diffusion constructed by Knudsen et al. (2005a) and Roman et al. (2006). This measure was based on counselors knowledge of the effectiveness of the treatment; a don t know answer was coded as indicating lack of diffusion. In predicting diffusion, these studies focused on treatment counselor attributes and reported that counselors who had received training about buprenorphine and were working in organizations that adopted buprenorphine were more likely to be knowledgeable about its effectiveness, as were counselors in recovery and with a longer history in the treatment field; they also found a significant negative association between a 12-step program orientation and buprenorphine adoption. In a separate study of the effects of workshop training and training quality in 32 treatment organizations, Simpson et al. (2007) showed that better training predicted more likely adoption and that staff communication and openness also played a role. In a study of outpatient substance abuse treatment centers in six New England states, Fuller et al. (2005) used data from three waves of surveys (1997, 1999, and 2001) to explore the correlates of naltrexone adoption over a five-year period. Although closures and mergers changed their sample size and composition over time, they uncovered stable relationships between organizational and staff characteristics and adoption. Using a structural equation model in the estimation, they found that participation in managed care arrangements was positively Page 9 of 79

10 associated with adoption of naltrexone, as were counselor education and the use of psychiatric medications. Oser et al. (2007) conducted event history analyses using four waves of the nationally representative National Treatment Center Study data to likewise examine factors influencing adoption of naltrexone over time, and they found that experience with pharmacological treatments and interconnectedness with other treatments organizations, as well as the normative pressures that accompany accreditation, promoted adoption of naltrexone. Heinrich and Hill (2008) reviewed other literature on naltrexone adoption, noting its focus on adopting (and non-adopting) organization characteristics and attitudinal data from physicians and clinicians. Roman and Johnson (2002), for example, analyzed information from interviews with administrators and clinical directors in private sector drug and alcohol treatment centers and found that naltrexone adoption was positively associated with facility age, administrators years of experience in the treatment field, staff professional/business and medical training, and the percentage of managed care patients and relapsed patients on the caseload. Mark, Kranzler, and Song (2003) showed that higher percentages of patients covered by Medicaid or with no insurance were associated with lower naltrexone prescription rates, and Miller et al. (2001) and Thomas and McCarty (2004) likewise concluded that both cost (the lack of parity in reimbursement) and the paucity of substance abuse-related training offered in medical schools were important factors limiting primary care physicians treatment of alcoholism and inhibiting the integration of alcohol and drug treatment with primary care and general health services. Although the studies we review suggest that there are a number of treatment center and intra-organizational attributes that consistently predict the adoption and diffusion of pharmaceutical agents (e.g., staff education, training and experience, organizational affiliations, Page 10 of 79

11 etc.), an important limitation of this prior research is its lack of attention to policy and environmental factors and how changes in external factors influence adoption and diffusion over time. Among the different variables identified above as important to diffusion (e.g., public policies and incentives for investments in innovations and their dissemination, monitoring and legitimization, environmental scanning and inter-organizational ties/networks, and other structural and environmental factors), we were challenged to find studies that had constructed measures of these factors at a level other than the organization/treatment center. For example, Knudsen and Roman (2004) defined environmental scanning based on treatment center administrators estimates of the extent to which their staff drew their knowledge of treatment innovations from publications and professional associations. In addition, as discussed earlier, the NCTS researchers measured the diffusion of buprenorphine using counselor reports of their knowledge of the effectiveness of the treatment at single point in time. We argue that there is considerable scope for advancement and improvement of the measures used in research on the adoption and diffusion of pharmaceutical agents. In fact, improving the measurement and analysis of state- and environmental-policy factors on adoption and diffusion was a central objective of this research. Researchers and practitioners have long pointed to the importance of state-level policy and environmental influences on organizational adoption of innovations and related changes in substance abuse treatment. Harris and Thomas (2004), for example, describe how states exercise their available discretion in Medicaid program provisions to influence treatment, such as charging co-pays, contracting with managed care programs, and imposing prescription limits. In addition, Drake et al. (2001), Goldman et al. (2001) and Simpson (2002) explicate how Medicaid coverage, managed care and related economic pressures strongly influence treatment practice. Other Page 11 of 79

