Robert Dubois, MD

 To explore formulary restrictions on noninsulin antihyperglycemic drugs (NIADs) in Medicare Part D plans and to estimate the impact of formulary restrictions on use of NIADs among low-income subsidy (LIS) recipient enrollees with type 2 diabetes (T2D) undergoing treatment intensification.

A cohort of 2919 LIS enrollees with T2D receiving metformin monotherapy during the first quarter of 2012 who intensified treatment later in the year was tracked to assess selection of and days’ supply with sulfonylureas, dipeptidyl peptidase-4 (DPP-4) inhibitors, and other NIADs. We tested whether being enrolled in a Part D plan with significant formulary restrictions on sole-source brand name NIADs reduced the likelihood of receiving such agents and, if so, what the impact was on days of therapy with the second agent. A 2-part regression model was estimated with explanatory variables for plan-level restrictions and individual covariates.

 We found that 63% of study subjects initiated a sulfonylurea, 25% a DPP-4 inhibitor, and 12% another NIAD. Greater restrictions on DPP-4 inhibitors as a class were associated with small reductions in initiation of DPP-4 inhibitors and a concomitant increase in use of sulfonylureas, but neither effect was statistically significant. For individual DPP-4 inhibitors, step therapy requirements on sitagliptin and formulary exclusion of saxagliptin resulted in significant reductions in uptake of the specific drugs but had no significant impact on total days’ supply of antihyperglycemic therapy.

Part D formulary restrictions on sole-source brand name NIADs had little impact on patterns of treatment intensification for T2D among LIS recipients enrolled in Medicare Part D plans in 2012.

The objectives of this study were: (1) to examine formulary restrictions (exclusion, prior authorization, step therapy) imposed on noninsulin antihyperglycemic drugs by Medicare Part D plans in 2012 and (2) to test whether the presence and type of formulary restriction influenced choice of second-line agent among low-income beneficiaries with type 2 diabetes undergoing treatment intensification from metformin monotherapy. To avoid confounding of cost sharing and formulary restrictions, we restricted the study to low-income subsidy recipients who faced the same nominal co-pays regardless of Part D plan. Our main findings include:

For glucagon-like peptide-1 (GLP-1) receptor agonists, 78% of all formularies analyzed excluded at least 1 of the 2 drugs available in 2012; 69% required prior authorization and 29% required step therapy.

For dipeptidyl peptidase-4 (DPP-4) inhibitors, 22% of formularies placed no restriction on any of the 3 drugs on market in 2012, whereas 20% either excluded or required step therapy for all 3 drugs in the class.

Generic sulfonylureas were the most commonly prescribed second-line agent (63% of subjects). Formulary restrictions had no statistically significant impact on selection of and days’ supply with GLP-1 receptor agonists and DPP-4 inhibitors.

Type 2 diabetes (T2D) is a progressive disease in which most patients who begin therapy with an antihyperglycemic agent will eventually require treatment intensification to maintain glycemic control. The American Diabetes Association recommends metformin as the drug of choice for initial treatment,

 but it does not make specific recommendations about which of many medications should be used in follow-up therapy.

The most commonly used second-line agents are sulfonylureas that, like metformin, are available as inexpensive generics. However, sulfonylureas increase the risk of hypoglycemia

 and have been associated with higher rates of cardiovascular disease and death.

 In recent years, there has been a trend toward prescribing newer brand name—only antihyperglycemic agents, including dipeptidyl peptidase-4 (DPP-4) inhibitors and glucagon-like peptide-1 and -2 (GLP-1 and GLP-2) receptor agonists, resulting in sharply higher diabetes-related pharmacotherapy costs.

Health insurers and their pharmacy benefit managers (PBMs) have reacted to growing drug spending by excluding selected high-cost brand name drugs from their formularies and by subjecting covered medications to utilization management (UM) restrictions, including prior authorization (PA), step therapy (ST), and quantity limits (QLs). Nowhere is this trend more evident than in Medicare Part D plans. In 2007, Part D plans covered, on average, 87% of all drugs on their formularies, with 18% requiring some form of UM.

 By 2016, the share of covered drugs had dropped to 77%, with 42% requiring UM.

Virtually all Part D formularies routinely exclude brand name drugs for multisource antihyperglycemic drugs, including metformin, sulfonylureas, α-glucosidase inhibitors, and thiazolidinediones (TZDs).

 Plans also commonly restrict access to various sole-source noninsulin antihyperglycemic drugs (NIADs), including DPP-4 inhibitors, GLP-1 receptor agonists, and GLP-2 receptor agonists for which there are no direct generic equivalents. Although restricting access to brand name drugs with identical generic equivalents is noncontroversial, placing constraints on sole-source brand name drugs may lead to suboptimal therapeutic substitution in some patients. The extent to which this occurs in conventional diabetes treatment is unknown.

