Meeting Summary. May 12-13, 2011 Hyatt Regency Baltimore, Maryland

Transcription

1 LARGE SIMPLE TRIALS FOR COMPARATIVE EFFECTIVENESS RESEARCH IN CHRONIC DISEASE WORKSHOP: MEDICARE DATA UNITED WITH SIMPLE CLINICAL EXPANDED NETWORK (MUSCLE) Meeting Summary May 12-13, 2011 Hyatt Regency Baltimore, Maryland

2 Large-scale simple clinical trial (LST) designs, also known as pragmatic clinical trials, involving simple inclusion criteria, randomization, and long-term surveillance with minimal monitoring burden are rare in the U.S. Coupling this important study design with readily available linked, longitudinal, administrative and pharmaceutical databases has high potential to provide important information about real-world comparative effectiveness in chronic disorders. Linking LSTs with Medicare creates a study design that is much more cost-effective than previously possible using more traditional cohort designs that require extensive follow-up via surveys or visits in a clinical setting. The large sample sizes of LSTs, leading to substantial statistical power and generalizability, necessitate the participation of community-based physicians. The recent availability of linked administrative and pharmacy data from the Medicare program, now including Part D prescription drug data, provide an ideal opportunity to conduct studies of chronic disorders that have a significant public health burden for older Americans. To address this public health challenge, the Medicare data United with Simple CLinical trials Expanded (MUSCLE) Network was created. MUSCLE is a collaborative network of networks that will provide the research infrastructure for conducting LSTs on comparative effectiveness of therapeutics in the treatment and management of chronic diseases using linked Medicare data. Osteoporosis therapeutics is an important example area where the options for prevention and treatment continue to increase rapidly but very little is known about the comparative effectiveness of these drugs and emerging biologics. A limited number of head-to-head randomized controlled trials (RCTs) of osteoporosis agents have focused only on bone surrogate outcomes but not on fracture outcomes. Further, safety considerations, particularly emerging concerns about longer-term bisphosphonate use have been inadequately studied in randomized designs with large enough numbers or with long enough follow-up to provide adequate risk estimates. An LST of osteoporosis agents is an ideal example of this type of research and a test for the MUSCLE network On May 12 and 13 th, 2011, a limited-attendance workshop on Large Simple Trials (LSTs) for Comparative Effectiveness Research in Osteoporosis was organized and held by the University of Alabama at Birmingham and the Center for Medical Technology Policy. The workshop convened a diverse group of approximately 60 stakeholders representing researchers, payers, patients/consumers, industry, government and policymakers. The goal of the workshop was to discuss the methods and strategies for conducting LSTs for Comparative Effectiveness Research on Osteoporosis and to provide feedback on the feasibility and challenges of developing a network to provide infrastructure for the successful implementation of this type of research. Among the topics discussed were: 1) Statistical Methods in LSTs / Accounting for Treatment Heterogeneity, 2) Participant Selection: Designing Inclusion/Exclusion Criteria, 3) Study Arms: Selecting Appropriate Comparators and Intervention Parameters, 4) Outcomes: Selecting Appropriate Primary and Secondary Outcome Measures and Defining Intensity of Follow-up, 5) Data Management and Medicare Data Linkage, 6)Practical Considerations for Conducting LSTs in Practice-Based Research Networks. Using a formal group process akin to the OMERACT meeting 1, 1 Tugwell P, Boers M, Brooks P, Simon L, Strand V, Iderda L. OMERACT: An international initiative to improve outcome measurement in rheumatology. Trials. 2007;8(38). PMIC. 2

3 working groups refined the questions followed by a plenary group discussion, using audience response to vote on propositions. The following is a summary of the key issues and messages discussed relating to each topic. Statistical Methods in LSTs As a large simple trial, some of the major statistical and design challenges faced by the ATLast trial include how to control for treatment heterogeneity, the extent to which subsequent care would be defined by the protocol and how to account for switching among treatment arms, ways to account for loss to follow up, and statistical issues with including multiple and composite outcomes for osteoporosis. The workshop also addressed more practical issues such as assuming one study arm is an IV drug, how might researchers address the practical issue that some study sites will not have an infusion center? What are the special considerations or responsibilities of a DSMB in a large simple trial? There was no agreement on how the primary analysis should address the issues of loss to follow up, death, switching health plans and/or switching medications. The workgroup suggested a non-inferiority trial, under the assumption that the trial design should be most relevant to policymakers. Following this design, it was suggested to make the primary analysis as-treated with windows, given the large amount of switching that would likely occur among the treatment groups. A window of one year was suggested to follow those who discontinue treatment, given the prolonged effects of the medications. The secondary analysis would be intent to treat. However, the larger group voted in favor of an intent to treat primary analysis, mainly with the purpose of determining happens with different treatment regimens (a completely naturalistic pragmatic clinical trial). It was recommended that all fracture rates and mortality be combined (hip fracture, non-hip/nonvertebral fracture, vertebral fracture, and death). Also, since some study sites may not have infusion centers, it was recommended that sites should have reasonable access to an infusion center, where reasonable access is defined as the physician having had a history of referring patients for infusion. To account for treatment heterogeneity, analysis should take into account variables such as prior fracture/no prior fracture, prior treatment for osteoporosis, smoking and other co-morbidities, and socioeconomic status. Lastly, the recommended role of a DSMB was to collect information on patients who refuse to be enrolled at baseline. Participant Selection: Designing Inclusion/Exclusion Criteria A major goal of a large simple trial is to produce results that are generalizable to the population most likely to use therapies in usual practice. A parsimonious list of inclusion/exclusion criteria is necessary to ensure participants are not exposed to known undue risks and that their underlying condition would not unduly affect trial outcomes. During the workgroup discussions, three main issues were addressed: How should eligible participants be identified? Are the inclusion/exclusion criteria proposed for this osteoporosis LST appropriate? How do the inclusion/exclusion criteria influence the practical nature of the study? How might researchers compare LST participants vs. non-participants to enhance generalizability? From the perspective of a real world study, requiring dual x-ray absorptiometry for confirmatory diagnosis of osteoporosis would limit the number of practices that would be eligible for inclusion; 3