12 researchers have also shown how public policies and other external environmental factors (such as increased cost containment pressures, shifts in ownership and funding, and changes in federal and state policies) interact with organizational characteristics to influence the internal environment of substance abuse treatment programs and treatment service approaches (Heinrich and Fournier, 2004; Schmidt and Weisner, 1998; Zarkin et al., 1995; Mardsen, 1998). Recently, Heinrich and Hill (2008) developed a state-level policy database, which we build on in this research, and found positive associations between facility-level naltrexone adoption and state-level policies such as Medicaid policies that support the use of generic drugs, reduce drug costs, and permit managed care organizations to establish policies encouraging use of generics. Alternatively, state policies that limit access to pharmaceutical technologies through Medicaid preferred drug lists, restrict access to pharmacy networks, and impose general limitations on use of Medicaid benefits for rehabilitation for substance abuse treatment were found to be negatively related to facility-level adoption of naltrexone. These findings indicate that policy levers available to state governments have the potential to influence the adoption of pharmaceutical technologies and thereby help to meet widespread need for access to clinicallyproven and cost-effective treatments for substance abuse. Data and Measures Data were compiled to facilitate the development of new models of adoption and diffusion of pharmaceutical agents that examine relationships over time and account for both external and internal factors that support or inhibit pharmaceutical use in addiction treatment. We obtained access to four years ( ) of data from the National Survey of Substance Abuse Treatment Services (N-SSATS) that include facility codes, allowing us to track and explain the adoption of pharmaceutical agents by treatment facilities and their diffusion to other Page 12 of 79

13 facilities within and across states over time. These data were linked with state-level data that were assembled by Heinrich and Hill (2008) and other state-level data were newly collected for this study. These linked data are used to examine whether particular factors are associated with the adoption and diffusion of specific pharmaceutical agents, or more generally with various agents, both within a given year and over time. Modeling diffusion over time is conceptually distinct from cross-sectional adoption decisions. Since 2002, the N-SSATS has added new questions that enable us to overcome some limitations of previous research on the adoption and diffusion of disulfiram, naltrexone and buprenorphine. Thus, we use the linked facility data over time, but we have also gathered or developed other measures of factors that have been identified as important to diffusion, including: policy-led diffusion, incentives for investment in innovation (including coverage and financing), direct exposure and legitimization of innovations, the number and proximity of adopters, inter-organizational ties and professional networks, administrative structures and managerial/technical capacity for adoption and diffusion, compatibility (with norms, values/culture, resources, etc.), and other environmental (political, social and economic) factors associated with adoption and diffusion. Appendix 1 Table A.1 presents descriptive statistics for facility-level measures we obtained from the N-SSATS, and Table A.2 in Appendix 1 shows descriptive statistics for statelevel measures that we derived and/or constructed from a number of different sources, including the Centers for Medicare and Medicaid Services, the Substance Abuse and Mental Health Services Administration, reports from the Kaiser Family Foundation, and other publicly available sources. The number of observations, mean, and standard deviation are shown for each Page of 79

14 measure for the years in which they are available. There were more than 54,000 facility observations over the four-year study period. We use the majority of facility-level measures as they appear in the original N-SSATS datasets. However, some recoding was required for a few of the state-level measures, and not every measure was available for each state. The most common reasons for data to be missing for a given state were: 1) the measure was not applicable to the state or the measure indicated the existence of a specific type of state law and the state did not have any law on the subject, or 2) the state did not participate in the survey used to collect the information. In the first case, we recoded data to generate measures indicating a given state had a law prohibiting a particular policy or allowing an exemption to a policy, while 0 in these cases indicated that the law did not prohibit the policy, did not allow an exemption, or was missing. In the second case, we assigned missing values when no information was available on a state policy for any year in a given state. However, in a few cases for variables derived from the Kaiser Family Foundation surveys (in 2003 and 2005), we inferred 2005 values based on 2003 data. In particular, for cases in which a state participated in the 2003 survey but not the 2005 survey, we assumed that a state would not reverse policies that lower the cost of Medicaid prescription benefits (as this was the trend observed for all other states with complete data). However, if a state did not have the cost saving policy in 2003, we did not make any assumptions about their 2005 policies. We recognize that we are modeling the diffusion of pharmaceutical technologies over a relatively short period of four years ( ). The identifiers for treatment facilities in the N- SSATS that enable us to link the facility data across these years were made available only for this period. At the same time, there is variation in the level and rate of change in the adoption of these different pharmaceutical agents over time. Descriptive analyses of adoption levels by Page 14 of 79