There is a substantial literature on the impact of formulary restrictions on medication use in health plans, including several general systematic reviews,

 as well as focused reviews on formulary exclusion policies,

 The general consensus is that formulary restrictions reduce utilization of the targeted drugs, but there is much less agreement on the impact of restrictions on costs and health outcomes. The literature is also limited in that few studies have analyzed formulary restrictions imposed on diabetes medications,

 and none of these have a Medicare focus. Our study was designed to fill these gaps in the literature.

We had 2 objectives. The first was to characterize formulary coverage for NIADs in the Medicare Part D market in 2012, and the second was to test whether the presence and type of formulary restrictions influenced the choice of add-on NIADs among Medicare beneficiaries with prevalent diabetes who intensified antihyperglycemic therapy in that same year. We focused on beneficiaries prescribed metformin monotherapy early in the year who then initiated an additional NIAD later in the year. Our expectation was that formulary restrictions on sole-source brand name drugs would reduce uptake of these medications while increasing the use of generic alternatives. We also expected to find that restrictions would reduce days’ supply of the restricted drug among patients who eventually did initiate the medication due to possible formulary-related delays in treatment initiation.

Data for the study were obtained from a random 5% sample of the Medicare population in 2012 from the Chronic Condition Data Warehouse (CCW) maintained by CMS. The inclusion criteria for the sample were: (1) a T2D diagnosis code in Medicare claims prior to 2012; (2) continuous LIS enrollment with coverage under Medicare Part A, Part B, and Part D throughout 2012 or up to date of death; (3) enrolled in a single stand-alone Part D prescription drug plan (PDP); (4) treated with metformin monotherapy during the first quarter of the year; (5) filled at least 1 prescription for a DPP-4 inhibitor, sulfonylurea, or other NIAD during the final 9 months of the year; and (6) refilled at least 1 prescription for metformin following the first date the second-line agent was filled. All study subjects were linked by encrypted beneficiary identification numbers to Part D plans. We restricted the sample to PDP enrollees because beneficiaries enrolled in Medicare Advantage Part D plans did not generate Part A and B claims data necessary to characterize subjects’ disease severity and comorbidities. We focused exclusively on LIS recipients because they paid the same nominal co-pays regardless of Part D plan, thereby avoiding bias associated with any correlation between formulary design and posted co-pays levied for those not receiving an LIS. The requirement that a metformin fill follow the first fill of a second NIAD ensured that the new medication represented treatment intensification rather than drug substitution.

The dependent variables of interest were utilization and days’ supply of NIAD drugs prescribed for treatment intensification in the 9 months following metformin monotherapy during the first quarter of 2012. Days’ supply for prescriptions extending into 2013 were truncated at December 31, 2012. The categories of NIAD drugs considered were: (1) sulfonylureas, (2) DPP-4 inhibitors, and (3) other NIADs (GLP-1 receptor agonists, TZDs, α-glucosidase inhibitors, amylinomimetics, and meglitinides). The sample sizes for specific drugs included in the third category (other NIADs) were too small to warrant separate analysis.

The CCW files include an annual end-of-year formulary file for each Part D sponsor that lists all covered medications with drug-specific UM indicators for PA, ST, and QLs. Using National Drug Codes in the CCW formulary file, we matched each plan’s drug list to the 2012 edition of the FirstData Bank drug dictionary to identify all off-formulary NIADs. We then created a set of binary variables indicating whether the drugs in each pharmacologic class were excluded or subject to PA or ST (we did not investigate QLs, as these policies typically place no limit on refills). Restrictiveness measures were computed for the 2 brand name—only classes of DPP-4 inhibitors and GLP-1 receptor agonists available during the study timeframe.

We categorized formulary restrictiveness as potentially meaningful at the drug level by type of restriction (exclusion, PA, or ST) and whether all forms of the drug were subject to some type of restriction. If at least 1 form of the drug was available without restriction, we considered the drug not meaningfully restricted. We used the term potentially meaningful because we had no a priori way to assess prescriber attitudes toward therapeutic substitution of antihyperglycemic agents. We did not consider restrictions on brand name products in multisource classes as meeting the threshold of meaningfulness, as most clinicians consider generics to be equivalent to brands with identical formulations. Similarly, among DPP-4 inhibitors, if Janumet (a combination of the DPP-4 inhibitor sitagliptin and metformin) was excluded from the formulary but Januvia (sitagliptin) and generic metformin were unrestricted, then this was not considered a meaningful restriction. However, if all forms of sitagliptin were restricted, we considered that a potentially meaningful restriction. Similar classification rules were applied to saxagliptin, linagliptin, and the 2 GLP-1 receptor agonists on the market in 2012 (exenatide and liraglutide). If all drugs in a pharmacologic class were restricted, we considered the entire class to be restricted.