4 therefore, using the WHO s self-reported Fracture Risk Assessment (FRAX) score was considered more acceptable for study entry. Those with osteoporosis at baseline would be included in the study. Given the reliance on Medicare data, those 65 and over would be included, without an upper age limit. Men who fell under these criteria would also be included due to their underrepresentation in trials. Baseline laboratory assessment would be mandated (current lab results would be acceptable), followed up by good clinical practice based on the specific treatments. Lastly, it was recommended that nonparticipants should be followed using Medicare data as well as some primary data upfront to help compare them with LST participants for outcomes and variables such as hip fracture rates, treatment/drug utilization, and age. Study Arms: Selecting Appropriate Comparators and Intervention Parameters The potential treatment arms to be included in an osteoporosis large simple trial (LST) were discussed at the meeting. Among the questions addressed were: What 2 or 3 osteoporosis treatments are most important to include in an Osteoporosis LST? Would potential need for an IND (Investigational New Drug) affect treatment choices? How long should study drugs be provided (keeping in mind variations in dosing intervals)? What is the appropriate role of the pharmaceutical industry and/or Medicare in providing paying for the drugs? How much deviation from the treatment protocol is considered reasonable? Should all study participants receive calcium and Vitamin D supplements? If so, what doses? The recommendations for potential treatment arms were: Alendronate (weekly oral) vs. Zolendronate (single, one-time infusion), or Alendronate (weekly oral) vs. Zolendronate (yearly infusion), or Alendronate (weekly oral) vs. Zolendronate (yearly infusion) vs. Denosumab (twice yearly subcutaneous injection), or Alendronate (weekly oral) vs. Zolendronate (yearly infusion) vs. Raloxifene (daily oral). It was agreed that IND would be required for once yearly Zolendronate as well as Denosumab. IND exemption should be requested for Alendronate and yearly Zolendronate and Raloxifene. The exemption request will allow the agency to review and make the final decision. Everyone echoed the sentiment that the study drugs should be paid for under the trial. The consensus was that the pharmaceutical industry can participate in the overall planning and provide funding, but should not be involved in the trial conduct and analysis. It was agreed that an intention to treat analysis should be conducted. Once randomized, all patients should be included in the primary analysis, regardless of what they receive. Also, some secondary analysis based on exposure was recommended. It was agreed that the physicians and patients should receive the best public health recommendations on supplementation. The IOM recommendations provide a good standard. Outcomes: Selecting Appropriate Primary and Secondary Outcome Measures and Defining Intensity of Follow-up Large simple trials aim to provide data that are more meaningful to making appropriate treatment decisions at point of care. For this reason, a major consideration in designing the ATLast trial is, What are the most relevant outcome measures for patients, clinicians and payers for making appropriate treatment and policy decisions? Participants also addressed: What are the best sources of data for 4

