Ask Mallard
Core concept

Choosing a dose-finding design

4 min read

Choose a dose-finding design by the safety target, activity goal, follow-up time and behavior under simulated scenarios.

Fit

When it matters, and when it does not

It matters when

  • You need to choose a dose or schedule before a larger efficacy trial.
  • You can monitor harm promptly and define rules for escalation, stopping and dose selection.

It matters less when

  • You need to model how an outcome varies across measured doses. Consider dose-response regression. Read: Choosing a dose-response analysis
  • You have chosen a dose and need to screen its response rate. Consider a single-arm phase II trial.

Before you start

What to check in your study

Decide these first
  1. The dose-selection goalSpecify toxicity, activity, exposure or their trade-off. The goal depends on the intervention.
  2. Dose levels, schedules and observation windowsSet starting doses, permitted jumps and when enough follow-up is available to escalate.
  3. Allocation, stopping and expansion rulesDefine the safety oversight and the complete decision process, including delayed harms and missing outcomes.
It rests on these
  1. The assumed dose ordering fits the intervention.How to check: Many oncology toxicity models assume harm rises with dose. Activity may plateau, and schedules may not share that ordering.
  2. Safety decisions account for incomplete follow-up.How to check: A pending toxicity outcome is not absence of toxicity. Use a method designed for delayed events or pause enrollment.
  3. The complete design has acceptable behavior under plausible scenarios.How to check: Simulate correct dose selection, exposure to unsafe doses and early stopping. Vary the dose-response curve and accrual rate.

Plain language

How it works

Rule-based, Model-based and Model-assisted approaches use different information to guide allocation. Compare their safety constraints, dose-selection target and ability to handle delayed harm. Use the 3+3, continual reassessment and BOIN pages for their procedures. Compare the complete designs through simulation before recruitment.

Selection is not confirmation

Escalation explores doses; expansion gathers more data at selected doses. Selection remains uncertain, and expansion alone does not confirm benefit. Toxicity-focused oncology designs often target a maximum tolerated dose. Other treatments may need an optimal biological dose, joint activity and safety goals, or a schedule comparison.

The structure of the approach
Prespecified safety rules govern each level decision. Pending follow-up is not absence of toxicity. The complete design needs simulation under plausible scenarios. Selecting or expanding a level does not confirm efficacy.Prespecified safety rules govern each level decisionDefine target and safetyrulesObserve outcomes andpending follow-upApply the chosen designand safety constraintsEscalate only whenpermittedRetain or reduce levelStop when requiredPending follow-up is not absence of toxicity. The complete design needs simulation under plausiblescenarios. Selecting or expanding a level does not confirm efficacy.
Conceptual diagram. It shows the structure of the idea, not data from any study.

Worked example

A worked example

Invented for illustration
Setting
An invented oncology cohort receives one dose with a prespecified toxicity window.
Question
What does the observed toxicity proportion tell you?
Approach
Wait for all patients to complete the window. Record 3 dose-limiting toxicities among 12 patients.
Result
The observed proportion is 25 in 100, from 3 of 12 patients.
What it means
That is a descriptive proportion, not proof that the dose meets the target. Follow the chosen design; do not invent an escalation rule.

In practice

Recent studies PubMed search · not vetted

Searching PubMed…

Watch out

Common mistakes

The one to remember

A dose selected from a small early-phase trial is uncertain. The highest tolerated dose need not give the best benefit-risk balance, and an expansion cohort does not prove efficacy.

  • Assuming the familiar 3+3 rule reliably selects the target dose.Instead: Compare its behavior with model-based or model-assisted designs through simulation.
  • Treating maximum tolerated dose as the best dose for every therapy.Instead: Oncology toxicity targets do not fit every intervention. Consider activity, exposure, chronic harm and patient burden.
  • Using a dose-response regression as a dose-allocation engine.Instead: A fitted curve alone supplies no safety oversight, escalation rule or stopping policy.

Inside Mallard

How Mallard uses this

Plans do not link to this concept yet. Mallard has no sample-size form for dose-finding and early-phase trials. A plan sized by simulation gets no headline sample size; Mallard does not run that simulation.

No Mallard code example points to this page.
Atlas node concept.choosing_dose_finding, the id a plan cue records.

Related Learn pages

Related designs and methods

Evidence

Sources & further reading

  1. Continual reassessment method: a practical design for phase 1 clinical trials in cancer (opens in a new tab) · Biometrics
  2. Sequential designs for phase I clinical trials with late-onset toxicities (opens in a new tab) · Biometrics · doi 10.1111/j.0006-341x.2000.01177.x
  3. Dose-finding based on efficacy-toxicity trade-offs (opens in a new tab) · Biometrics · doi 10.1111/j.0006-341X.2004.00218.x
  4. Bayesian Optimal Interval Design: A Simple and Well-Performing Design for Phase I Oncology Trials (opens in a new tab) · Clin Cancer Res · doi 10.1158/1078-0432.CCR-16-0592
  5. Enhancing reporting quality and impact of early phase dose-finding clinical trials: CONSORT Dose-finding Extension (CONSORT-DEFINE) guidance (opens in a new tab) · BMJ · doi 10.1136/bmj-2023-076387
Prepared with AI assistance. Source titles, authors, publication years and available DOI metadata were checked against PubMed on 2026-11-06. No independent methodological approval is claimed. O'Quigley et al. (1990) introduce continual reassessment; PubMed lists no DOI for that record, so none is supplied. Cheung and Chappell (2000) address delayed toxicities. Thall and Cook (2004) develop joint efficacy-toxicity dose selection. Yuan et al. (2016) give the BOIN decision boundaries and simulation comparisons. Liu S and Yuan Y (2015), Bayesian optimal interval designs for phase I clinical trials, Journal of the Royal Statistical Society: Series C, is non-indexed in PubMed and has no PMID; it is named here as the original BOIN paper, not as a PubMed-verified source. Reporting guidance: CONSORT-DEFINE (Yap et al., 2023) addresses early-phase dose-finding reports. A reporting checklist is not a design manual. Record verified, confirm it supports this use. Source records rechecked against PubMed on 2026-11-07 UTC.
Last source review: 2026-11-07
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.