In the analytic models described below, we included a set of categorical covariates capturing individual-level variation in characteristics that could be correlated both with our formulary restrictiveness measures and with patterns of beneficiary drug utilization. These variables included demographic characteristics (age ranges, race, and sex); measures of diabetes severity (uncontrolled diabetes, short- and long-term diabetes complications, and hypoglycemia); diabetes management procedures (fasting plasma test, glycated hemoglobin test, low-density lipoprotein cholesterol test, diabetes management class, diabetic eye exam, and influenza vaccination); comorbidities common among patients with diabetes (cancer, chronic kidney disease, heart failure, and chronic obstructive pulmonary disease); and contacts with the health system (inpatient admissions and physician visits). All characteristics associated with disease states and medical utilization were captured during the 3-month baseline period during which each study subject used metformin only. 

 lists all covariates, identifies how each variable was measured, and provides source notes for diagnostic codes.

We first produced descriptive statistics for individual beneficiary characteristics by class of second-line therapy taken. Next, we characterized restrictions on DPP-4 inhibitors and GLP-1 receptor agonists across all Part D formularies applicable to the study sample in 2012, calculating the percent of beneficiaries and mean days’ supply among users subject to each type of restriction. In each case, we used 

 tests to determine whether utilization rates differed significantly between formularies with and without specific restrictions. Finally, we estimated a 2-part regression model to obtain conditional effects of formulary restrictions on the probability of using DPP-4 inhibitors or other NIADs versus sulfonylureas (part 1) and mean days’ supply of drugs restricted to users of sulfonylureas, DPP-4 inhibitors, and other NIADs (part 2). For the part 1 equation, we used a multinomial logistic regression comparing characteristics of subjects initiating DPP-4 inhibitors and other NIADs with those of subjects initiating sulfonylureas. For the second part of the model, we estimated 3 ordinary least squares regressions on days’ supply restricted to users of DPP-4 inhibitors, sulfonylureas, and other NIADs, respectively. Variables for number of DDP-4 inhibitors with meaningful restrictions were included in all models to test for the effect of formulary restrictiveness on drug utilization patterns among study subjects (formulary restrictions on the GLP-1 receptor agonists had no measurable impact on use of other NIADs in preliminary analyses and thus were dropped in the final models).

Table 1 presents characteristics of the 2919 LIS recipients who met the study inclusion criteria, arrayed by choice of second-line antihyperglycemic agent following metformin monotherapy. A total of 63% (1848) initiated a sulfonylurea, 24.5% a DPP-4 inhibitor (716), and 12.2% another NIAD (355). The 3 groups were broadly similar with some notable exceptions. Specifically, a much higher proportion (53.2%) of those using other NIADs were younger than 65 years (ie, they were Medicare beneficiaries due to disability rather than age) compared with between 37% and 38% for the other 2 cohorts. Users of other NIADs also had the highest rates of uncontrolled diabetes and long-term diabetes complications.

Our analysis of 153 Part D formularies found no meaningful restrictions on metformin, sulfonylureas, α-glucosidase inhibitors, or TZDs (results not shown). Brand name drugs in these classes were almost universally excluded, but because there were generic versions available without restriction, we did not consider such exclusions to be meaningfully restrictive.

Among GLP-1 receptor agonists, either exanatide or liraglutide was off formulary in 120 formularies, subject to PA in 105 formularies, and subject to ST in 44 formularies. However, because fewer than 3% of all study subjects initiated 1 of these medications as second-line therapy, the effects on overall NIAD intensification patterns were minimal and none of the bivariate tests of GLP-1 restrictions on uptake and days’ supply was statistically significant (results not shown).

All formulary restrictions on DPP-4 inhibitors that were judged to be meaningful using the criteria specified earlier are presented in the left-hand column of 

] that outlines the entire set of DPP-4 inhibitor restrictions, meaningful and not.) The top panel of Table 2 shows the extent of formulary restrictions across all DPP-4 inhibitors together with statistics on initiation and days’ supply for DPP-4 inhibitors, sulfonylureas, and other NIADs. Thirty-four formularies (436 study subjects) included at least 1 form of all 3 DPP-4 inhibitors with no restrictions. Sixty-one formularies (1811 subjects) restricted 1 DPP-4 inhibitor, followed by 27 restricting 2 DPP-4 inhibitors (215 subjects) and 31 restricting all 3 DPP-4 inhibitors (457 subjects). Consistent with our hypothesis, the initiation rate for DPP-4 inhibitors was generally lower among formularies restricting 2 or 3 DPP-4 inhibitors. Likewise, initiation of sulfonylureas was higher in formularies with the most restrictions on DPP-4 inhibitors, but neither of these results was statistically significant. DPP-4 inhibitor restrictions had no discernable impact on uptake of other NIADs or days’ supply of any second-line agent.

