Research in Social and Administrative Pharmacy j (2015) j–j

Commentary

Prior authorization policies in Medicaid programs: The importance of study design and analysis on findings and outcomes from research Shellie L. Keast, Ph.D.a,*, Kevin Farmer, Ph.D.a, Michael Smith, Ph.D.a, Nancy Nesser, Pharm.D., J.D.b, Donald Harrison, Ph.D.a a

University of Oklahoma College of Pharmacy, ORI-W4403, P.O. Box 26901, Oklahoma City, OK 73126-0901, USA b Oklahoma Health Care Authority, Oklahoma City, OK, USA

Summary U.S. State Medicaid programs for the medically indigent strive to deliver quality health care services with limited budgets. An often used cost management strategy is prior authorization of services or prescription medications. The goal of this strategy is to shape the pharmaceutical market share in the most efficient manner for the particular state Medicaid program, much like commercial managed care organizations. These policies are often scrutinized due to the population Medicaid serves, which in the past was largely composed of individuals with vulnerable health status. Unintended consequences can occur if these policies are not carried out in an appropriate manner or if they greatly restrict services. The data used for policy implementation research is prone to certain problems such as skewness and multimodality. Previous guidelines have been published regarding the best practices when analyzing these data. These guidelines were used to review the current body of literature regarding prior authorization in Medicaid. Further discussed are additional characteristics such as therapeutic areas researched and the outcomes identified. Finally, the importance of considering state-specific characteristics when reviewing individual policies and the usefulness of these results for other programs are also considered. Ó 2015 Elsevier Inc. All rights reserved. Keywords: Medicaid; Prior authorization; Methodology; Statistics; Outcomes; Unintended consequences

Background U.S. State Medicaid programs continuously struggle to balance limited budgets with increasing demand for services. Cost management strategies for state Medicaid programs typically center around three “short-term” control measures: provider reimbursement, enrollment limitations, and benefit reductions and/or restrictions.1

The majority of Medicaid cost management efforts center on benefit reductions strategies. The pharmacy benefit component of expenditure control strategies is focused on the cost of the products and the utilization of the products.2 Strategies to control product cost or utilization include: generic mandates, pricing restrictions, step-therapy, prior authorizations, formulary

* Corresponding author. Tel.: þ1 405 271 8222; fax: þ1 405 271 6602. E-mail address: [email protected] (S.L. Keast). 1551-7411/$ - see front matter Ó 2015 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.sapharm.2015.04.003 RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

2

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

restrictions, manufacturer rebates, reduced professional fees (provider reimbursement), and increased patient cost share. For Medicaid programs, the mainstay costcontainment measure for non-generic medications has become some form of prior authorization (PA) program, such as step-therapy or preferred drug lists (PDLs). Most programs rely on an approval process or prior authorization strategies to encourage use of preferred prescription products. For various reasons, Medicaid programs have not utilized significant copay differentials to discourage the use of nonpreferred products. Prior authorizations can be manual or automated system processes. Step therapy programs utilize an evidence-based tiering of available products and typically place the most costeffective therapies on the lowest tier, or step. Higher tiers or steps require at least one trial of a lower tier medication or an approval process. As of 2004, all but 7 of the contiguous states had or planned to have a preferred drug list.3 However, there is significant controversy over the ability of these programs to provide quality care while controlling costs; nevertheless, these programs have become a permanent component of Medicaid pharmacy benefits.4 The goal of prior authorization policies within Medicaid programs is to actively shape the market share or utilization of drug products in the most efficient manner for that particular state Medicaid program. Each state brings its unique population demographics, reimbursement strategies, regional prescribing influences, and political climate to the table when determining pharmacy benefit policies. Thus, each state’s prior authorization policies are uniquely their own and may not be comparable to other state programs. Furthermore, because Medicaid programs often serve individuals in vulnerable physical and mental states of health, it is important to formally review these policies to ensure the intended results were achieved while limiting unintended consequences.5 In the majority of cases, the intended result of a Medicaid prior authorization policy is to decrease spending on pharmaceutical products while maintaining quality of care. Unintended consequences occur when other aspects of an individual’s health care is negatively impacted due to suboptimal therapeutic outcomes and may be seen as increased emergency room visits, physician office visits, adverse events, or utilization of other health care services. It may also be important to consider the therapeutic category which has been included in the policy and the sensitivity of the population to changes in therapies for the applicable disease states.

