A Large Pragmatic Trial is the Right Solution for Testing Anti-Amyloid Therapies for Alzheimer’s Disease

The Centers for Medicare and Medicaid Services’ (CMS) recently released a proposed National Coverage Decision (NCD) specifying that it would only cover the cost of treatment with monoclonal antibodies directed against amyloid, including aducanumab, the new Food and Drug Administration (FDA) approved Alzheimer’s disease drug (AD), for Medicare beneficiaries enrolled in an approved randomized clinical trial.¹ In spite of the outcry from some advocates and anti-regulation sources,² the CMS decision is the right one. Instead of waiting a decade or more for the results of an FDA-mandated follow-up study, we could have definitive findings in three to four years. Instead of relying upon a registry of patients with no convincing comparison group, a randomized trial will provide rigorous quantification of both harms and benefits.

By requiring a random assignment trial under its Coverage with Evidence Development (CED) authority, CMS is upholding its promise to patients that “Medicare intends to provide access to high-quality health care”.³ Historically, CMS counted on the FDA to assure that drugs covered under Medicare were safe and effective. However, in our opinion FDA inappropriately used the accelerated approval pathway for aducanumab,⁴ offering a shortcut to bring a medication with unproven efficacy to broad clinical use. Mild Cognitive Impairment (MCI) and early AD,⁵ for which aducanumab was approved, are not immediately life-threatening. The estimated life-expectancy of a 70-year old person with MCI is 13.8 years, only around 3.7 years less than life expectancy for a 70-year old without MCI.⁶ This is not a population for which last-chance medications with documented risks and no proven benefit should be routinely prescribed.

In applauding the CMS provisional ruling, we highlight the two most important features required of any CMS approved study: a randomized comparison group and enrollment of a diverse sample reflecting the population affected by AD. Critics claiming CMS’s decision is discriminatory are misguided and efforts to weaken the CMS determination would be a setback for people affected by AD.² Here we outline a rigorous study design that would foster fairness and equity. Such a study needs a pragmatic design, that facilitates testing for heterogeneity of treatment effects and the long-term consequences of treatment, and fosters collection of biomarkers for proposed surrogate outcomes, at least in sub-samples. Such a study would help address inadequacies in the evidence to date and guide interpretation of forthcoming results for other amyloid targeting therapies. This would create a robust basis for future coverage decisions for this medication class.

The many types of pragmatic randomized trials retain the validity of randomization but relinquish some features of traditional double-blinded RCTs in order to achieve speed, representativeness, and real world context.⁷ The CMS CED decision covers the entire class of amyloid-targeting monoclonal antibodies. Trial designs should be sufficiently broad to address all critical policy and scientific questions which the FDA accelerated approval of aducanumab essentially short circuited. One pragmatic approach to an RCT of aducanumab is the stepped wedge design, in which all interested patients eventually receive the medication.⁸ The stepped wedge cluster randomized design staggers rollout of aducanumab treatment⁹ by randomly assigning participating sites to the date when treatment would begin (Figure 1).

Figure 1. Example of a pragmatic trial of anti-amyloid treatment using a stepped wedge design.

Study participants complete routine cognitive assessments (Cog 0 through Cog 8); amyloid PET measures for study eligibility determination and biomarker assessments; and MRI assessments to monitor safety and adverse effects. Primary analysis at 22–24 months are intent-to-treat comparisons of participants at sites randomly assigned to varying treatment initiation dates. Long term follow-up data can accrue indefinitely to evaluate if benefits emerge for some subgroups 2+ years after treatment.

Since the aducanumab treatment requires monthly clinic visits for infusion, group level random assignment is preferable to individual random assignment in which control patients would be given sham infusions or, in the absence of blinding, would merely be routinely measured for cognition. Participating sites would recruit beneficiaries, creating a waitlist of eligible individuals interested in receiving aducanumab treatment and collecting relevant background data as well as baseline cognitive functioning. With only Medicare beneficiaries participating, all past Medicare medical utilization history could be rapidly matched to trial participation data using National Institute on Aging (NIA) MedRIC resources.¹⁰ Baseline and regular follow up measurement of cognitive status should use an assessment instrument that is not influenced by patients’ or caregivers’ optimism about receiving a medication.

To improve feasibility, a stepped wedge design could leverage existing Alzheimer’s Disease Research Centers (ADRCs), major care delivery networks such as Kaiser Permanente or Geisinger Healthcare, or other care networks with appropriate hospital-based outpatient clinical infrastructure with infusion capacity, MRIs for safety monitoring, and amyloid Positron Emission Technology (PET) scans or other biomarkers. Working with ADRCs takes advantage of patients already enrolled in research, many of whom have provided detailed background data. The recruitment of diverse patients via the ADRCs is an ongoing challenge and the geographic coverage of the ADRCs is incomplete.^11,12 ADRC’s recruitment would ideally be augmented with a comprehensive outreach and recruitment plan partnering with trusted community-based organizations. Additionally, direct outreach by other networks, such as pharmaceutical chains or by CMS itself, to their potentially eligible customers could be used. CMS beneficiary lists are routinely used for research recruitment, for example for the Medicare Current Beneficiary Survey (MCBS) and the National Health and Aging Trends Study (NHATS).¹³ Given the egregious under-representation of racial/ethnic minority groups in prior research,¹⁴ strong incentives or quotas to ensure diversity targets are met should be adopted for each site.