The next 3 panels of Table 2 present descriptive results on initiation patterns of second-line agents as a function of formulary restrictions on individual DPP-4 inhibitors. Most formularies (122) placed no meaningful restriction on sitagliptin, whereas 29 imposed ST and 2 required PA. Initiation of sitagliptin was significantly higher (

<.05) in unrestricted formularies (19.7%) compared with those imposing ST (15.6%), with correspondingly fewer subjects initiating a sulfonylurea in plans with no sitagliptin restrictions (62.2%) compared with 67.8% (

<.05) in plans requiring ST for sitagliptin. Similar patterns were seen with respect to restrictions on saxagliptin and linagliptin, albeit the number of subjects initiating these drugs was much lower than in the case of sitagliptin. None of the drug-specific restrictions had any measurable effect on days’ supply among users.

 presents findings from the multinomial regression. As in the unadjusted results, restrictions of 1 or more DPP-4 inhibitor were associated with lower initiation rates for DPP-4 inhibitors and also lower rates for other NIADs, but in neither case were the results statistically significant. The second part of the 2-part models (

) also failed to identify any significant effects of DPP-4 inhibitor formulary restrictions on days’ supply among users of DPP-4 inhibitors, sulfonylureas, and other NIADs.

Our evaluation of formulary restrictions on antihyperglycemic drugs demonstrated the breadth of restrictions imposed by Part D plans in 2012. The most common type of restriction was formulary exclusion. Part D sponsors excluded virtually all brand name medications in multisource classes (metformin, sulfonylureas, α-glucosidase inhibitors, and TZDs). We did not consider these to be meaningful restrictions because generic equivalents were available without restriction in every case. Among brand name—only products, 119 of 153 formularies placed a meaningful restriction on 1 or more DPP-4 inhibitors. Formulary exclusion was the most common restriction, followed by ST. Very few plans required PA for DPP-4 inhibitors, but the practice was more common for GLP-1 receptor agonists.

Our analysis provides mixed evidence that formulary restrictions influence initiation of second-line NIADs. As hypothesized, restricting more DPP-4 inhibitors was associated with lower initiation rates for DPP-4 inhibitors and higher initiation levels with sulfonylureas, but the results were not statistically significant. At the individual drug level, ST requirements on sitagliptin reduced uptake and increased the likelihood of sulfonylurea use. Likewise, excluding saxagliptin significantly reduced uptake among the 26 formularies excluding all forms of this drug. However, the fact that these restrictions did not wholly eliminate access to either sitagliptin or saxagliptin suggests that prescribers who favored a particular drug were able to override barriers to access.

Taken together, these findings suggest that Part D plan formulary designs had little net impact on how prescribers intensified antihyperglycemic therapy for Medicare beneficiaries with diabetes in 2012. The dominant choice among prescribers was a generic sulfonylurea, followed by sitagliptin. The newer DPP-4 inhibitors, saxagliptin and linagliptin, were prescribed infrequently. We estimate that the reduction in prescribing of saxagliptin because of formulary exclusion was just 2 percentage points. These patterns would thus appear to have less to do with formulary restrictions than with clinician prescribing habits. Sitagliptin was the first DPP-4 inhibitor to be approved by the FDA, and later entrants had difficulty gaining market share. In our study, GLP-1 receptor agonists were the least frequently prescribed second-line brand NIADs, but they had too few initiators to reliably estimate formulary effects.

It is difficult to compare these findings with those of previous studies given the slim literature on formulary design effects in diabetes treatment. Williams et al

 in a 2012 study found that ST restrictions on second-line NAIDs had no effect on reducing diabetes-related costs compared with unrestricted use of fixed-dose combination products. By contrast, Gamble et al

 found that PA requirements significantly reduced prescribing rates for TZDs and sitagliptin, respectively. The larger literature also suggests that ST

 may have less predictable outcomes compared with PA,

 but it is worth noting that during our study period few Part D sponsors required PA for any brand name NIADs.

A review of second-line diabetes medications conducted by an expert panel of editors for 

 determined that all of the medications we analyzed are clinically effective in most patients, although the risk-to-benefit profiles for newer NIADs are less well understood than those of older agents.

 The panel concluded that choosing a specific agent for treatment intensification is less important than ensuring that such intensification is implemented as soon as clinically indicated.

 From this perspective, our finding that formulary restrictions had little net impact on diabetes treatment intensification patterns should help assuage potential policy maker concerns that Part D sponsor formulary decisions reduce patient access to needed medications—at least with respect to treatment of patients with T2D.