5 those outcomes? What are the most important confounders to measure? and What is the most appropriate intensity of follow up? The small work group proposed the primary outcomes should be a composite of the four components of WHO s Fracture Risk Assessment Tool (FRAX), including fractures of the: Hip; Vertebrae (clinical) Radius ulna; and Humerus. The proposed secondary outcome was all clinical fractures less fingers, skull, and toes. Recommended safety measures included: all-cause hospitalization, serious infections (pertinent to Denosumab), atypical fractures, osteonecrosis of the jaw, malignancy, and atrial fibrillation. Other secondary measures include adherence to medication, quality of life, and mortality. When this proposal was discussed in the larger group, more than 2/3 of the large group felt hip fracture should be a co-primary outcome. This is the most compelling outcome to Medicare beneficiaries due to its significant impact on cost, quality of life and mortality. Some participants were concerned about whether the study would be adequately powered to detect significant differences in hip fractures alone and cautioned against including it if statistically significant differences were unlikely to be found. There was strong disagreement with the recommendation that traumatic fractures should be eliminated from the secondary outcome. Many participants agreed that Patient Reported Outcomes (PROS) are important and that participants should agree to provide these data as a condition of participation. Recommendations for the periodicity of data collection varied between every 6 months to every 12 months. Validation and adjudication of the claims-based outcomes measures will be important. In order to pose minimal disruption to care participants recommended to adjudicate safety and fracture outcomes in a subset of enrollees with internet access. Data Management and Medicare Data Linkage The ATLast trial plans to use Medicare data to capture long-term fracture and safety risk with 36 months follow up. As Medicare data are critical to providing long-term outcomes, enrollees are restricted to Medicare beneficiaries with Part A and B benefits that will have utilization data in the claims (non Medicare Advantage (MA)). Participants will also be required to be enrolled in the Medicare prescription drug benefit (Part D). This design feature raises several issues that were explored at the meeting. Two major issues discussed were how to mitigate study biases by exclusion of MA enrollees and how one could determine Medicare Part D or Medicare Advantage coverage at or near time of enrollment. Participants also explored the reliability of claims data for capturing safety risk. Participants noted eliminating managed care patients from the study would substantially reduce its generalizability. The penetration of Medicare Advantage enrollment ranges from 12 to over 40 percent depending on the state. The consensus was it would be highly desirable to augment Medicare data by 5

6 including data from private MA plans that maintain encounter data, such as Wellpoint. The lag in the availability of claims data (9 to 18 months) makes it infeasible to use it for determining eligibility at point of enrollment. Participants recommended filtering practices within the networks on the basis of percentage of women meeting benefit requirements (e.g., Part A, B and D and no MA) to increase the probability patients would be eligible. One participant recommended exploring whether it would be feasible to query Medicare s pharmacy database to determine Medicare Part D coverage at time of enrollment. This would provide real-time information. Another working group member who could not attend in person suggested that we could also query the main enrollment database to ascertain whether someone was in a Medicare Advantage plan. To ensure efficient data linkage it was recommended that the Medicare Health Insurance Claim (HIC), social security number, date of birth and gender be collected in a baseline survey. Enrollees will not be able to reliably report their participation in MA plans or Part D, but study coordinators could request insurance cards. Participants also noted that most adverse events cannot be reliably obtained from claims data and would need follow up surveys. Validation of claims-based events should be built into the design. Practical Considerations for Conducting LSTs in Practice-Based Research Networks The ATLAST trial is leveraging practice-based research networks (PBRNs) in the effort to develop the infrastructure to support LSTs in chronic diseases more generally. ATLAST will recruit PBRN sites to enroll participants who reflect patients treated in usual care settings, which raises unique questions regarding recruitment and oversight. Among the questions addressed included: What are the criteria that should be used to identify PBRN sites? What partners should we engage to identify PBRN sites and how should we recruit sites? What is the appropriate compensation for doctors, nurses, office staff, patients and PBRN directors? Who should be responsible for technology management? What are the options for the consent process and which can be used best in a community practice? Can multimedia (web/video) technologies be used in the consent process? How does required training of local community physicians and nurses factor into the consent process? It was agreed that several criteria could be used to recruit PBRN sites: sites that have a relationship with an infusion center, sites that mainly treat patients who are age 65 and older, sites that are over a certain size (one way to do this would be to set a fair reimbursement rate for sites and then ask the network to identify sites that can recruit over 200 patients), PBRNs that have prior experience conducting randomized control trials, and PBRNs that have electronic health records. However, the number of criteria to be used should be chosen prudently as it influences the trial s pragmatism. Also, there is a need to decide on whether we should focus on proven research entities. To recruit sites, one could research out to various state chapters, contact the resource center at the Agency for Healthcare Research and Quality, attend the annual PBRN meeting to network, and/or work with networks such as the HMO Research Network (HMORN). The PBRN resource centers should be used to engage all the PBRNs around the country. Involving the hospital systems and health systems was considered to provide substantial value. 6

7 Regarding compensation, the consensus was that the PBRNs cannot implement a study at a loss and so they need to be compensated to at least cover their expenses. Several models to structure their payment were discussed: structure payment around the screening process or provide sites with an upfront amount to get them started and then pay additional money per visit. Sites cannot be compensated on a per person basis. The workgroup participants echoed the sentiment that it would be unethical to ask people to be involved in this trial without providing them any compensation for the additional time they would spend in the trial with respect to going through the consent process and travel required for study-related follow-up. Most participants in the group agreed that it was important to ensure that drug-related costs were covered for trial participants. Once a trial is implemented, technology management should be the responsibility of central study staff and can be supported by local study staff where it exists. The group recommended that some technology (such as a video) be used in the consent process to provide information and assess understanding. Face-to-face contact could be used for the authorization component. In order to conduct informed consent, community physicians and nurses must be HIPAA certified, and must receive training in human subjects protections and conflict of interest, as well as training on the trial protocol. 7