Hazards in health care data analysis When analyzing health care resources and costs, there are inherent characteristics of the data that must be considered. Health care data are rarely normally distributed. The data typically are non-negative and skewed positively to the right. The data can also be multimodal with many peaks and troughs; or have an excess of zeros.6,7 Most data used for policy analysis are from paid administrative claims that were designed for payment, not for research purposes. Most studies are observational and historical. Lack of randomization allows for potential biases and confounding.7 Many evaluations of prior authorization policies in state Medicaid programs have been performed; however, as statistical methods become more sophisticated and the results of these assessments garner greater scrutiny, it is important to review the current body of literature regarding the implementation of prior authorization in Medicaid programs and to discuss whether the literature meets currently published guidance for analyses of these data.

Published guidelines for handling data Two published papers provide guidelines for analysis of health care utilization and resources, Diehr (1999)6 and Mihaylova (2011).7 Diehr and colleagues outlined methods for working with these types of data.6 Adjustments for patient characteristics such as age and gender are often only done with a simple linear method and do not account for potential interactions. Additionally, adjustments made using an individual’s past utilization (which is a strong predictor of utilization to come) could mask differences. While onepart models are appealing due to their ease of use, these can be less informative when dealing with multi-nodal data, particularly those with a zero intensity, in which case a two-part model is preferred. Transformation of data is often required to overcome the normality issues, unless the data set is large enough to overcome these problems. Diehr offers recommendations on which model to use: for understanding systems, a two-part model is most appealing and preferred; or for analyzing individual effects on costs, a onepart model may be more appropriate.6 Besides discussion regarding the distributions of these data, Diehr also discusses often neglected issues related to costs such as whether the billed charges are representative of the true costs.

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

Mihaylova and colleagues offered a colorcoded scheme for determining the most robust methodology: green, amber, and red orbits.7 Data lie within in the green orbit when there is a large sample size (defined as hundreds to thousands in each group) for the Central Limit Theorem to apply and normalcy issues to dissolve. For data in the amber orbit, the sample size is insufficient to outweigh normalcy issues, or the data are too complex. For these types of data-sets Mihaylova recommends using alternate distributions with inverse gamma being the most highly recommended. Transformations are also recommended, but back-transformations are required. Generalized linear models are also an option, but require appropriate distribution choices and may suffer from loss of precision. Finally two-part models may be appropriate for amber orbit data particularly if there is a need to appropriately handle zero costs. If data fall within the red orbit, only those with significant experience should perform the modeling. These models include using Bayesian or mixture models.7 Studies meeting appropriate criteria were examined for their application of these guidelines.

Applying guidelines to published studies on prior authorization policy A search of MEDLINE was performed to locate relevant publications on the use of prior authorization for medications in state Medicaid programs during the past 15 years (1995–2014). ‘Prior authorization’ and ‘Medicaid’ were the primary search terms used and resulted in 36 articles for detailed review. An additional 8 articles were located as a result of citation reviews. In order to be included in the analysis, the articles had to 1) be published in a peer-reviewed journal, 2) review a prior authorization program implementation for clinical or cost-related outcomes, 3) be performed in a state Medicaid population, 4) be published between 1995 and 2014, and 5) be available in the English language. Excluded studies were 1) prior authorization reviews of commercial plans, 2) discussion of policies and not an implementation review, and 3) published prior to 1995 or after 2014. From the studies selected for detailed review, the study design, population included, outcomes measured, drug classes included, and results were extracted. A total of 26 studies met all inclusion criteria and were included in this review.8–33