Our focus on LIS recipients represents both a study strength and a limitation. On the one hand, we can rule out any confounding of formulary impacts associated with cost sharing because all LIS recipients face the same nominal generic and brand co-pays regardless of formulary configuration or choice of drug. On the other hand, LIS recipients tend to cluster in basic benchmark Part D plans in which they face no monthly premiums. Non-LIS beneficiaries tend to favor enhanced plans with higher premiums and broader formulary coverage. Thus, our results cannot generalize to these beneficiaries.

Other limitations of our study design warrant mention. First, we built the study around treatment intensification in diabetes because this is a point at which many patients with diabetes first face the issue of brand name—only versus generic treatment options. Our thinking was that formulary effects would be most evident in this situation. However, the sample selection decisions we made to reduce potential confounding (limiting to LIS recipients, requiring a set period on metformin monotherapy, restricting to patients adding a second-line drug) also reduced the available sample size, which could have contributed to the lack of statistically significant findings. Our focus on LIS recipients also limits generalizability of study findings to Medicare beneficiaries who confront both formulary restrictions and cost sharing.

Second, we were blind to the administrative machinery surrounding each type of formulary restriction. Anecdotal conversations with PBMs and pharmacy and therapeutics committee members suggest that there is considerable market heterogeneity in the application and override provisions relating to all of the formulary restrictions we investigated. By contrast, the early literature on PA requirements in state Medicaid programs suggests that some state authorities aggressively used PA as a method for restricting drug access,

 but the increased use of electronic health records may substantially alleviate the burden of PA on medical practices and thus reduce the disincentive to prescribe drugs on PA lists.

A third methodological issue is that formulary influence may simply be diluted by the plethora of antihyperglycemic choices available to clinicians: More choices mean that any single restriction is likely to have less of an impact than it would in therapeutic classes with fewer alternatives. Finally, the most important caveat of all is that our findings pertain to just 1 therapeutic class used to treat a single disease within a sample of Medicare Part D plans.

Part D formulary restrictions on sole-source brand name NIADs had little impact on patterns of treatment intensification for T2D for LIS recipients enrolled in Medicare Part D plans in 2012. More research is necessary to understand how formulary restrictions and UM tools work in other therapeutic areas and in other organizational settings.

 University of Maryland Baltimore (BCS, JFS, J-DR, CC, XS), Baltimore, MD; PhRMA (PR), Washington, DC; National Pharmaceutical Council (MC, RD), Washington, DC.

 Drs Stuart, Slejko, and Cooke received a grant from the National Pharmaceutical Council that funded this research. Mr Ciarametaro and Dr Dubois are employees of the National Pharmaceutical Council, an industry-funded health policy research group that is not involved in lobbying or advocacy. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (BCS, JFS, CC, XS, PR, MC, RD); acquisition of data (BCS, XS); analysis and interpretation of data (BCS, JFS, J-DR, CC, XS, PR, MC, RD); drafting of the manuscript (BCS, JFS, MC, RD); critical revision of the manuscript for important intellectual content (BCS, J-DR, CC, PR, MC, RD); statistical analysis (BCS, JFS); obtaining funding (BCS); administrative, technical, or logistic support (J-DR); and supervision (JFS).

 Bruce C. Stuart, PhD, Department of Pharmaceutical Health Services Research, University of Maryland Baltimore, 220 Arch St, 12th Fl, Baltimore, MD 21201. Email: bstuart@rx.umaryland.edu.

1. American Diabetes Association. Standards of medical care in diabetes—2013. 

 2013;36(suppl 1):S11-S66. doi: 10.2337/dc13-S011.

Articles

Formulary restrictions on brand name noninsulin antihyperglycemic drugs have little impact on treatment intensification patterns among low-income patients with diabetes in Medicare Part D.

Impact of Formulary Restrictions on Medication Intensification in Diabetes Treatment

Unprecedented funding for comparative effectiveness research (CER) to help provide better evidence for decision making as a way to lower costs and improve quality is under way. Yet how research findings are adopted and applied will impact the nation’s return on this investment. We examine the relationship between the publication of findings from 4 seminal CER trials, the release of subsequent clinical practice guidelines (CPGs), and utilization trends for associated surgical interventions, diagnostic interventions, or medications.

Using a large national administrative claims database, we examined time series utilization trends before and after publication of findings from 4 CER trials published within the last decade.

We found no clear pattern of utilization in the first 4 quarters after publication. However, we found that results for 2 of the studies were in concert with the release of CPGs and publication of study results. The trend in intensive statin therapy rose rapidly starting at the end of 2007, while the trend in standard therapy remained relatively constant (PROVE-IT). And, 9 months after trial publication, breast magnetic resolution imaging (MRI) utilization rates rose 43.2%, from 0.033 to 0.048 per 100 enrollees (Mammography With MRI).

Our analysis of 4 case studies supports the call others have made to translate and disseminate CER findings to improve application of research findings to clinical practice and the need for continued development and dissemination of CPGs that serve to synthesize research findings and guide practitioners in clinical decision making. Further research is needed to determine whether these findings apply to different medical topics.

