Lead Investigator: Armando Turchetta, McGill University
Title of Proposal Research: Forecasting recruitments in multicenter clinical trials via the time-dependent Poisson-Gamma model
Vivli Data Request: 8767
Funding Source: My research is funded by the Fonds de Recherche du Québec Nature et technologies (FRQNT).
Potential Conflicts of Interest: I am currently employed at F. Hoffmann-La Roche Ltd, i.e. the sponsor of the trials I am requesting access to, however, this research project is only part of my PhD and is fully funded by the Fonds de Recherche du Québec Nature et technologies (FRQNT). Roche (via internal discussions) has kindly agreed to share with me the data of some of their trials to help me complete this project, but it will not fund this research. Therefore, no conflict of interest is expected to occur. A member of Roche will be included in the paper as a co-author (if they wish so) and the funding and affiliation information will be disclosed in the manuscript.
Summary of the Proposed Research:
One of the key challenges in the planning and monitoring phases of multicenter clinical trials is the assessment of the recruitment time as this informs the feasibility of the proposed study. Yet, although an inaccurate estimation of this time may result in a substantial loss of resources, most of the traditional techniques used in recruitment planning are either deterministic or rely on assumptions that are often unrealistic, such as constant recruitment intensity over time. Building on an existing recruitment forecasting model, the Poisson-Gamma model, we developed a flexible generalization of this methodology called the time-dependent Poisson-Gamma model (tPG), which is suited to estimate a wide range of recruitment behaviors over time. We have already validated this model using recruitment data from a cohort study, showing a significant improvement over the standard Poisson-Gamma model.
A realistic estimate of the time needed to achieve the target sample size is an important piece of information that can drive significant financial and practical decisions in multicenter studies, such as the opening or closure of recruitment centers and the allocation of resources. The main goal of this project is to show the applicability of the time-dependent Poisson-Gamma model for forecasting enrollments in multicenter clinical trials, as well as its advantages over the commonly used methodology based on constant rates over time. In turn, this study will help with the monitoring phase of future clinical trials, as the resulting paper will also outline the functionalities of the R package we built to implement this methodology and how to use them in practice.
Statistical Analysis Plan:
The proposed recruitment forecasting model is an adaptive technique whereby the predictions on future enrollments and the remaining recruitment time are updated based on the data of subjects already recruited. This model is built on the Bayesian Poisson-Gamma model, whereby participants are assumed to arrive at the recruitment centers according to a Poisson process with center-specific random rates originating from a Gamma distribution. Since this model assumes that the recruitment speed is constant over time, we developed a flexible generalization of this methodology called the time-dependent Poisson-Gamma model (tPG), which is suited to estimate a wide range of recruitment behaviors over time.
To assess this methodology on the recruitment data from the four clinical trials, for each trial, we will consider several interim times and, at each of these time points, we will evaluate the predictions produced by the algorithm on future enrollments by comparing them with the observed recruitment process. In addition, we will evaluate the incorporation of center-specific covariates into the forecasting model to improve its accuracy. All the results will be presented at the aggregate level with no possibility of retrieving the individual dates of the enrollments.
The four trials were chosen from the list of available studies sponsored by Roche, as this company is my current employer and this will be a joint project with my University (McGill). The choice of trials was purely based on the type of study (randomized clinical trial) and the number of centers involved in the recruitment in relation to the trial’s sample size, as our methodology works better with a higher average number of recruitments per center. Although I have added my PhD supervisors Drs. Erica Moodie and David Stephens as additional researchers for this project, I will be the only one who needs access to the data.
Requested Studies:
A Randomized Three-Arm, Multicenter Comparison of 1 Year and 2 Years of Herceptin Versus No Herceptin in Women With HER2-Positive Primary Breast Cancer Who Have Completed Adjuvant Chemotherapy
Data Contributor: Roche
Study ID: NCT00045032
Sponsor ID: BO16348
A Randomized Multicenter, Double-Blind, Placebo-Controlled Comparison of Chemotherapy Plus Trastuzumab Plus Placebo Versus Chemotherapy Plus Trastuzumab Plus Pertuzumab as Adjuvant Therapy in Patients With Operable HER2-Positive Primary Breast Cancer
Data Contributor: Roche
Study ID: NCT01358877
Sponsor ID: BO25126
A Randomized, Double-blind Study of Safety and Reduction in Signs and Symptoms During Treatment With Tocilizumab Versus Placebo, in Combination With Methotrexate, in Patients With Moderate to Severe Rheumatoid Arthritis
Data Contributor: Roche
Study ID: NCT00106548
Sponsor ID: WA17822
A Randomized, Double-blind Study of the Effect of Tocilizumab on Reduction in Signs and Symptoms in Patients With Moderate to Severe Active Rheumatoid Arthritis and Inadequate Response to DMARD Therapy
Data Contributor: Roche
Study ID: NCT00106574
Sponsor ID: WA18063
Summary of Results:
Unfortunately, no analysis was conducted. In the beginning, there was a logistic issue as we received the data too close to my thesis deadline so I had to use a different data set. We waited to see if we could perform the analysis for an additional manuscript, however, in the meantime I left academia, and I was unable to find the time to work on this.
Since there is no one in my former lab working on this topic, I feel it could be a waste of resources to ask for an extension. As far as I know, there is no one in my former department working in this research area who could continue developing this project.