3

For the comparison with previous guidelines for health care utilization and resource methods, we applied Mihaylova and colleagues7 orbits to classify the studies. Both Diehr6 and Mihaylova7 offer similar methods, but the latter provides an easy system for classification. Each study was classified into an orbit and then postulated on whether the main analysis method was appropriate based on the recommendations for each orbit. Appropriateness was further separated into 3 levels: acceptable (A), mixed (M), not discussed in guidelines (N). Additionally, we looked for adjustments made for patient-specific characteristics as suggested by Diehr and colleagues.6 All classifications are based on the opinions of the authors. Green orbit To be classified in this orbit, studies should be significantly large with hundreds to thousands of observations in each grouping. They should also be relatively simple analyses with minimal groups or adjustments. Seventeen studies fell within this category due to their very large sample sizes (greater than 1000 per group) and/or use of aggregated data (averages) for the analyses (Table 1),8–10,13–18,20,21,23–27,29 with the majority of these studies occurring from 1995 to 2010. Amber orbit Studies classified in the amber orbit may have large numbers of observations, but also required more complex analyses, especially if adjustments were made for patient characteristics. A total of 9 studies were considered in the amber orbit (Table 1).11,12,19,22,28,30–32,34 Red orbit For a study to be classified in the red orbit, sufficient complexity requiring statistical analyses which are less standard and often not included in standard software packages would be used. Of the studies we included for review, we felt none met this criteria. Appropriate methods All but 4 studies had methods which were considered appropriate for their assigned orbits (Table 1).12,16,26,30 Three studies had methods which were not discussed in either Mihaylova7 or Diehr’s6 framework.16,26,30 This does not mean that the methodology was inappropriate, however it did not fall within the scope of the

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

4

Table 1 Evaluated studies’ application of orbits guidelines Article

Orbits7 Green

Smalley (1995)8 Kotzan (1996)9 Hartung (2004)10 Fischer (2004)11 Delate (2005)12 Roughead (2006)13 Carroll (2006)14 Fischer (2007)15 Siracuse (2008)16 Fischer (2008)17 Morden (2008)18 Farley (2008)19 Law (2008)20 Margolis (2009)21 Adams (2009)22 Zhang (2009)23 Walthour (2010)24 Lu (2010)25 Simeone (2010)26 Law (2010)27 Lu, Adams (2011)28 Lu, Law (2011)29 Kloepfer (2012)30 Constantine (2012)31 Keast (2014)33 Clark (2014)32 a b

Amber

Red

X X X X X X X X X X X X X X X X X X X X X X X X X X

Methods appropriate for orbita,6,7

Adjustments for patient characteristicsb,6

A A A A M A A A N A A A A A A A A A N A A A N A A A

No No No Partial Partial No Yes No No No No Partial No Yes Partial Yes No Yes No No Yes No No Yes Yes Partial

Acceptable (A), Mixed (M), Not Discussed in Guidelines (N). Partial indicates adjustments were made to sub-analyses.

guidelines. The final study, which was considered in the amber orbit due to a smaller sample size and more complex analyses, was classified as mixed.12 This study, which was one of the few studies in the early years to utilize individual patient-level data and performed sub-analyses which adjusted for patient characteristics including two-part modeling, added a constant to the zeros for at least a portion of the analyses. The adding of a constant to the zero responses is not recommended by Mihaylova.7 Adjustments for patient characteristics Almost half of the studies included adjustments for patient characteristics for all or part of the analyses.11,12,14,19,21–23,25,28,31,33 Most often adjustments were made as part of smaller sub-analyses which were performed alongside more population-based analyses. As indicated

previously when considering the orbit classifications, adjustments for patient characteristics increased as use of individual patient-level data increased.