Time series utilization trends before and after publication of findings from 4 CER trials published within the last decade revealed no clear pattern of utilization in the first 4 quarters after publication. Results for 2 of the studies (PROVE-IT, Mammography With MRI) were in concert with the release of clinical practice guidelines (CPGs) and publication of study results.

		Our findings support a continued effort to translate and disseminate CER results to improve application of research findings to clinical practice.

		Continued development and dissemination of CPGs to synthesize research findings and guide practitioners in clinical decision making is necessary.

With the Affordable Care Act, public funding for comparative effectiveness research (CER) expands beyond the Agency for Healthcare Research and Quality (AHRQ) and the National Institutes of Health to include the Patient-Centered Outcomes Research Institute (PCORI).

 According to the Institute of Medicine of the National Academies, “The purpose of CER is to assist consumers, clinicians, purchasers, and policy makers to make informed decisions that will improve healthcare at both the individual and population levels.”

The return on this investment in CER will be measured by if, and how quickly, results from the research are used in decision making and translated into clinical practice.

One of the most notable examples of the relationship between findings and practice is the Estrogen Plus Progestin trial, a component of the Women’s Health Initiative.

 The trial was stopped in May 2002 after investigators found that the associated health risks of combination hormone therapy among postmenopausal women outweighed the benefits. Almost immediately following publication, rates of hormone therapy in postmenopausal women were shown to drop precipitously, from 90 million prescriptions to 57 million—a 36.7% decrease.

The Estrogen Plus Progestin trial is an example of a study where the evidence resulted in near-term changes in clinical practice. However, beyond a few notable studies,

researchers have found that it can take several years for study findings to translate to practice.

 Additionally, fewer than 1 study in 1000 is reported by the mainstream media.

 Historically, one way that research findings have impacted medical practice has been through the development, dissemination, and use of clinical practice guidelines (CPGs).

In this study, we used a large national administrative database to examine real-world utilization trends before and after publication of CER findings (and release of relevant CPGs) from 4 high-profile CER studies published within the last decade. We analyzed changes in utilization rates of procedures and treatments associated with widely communicated CER evidence and CPGs for up to 6 years following the evidence of the publication. Our approach was not normative or prescriptive, but was intended to examine historical trends and resulting changes in practice patterns within a large sample of enrollees with employersponsored health insurance and to discuss the implications of our findings for those seeking to increase the translation of research findings into practice.

To distinguish case studies and conditions for our areas of focus, we performed an environmental scan via a literature review. We queried MEDLINE/PubMed,

Congressional Budget Office publications,

 studies to find CER evidence reported between January 1, 2000, and December 31, 2009.

Studies (including randomized controlled trials, metaanalyses, or observational studies) which were highly cited in high-impact medical journals (

) that compared 2 or more treatment options or reviewed the safety of a drug were selected. Based on these criteria, 23 studies were selected, characteristics of each study were abstracted (eg, type of comparators, primary aim as safety vs efficacy, study design), and then each was reviewed by the study team to determine which should be included in the quantitative analysis (

). The main criteria for the selected case studies were (1) high-profile study of a medication, surgical, or diagnostic intervention (published in a high-impact journal as ranked by ISI Web of Knowledge)

; and (2) widely cited results (ISI Web of Knowledge

 and Google Scholar citations). We also focused on quality criteria that included, as reviewed by the study team: (1) study design (adequate sample size with an experimental design); (2) clear findings (lack of ambiguity in the implications); and (3) findings that were not reversed or contradicted by subsequent evidence. Cost-effectiveness studies were excluded.

Of these, the study team selected 4 case studies that represented varying types of trials: 1 study compared medical therapies (PROVE-IT TIMI 22), 2 compared surgical versus nonsurgical treatments (COURAGE and SPORT) and 1 compared diagnostic screening procedures (called Mammography With MRI here for simplicity; see reference 20 for actual title).

We also examined the first release of related CPGs, following the publication of trial results. To determine when guidelines might have started to influence uptake or discontinuation of a specific practice related to the clinical trial, we conducted an environmental scan via a literature review. The criteria for the scan limited results to articles that focused on clinical guidelines and that specifically cited 1 of the 4 seminal studies in relationship to guideline development. We queried Google Scholar for publications following the release of the trial(s) through the end of our study period in 2010. We then eliminated nonguidelines. We also conducted a keyword search of guidelines published by the AHRQ National Guidelines Clearinghouse for the same time period. From the combined list we retrieved the first clinical guideline (from search results), corresponding to each trial (

