crmPack 2.0: A Major Milestone for Adaptive Dose Escalation Trials

package
statistics
Author

Daniel

Published

December 1, 2025

Introduction

Adaptive dose escalation trials are a key tool for Phase 1 development both in oncology and non-oncology settings. The R package crmPack has been developed over the last 11 years to facilitate the design and analysis of these trials. crmPack implements a wide range of dose escalation designs, ranging from classical and modern continual reassessment methods (CRMs), based on dose-limiting toxicity endpoints and dual-endpoint designs (e.g., plus biomarker or clinical endpoints), and permitting Bayesian and non-Bayesian inference (Sabanés Bové et al. 2019).

On 29th November 2025, a major milestone was reached. crmPack version 2.0 was released to CRAN, featuring a complete refactoring of the code base, extensive new functionality, comprehensive tests, and improved documentation - please have a look at the release announcement for details. This release marks the culmination of years of collaborative development by a team of statisticians from multiple pharmaceutical companies and academic institutions. We had previously the joy of sharing a bit of the story of the collaborative crmPack development at the ISCB 2022 conference (Boix and Günhan 2022), and we are going to describe this again here to celebrate the team’s success!

We also describe the innovative funding model to ensure the sustainability, documentation and training of crmPack and to develop new functionalities, which has the potential to make crmPack the new standard for adaptive dose escalation trials. The model is along the lines of the rpact funding model (see RPACT SLA), which has been successful for over 8 years now.

Acknowledgments

First, we want to acknowledge all the contributors who have made crmPack what it is today:

From Bayer: Clara Beck, Oliver Boix, Prerana Chandratre, Robert Adams, Dimitris Kontos (ClinBAY/Bayer)

From Genentech: Jiawen Zhu, Ziwei Liao

From Roche: John Kirkpatrick (now Astellas), Giuseppe Palermo, Guanya Peng, Doug Kelkhoff, Wojciech Wojciak (now independent), Uli Beyer

From Merck: Marlene Schulte-Goebel, Burak Kuersad Guenhan

From Academia: Wai Yin Yeung (University of Lancaster, now Roche), Thomas Jaki (University of Lancaster/Cambridge/Regensburg)

Adaptive Dose Escalation Trials

Adaptive dose escalation trials play a crucial role in the early stages of drug development. They aim to determine the appropriate dosage for a new drug in humans by gradually increasing the dosage in a stepwise manner. These trials gather data on both efficacy and safety of the drug, balancing therapeutic effect with the avoidance of excessive toxicity.

Conducted with a small overall sample size (typically 20-50 patients), these trials allow close monitoring and analysis of the drug’s performance. They are applied across various therapeutic areas and are not restricted to oncology.

Common design approaches include:

  • Rule-based designs: such as the traditional 3+3 design
  • Model-assisted designs: like the Bayesian Optimal Interval Design (BOIN)
  • Model-based designs: including the Continual Reassessment Method (CRM)

crmPack is focused on model-based designs, but also includes rule-based designs for comparison. We also plan to add model-assisted designs in future releases.

Illustration of a Dose Escalation Trial

We are using crmPack here to illustrate a typical dose escalation trial’s course:

library(crmPack)
myData <- Data(
    x = c(1, 1, 1, 3, 3, 3, 6, 6, 6),
    y = c(0, 0, 0, 0, 0, 0, 0, 1, 0),
    doseGrid = c(
        1, 3, 6, 12, 24
    )
)
plot(myData, blind = TRUE) + ylab("Dose")

Example of a dose escalation trial’s course

We see that 3 patients each were treated sequentially at the dose levels 1, 3, and 6, with one dose-limiting toxicity (DLT) observed at dose level 6. Dose escalation designs answer the question of which dose to recommend for the next cohort of patients based on the observed data, and how to estimate the maximum tolerated dose (MTD) at the end of the trial.

If you’d like to try crmPack to do so, please check out our introductory example!

The R Package crmPack

The crmPack package is a specialized R package for dose escalation trials. Its initial CRAN release was in 2016, making it widely accessible to the R community. The package offers higher flexibility compared to other software, thanks to its modular design principles using S4 classes.

Key features include:

  • Easy extensibility and adoption of new designs
  • Visual and numeric output for clear and intuitive presentation of trial results
  • Simulation capabilities for evaluating various scenarios and assessing the performance of different dose escalation strategies

Framework Architecture

crmPack provides a highly flexible framework for the design and analysis of dose escalation trials. The framework’s key components include data handling, model specification, design rules, and stopping criteria, all working together to provide comprehensive trial design capabilities.

crmPack Framework

Development History

The development of crmPack has been a collaborative effort spanning over a decade:

  • 2014: Development started at Roche (Daniel Sabanés Bové)
  • 2014-2016: Collaboration with Prof. Thomas Jaki and Wai Yin Yeung (University of Lancaster)
  • 2016: Open-sourced on CRAN (link)
  • 2019: Academic paper published in Journal of Statistical Software (link)
  • 2021: Open source team collaboration began (Bayer, Merck, Roche, Cambridge University)
  • 2024: Daniel left Roche and co-founded RCONIS
  • 2025: Start of the new collaborative funding model and release of version 2.0 on CRAN