Additional characteristics of prior authorization studies Five of the included studies were conducted at the national level using data from all state Medicaid programs.11,13,15,17,18 Georgia,9,19,24 Maine,23,25,29 and Michigan22,27,29 had 3 published studies of their State program (with some combined), while Indiana had 2 published studies.27,29 There were 14 studies that included a comparator group or state. All but one study was conducted using secondary databases of administrative claim records; the final study reviewed the actual prior

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

authorization requests submitted.31 Almost all studies utilized a quasi-experimental design which was longitudinal in nature (Table 2). The most common analytical approach for evaluating PA policies was in a pre/post manner. Five studies used methodologies which were not longitudinal.14,16,21,30,31 It is of interest to note that all but 6 studies had access to individual person level data, however only 9 studies14,16,21,23,28,30–32,34 actually analyzed the data on an individual level (rather than on an aggregate level). Four studies that used aggregated data for the time-series analysis used individual level data for subanalyses.12 Statistical methodology The statistical methods used for these prior authorization reviews ranged from simple descriptive techniques (Chi-square and T-tests) to more complicated methods such as negative binomial

5

and Kaplan–Meier survival curves. At least 16 studies utilized some form of time-series or segmented regression based on the pre- and post-implementation design (Table 2). This method was used almost exclusively in these studies from 1995 through 2008. After 2008, the majority of the studies continued to use this method; however more diversity can be seen. It is important to note that some of the most recent analyses also included less complex strategies (e.g., Chi-square, T-tests, and Wilcoxon sum rank tests). Therapeutic areas reviewed The therapeutic areas reviewed can be grouped into three main categories: 1) mental health, 2) chronic disease, and 3) pain relief (Table 3). There were 8 studies8–11,13,16,18,21 that focused on pain relief, with the majority related to the prior authorization of non-steroidal anti-inflammatory agents

Table 2 Statistical methods employed in the studies evaluated Study

Multivariate statistics

Statistical inference

Individual Time-series Generalized Survival Parametric Non-parametric level data available or segmented linear analysis for all time regression regression points

Aggregated data for primary analysis

Smalley (1995)8 Kotzan (1996)9 Hartung (2004)10 Fischer (2004)11 Delate (2005)12 Roughead (2006)13 Carroll (2006)14 Fischer (2007)15 Siracuse (2008)16 Fischer (2008)17 Morden (2008)18 Farley (2008)19 Law (2008)20 Margolis (2009)21 Adams (2009)22 Zhang (2009)23 Walthour (2010)24 Lu (2010)25 Simeone (2010)26 Law (2010)27 Lu, Adams (2011)28 Lu, Law (2011)29 Kloepfer (2012)30 Constantine (2012)31 Keast (2014)33 Clark (2014)32

X X X X X X

X X X X X X X X X X X X X

X X X X

X

X

X

X X X X X X

X X X

X X X X

X X X

X X X

X

X X X X

X

X X X

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

X X X X X X X X X X X

X X X X X X

X X

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

6

Table 3 Therapeutic areas and outcomes evaluated Study

Therapeutic area Mental health

Smalley (1995)8 Kotzan (1996)9 Hartung (2004)10 Fischer (2004)11 Delate (2005)12 Roughead (2006)13 Carroll (2006)14 Fischer (2007)15 Siracuse (2008)16 Fischer (2008)17 Morden (2008)18 Farley (2008)19 Law (2008)20 Margolis (2009)21 Adams (2009)22 Zhang (2009)23 Walthour (2010)24 Lu (2010)25 Simeone (2010)26 Law (2010)27 Lu, Adams (2011)28 Lu, Law (2011)29 Kloepfer (2012)30 Constantine (2012)31 Keast (2014)33 Clark (2014)32

Chronic disease

Outcome measure Pain

Utilization based

Cost based

Clinical based

X X X X

X X X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X

X X X X X X X X X X X X X X X X X X X X X

X

X X

X

X

X

X X X X X X

X X

X

X X X

Overall findingsa þ þ þ/ þ þ þ/ þ þ þ þ/ þ/ þ/ þ/ – þ/ þ þ – þ þ þ þ þ þ þ þ/

a

The study was classified with a ‘þ’ if the outcomes were identified by the authors to be positive, such as cost savings were achieved or no increases in emergency department visits. The study was classified with a ‘’ if the outcomes were identified by the authors to be negative, such as increased costs or increased hospitalizations. If there were both positive and negative findings found, then the study was classified with both (‘þ/’).