15 treatment facilities in our sample show ranges of 16.3 to 17.5 percent for disulfiram, 11.7 to 12.9 percent for naltrexone, and 5.4 to 7.8 percent for buprenorphine (the latter for the years ). More importantly for this study, however, these apparently stable year-to-year levels of adoption reported above mask considerable variation in facility-level decisions to adopt and continue (or stop) using a given pharmaceutical agent from year to year. Table 1, which reports how treatment facilities offering of these different pharmaceutical agents changed over time, suggests a number of interesting patterns. First, in each year and for each pharmaceutical agent, a non-negligible fraction (28-42%) of facilities that were offering a given treatment in the prior year had discontinued offering it in the subsequent year. Disulfiram is the pharmaceutical agent that has been available the longest, and this is the most stable of the three treatments in terms of the number of facilities that continue use from year to year (approximately 70%). A positive finding is that across all three pharmaceutical agents, the percent of facilities continuing to offer treatment from year to year is increasing over time, especially for buprenorphine, the most recently available of the three treatments. Alternatively, Table 1 shows that the fraction of treatment facilities that are new adopters from year to year is more variable. For disulfiram, there is a slight decline in the number of new adopters over time (from 30.2% to 27.6%), while naltrexone is up, down and then up again by a few percentage points. For buprenorphine, we can only compare changes across two years, but Table 1 shows a comparatively large fraction that were new adopters in 2004 (56.3%), which drops to about 40 percent in 2005 (and yet is still higher than for the other two treatments). Although we also uncovered patterns in which facilities adopted, stopped and then continued offering treatment (or vise versa) for all three treatments, these highly varying patterns of use, discontinuation and/or re-adoption involved only about 2 percent of facilities in each case. Page 15 of 79

16 The period over which we examine the diffusion of pharmaceutical agents also includes important variation in state Medicaid policies that we take advantage of in our analyses to assess policy-led diffusion. While we were limited in our policy measures to years 2003 and 2005 for most of these variables, the information in Table 2 confirms that significant changes were occurring in state policies over this period. Importantly, given the limitations of our data, the Medicaid policy changes we observed were unidirectional; that is, we did not see any reversals in the policy changes that states made such as requiring generics, lowering co-pays or paying generic rates for brand name drugs and the policy changes were also all heading in the same direction. In general, states were moving to make pharmaceutical agents more readily available in generic form and with lower co-pays, while at the same time limiting refills and access to brand name drugs (presumably to hold down Medicaid costs). Other state policies that we examine in this study include measures of state health insurance parity laws (for alcohol-related treatment) that we expect would influence incentives for coverage and financing of substance abuse treatment. Prior research has shown lower rates of health insurance coverage for individuals discharged from hospitals with a principal diagnosis of alcohol-related mental and behavioral disorder. At the same time, more recent research by Yi et al. (2007) found that while the risk of having no insurance coverage is five times higher for these individuals, this risk is reduced by approximately half for discharges in states that have must cover parity provisions. Table 2 shows measures of five state health insurance parity laws for alcohol-related treatment and their frequency of adoption in the states. Unlike the Medicaid policy measures shown above them, these policies changed little between 2002 and 2005 (based on data tabulations for all four years, not shown in the table). Page 16 of 79

17 Some of the state-level variables that we identified to measure external pressures to conform or encourage the adoption of innovations were only available for one or two years. For example, we measure the numbers or presence of organizations or support networks in the states that might influence the adoption of pharmaceutical agents by treatment facilities, such as the number of Drug Free Coalition Grantees and Clinical Trials Network Members in 2005, community treatment programs in 2004, and operating drug courts and Substance Abuse and Mental Health Services Administration (SAMHSA) Screening and Brief Intervention Grantees in 2004 and The implication is that we are limited in our ability to examine the role of these factors in the diffusion of pharmaceutical agents, although we might still explore their association with the level of adoption by treatment facilities in the states. Finally, as discussed above, we also sought to construct measures of the geographical proximity and density of treatment facilities adopting pharmaceutical agents over time (to measure conformity pressures or supports of diffusion), and to explore these relationships with geographical information system (GIS) modeling. Figures 1-9 present maps that are colored to show density of adoption/diffusion from year to year. To create the maps in these figures, we calculated (by year) the percentage of facilities per county that offered each pharmacotherapy. We merged these percentages, by state and county FIPS code, to the 2009 county TIGER/Line Shapefiles from the U.S. Census Bureau. We then mapped the percentages for each year and each pharmacotherapy separately, using GRASS GIS Version 6.4.0svn and Quantum GIS Version 1.0. In addition, we used the 2009 state TIGER/Line Shapefiles to outline state borders. Counties with facilities offering the pharmacotherapy are shaded on a green to blue scale in increments of 10 percent (according to the percentage of facilities offering the pharmacotherapy). Counties with no facilities offering the pharmacotherapy are in white. Page 17 of 79

