The Statistical Group was comprised of select members of the Forum, statisticians from the FDA and Medicines and Healthcare Products Regulatory Agency (MHRA), and a patient representative (IL). The chairperson (JS) led a series of meetings to assess MDE suitability for AL amyloidosis trials. The Statistical Group also reviewed publications and applications of MDEs in other therapeutic areas [16, 17] and consulted regulatory guidance from the FDA [9], European Medicines Agency [18], and the European Network for Health Technology Assessment [19] for the development and use of composite endpoints in drug evaluation. Each organ-specific working group presented available evidence and recommendations for appropriate endpoints for use in AL amyloidosis trials (Table 1). The Statistical Group considered several candidate MDE approaches along with recommended next steps to inform MDE development and to evaluate performance using natural history and available clinical trial data.

Table 1 Outcome Measures Prioritized by Amyloidosis Forum Working GroupsTypes of Multi-domain Composite Endpoints and Statistical Methodology

Different types of MDEs used in drug development across therapeutic areas were considered in terms of advantages and disadvantages for AL amyloidosis (Table 2). Additional approaches to develop a MDE may also be viable based on the mechanism of action of the intervention, target indication, study population, and trial design.

Table 2 Types of Composite EndpointsComposite Responder Criteria

Composite responder criteria, such as the American College of Rheumatology/European League Against Rheumatism (ACR/EULAR) classification criteria for rheumatoid arthritis [20], combine biomarkers and clinical outcomes important for near-term relief and longer-term disease control. Response is defined at the patient level as reaching or exceeding a given degree of improvement from baseline across the included domains [20, 28]. The timing of response assessment should be justified, based on the underlying disease biology and considering the multiple included domains, along with specificity on whether responder criteria must be satisfied at a specific time point or at any time up to and including that time point. Specificity is particularly important when underlying measures (e.g. NT-proBNP in AL amyloidosis) fluctuate over time. A response based MDE can be sensitive to drug effects on near-term improvements, but less sensitive to drug effects that slow or stabilize disease progression without improvement. A response based MDE offers easily interpretable results if the responder status is well established with clear clinical meaning. Caution should be applied for response outcomes that are derived by dichotomizing a continuous or ordinal measure. Such dichotomizations can complicate the interpretation of the overall treatment effect and may not be best practice. In some situations, a large responder effect may not necessarily reflect a clinically important treatment effect in a continuous or ordinal measure [19].

Composite Progression Endpoints

Composite progression endpoints have established precedent in cardiology, hematology, and oncology; progression-free survival (PFS) is a widely used example. In AL amyloidosis, the composite of major-organ deterioration progression-free survival (MOD-PFS), a secondary endpoint in the ANDROMEDA trial, supported accelerated FDA approval of daratumumab when added to bortezomib, cyclophosphamide, and dexamethasone (VCD) in patients with newly diagnosed AL amyloidosis [25, 29]. MOD-PFS is comprised of clinical endpoints including death, clinical manifestation of cardiac failure (defined as need for cardiac transplant, left ventricular assist device, or intra-aortic balloon pump), clinical manifestation of renal failure (defined as development of end stage renal disease evidenced by need for hemodialysis or renal transplant), or development of hematologic progressive disease per consensus guidelines [25]. Composite progression endpoints are often studied as the time to a subject’s first event, but recurrent events may also be applicable. Progression endpoints are directly sensitive to drug effects that slow the rate of progression, and indirectly sensitive to improvements persistent enough to delay progression. Composite progression endpoints are widely used and accepted in the regulatory setting when the progression criteria are clinically meaningful.

Hierarchical Composite Endpoints

The Finkelstein-Schoenfeld (FS) test [30] and the win ratio test [31] provide additional options for analyzing composite endpoints in the context of a randomized controlled trial design. Such tests are based on comparisons between all possible pairs of treated and untreated individuals. For each pair, the question is asked: who did ‘better,’ the treated or untreated individual? The individual with the better outcome is identified using a pre-defined hierarchy across multiple domains and can accommodate outcomes on different scales/types (e.g., time to event, categorical outcomes, continuous outcomes). Finkelstein and Schoenfeld first proposed a test statistic by performing pairwise comparisons on longitudinal and survival outcomes of all patients hierarchically. The FS test was used as the primary efficacy analysis in the ATTR-ACT trial of tafamidis in patients with transthyretin amyloid cardiomyopathy [32]. The approach works best when the component in the higher hierarchical order has a large treatment effect and a high event rate. The methodology can give priority to the more clinically important events and can also combine different types of components (longitudinal, outcome, categorical). However, caution needs to be taken when continuous measurements are included as components in the hierarchical composite endpoint. In such situations, the win-lose algorithm can be driven by a trivial difference from the continuous measurement and obscure the clinical meaning of the composite endpoint. The algorithm is computationally intensive, and the handling of censoring and missing data can be complicated. Other variants, generalizations, and approaches to combining multiple domains via statistical analysis have been proposed [33].