(NSAIDs) and cyclooxygenase-2 inhibitors (COX2). A total of 7 studies12,15,17,27,29,30,34 focused on a chronic disease. There were 6 diseases reviewed: gastrointestinal, rheumatism, hypertension, hyperlipidemia, allergic rhinitis, and asthma. A final 10 studies19,20,22–26,28,31,32 were in the mental health arena. These studies mainly focused on schizophrenia, bipolar disease, and depression. Outcomes evaluated The outcomes reviewed in these studies can also be roughly grouped into 3 main categories: 1) utilization, 2) cost, and 3) clinical (Table 3). Utilization outcomes included prescription volume, defined daily doses, market share, and medical service utilization. Cost outcomes included the changes in prescription costs, health care service costs, and administrative costs. Finally, clinically

based outcomes included review of medical service utilization such as emergency room visits or other utilization measures used as a proxy for clinical control or outcome, especially to determine if unintended negative clinical outcomes were incurred as a result of the prior authorization implementation. As expected, all studies included a review of utilization ranging from a narrow focus on the single category being reviewed to a wide range of utilization which included additional drug categories and multiple health services utilization. Not all studies included cost in the analyses; 9 studies did not consider any costs associated with the prior authorization implementation.22,24,25,28,31,34 A total of 17 studies included outcomes which served as estimates for the occurrence of clinically relevant unintended consequences attributed to the prior authorization implementation.8–10,12,16,19,22–26,28,31,32,34

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

Discussion Evaluation of policy implementation is a crucial component of the evidence base for providing cost-effective care. It is also an important process in both the summative and formative programmatic evaluation processes. It may seem that with the large number of published evaluations uncovered, that there is little additional work needed in this area. However, as State Medicaid programs potentially expand the number of beneficiaries beyond the traditional recipient population (due to the U.S. Affordable Care Act), it is necessary to continually review policy implementation for efficiency in light of fragile patient populations and in high-cost disease states. This review attempts to provide a snapshot of the current published peer-reviewed literature on prior authorization implementation in Medicaid programs and provide insight into future research directions which may help shape both cost-effective and clinically safe policies. The authors acknowledge that the classifications and results of this review are based on the observations made from review of the individual studies and subject to the information available in each publication. As the purpose of the data are claims payment, they are not designed for research purposes; thus, all research undertaken with these data has the potential for errors of commission and omission. However, these data are generally considered acceptable for retrospective analyses. Unfortunately, serious errors of interpretation can be made when these data are utilized by researchers who do not thoroughly understand historical events and state-specific nuances that provide background for the data. It is possible for the uninitiated researcher to unknowingly experience confounding errors. This may occur when there is an undocumented policy or benefit change that occurs simultaneously or is closely linked with the policy modification being studied. Without knowledge of the subtle nuances in the data, the researcher may reach unreliable and invalid conclusions. In reviewing the collected studies, it is possible that this may have occurred in studies which utilized national data. For instance, in Roughead et al,13 the authors included a list of characteristics and timing of adoption of PA policies for Cox II inhibitors in 19 states. Two pairs of states had criteria listed which were exactly alike: 1) “PA required if younger than 60 years”, and 2) “PA required”.18 Other states had more extensive descriptions of