This analysis was based on data contained in the Truven Health MarketScan Commercial Claims and Encounters Database and the MarketScan Medicare Supplemental and Coordination of Benefits Database for the period of January 1, 2003, to June 30, 2010. The MarketScan Database includes the enrollment, inpatient, outpatient, and outpatient pharmacy claims experience of tens of millions of individuals across the nation with employer-sponsored insurance or employer-sponsored Medicare Supplemental insurance. Health insurance was offered through a variety of capitated and fee-for-service health plan types, and prescription drug coverage was offered in conjunction with the medical benefit. Its sample size is large enough to allow creation of a nationally representative data sample of Americans with employer-provided health insurance. The enrollment characteristics in MarketScan are largely similar to nationally representative data for employer-sponsored insurance in the Medical Expenditure Panel Survey (MEPS), although a higher percentage of MarketScan enrollees reside in the South census region. The data cover all 50 states and the District of Columbia. Used primarily for research, these databases are fully Health Insurance Portability and Accountability Act compliant.

To study trends in the utilization rates for each case study, we calculated the quarterly utilization rates of each procedure (by calendar quarter). First, we selected the continuous cohort of firms contributing data to MarketScan each year from 2003 through 2010 Q1, representing approximately 4 million enrollees annually (eAppendix C), so trends would not be impacted by firms entering and leaving the sample. Within these firms, each enrollee selected for inclusion in the study had medical and pharmacy claims data available, was 18 years or older, was continuously enrolled in the 4 calendar quarters prior to the quarter of interest, and was not pregnant (no diagnosis of pregnancy).

Second, within this cohort of enrollees, we replicated (as was possible using claims data) the inclusion and exclusion criteria used in each clinical trial (based upon publication or online trial-specific materials) for patient selection (See the Table for summarized criteria and 

 for detailed criteria and codes). To determine these sets of criteria, a nosologist (certified clinical coding expert) provided clinical codes (

International Classification of Diseases, Ninth Revision, Clinical Modification

] procedure codes, and Current Procedural Terminology, 4th Edition [CPT-4] procedure codes) for each of the inclusion and exclusion criteria and the procedures listed in the clinical trial. As part of an iterative process, these codes were reviewed by a clinician expert in clinical coding and the resulting list was reviewed a second time by the nosologist. Through a consensus-driven process, any differences between the clinician and the nosologist were resolved in consultation with the project manager, and by 2 other clinicians. A similar process was employed for creating a list of pharmaceutical codes, incorporating a review of National Drug Codes by a licensed pharmacist.

The study clinicians relaxed some of the criteria (see 

 for a description of how these criteria were relaxed) designated in the trial to focus on patients who were similar to those targeted by very specific study criteria for 2 reasons. First, clinical coding in administrative data sets can lack precision.

 Second, we hypothesized that these results would have spillover effects on clinical practice beyond patients meeting the restrictive clinical inclusion criteria used in the respective clinical trial. Therefore, we focused on patients highly similar to those included in the trial. For example, for the Mammography With MRI study, we did not require women to have a diagnosis on a claim indicating that they carried the BRCA1 or BRCA2 mutation, because test results are infrequently available in claims analyses.

 For the SPORT case study, we did not exclude enrollees if they had a previous lumbar surgery. The relaxed criteria were then applied to the data set to find the set of enrollees in each calendar quarter meeting the patient selection criteria (the denominator). For the pre-period for each study, we required patients to be continuously enrolled at least 1 year prior to the quarter, to check for a number of inclusion and exclusion criteria. Claims incurred during this year were used to evaluate inclusion and exclusion criteria (eAppendix D). Once meeting study requirements, patients were followed until the end of their continuous enrollment, or until evidence appeared that they met the exclusion criteria, or until completion of 2 years of follow-up (except in the Mammography With MRI study, where patients were followed through the end of their continuous enrollment).

Finally, a list of the treatments or procedures of interest (eg, breast magnetic resonance imaging [MRI]) was created (numerator of each measure). Within the enrollees meeting the selection criteria, enrollees receiving each treatment or procedure were flagged in the database and the number of occurrences of the treatment or procedure of interest (the numerator) was recorded (note: eAppendix D contains the clinical criteria used in each case study). Using this information, we calculated aggregate quarterly trends in utilization rates for each treatment or procedure.

We present the descriptive findings in 2 formats. First, we display graphical trends before and after the publication of the CER evidence and guideline release. We show the timing of the first national CPG that referenced the study. Second, we display utilization rates in tabular form at 3- and 6-month intervals and calculate the percent change from the prior value at each time interval. As a separate analysis, we replicated the study results using a stricter set of criteria using claims data, described in the clinical trial (without relaxation). These results represent a smaller group of enrollees (

A consistent cohort of 66 data contributors (large firms and insurers) contributing data throughout the entire study period was selected for analysis, representing approximately 7 million enrollees each year. Between 2003 and 2010, the number of patients meeting the clinical trial inclusion criteria ranged from 86,080 for PROVE-IT to 5,159,253 for Mammography With MRI. Patient characteristics of these subjects are detailed in eAppendix C. All reported results reflect the relaxed version of the original study criteria.