Multi-domain Responder Index (MDRI)

The MDRI encompasses multiple endpoints via pre-specified responder thresholds for each. Scores are assigned based on the outcome in each domain, then summed across all domains. For example, a score of + 1, 0, or −1 may be assigned for clinically significant improvement, no significant change, or clinically significant worsening, respectively, to an individual’s outcome in each domain. The total score summed across all domains represents the net number of domains improved vs. worsened for an individual, which can then be averaged and compared across treatment groups [34, 35]. MDRI was used in a post-hoc analysis of data from the pivotal trial of laronidase for the treatment of mucopolysaccharidosis I [36]. By design, MDRIs are sensitive to treatment effects that lead to improvement and/or the prevention of worsening across domains. In practice, MDRI can be difficult to implement and has not been accepted for pivotal trials in the regulatory setting. The responder threshold for each component should be clinically meaningful, which can be challenging to determine.

Composite Statistical Tests

There are also approaches that combine across component endpoints into a composite test. For example, the O’Brien Rank-Sum test considers the ranks of each outcome across both domains and patients, and assesses whether outcomes in the treatment group tend to be better ranked than those in the control group [37]. The Wei-Johnson test combines mean treatment effects across endpoints, including time-to-event endpoints, by weighting the contribution of each domain based on variance and correlation to produce a composite test [38]. These composite tests may also be viable approaches and should be further explored.

Pathway for Development and Evaluation of Composite Endpoints for Use in AL Amyloidosis TrialsIdentification of Principal Organ Systems, Domains, and Candidate Component Endpoints

The first step in development of a composite endpoint for use in AL Amyloidosis trials (or for any rare therapeutic indication characterized by multi-systemic involvement) involves identification of domains most indicative of disease progression and of disease aspects that patients consider most important for demonstration of treatment benefit (Fig. 1). Initially, the organ system-specific Working Groups reviewed available literature to identify available published reports on candidate component endpoint validation. If deemed insufficient, the Working Groups utilized clinical expert input and input from patients [39,40,41]. Considerations for each candidate endpoint included: (1) clinical relevance, (2) available natural history data and/or clinical experience, (3) time horizon to detect change based on natural history and timing of impact of effective treatment, (4) meaningful thresholds, and (5) gaps in knowledge.

Ensuring a Foundation for Clinical Meaningfulness

For an MDE to be considered clinically meaningful, each component of the MDE should be either:

A clinical endpoint that measures how a patient feels, functions, or survives; or

A validated surrogate endpoint predictive of treatment effects on how a patient feels, functions or survives

Surrogate endpoints must be validated separately; predictive associations between a potential surrogate and clinical outcomes are highly supportive, but not sufficient. Associations between treatment effects on surrogate response and treatment effects on overall survival, or on other clinical outcomes, should be established across multiple randomized trials [42,43,44]. Within-trial estimates of patient-level associations and proportions of the treatment effect on the clinical outcome explained by effects on the surrogate can complement cross-trial analyses [45].

The Prentice criteria provides a helpful framework for assessing surrogacy though a combination of statistical assessments and clinical/biological judgement about the pathways by which treatment might affect the surrogate and the ultimate clinical outcome [46]. Subsequent work has further extended and modified these criteria and emphasized the risks of ill-formed or un-validated surrogates [43, 44, 47, 48]. The approach of Buyse et al. [49] and its extensions has provided valuable empirical evidence of surrogacy supportive of validation alongside clinical and biological plausibility.

The FDA’s PFDD guidance provides recommendations on how to develop and validate COA based endpoints that reflect patients’ experiences and priorities and how to determine the clinical meaningful change in a COA based endpoint [10]. The interpretability of COA endpoints depends on how closely the measurement reflects patients’ experience and the COA metric used. To find the meaningfulness of a treatment effect in a COA measure, one can consider anchor-based methods by mapping an anchor, which is some external variable with direct interpretable difference (e.g., a patient or physician global assessment), to differences of the COA scores. Multiple anchors can be used to inform a plausible range of meaningful score difference (MSD).