7

their PA policies and all varied to some degree. Although the methods employed in this study are appropriate, we do not know if differences within each state’s PA policy had unique influences on their results. Within the research context, it is preferred to have a counterfactual comparator group to reduce potential confounding35,36 variables. In selecting PA implementation published research studies in the context of this process, only 14 studies used some form of comparator or control group. Several of the studies we reviewed attempted to include multiple states, or use comparative states for control groups as a method to increase the internal rigor (internal validity) of the research. Also, it is typically expected that research should be accomplished in a manner that helps maximize generalizability while strengthening the study’s internal validity. If the study was undertaken with sufficient internal and external validity, the results of a PA implementation in one state Medicaid program should be generalizable to another state Medicaid program. However, it may be necessary to accept that there are influences (i.e., confounding factors) on a single state’s policy implementation that may create a unique situation in that state Medicaid program, and thus the results (whether positive or negative) cannot be generalizable to every state program or the overall U.S. Medicaid population. Rather, each policy should be reviewed within the context of the state itself regardless of reported results; all possible influences (confounders) on the policy within the state should be presented and evaluated. Finding a true counterfactual comparison population between state Medicaid programs is challenging. Differences in copay amounts, prescription limits, and strictness of prior authorization approval processes all create unique settings and hinder the application of a true counterfactual setting. Yet it is important to continue to improve research investigating policy implementation by utilizing methods which attempt to control for confounding variables between populations and the complexity of health care utilization data. At least 19 studies8–13,15,17–20,22–25,27,32,33 examined in this inquiry utilized some form of timeseries or segmented regression based on the basic pre- and post-implementation design (Table 2). When sufficient variables are controlled, this is an acceptable, if not preferred statistical method for evaluation of policy implementations.37 However, there are two important concepts that must

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

8

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

be addressed: 1) use of individual level data in an aggregated fashion, and 2) minimal use of twostep modeling or alternative distribution curves. Recommendations for future research Health care utilization data have unusual characteristics that often make their analysis problematic. As previously mentioned, health care data are usually positive and skewed to the right. In some cases, this can be very extreme. Among the literature included in this inquiry, a popular method for dealing with this uniqueness has been to aggregate the data for policy review into monthly or quarterly averages, thus eliminating some degree of skewness. When health care data are aggregated, review of within-individual changes/characteristics is no longer possible. In order for the researcher to evaluate withinindividual changes, alternate analytical methods such as those proposed by Mihaylova7 and Diehr6 are acceptable methods that could be considered. Of particular interest is incorporating processes for managing the phenomenon of ‘zero-clumping’ (when a patient has zero utilization of a health service at a given time point) and long positive tails.38,39 Of the studies examined, only one study12 utilized this method on the available individual level data. Based on information included in the articles, it is impossible to determine if other studies observed this phenomenon, however it is highly suspected that had individual-level data been utilized, this may have occurred in more than the single study which accounted for it. By utilizing two-part modeling, both the changes in the costs or utilization and the correlation between an individual with high values and frequent occurrences can be taken into account.38 This information could be useful to policy makers who need to determine the most efficient use of limited resources. And even if less complex methods are the most appropriate for the data at hand, the modeling of individual level data focuses on the changes to the individuals themselves. While this may reduce generalizability to other payer situations, it does allow the policy maker to determine if the policy implementation was successful for the population being studied. For example, Constantine et al,31 were able to notice changes in prescriber patterns, which could indicate a state-specific problem that needs to be addressed. By maintaining the individual level approach, subtle differences which are important to specific populations can be determined. These

differences might then result in policy change or information which other payers can use when developing their own programs. It may be more important to determine why an individual policy implementation was a success or failure than to determine across the board that all policies for a particular disease state are effective.

Conclusion Each research endeavor should use the most efficient and robust method available based on the data collected. It is acknowledged that each researcher is subject to the limitations of each unique data set and the hypotheses being tested. It is imperative that research continues to investigate cost-effective delivery of prescription drug products, particularly with U.S. Affordable Care Act health care coverage expansions. It is also necessary that each policy implementation is reviewed in the light of the unique influences which are affecting the decisions being made on an individual geographic area and that the lessons learned from these situations be translated into improved efficiency and quality of care for other payers or locations. Further it is imperative that appropriate methodology and statistical analyses continue to be performed to ensure results are accurate, meaningful, and informative.