After the baseline enrollment and assessment period, sites would be randomly assigned to the order in which they initiate patient treatment. Our model in figure 1 assumes 20 participating sites rolled out in 4 waves staggered by 3–6 months: the latest treatment initiation occurs a year after the earliest. Depending upon the final number of sites participating, some site stratification might be desirable to ensure adequate minority (or geographic) representation in each wave and improve statistical power.

Each site contacts participants to begin monthly infusion treatments at the site’s randomly assigned initiation wave, accompanied by ongoing cognitive assessments and periodic biomarker measurements. Prior to the randomly assigned initiation wave, sites maintain ongoing contact with waitlisted participants completing periodic cognitive assessments. Repeated assessments of the primary outcome of cognitive performance improves statistical performance, mitigates bias from mortality or loss-to-follow-up, and practice effects, and can identify differences in cognitive decline. Additional biomarker measures would help evaluate potential surrogates for clinical benefit, but depending upon the overall sample size, these could be restricted to a sub-sample. For safety purposes, all patients undergoing treatment must receive an MRI prior to their 7th and 12th doses of aducanumab.

Interim analyses could begin at month 16, when the earliest treated individuals would have been on a full dose for 6 months and primary analyses could begin in month 22, approximately 18 months after the first sites begin treating patients. The analyses are premised on the assumption that the key determinant of effect is the cumulative dose of aducanumab. Analyses would preserve intention-to-treat (ITT) principles based on randomization to timing of treatment initiation. If the wait list timing ranges from 0 to 12 months, analytically we would compare individuals in sites 1–5, where participants would have been receiving full dose for 12 months, to participants in sites 15–20, who would be just reaching titration to full dose. An important strength of leveraging existing points of care (as well as matching to Medicare claims) is the likelihood that data on long-term benefits could accrue for years to come. Failures of amyloid-targeting therapies may be due to recruiting the wrong patient base or not continuing follow-up long enough to see benefits.¹⁵

Although simplicity is an advantage in pragmatic studies, evidence on the validity of surrogates for clinical outcomes is urgently needed. The FDA approved aducanumab because it removes brain amyloid. However, evidence to date casts doubt on whether amyloid reduction is a valid surrogate for clinical benefit. Good surrogate measures should have a correlation of at least .85 with clinically relevant outcomes.¹⁶ In four studies which reported individual level analyses, the correlations between amyloid reduction and clinical improvement were all under 0.2. ^17,18 A meta-analysis of grouped data found almost no association between amyloid reduction and cognitive outcomes.¹⁹ These findings do not support amyloid reduction as a surrogate for clinical outcomes, but are not conclusive because of limitations in the available evidence. Because the CMS decision applies to a class of drugs, not just aducanumab, it is important to evaluate subgroup effects and to evaluate the proposed surrogate outcomes likely to be reported to justify future medication approvals.

One major limitation of conventional trials has been the statistical power established to detect an average effect. Growing evidence indicates AD is a heterogeneous disease, and that even a successful medication may only benefit some patients. Most studies are not powered to detect such heterogeneity. We need evidence that can rigorously evaluate whether only some subgroups of patients benefit from treatment. Such evidence would require comprehensive pre-randomization information on individuals in the studies, and studies large enough to provide power to detect heterogeneity. This is an additional advantage to leveraging existing, ongoing studies and cohorts for a pragmatic trial: the data already collected would allow us to evaluate heterogeneity in effects without imposing a major additional burden on study participants.

A large trial, powered to detect sub-groups of people who benefit from anti-amyloid therapy while preserving reasonable type 1 error rates will be expensive. If Biogen does not substantially drop the price of aducanumab, this will be a non-trivial expense for CMS. When considering the costs and benefits of a large trial, the concerns of patients must come first. Provided patients are empowered to make an informed decision about participating and safety is monitored as closely as possible, increasing the chance of identifying a sub-group that benefits should be a top priority. Indeed, if Biogen can identify a subgroup of patients they expect to benefit from aducanumab based on their as-yet-unpublished trial data, subgroup hypotheses can be specified a priori in the CMS sponsored trials.

In summary, the CMS proposal to cover aducanumab only for Medicare beneficiaries participating in an approved trial is one of the most promising developments in AD research in recent years and we encourage CMS to finalize its proposal. We have outlined one of many possible pragmatic approaches that could ensure we quickly establish the safety and efficacy of anti-amyloid therapies. Small edits to CMS’ proposed language would allow for pragmatic designs, foster large sample sizes with detailed data collection to evaluate heterogeneity of patient responses, and support biomarker assessments. Instead of staggering in a quagmire of uncertainty for years to come, the CMS decision, if finalized, promises clarity and insight. Large rigorous studies will either vindicate amyloid targeting therapies or allow us to confidently redirect resources to other targets. We urgently need to stimulate progress on AD treatments, but such progress can only come if we know that we are traveling in the right direction. The CMS required trials will give us that insight.