When statistics and pharmacometrics join forces for advanced quantitative drug development: A case study

Aims: The design of full development trials requires a number of features to be defined. Those include the population, treatment regimen, endpoint with clinical relevance and sensitivity to drug effect, assumptions on expected outcomes and their variability, and statistical methods to control errors in decision making and to avoid bias in estimation. In novel drug development settings, relevant information may be scarce to set those features. It becomes essential to inform choices by a synthesis of all available data, to identify the assumptions made on the less known parameters, and to assess their impact on the trial operating characteristics. The modelling framework of pharmacometrics, together with trial design methodologies and the statistical assessment of risk, constitute a fundamental asset to guide drug development with advanced quantitative insights. This is illustrated by two case studies of development programs in ophthalmology and neuroscience: a sustained drug delivery device in neovascular (wet) age-related macular degeneration (nAMD), and compounds targeting the amyloid pathway in Alzheimer disease (AD) prevention. In both situations, an integrated approach between pharmacometrics and statistics enabled the design of science-based innovative phase 2 and 3 trials.

Methods: The ophthalmology example concerns the development of a new mode of administration for a treatment of nAMD. This treatment is currently administered by repeated intravitreal (IVT) injections. Frequent re-treatment by IVT injections may cause undesired effects and become burdensome to patients and caregivers. A novel implantable and refillable device allowing sustained drug delivery into the vitreous may enable less frequent re-treatment, for a similar clinical outcome. A phase 2/3 development plan was set up to confirm this hypothesis. Due to the lack of prior efficacy data with the device, the phase 2 trial design was highly challenging. Key questions arose such as the relevant endpoint and statistical design, the design parameters to optimize in a multi-dimensional decision space, and the sample size to achieve a target power. A model was developed by pharmacometricians and statisticians to inform these questions on the basis of available data on the disease progression, the compound and best corrected visual acuity (BCVA), the standard clinical outcome measure in this disease. The model incorporates available information and assumptions on all aspects, from disease progression, device release performance and drug concentration profiles in the vitreous (based on pre-clinical experiments), patient inclusion criteria including number of loading IVT doses, device refill criteria, enrollment period and study duration, up to the clinical outcome of BCVA.

The neuroscience example focuses on AD prevention, for which the development of disease-modifying therapies faces many challenges. Trial outcomes and designs need to be established for pre-symptomatic stages of AD. Traditional measures of cognitive changes developed for studies in later stages such as mild cognitive impairment or AD dementia are of limited value (i.e. due to the psychometric properties of the tests). Extensive model-based exploratory analyses of real world data allowed to describe the natural time course of novel composite cognitive endpoints prior to and up to actual disease diagnosis. A simulation approach of the two primary endpoints selected for the planned phase 3 study, accounting for their correlation, was then established and used to assess different assumptions and design options.

Results: In both case studies, the model behavior was assessed by reproducing existing data, including time profiles of outcomes such as BCVA in nAMD or the composite cognitive endpoint in AD prevention. Statistically more elaborate time-to-event clinical endpoints were also well described. Using the models, a range of design parameters and uncertainties were explored by simulations and their impact on study power were then examined. Decisions were made to specifically optimize the study designs, leading to more realistic and scientifically driven plans and a greater probability of success. Features investigated included the population definition, the selection of endpoints, study duration, assessment schedule, re-treatment criteria (in nAMD), statistical testing strategy, sample size and power.

Conclusion: Advanced quantitative guidance of two difficult development plans was provided, through a joint pharmacometric and statistical modeling approach. This determined fundamental trial design features for both programs. This integrated quantitative approach to study design illustrated the benefit of combining the skill sets of pharmacometricians and statisticians. Success of such collaborations requires an engagement to understand each other's perspectives, skills and language, for a more comprehensive view on drug development. It also requires a nurturing environment that encourages and builds effective multidisciplinary teams.