Landscape of Available Software

Several alternative options exist for implementing dose escalation trial design:

Proprietary software: FACTS, East Horizon

Websites: TrialDesign.org, MD Anderson Biostatistics Software

Open source R packages:

  • Model-based (CRM designs): bcrm, blrm, dfcrm, OncoBayes2
  • Model-assisted: BOIN
  • Wrapper packages: escalation

Advantages of crmPack

crmPack stands out by:

  • Combining model-based and rule-based designs in one package
  • Offering more flexibility and extensibility
  • Allowing users to easily define custom models, designs, and stopping rules

Package Usage and Impact

The following chart shows the median daily downloads of crmPack from CRAN each month since 2016:

Monthly medians of daily downloads of crmPack from CRAN (obtained via cranlogs)

We aim to increase the visibility and usage of crmPack further through improved documentation, training, and support - as well as adding significant new features through the collaborative funding model.

User Testimonials

The impact of crmPack is best illustrated through feedback from its users:

Burak and Marlene (Merck):

“We already used and continuously use successfully for clinical trials in regulatory context, and we appreciate the flexibility that you can adapt it to your needs. As we were part of development team, we are aware that development version has a high software quality. We have very positive experience collaborating and working with the lead developer (Daniel) over the years.”

Dimitris and Oliver (Bayer):

“Implementation of a dose escalation study can be achieved without requiring in-depth knowledge of R programming. Comprehensive supporting documentation, available vignettes and well structured code base enables easy implementation of new design features. Collaboration with the crmPack development team has been rewarding over the past few years.”

Uli (Roche):

“We started to use CRM dose escalation designs in early development oncology and non-oncology more than 10 years ago. At the beginning crmPack was not yet available and especially the simulation part was a little cumbersome. In the meantime, crmPack is our standard package used for CRM dose-escalation methods. Based on crmPack we also developed templates for the protocol sections and appendix, which includes simulations for different assumed dose-toxicity relationships. With the help of crmPack a model based dose escalation design is no longer a burden also for new-comers and colleagues who have not worked in early development before.”

Anonymous (CRO):

“Running a CRM trial rather than a 3+3 saved us a year in development time and almost 1 million USD (!) as well as providing greater insight to the relationship between dose and toxicity.”

Current Package Status

The current CRAN release 2.0 of crmPack includes an impressive array of features:

  • Outcomes: Binary, time-to-event, dual efficacy/safety, ordinal outcomes
  • Models: 15 different model options
  • Recommendation options: 15 different approaches
  • Stopping rules: 19 different criteria
  • Cohort size rules: 9 options
  • Increment rules: 9 different approaches
  • Reporting: Full functionality using knitr
  • Documentation: 9 comprehensive vignettes

The Funding Model

The now active funding model follows the successful rpact approach via a Service Level Agreement with RPACT:

  • Proven track record: This model has worked successfully for 8 years with rpact
  • Simple procurement: Short contract (couple of pages) that is easy for company procurement processes
  • Collaborative approach: Multiple pharma companies and CROs together fund the maintenance and development
  • Existing infrastructure: Utilizes the existing RPACT company and RCONIS joint venture
  • No new vendors needed: Uses existing validation infrastructure

Currently, 4 pharmaceutical companies have joined the crmPack funding initiative - and of course there is room for more!

Service Level Agreement Details

Included services:

The annual flat fee includes the following services:

  • Maintenance of the package with regular updates
  • Comprehensive support for the package
  • Development of new open source features
  • Training of new users in participating companies
  • Formal GxP-compliant validation documentation
  • Installation Qualification on company servers

Optional services:

  • In-house graphical user interfaces
  • Automation support
  • Private (not open source) extensions
  • Customer-specific programs

Conclusion

The crmPack package represents a decade of collaborative development and has proven its value in real-world clinical trial applications. With the collaborative funding model, we can ensure its continued development, maintenance, and growth as the standard tool for adaptive dose escalation trials.

The success of this collaborative model with rpact demonstrates that this approach works effectively for the pharmaceutical industry. By participating in this funding initiative, companies can secure access to cutting-edge dose escalation methodology while contributing to the broader statistical community.

We invite interested pharmaceutical companies and CROs to join this collaborative effort to secure the future of crmPack and advance the field of adaptive dose escalation trial designs - please reach out to if you are interested in a first informal chat!

References

Boix, Odile, and Burak Kürsad Günhan. 2022. “A Collaborative Approach to Software Development; the crmPack Experience.” In ISCB 2022 Conference. Newcastle, UK. https://www.burakguenhan.com/talk/crmpack/.
Sabanés Bové, Daniel, Wai Yin Yeung, Giuseppe Palermo, and Thomas Jaki. 2019. “Model-Based Dose Escalation Designs in R with crmPack.” Journal of Statistical Software 89: 1–22. https://doi.org/10.18637/jss.v089.i10.