18 Overall, our simple descriptive analyses and mapping of adoption density confirm that there is diffusion of these pharmaceutical technologies taking place over time, and that the rate of new adoptions and continuations (or discontinuations) varies both over time and across the three pharmaceutical agents we study. In addition, there is important variation in state policies over time and in their adoption across states that we can explore in relation to the use of these pharmaceutical agents by treatment facilities. It is this variation that we exploit in our analysis to advance our understanding of the various factors that might play a role in aiding or inhibiting the diffusion of evidence-based pharmaceutical technologies in addiction treatment.. Methodology The relationships we explore in our analysis of the diffusion of pharmaceutical agents span multiple levels, that is, with substance abuse treatment facilities nested within states. In addition, given that we are modeling these multilevel relationships over time, we also investigated using multilevel mixed models with repeated measures (or multilevel longitudinal modeling). However, in light of the fact that we have only four years of data, and some of the state-level variables were only available for one or two these years, a simpler, multilevel mixed model specification that examines the relationship of state and treatment facility characteristics to the number of years that facilities offered these pharmaceutical agents was a better fit with the data. The technical appendix presents additional details on the multilevel models. In addition, we estimate these same specifications with ordered and multinomial logit models that adjust the standard errors for the clustering of facilities within the states. In the models in which we predict the number of years that facilities offered these medical treatments, we use the values of these variables for the last year available (2005); facility-level characteristics are relatively unchanged over the study period ( ), and 2005 is more Page 18 of 79

19 likely to capture the direction in which policies were evolving, as well as most of the state-level characteristics we measure. Estimating these models with a multinomial logit specification (adjusting for clustering of facilities within states) allows us to see whether the policy and environmental variables at the state-level are more (or less) important in predicting facilities initial adoption or adoption for just one year vs. continuation (i.e., multiple years) of making the pharmaceutical agent available to clients. In general, the different models that we estimate in this study multilevel mixed, ordered logit and multinomial logit models of the number of years facilities offered the pharmaceutical agents produce substantive findings that are fairly consistent across the various approaches and specifications. Although prior research on diffusion (Oser et al., 2007) has employed event history models that examine time to adoption, in light of the patterns in Table 1 that show approximately one-third of facilities discontinuing use of these pharmacotherapies after adopting (in any given year) and then sometimes re-adopting, we think event history models might miss important variation that we will try to capture in the different empirical modeling approaches described above. Study Findings Treatment facility-level findings In discussing the findings of these models, we begin with the ordered logit model (the most basic specification), in which we examine factors that predict the adoption of these pharmacotherapies for one, two, three or four years, relative to the reference category of zero years (i.e., not offered over the period ). The results, shown in Table 3, largely confirm expectations (based on prior research) for the relationships of a number of the facilitylevel variables to the adoption and diffusion of innovations in substance abuse treatment Page 19 of 79

20 (Friedman, Alexander and D Aunno, 1999; Fuller et al., 2005; Knudsen et al., 2005, 2006; Roman et al., 2006; Ducharme et al., 2007; Heinrich and Hill, 2008), and these relationships are fairly consistent for the three pharmaceutical agents, disulfiram, naltrexone and buprenorphine. For example, one of the strongest, statistically significant relationships we observe across the three models is a positive association of accreditation of substance abuse treatment facilities by the Joint Commission on Accreditation of Health Care Organizations (JCAHO) and the adoption of these pharmaceutical agents for one or more years. JCAHO accreditation, which increases the odds of facilities offering these pharmacotherapies an additional year by percent, is seen as a signal of higher-quality services and may contribute to the perceived legitimacy of making these pharmaceutical treatments available (Oser et al., 2007). Licensing/certification by a hospital licensing authority (relative to a state substance abuse licensing agency) is likewise positively associated with adoption of all three pharmacotherapies over these years, which is expected given the well-known, important role of medical personnel in supporting the use of these treatments. Other facility-level factors that are influential for all three pharmacotherapies include the availability of relapse prevention counseling and outpatient substance abuse services, which appear to be used in combination with or support of these treatments, as commonly recommended by medical professionals. Facility-level treatment financing arrangements are also important predictors of adoption of these pharmacotherapies, as shown in Table 3. Consistent with prior research on participation in managed care arrangements and naltrexone adoption (Fuller et al., 2005), we find that facilities that accept Medicare and contract with managed care organizations have significantly higher odds (up to 197% and 123% greater, respectively) of Page 20 of 79