The organ-specific working groups identified several candidates of high interest for constructing the MDEs, e.g. NT-proBNP, 6MWD, eGFR, and hematologic responses (Table 1). The applicability of these endpoints in AL amyloidosis needs further investigation and validation. For AL amyloidosis, understanding the potential surrogacy of NT-proBNP and hematologic response or progression outcomes is a priority. While changes in eGFR (including slope) have been accepted as a clinical endpoint in other therapeutic areas, applicability in AL amyloidosis needs further validation. With sufficient data from multiple randomized trials now available, research is underway within the Forum to conduct these assessments.

These steps provide a necessary foundation for identifying meaningful components of a composite MDE but are not alone sufficient for determining the clinical meaningfulness of the composite. For a composite, the method for combining information across domains also needs to be assessed for clinical meaning. Each individual component in a composite endpoint should be clearly understood in the analytic methods. When a composite endpoint shows a treatment effect, it’s necessary to also examine the treatment effect of each individual component.

Evaluating Component Endpoints

The evaluation of candidate endpoints for use as components has been discussed at public workshops [50]. Evaluation should assess clinical meaningfulness and reliability, as well as natural history in terms of rates of change and levels of variability over time in relevant patient populations and the degree of correlation among component endpoints.

Broadly, assessments of natural history provide drug developers with a quantitative understanding of component endpoint performance needed to evaluate suitability for incorporation into a composite clinical trial endpoint. Natural history of each component may be established using retrospective data from natural history studies/registries, post-hoc analysis of placebo data from completed clinical trials, or prospectively though pilot studies or during early phases of a clinical development program. Such assessments of endpoint components provide a foundation for regulators and other decision-makers to assess suitability, adequately power a trial, and to interpret trial results based on the endpoint composite.

The characteristics of candidate endpoints (e.g., 6MWD, eGFR, NT-proBNP, and neurological measures) should be further evaluated, including the rates of change and intra-subject variability. For potential surrogate endpoints, in-depth understanding of the causal pathway of the disease process would be ideal; as such, an extensive overview on available studies with reliable estimates on both the clinical outcomes and the biomarkers is recommended. For proposed response or progression thresholds, rates of response and progression—and rates of reversal of response or progression—should be quantified. Such data are necessary in making evidence-based choices for the combination of clinical trial enrollment criteria, endpoint selection, and trial follow-up duration to detect meaningful effects in the development of investigational products.

Development of Candidate Composite Multi-domain Endpoints

The Statistical Group discussed four ways of constructing multi-domain composite endpoints: (1) a response criterion (similar to the ACR); (2) time to a composite progression event (e.g., MOD-PFS); (3) an MDRI, and (4) a multi-domain hierarchical composite endpoint.

To inform the statistical performance and interpretation of composite MDEs, overlap across domains should be assessed both statistically and based on clinical judgement. If two component endpoints are very highly correlated and represent similar clinical concepts, this could unintentionally lead to effective “double counting” of the underlying domains. When reporting treatment effects on MDE outcomes, the treatment effect of each component should be reported in parallel to support interpretation [18].

In addition to establishing the clinical meaning of each component, it is necessary to characterize the clinical meaning of the composite MDE, particularly in the context of AL amyloidosis where different organ systems may be affected to varying degrees in different patients. Composite MDE outcomes may be quantified and qualified using natural history or placebo arm data in populations that could be enrolled in clinical trials. In particular, response rates and times to progression—and the stability of response or progression status—need to be assessed across different sets of inclusion/exclusion criteria. MDEs may combine different types of component endpoints, e.g., physician-assessed outcome assessments, patient-reported outcome assessments, performance outcomes, or drug-induced changes in biomarkers obtained from laboratory assays or imaging. When combining several types of components, it is not recommended to combine components with drastically different clinical importance. Such MDEs would often be challenging to interpret and determine clinical meaningfulness.

Ultimately, the performance of an MDE for AL amyloidosis clinical trials must be evaluated within the context of its intended use, including the specific population (e.g., heterogenous organ involvement), hypothesized multi-domain treatment effect (i.e., for a plasma-cell or amyloid-directed therapy), and the regulatory decision/claim being sought. The pathway proposed above will provide a foundation for endpoint selection and development of composite MDEs in AL amyloidosis. Specifically, clinical trial simulations, and power and sample size calculations will rely on evidence derived from natural history data. While pre-existing evidence and the pathway to develop composite MDEs can provide a strong foundation for endpoint selection in AL amyloidosis, there may be value in updating and tailoring this approach for specific drug development goals, e.g., for drugs with novel mechanisms of action, for populations defined by novel biomarkers, or to adapt to future changes in the standard of care. For these reasons, availability of up-to-date and high-quality natural history data is important for drug development in AL amyloidosis.