Planning and ANalyzing OPTImal Clinical trials with Adaptive Design

• Applicant: Professor Meinhard Kieser, Ph.D.
• Subject Area: Medical Informatics and Medical Bioinformatics ; Software Engineering and Programming Languages
• Term: from 2021 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 443177220

• Final Report Year: 2024

Final Report Abstract

Sample size calculation is an important aspect when planning clinical trials. It involves determining how many patients need to be recruited for a trial to prove therapeutic effects. A novel method for sample size planning are the so-called optimal adaptive designs. In adaptive designs, the characteristics of the trial design may be modified during its course based on observed data. In optimal adaptive study designs, the rules for changing the design parameters (e.g. the sample size) are determined by solving an optimization problem. The criterion being optimized can be composed modularly of various values, such as the average and maximum sample size. Determining an optimal adaptive design is challenging and requires complex numerical and statistical methods. This approach was first developed and investigated in the DFG project ORACLE, during which the R package adoptr was created. This package allows calculating specific optimal adaptive design parameters given an optimization criterion. The current PANOPTICAD project aimed at improving quality assurance and at expanding the software for calculating optimal adaptive trial designs. adoptr was extended by various functionalities, making it possible to plan not only studies with normally distributed endpoints but also with binary or time-to-event endpoints. Furthermore, methods were implemented to determine optimal adaptive designs for trials comparing more than two treatment groups. A particular focus of the PANOPTICAD project was to provide a thorough documentation in order to ensure the quality and sustainable usability of the software package. Clinical trials are heavily regulated, and various guidelines require the use of validated software. For these reasons, an extensive quality assurance concept was pursued, including measures for archiving, versioning, error reporting, and code documentation. One example is the creation of a validation report, which demonstrates possible use cases of the methods implemented in adoptr in various scenarios.

Publications

• Two-stage designs with small sample sizes. Journal of Biopharmaceutical Statistics, 33(1), 53-59.
  Kieser, Meinhard; Rauch, Geraldine & Pilz, Maximilian
  
• adestr: Estimation in Optimal Adaptive Two-Stage Designs. CRAN: Contributed Packages. The R Foundation.
  Meis, Jan
  
• baskspline: Extending baskexact for weighting via monotonic splines
  Thalmann P. & Baumann L.
  
• adoptr: Adaptive optimal two-stage designs for clinical trials. CRAN
  Kunzmann K., Pilz M., Bruder N. & Meis J.
  
• Point estimation, confidence intervals, and P‐values for optimal adaptive two‐stage designs with normal endpoints. Statistics in Medicine, 43(8), 1577-1603.
  Meis, Jan; Pilz, Maximilian; Bokelmann, Björn; Herrmann, Carolin; Rauch, Geraldine & Kieser, Meinhard
  
• Validation report for the adoptr package
  Kunzmann K., Pilz M. & Bruder N.