21 offering all three of these medical treatments. However, if they use a sliding fee scale for determining costs, the odds of adoption are about percent lower. One other facility-level finding that bears mention and differs across these pharmaceutical agents is the role of organizational form, i.e., private for-profit or nonprofit facilities vs. public (federal, local, county, or community government) organizations in treating addiction. Past research has produced mixed findings, with some studies suggesting positive roles of for-profit organizations in pharmacotherapy adoption (Knudsen et al., 2006; Roman et al., 2006; Oser et al., 2007), and other studies pointing to more limited payment arrangements and/or service offerings in for-profits relative to public or nonprofit organizations (Friedman, Alexander, and D Aunno, 1999; Wheeler and Nahra, 2000; Heinrich and Lynn, 2002). The results in Table 3 (as well as in other empirical models we discuss below) show that for-profit treatment facilities are significantly less likely (compared to public facilities) to offer disulfiram or naltrexone, while they are significantly more likely to adopt buprenorphine. Recall that during the time period of this study, the use of buprenorphine by medical professionals for treating opioid dependence had just been approved by the FDA (in 2002), whereas previously only a limited number of specially registered drug treatment centers were allowed to offer this pharmacotherapy. One possible explanation for the mixed findings on the adoption of naltrexone (in earlier vs. more recent studies) and the differences in these findings for naltrexone and disulfiram vs. buprenorphine might relate to the timing of the research investigations relative to the approval of these medical treatments. For example, prior to states taking policy actions to make these treatments more affordable (e.g., requiring generics), these treatments might be more likely to be offered in private for-profit treatment facilities that receive more of their revenues from patient fees and private insurance. Oser et al. (2007) also suggest that for-profit facilities Page 21 of 79

22 are more likely to possess the financial slack to absorb the start-up costs of early pharmacological adoption, in addition to the human resources required to pursue innovation and experimentation. Before considering the findings on state-level factors, we first look to the results for the multinomial logit models (see Table 4) that allow us to examine whether these facility-level factors differentially predict adoption/continued use of these pharmacotherapies for one, two, three or four years. Indeed, there are some important differences. For example, the odds of facilities adopting these treatments for just one (vs. zero) years are percent higher if they are JCAHO-accredited; for two years vs. zero years the odds are percent higher, and for adoption over three (or four) years vs. zero years, the odds are percent higher for JCAHO-accredited facilities. For other facility-level factors, including the offering of outpatient substance abuse services, accepting Medicare, licensing by a hospital licensing authority and forprofit ownership, we likewise observe a stronger role for these factors in adoption and continued use of these pharmacotherapies over multiple years. Findings on state policies and other state-level factors Clearly, however, state-level policy and other factors are also playing an important role in influencing whether facilities adopt disulfiram, naltrexone and buprenorphine and continue to offer them (and possibly encourage adoption among other facilities). The results in both Tables 3 and 4 show that facilities in states with Medicaid preferred drug lists are significantly less likely to adopt disulfiram, naltrexone and buprenorphine and to do so for multiple years, with odds as much as 46 percent lower for buprenorphine and naltrexone (see Table 4). Table 4 also shows that state Medicaid policies have the strongest effects on the adoption and continuation of buprenorphine over three years; while lowering generic co-pays and including generics on the Page 22 of 79