References 1. Hansen M. Confronting costs. State Legis 2012;38: 30–32. 2. Principles and Practices of Managed Care Pharmacy. Alexandria, VA: Academy of Managed Care Pharmacy; 1995. 3. King M, Gordon D, Cauchi D, et al. Medicaid: 10 fixes that work. State Legis 2004;30:14–18. 4. Hamel MB, Epstein AM. Prior-authorization programs for controlling drug spending. N Engl J Med 2004;351:2156–2158. 5. Soumerai SB. Benefits and risks of increasing restrictions on access to costly drugs in Medicaid. Health Aff 2004;23:135–146. 6. Diehr P, Yanez D, Ash A, Hornbrook M, Lin DY. Methods for analyzing health care utilization and costs. Annu Rev Publ Health 1999;20:125–144. 7. Mihaylova B, Briggs A, O’Hagan A, Thompson SG. Review of statistical methods for analysing healthcare resources and costs. Health Econ 2011;20:897–916. 8. Smalley WE, Griffin MR, Fought RL, Sullivan L, Ray WA. Effect of a prior-authorization requirement on the use of nonsteroidal antiinflammatory drugs by Medicaid patients. N Engl J Med 1995;332:1612–1617.

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10 9. Kotzan JA, McMillan JA, Jankel CA, Foster AL. Initial impact of a medicaid prior authorization program for NSAID prescriptions. Stud Pharm Econ; 1996:113. 10. Hartung DM, Touchette DR, Ketchum KL, Haxby DG, Goldberg BW. Effects of a priorauthorization policy for celecoxib on medical service and prescription drug use in a managed care Medicaid population. Clin Ther 2004;26:1518– 1532. 11. Fischer MA, Schneeweiss S, Avorn J, Solomon DH. Medicaid prior-authorization programs and the use of cyclooxygenase-2 inhibitors. N Engl J Med 2004; 351:2187–2194. 12. Delate T, Mager DE, Sheth J, Motheral BR. Clinical and financial outcomes associated with a proton pump inhibitor prior-authorization program in a Medicaid population. Am J Manag Care 2005;11: 29–36. 13. Roughead EE, Zhang F, Ross-Degnan D, Soumerai S. Differential effect of early or late implementation of prior authorization policies on the use of Cox II inhibitors. Med Care 2006;44:378–382. 14. Carroll NV, Smith JC, Berringer RA, Oestreich GL. Evaluation of an automated system for prior authorization: a COX-2 inhibitor example. Am J Manag Care 2006;12:501–508. 15. Fischer MA, Choudhry NK, Winkelmayer WC. Impact of Medicaid prior authorization on angiotensin-receptor blockers: can policy promote rational prescribing? Health Aff 2007;26:800–807. 16. Siracuse MV, Vuchetich PJ. Impact of Medicaid prior authorization requirement for COX-2 inhibitor drugs in Nebraska. Health Serv Res 2008;43: 435–450. 17. Fischer MA, Polinski JM, Servi AD, Agnew-Blais J, Kaci L, Solomon DH. Prior authorization for biologic disease-modifying antirheumatic drugs: a description of US Medicaid programs. Arthritis Rheum 2008;59:1611–1617. 18. Morden NE, Zerzan JT, Rue TC, et al. Medicaid prior authorization and controlled-release oxycodone. Med Care 2008;46:573–580. 19. Farley JF, Cline RR, Schommer JC, Hadsall RS, Nyman JA. Retrospective assessment of Medicaid step-therapy prior authorization policy for atypical antipsychotic medications. Clin Ther 2008;30:1524– 1539. discussion 06–7. 20. Law MR, Ross-Degnan D, Soumerai SB. Effect of prior authorization of second-generation antipsychotic agents on pharmacy utilization and reimbursements. Psychiatr Serv 2008;59:540–546. 21. Margolis JM, Johnston SS, Chu B-C, et al. Effects of a Medicaid prior authorization policy for pregabalin. Am J Manag Care 2009;15:e95–102. 22. Adams AS, Zhang F, LeCates RF, et al. Prior authorization for antidepressants in Medicaid: effects among disabled dual enrollees. Arch Intern Med 2009;169:750–756.