The rest of this paper summarizes the main findings from each case study and the temporal trends in each study, in chronological order of publication. Figures 1 through 4 show the quarterly utilization trends for each case study.

 investigated whether standard therapy with 40 mg of pravastatin daily or intensive therapy with 80 mg of atorvastatin daily decreased a composite outcome, including all-cause mortality, myocardial infarction (MI), unstable angina, revascularization, and stroke among patients who had been recently hospitalized for acute coronary syndrome. Patients who obtained intensive lipid-lowering statin therapy had a lower risk of death or cardiovascular events than those on a standard statin regimen. The literature search yielded 24 CPGs published during the review period (conclusion of the study through 2010). The first guideline was published in 2004 by the National Cholesterol Education Program.

 The CPG called for more aggressive lowering of low-density lipoprotein-cholesterol, and was published in response to 5 new statin trials, including PROVE-IT. Based upon these results, we hypothesized that after the release of the PROVE-IT trial results and publication of the CPG, prescription utilization rates (days supplied per person per quarter) would increase for patients on intensive statin therapy and decrease with standard statin therapy.

We found the use of intensive statin therapy increased; however, the increase in utilization started prior to the publication of PROVE-IT in 2004. After the initial rise, the trend in intensive therapy flattened for the next 3 years (after publication of results and release of the subsequent CPG) and rose rapidly starting at the end of 2007. In contrast, the trend in standard therapy remained relatively constant prior to publication and through the 3 years after publication of results and release of the CPG. Utilization of standard therapy declined starting at the end of 2007 and then leveled off after early 2009 (Figure 1).

evaluated the sensitivity and specificity of 4 breast cancer surveillance methods: mammography, ultrasound, magnetic resonance imaging (MRI), and a clinical breast exam among carriers of BRCA1 or BRCA2 mutations. The sensitivity of MRI to detect breast cancer was highest, with 77% of all cancers detected found by MRI compared with 36% by mammography, 33% by ultrasound, and 9.1% by clinical breast exam.

The literature search yielded 15 CPGs published during the review period. The first CPG was published in 2007 by the American Cancer Society.

 This updated CPG added annual MRI screening to mammography for women with greater than or equal to a 20% to 25% lifetime risk. Based on the results of this trial, we hypothesized that MRI utilization rates among women would increase.

We found that mammography rates (number of procedures per 100 enrollees) remained relatively consistent (with seasonal variation) and breast ultrasound utilization rates rose steadily over time. Compared with the other imaging technologies of mammography and ultrasound, there was a steady increase in MRI utilization prior to and following publication in 2004 through the end of 2006, and then a sharper increase after the CPG was released in 2007 Q2. In mid-2005, 9 months after publication, breast MRI utilization rates rose 43.2% from 0.033 to 0.048 per 100 enrollees. Despite the increase in utilization, the absolute rate of MRI utilization was far less than the utilization of ultrasound or mammography (0.048 vs 1.26 and 9.91 per 100 enrollees at 9 months after publication of the results and 9 months after release of the CPG) (Figure 2).

The Spine Patient Outcomes Research Trial (SPORT)

 was designed to assess whether a standard open diskectomy was more effective than non-operative treatment at decreasing pain and improving physical function among patients with intervertebral disk herniation. Additional outcome measures were patient self-reported improvement, work status, symptom satisfaction, and care satisfaction. Intent-to-treat analysis did not find statistically significant differences in pain reduction or physical function, but found differences in measures of sciatica severity and self-reported improvement. However, nonadherence to the assigned treatment was high. After 2 years of follow- up, 60% of the patients randomized to the surgical intervention and 45% of the patients randomized to usual care had surgery. Larger beneficial effects from surgery were seen for the as-treated analysis.

The literature search yielded 4 guidelines citing trial evidence. In 2007, the College of Physicians and the American Pain Society published the first CPG.

 The SPORT trial was one of many back pain studies cited in this guideline release.

Based on the findings of this trial, we hypothesized that standard diskectomy utilization rates (percent of eligible enrollees with the procedure) would increase after publication of the results in 2006 Q4 and release of the CPG in 2007 Q2. However, we found that standard diskectomy rates remained relatively consistent (remaining around 1 diskectomy per 100 enrollees) after release of the SPORT trial results (Figure 3).

The authors found no consistent pattern in the concordance between CER evidence and subsequent utilization patterns.

Robert Dubois, MD

Articles

Impact of Formulary Restrictions on Medication Intensification in Diabetes Treatment

Real-World Impact of Comparative Effectiveness Research Findings on Clinical Practice