Statistics sit at the heart of every early-phase clinical trial. From dose escalation to determining the Maximum Tolerated Dose (MTD), statistical decisions guide how evidence is generated, interpreted, and acted upon. These choices influence patient safety, trial efficiency, and the likelihood of identifying the right dose to take forward.
In this blog, we explore some of the key statistical concepts that underpin early-phase design, including model-based/model-assisted methodology, operating characteristics, and the importance of thorough documentation and collaboration.
Beyond the 3+3: Moving Towards Model-Based and Model-Assisted Designs
Traditional rule-based designs, such as the 3+3 method, have long been used to determine the MTD. While simple, they can be inefficient and less accurate, particularly when investigating complex or targeted therapies.
In practice, statisticians consider a range of options when selecting a dose-escalation design. Rule-based methods are straightforward and easy to implement, but may not make the best use of accumulating data. Model-based designs, such as CRM, Escalation With Overdose Control (EWOC), or Bayesian logistic regression models (BLRM), use statistical models to estimate the relationship between dose and toxicity or efficacy, updating these estimates as new data emerge.
A hybrid approach, known as model-assisted design, combines the logistical simplicity of rule-based methods with the statistical rigor of model-based approaches. Examples include the BOIN (Bayesian Optimal Interval) and Keyboard designs. These often offer strong performance while being easier to operationalize during a live trial.
Simulation studies comparing CRM with 3+3 typically show that CRM identifies the correct dose more often and does so with fewer participants. Although CRM may slightly increase the probability of overdosing in some cases, this is often outweighed by its improved ability to select the correct MTD, making it a valuable choice for many modern trials.
Evaluating Operating Characteristics
When designing an early-phase study, it is not enough to simply select a method. Each design should be evaluated through simulation to understand its operating characteristics, how it performs under different assumptions and scenarios.[1]
These characteristics include the probability of identifying the correct dose, the likelihood of overdosing or underdosing, and the expected number of patients exposed at each level. Reviewing these results helps sponsors and statisticians understand trade-offs and agree on what is acceptable in practice. Across many early phase programs, our experts have supported design decisions by critically assessing multiple methodological options and their implications for safety, efficiency, and interpretability.
This process requires close collaboration between statisticians and clinicians to ensure the chosen design reflects both scientific ambition and ethical responsibility.
Documenting the Statistical Approach
Once the design has been chosen, it is vital to clearly document the rationale, methods, and decision rules. Statistical detail should be captured in both the protocol and the statistical analysis plan (SAP), with clear guidance on how dose-escalation decisions will be made and how data will be summarized.
Two key references provide helpful direction here:
- The SPIRIT extension for dose-finding trials outlines additional items that should be included in the protocol. [2]
- Early phase SAP guidance, developed with input from the MHRA, provides best practices for writing analysis plans in early-phase settings. [3]
Comprehensive documentation ensures transparency, regulatory compliance, and reproducibility, setting the foundation for confident decision-making as the trial progresses.
Conclusion
Statistics are not an afterthought in early-phase development, they are integral to its success. From selecting designs that balance safety and efficiency, to understanding operating characteristics and maintaining transparent documentation, robust statistical thinking drives better outcomes.
Phastar’s statisticians have extensive experience designing and analyzing early-phase studies across therapeutic areas. We support sponsors in exploring innovative methods, running simulations, and ensuring every design choice is backed by solid evidence and clear reasoning.
References
1. Wages, N. A., Horton, B. J., Conaway, M. R., & Petroni, G. R. (2021). Operating characteristics are needed to properly evaluate the scientific validity of phase I protocols. Contemporary Clinical Trials, 108, 106517. https://doi.org/10.1016/j.cct.2021.106517
2. Yap, C., Rekowski, J., Ursino, M., Solovyeva, O., Patel, D., Dimairo, M., Weir, C. J., Chan, A.-W., Jaki, T., Mander, A., Evans, T. J. R., Peck, R., Hayward, K. S., Calvert, M., Rerhou Rantell, K., Lee, S., Kightley, A., Hopewell, S., Ashby, D., Garrett-Mayer, E., Isaacs, J., Golub, R., Kholmanskikh, O., Richards, D. P., Boix, O., Matcham, J., Seymour, L., Ivy, S. P., Marshall, L. V., Hommais, A., Liu, R., Tanaka, Y., Berlin, J., Espinasse, A., & de Bono, J. (2023). Enhancing quality and impact of early phase dose-finding clinical trial protocols: SPIRIT Dose-finding Extension (SPIRIT-DEFINE) guidance. BMJ, 383, e076386. https://doi.org/10.1136/bmj-2023-076386
3. Homer, V., Yap, C., Bond, S., Holmes, J., Stocken, D., Walker, K., Robinson, E. J., Wheeler, G., Brown, S., Hinsley, S., Schipper, M., Weir, C. J., Rantell, K., Prior, T., Yu, L.-M., Kirkpatrick, J., Bedding, A., Gamble, C., & Gaunt, P. (2022). Early phase clinical trials extension to guidelines for the content of statistical analysis plans. BMJ, 376, e068177. https://doi.org/10.1136/bmj-2021-068177
