The Continual Reassessment Method (CRM) is a model-based dose-finding approach for early clinical trials. It combines a pre-specified dose–toxicity relationship assumed before the study starts with the toxicity data newly available after each patient cohort. On this basis, the next dose is selected so that its estimated probability of toxicity is as close as possible to a pre-defined target toxicity.
Model-based dose selection
Before recruitment, an ordered dose sequence is defined and linked to an initial assumption about its toxicities. The model updates this relationship as soon as evaluable dose-limiting toxicities are available. The next cohort does not automatically receive one level higher or lower; instead, it receives the dose whose updated toxicity probability is closest to the target and that meets the specified safety rules.
CRM thus identifies a dose that is close to a maximum tolerated dose in terms of the target toxicity. It does not demonstrate that this dose is the best biological or therapeutic dose. Pharmacokinetics, pharmacodynamics, antitumour activity, and long-term safety can provide additional, separately planned information for further dose decisions.
Prerequisites for safe conduct
The target value for dose-limiting toxicity, the observation window, the definition of a DLT, and the permitted dose transitions must be specified before the first participant. If a DLT cannot yet be definitively assessed, dose assignment for the next cohort cannot be handled as if complete toxicity data were available. The rules for pending observations are part of the design.
Additional safeguards can prevent skipping untested doses, provide overdose control, or pause recruitment in the event of an unfavourable toxicity profile. Their specific design follows the investigational product, the patient population, and the risk, rather than a universal CRM standard. Pre-study simulations assess how the chosen design responds under different plausible dose–toxicity trajectories.
Distinction from the 3+3 design
CRM is the counterpart to the rule-based 3+3 design. A 3+3 design uses predefined counting rules within small cohorts to decide on escalation, repetition, or de-escalation and does not use a statistical dose–toxicity model. CRM, by contrast, uses all toxicity information available up to that point to update the entire assumed curve.
The existing entries dose-escalation, dosiseskalation, and maximum-tolerated-dose describe the process or objective of dose finding. CRM refers to the specific decision logic behind selecting the next dose. Model-assisted methods such as BOIN are methodologically between rigid rules and a fully model-based approach, but are not identical to CRM.
A CRM must also specify whether and how previously treated patients with an incomplete DLT window are included. If such data are ignored, the dose decision is delayed; if they are treated as complete without a rule, toxicity may be underestimated. Time-weighted variants or recruitment restrictions are decisions of the specific design and must be described in advance.
The initial dose–toxicity model is not a substitute for a rational starting dose. Preclinical data, existing human experience, and the risk assessment still determine which doses are included in the study at all. CRM controls assignment within this evaluated set.
Relevance for clinical trials
For a CRM study, the clinical team, safety committee, and statisticians must promptly clarify whether an observed toxicity meets the DLT definition and whether its observation window has been completed. The dose decision is documented for each cohort according to the pre-agreed rules. This documentation must show which data were available, which model update was performed, and which safety rule limited the dose actually administered.
Full-service CROs such as Mediconomics support the development of the dose-finding protocol section, the definition of DLT capture processes, and the statistical simulation and programming of dose decisions. Project management, medical monitoring, and biostatistics can coordinate cohort release and create a transparent decision record for safety committees and subsequent regulatory assessment.
Before each dose assignment, the responsible roles, data updates, and documentation timelines must be operationally defined. A retrospective model calculation cannot make a cohort decision taken under different information traceable.
Frequently Asked Questions (FAQ)
Does each new cohort receive a higher dose in CRM?
No. After the model update, the same dose may be repeated, a lower dose selected, or escalation limited. The direction follows the estimated toxicity and the pre-defined safety rules.
Is the dose selected by CRM automatically the recommended Phase II dose?
No. Proximity to the target toxicity is an important selection criterion, but it is not, on its own, proof of a favourable benefit–risk profile. Additional data on exposure, activity, and tolerability may influence the development decision.
Why are simulations important for CRM design?
They assess, before the study, how often the design selects an appropriate dose or assigns patients to an overly toxic dose under different true dose–toxicity curves. This allows the model choice and protection rules to be verified before use.
Regulatory References
- ICH E20, Adaptive Designs for Clinical Trials, Step 2b – describes pre-planned adaptations based on interim analyses as a draft.
- CHMP/EWP/2459/02, Reflection Paper on Methodological Issues in Confirmatory Clinical Trials Planned with an Adaptive Design – addresses prerequisites and risks of adaptive changes.
- ICH E9(R1), Addendum on estimands and sensitivity analyses – calls for a precise, pre-defined analysis and endpoint.