23 preferred drug lists/formularies significantly increase the odds of facilities offering buprenorphine all three years (by 71-89%), requiring generics significantly lowers the odds of adoption by more than 90 percent. We speculate that these relationships are stronger for buprenorphine (compared to disulfiram and naltrexone) over this period ( ) because it had just received FDA approval for expanded use in addiction treatment. In addition to state Medicaid policies, state policies requiring health insurance parity for alcohol-related treatment are also significantly related to facilities adoption and continuation of these pharmacotherapies. Particularly for adoption of buprenorphine and naltrexone, state policies that require insurers to provide coverage for alcohol-related treatment significantly increase the odds of adoption and continued offering of these treatments over three to four years (by as much as 181%). On the contrary, state policies that restrict facilities from imposing greater copayments or coinsurance on alcohol-related treatment significantly reduce the odds that facilities offer buprenorphine and naltrexone, again particularly for multiple (3-4) years. 2 Exempting health insurers from some or all of the parity provisions is also negatively related (and statistically significantly, primarily for buprenorphine) to facilities adoption and continuation of these medical treatments. Another interesting but mixed finding on state policies affecting treatment financing is the relationship between state alcohol tax collections dedicated for treatment and the adoption of these three pharmaceutical agents. Facilities in states with alcohol tax collections dedicated for treatment have up to 121 percent higher odds of adopting buprenorphine (the direction of the 2 The multinomial logit model results suggest that the odds of adoption of disulfiram for just one (vs. zero) years are higher for facilities in states that restrict facilities from imposing greater copayments or coinsurance on alcoholrelated treatment; however, this relationship is reversed for facilities offering disulfiram all four years (with the odds 24% lower for facilities in states that impose this restriction). For a pharmacotherapy that has long been available (disulfiram), adoption for just one year over the four-year period of this study might reflect a late or trial adopter that is unsuccessful in continuing to offer the medical treatment. Page 23 of 79

24 relationship we expected); however, the odds of them adopting disulfiram or naltrexone are reduced by about half if the state dedicates some portion of tax revenues to treatment. States that dedicate tax collections for treatment are also significantly more likely to restrict facilities from imposing greater co-payments and to require generics, but they are less likely to allow lower copays for generics, suggesting that they may shift a larger share of the burden for making treatment affordable to the facilities. We also find some statistically significant relationships among inter-organizational or networking and other external/environmental factors that theory suggests may be potentially important in explaining the adoption and diffusion of substance abuse treatment innovations. For example, as discussed previously, early knowing and exposure to innovations that support diffusion are expected to increase through both direct and indirect ties with other organizations. Among the numerous measures of such factors that we constructed and tested in our models, several were statistically significant predictors of adoption of pharmaceutical agents, although not in all models. For example, the number of Drug-Free Coalition grantees in the states increases the odds of adoption of all three pharmacotherapies, and this variable is statistically significant for naltrexone and buprenorphine, with approximately 2 percent greater odds of adoption for each additional grantee (see Table 3). Initially, we were surprised to see some negative relationships between the number of Clinical Trials Network members and recipients of SAMHSA Screening and Brief Intervention grantees in the states and facilities adoption and continued offering of pharmacotherapies. However, a comparison of their locations with the maps in Figures 1-9 showing geographical clusters of adoption suggests that some of these grants and networking efforts were being targeted to counties and (large) states where adoption and diffusion rates were relatively low, such as Texas, Illinois, Ohio and Page 24 of 79

25 Pennsylvania. As we first observe the allocation of Screening and Brief Intervention grants in our data in 2004, we do not expect to observe effects of these interventions over the study period. The data used to construct the maps showing the distribution and density of treatment facilities offering disulfiram, naltrexone and buprenorphine were also included as variables to capture the role of conformity pressures or the proximity of leading/legitimizing behavior in the adoption and diffusion of pharmaceutical agents. In all of the models estimated, including the multilevel models that have not yet been discussed, these were among the most powerful (statistically significant) and precisely estimated predictors of adoption/diffusion. The ordered logit results in Table 3 suggest that for each additional percentage of facilities in a given county that are offering pharmaceutical agents, the odds of a facility adopting the medical treatment are 4.4 to 7.6 higher. Furthermore, the multinomial logit results in Table 4 show that proximity to other adopting facilities is even more important for continuing to offer the treatment in additional years the odds of offering buprenorphine all three years are 11 percent higher for each additional percentage of facilities in a county offering buprenorphine, and the odds of offering disulfiram and naltrexone for all four years in the study period are almost double those shown in Table 3 for each additional percentage of facilities offering these pharmacotherapies. Findings of multilevel model specifications Table 5 presents the results from the multilevel (mixed) models that predict the number of years a pharmaceutical agent was offered (over the study period). These models correspond most closely to the ordered logit models in Table 3. Indeed, there are no major substantive differences in the results of these simple mixed models that employ an alternative approach to adjusting for facilities clustering or nesting within states, although a few of the relationships no longer attain statistical significance (e.g., the number of Drug-Free Coalition grantees in the Page 25 of 79