Application of improved statistical methodologies for clinical trials in Progressive Supranuclear Palsy

Lead Investigator: Günter Höglinger, Hanover Medical School
Title of Proposal Research: Application of improved statistical methodologies for clinical trials in Progressive Supranuclear Palsy
Vivli Data Request: 6495, 5485
Funding Source: Funded by the European Joint Programme on Rare Diseases (EJP RD), see attached proposal application form
Potential Conflicts of Interest: Günter Höglinger has served on the advisory boards for AbbVie, Alzprotect, Asceneuron, Biogen, Novartis, Roche, Sanofi, UCB; has received honoraria for scientific presentations from Abbvie, Biogen, Roche, Teva, UCB, has received research support from CurePSP, the German Academic Exchange Service (DAAD), German Parkinson’s Disease Foundation (DPG), German PSP Association (PSP Gesellschaft), German Research Foundation (DFG) and the German Ministry of Education and Research (BMBF), International Parkinson’s Fonds (IPF); has received institutional support from the German Center for Neurodegenerative Diseases (DZNE).
Potential conflicts of interests will be stated in any publication connected with this research proposal.

Summary of the Proposed Research:

Background: Progressive supranuclear palsy (PSP) is a neurodegenerative disease with a prevalence of 1-9/100,000 and an average disease duration of approximately 8 years. Effective symptomatic treatments and disease-modifying therapies are not available. Therefore, development of innovative therapies to slow down or halt disease progression are urgently required. Numerous investigational new drugs are entering the clinical trial pipeline for evaluation of their safety and efficacy. The experience and data generated by the consortium coordinator with the past clinical trials in PSP has demonstrated limitations of previously used methodologies, including heterogeneity in disease progression rates, limitations of the primary endpoint, limitations resulting from the short observation period and reluctance to be randomized to placebo. Identification of these limitations opened up ways forward to improve methodologies for future trials.

Aim: The applicants therefore aim to optimize clinical trial methodology for PSP.

Methods: The consortium will develop more sensitive and more relevant endpoints (modified outcome measures, composite endpoints), develop prediction models, more efficient trial designs (adaptive designs, delayed-start designs, longer duration trials, platform trials, basket trials), and use historical longitudinal data for multiple outcome measures to improve statistical power.

Perspective: Applying the statistical methods developed by the applicants will help to improve the trial design and the statistical analysis methods for future trials in PSP.

Statistical Analysis Plan:

Task 1: Development of joint models: Our analysis will utilize the existing clinical trial data to develop a parametric model in the item response theory (IRT) framework. IRT is a
paradigm commonly used in psychometrics for the analysis of questionnaire data. It describes the probability of responses to individual items in an assessment as the function
of item-specific parameters and subject-specific latent variables. Longitudinal dynamics are integrated as, potentially non-linear, changes of the latent variable Ψ𝑖 due to disease
progression or treatment. The parameters in the resulting non-linear mixed effect model are estimated using marginal maximum likelihood estimation.
This type of approach allows the integration of different assessments in a joint model. In this context, we intend to develop a model integrating PSPRS, SEADL, and CGIDS scores.
Volumetric MRI data and survival outcomes will be integrated as additional outcomes and linked to the latent variable through correlated random effect parameters. The inclusion of subject-specific covariates will allow us to describe the multiple endophenotypes PSP-RS, PSP-P, and PSP-CBS in a single parametric model.
The resulting integrated model can then be used to (i) better understand relationships between different endpoints, (ii) individualized prediction of progression, and (iii) planning of future studies by comparing different innovative trial designs. Understand relationships between different endpoints: The increased understanding of different endpoints, will be enabled by simultaneously capturing the correlations between rating scales and survival endpoints. This will allow us to investigate distinctions and communalities between scales and read-outs as well as to explore relationships between scales and mortality.

Task 2: Individualized prediction: The model will be formulated on an individual patient level and, therefore, enable individualized prediction of progression and delayed onset of
efficacy. The individualization can occur based solely on baseline covariates or using observed scores up to a specific time point in a Bayesian forecasting framework.

