Umbrella trial
Profile one disease, then route patients to marker-defined treatment substudies whose allocation and comparisons are planned separately.
Why this level?
The design family and its defining features have not been specified for this hierarchy. NHMRC 2009 sourceUmbrella describes the routing structure, not one allocation method. A randomized substudy with concurrent controls matches intervention level II; an uncontrolled substudy matches level IV.
Where the NHMRC 2009 hierarchy names this design, it places it here:
| Research question | Level | When |
|---|---|---|
| Intervention | II | Participants or clusters are randomly allocated to concurrent comparison groups. |
| Intervention | IV | A single case series has post-test or pre-test/post-test outcomes, without a comparison group. |
Fit
When to use it, and when not to
Use it when
- One disease has several plausible targets and matching treatments.
- You can deliver timely profiling and run several defined substudies.
Not when
- You have one marker across several tumor types. Use a basket trial.
- You need only one fixed treatment comparison. Use a parallel randomized trial. Read: Randomized controlled trial
Before you start
What you need
- The disease population and profiling assayState disease stage, previous care and tissue requirements.
- The substudy assignment rulesDefine marker thresholds, multiple matches, assay failure and no-match pathways.
- The control and allocation within each substudyassignmentSpecify which substudies randomize and which only describe activity.
- The screening-to-enrollment pathwayPlan for marker rarity, tissue failure, delays and patients becoming ineligible.
- Assumptions and methods for missing dataCompare missing completely at random, missing at random and missing not at random. Describe which data are absent.
- The assay selects the intended marker groups.How to check: Check test quality and record missing or failed profiles.
- Each substudy comparison supports its stated claim.How to check: Randomize within marker groups for a treatment effect where feasible. Across-group comparisons need stronger assumptions.
- Screening losses are visible.How to check: Report all screened patients, matches, failures and enrolments. Excluding missing profiles can select a different population.
- Missing at random may be plausible after recorded predictors are included.How to check: Consider multiple imputation for missing covariates under missing at random. Multiple imputation is not inherently better.
- Missing not at random remains possible.How to check: Use a delta or tipping-point sensitivity analysis for missing not at random.
Plain language
How it works
Start with one disease population and test for several markers. Use planned rules to route patients to the relevant treatment substudy. Define what happens when patients have several matches or none.
Each substudy has its own allocation, control, outcome and stopping plan. Molecular matching does not create randomization. Report the full screening pathway so readers can see who was tested but never enrolled. Interpret each treatment result using its own comparison design. Use CONSORT for randomized substudies. State the allocation and control for every other substudy.
Worked example
A worked example
- Setting
- A lung cancer program screens 100 people for treatment substudies.
- Question
- How many reach a rare-marker substudy, and what activity is seen?
- Approach
- Ten match the marker and eight start treatment. Follow the eight with a prespecified response assessment.
- Result
- There are 2 responses among 8 treated patients, or 25 in 100.
- What it means
- Only 8 of 100 screened people started treatment. This describes one selected substudy, not a comparison with another marker group.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Marker matching is not randomization. Each substudy needs its own control and evidence appraisal; differences between marker groups cannot be read as treatment effects.
- Comparing response rates across marker groups as if treatments were randomized.Instead: Compare treatment with its control within the relevant substudy.
- Counting all screened people as available for each arm.Instead: Use marker prevalence, assay success and enrollment losses in recruitment planning.
- Dropping patients whose exposure or predictors could not be measured.Instead: This can cause selection bias. Retain people with multiple imputation when missing at random is plausible; assess departures.
Inside Mallard
How Mallard uses this
Plans do not link to this design yet. Mallard has no dedicated sizing form for umbrella trial.
design.umbrella_trial, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- The BATTLE trial: personalizing therapy for lung cancer (opens in a new tab) · Cancer Discov · doi 10.1158/2159-8274.CD-10-0010
- Biomarker-driven therapies for previously treated squamous non-small-cell lung cancer (Lung-MAP SWOG S1400): a biomarker-driven master protocol (opens in a new tab) · Lancet Oncol · doi 10.1016/S1470-2045(20)30475-7
- CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials (opens in a new tab) · BMJ · doi 10.1136/bmj.c332
- Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls (opens in a new tab) · BMJ · doi 10.1136/bmj.b2393