9

23. Zhang Y, Adams AS, Ross-Degnan D, Zhang F, Soumerai SB. Effects of prior authorization on medication discontinuation among Medicaid beneficiaries with bipolar disorder. Psychiatr Serv 2009;60:520–527. 24. Walthour A, Seymour L, Tackett R, Perri M. Assessment of changes in utilization of health-care services after implementation of a prior authorization policy for atypical antipsychotic agents. Ann Pharmacother 2010;44:809–818. 25. Lu CY, Soumerai SB, Ross-Degnan D, Zhang F, Adams AS. Unintended impacts of a Medicaid prior authorization policy on access to medications for bipolar illness. Med Care 2010;48:4–9. 26. Simeone JC, Marcoux RM, Quilliam BJ. Cost and utilization of behavioral health medications associated with rescission of an exemption for prior authorization for severe and persistent mental illness in the Vermont Medicaid Program. J Manag Care Pharm 2010;16:317–328. 27. Law MR, Lu CY, Soumerai SB, et al. Impact of two Medicaid prior-authorization policies on antihypertensive use and costs among Michigan and Indiana residents dually enrolled in Medicaid and Medicare: results of a longitudinal, population-based study. Clin Ther 2010;32:729–741. discussion 16. 28. Lu CY, Adams AS, Ross-Degnan D, et al. Association between prior authorization for medications and health service use by Medicaid patients with bipolar disorder. Psychiatr Serv 2011;62:186–193. 29. Lu CY, Law MR, Soumerai SB, et al. Impact of prior authorization on the use and costs of lipid-lowering medications among Michigan and Indiana dual enrollees in Medicaid and Medicare: results of a longitudinal, population-based study. Clin Ther 2011; 33:135–144. 30. Kloepfer KM, Helm ME, Perry TT, Hu P, Jones SM, Vargas PA. Preferred drug utilization: treating allergic rhinitis with less-sedating antihistamines. Am J Pharm Benefits 2012;4:e130–e137. 31. Constantine R, Bengtson MA, Murphy T, et al. Impact of the Florida Medicaid prior-authorization program on use of antipsychotics by children under age six. Psychiatr Serv 2012;63:1257–1260. 32. Clark RE, Baxter JD, Barton BA, Aweh G, O’Connell E, Fisher WH. The impact of prior authorization on buprenorphine dose, relapse rates, and cost for Massachusetts Medicaid beneficiaries with opioid dependence. Health Serv Res 2014;49:1964–1979. 33. Keast SL, Thompson D, Farmer K, Smith M, Nesser N, Harrison D. Impact of a prior authorization policy for montelukast on clinical outcomes for asthma and allergic rhinitis among children and adolescents in a state Medicaid program. J Manag Care Pharm 2014;20:612–621. 34. Keast SL. Impact of Prior Authorization of Montelukast on Resources and Outcomes for Asthma and Allergic Rhinitis Amongst Children and Adolescents in a State Medicaid Program. Oklahoma City, OK: University of Oklahoma; 2013.

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

10

Keast et al. / Research in Social and Administrative Pharmacy j (2015) 1–10

35. Maldonado G, Greenland S. Estimating causal effects. Int J Epidemiol 2002;31:422–429. 36. Ward A, Johnson P. Addressing confounding errors when using non-experimental, observational data to make causal claims. Synthese 2008;163: 419–432. 37. Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time

series studies in medication use research. J Clin Pharm Ther 2002;27:299–309. 38. Tooze JA, Grunwald GK, Jones RH. Analysis of repeated measures data with clumping at zero. Stat Methods Med Res 2002;11:341–355. 39. Liu L, Strawderman RL, Cowen ME, Shih Y-CT. A flexible two-part random effects model for correlated medical costs. J Health Econ 2010;29:110–123.

RSAP603_proof ■ 12 May 2015 ■ 11:09 pm

Prior authorization policies in Medicaid programs: The importance of study design and analysis on findings and outcomes from research.

U.S. State Medicaid programs for the medically indigent strive to deliver quality health care services with limited budgets. An often used cost manage...
343KB Sizes 0 Downloads 6 Views