Task 3: Alternative designs and analysis strategies: Testing Strategies for Multiple Endpoints: The improved understanding of the correlation and joint distribution of different endpoints, derived from the available clinical trial data, will be used to identify efficient hypothesis testing and estimation procedures. Especially, the operating characteristics of a range of hypothesis testing strategies will be compared for trials in PSP: approaches to summarize multiple endpoints in a single composite endpoint will be compared to several multiple hypotheses tests for the individual endpoints. As endpoints we will consider progression in scales as the Progressive Supranuclear Palsy Rating Scale (PSPRS) or the Schwab and England Activities of Daily Living Scale (SEADL) as well as mortality. Based on the estimand framework (ICH E9 (R1)), endpoints will be defined and corresponding statistical testing and estimation procedures in randomized clinical trials will be assessed with respect to the statistical power, handling of missing data and intercurrent events using simulations from the model. The identified testing strategies will be applied to re-analyse the clinical trials and according treatment effect estimates and confidence intervals will be computed. Adaptive trial designs and Basket Trials: Adaptive designs allow one to adapt the design of an ongoing trial based on unblinded interim information. We will use the available data from the RCTs to emulate adaptive designs by sampling of observations and parametric bootstrap. First, we will use the longitudinal data structure to improve the estimation of the conditional power based on both short- and long-term endpoints at an adaptive interim analysis. E.g., instead of using only the change in PSPRS at 52 weeks, for patients that did not reach week 52 yet, in addition measurements at earlier time points will be used to improve conditional power estimates. For survival data, also the information obtained in the rating scales will be used to determine the conditional power in a joint model. Especially in situations where the treatment effect quickly manifests in the rating scale but has a delayed-onset for survival, this shall improve decision making to perform trial adaptations. We will consider adaptations such as sample size reassessment and enrichment of subgroups at interim. However, to ensure proper control of the type 1 error rate required for confirmatory RCTs, special techniques are needed to combine the information obtained before and after the adaptive interim analysis. We will assess the application of adaptive enrichment designs for clinical trials in PSP that include, in addition to the main clinical manifestation Richardson’s syndrome also patients with a broader phenotypic spectrum. Based on data from two independent observational natural history studies with longitudinal follow-up of PSP patients with a broader phenotypic spectrum we will explore scenarios where the treatment effect may generalize to a broader population. We will compare the operating characteristics of adaptive designs allowing selection and testing of both the full study population and a predefined subgroup(s). This will also include Basket trials, where promising subgroups can be selected and combined for confirmatory efficacy testing at final analysis.

Task 4: Use of external data: To increase the power of randomized controlled trials in settings where large sample sizes are infeasible and to minimize the number of patients randomized to control groups in the presence of substantial unmet medical need, it has been proposed to use data from historical controls and thresholding approaches for statistical inference. We will evaluate Bayesian approaches and frequentist methods to incorporate data from historical controls in clinical trials in PSP for statistical testing and estimation and will evaluate the gain in power, type I error rates. Especially, taking into account the inclusion and exclusion criteria in the different trials, we will aggregate control group estimates from several trials (as well as registry data) to obtain a more powerful statistical test. As historic data may introduce bias due to seasonal trends or different distributions of covariates, we will address adjustments based on time trends and the use of co-variate information to adjust for differences in the populations between the historical controls and the trial population.

Requested Studies:

A Randomized, Double-Blind, Placebo-Controlled Multiple Dose Study to Assess Efficacy, Safety, Tolerability, and Pharmacokinetics of ABBV-8E12 in Progressive Supranuclear Palsy
Sponsor: AbbVie
Study ID: NCT02985879
Sponsor ID: M15-562

(Note: Title and Summary Updated as part of Vivli Data Request 6495)

Public Disclosures:

  1. Yousefi, E., Gewily, M., König, F., Höglinger, G., Hopfner, F., Karlsson, M.O., Ristl, R., Zehetmayer, S. and Posch, M., 2023. Efficiency of Multivariate Tests in Trials in Progressive Supranuclear Palsy. arXiv preprint arXiv:2312.08169. Doi : 10.48550/arXiv.2312.08169
  2. Krotka, P., Posch, M., Gewily, M., Höglinger, G. and Roig, M.B., 2024. Statistical modeling to adjust for time trends in adaptive platform trials utilizing non-concurrent controls. arXiv preprint arXiv:2403.14348. Doi : 10.48550/arXiv.2403.14348