EN
Русский English Français
Sign in Create account
YouEDC Capture

Randomisation

Methods, the scheme editor and the simulator, signatures, randomising a patient, blinding, emergency unblinding and errors.

The Randomisation section of YouEDC Capture assigns patients to treatment arms the way the protocol asks: method, allocation ratio, stratification factors, blinding. Everything is recorded, signed and can be checked by replaying it.

The path of a scheme 1 Draft of the scheme arms, method, factors, conditions, parameters 2 Simulation hundreds of invented studies on your factors 3 Two signatures the author signs, a second person approves 4 Test mode while the study is set up: numbers TEST-0001 5 Study launch test allocations removed; real numbers R0001
A scheme changes only through a new version with a reason and two signatures again; randomised patients stay in their arms.

Methods

  • Minimisation (Pocock–Simon): each new patient goes to the arm that least disturbs the balance by their site, sex and other factors, with probability p (usually 0.8). Good with many factors and few patients.
  • Permuted blocks, with or without strata: each block holds the arms in the set ratio, the block size is drawn from a few sizes. A stratum is a combination of the levels of all factors.
  • Simple randomisation: a coin weighted by the ratio each time. For large studies.
  • The statistician’s list: a ready list (CSV or Excel), used in order within each stratum.

Any ratio (1:1, 2:1, 1:1:1), up to six arms and six factors. A factor comes from the patient record (site, sex, age) or from an eCRF field (choice, yes/no, or a number with cut points such as HbA1c ≤ 8 and > 8).

Setting up the scheme

  1. Open Randomisation → Set up the scheme. Seven steps: arms and ratio, method, factors, conditions before randomisation, parameters, simulation, signatures.
  2. Under conditions, tick the forms that must be complete with no open queries, and field values (for example, “eligible = Yes”). Until a condition holds, the randomise button does not work.
  3. Press Run the simulation: the system plays hundreds of invented studies with your factors and compares the methods: how far the arms drift apart overall and in the worst stratum, and how often the next arm can be guessed. Advice on the method sits beside it.
  4. The author presses Sign as the author with their password; a second person approves. In a blinded study only an unblinded member (usually the statistician) approves. The approved blinded specification goes to the eTMF inbox on its own.
The scheme editor and the simulation
The scheme editor: the method, the factors and the methods compared in the simulation.
Minimisation: a worked example New patient: site 01, female, HbA1c 9.1% (level L2, factor weight 2). Arms A and B 1:1. Patient’s level A B if in A if in B SITE=01 2 0 SEX=F 1 1 HBA1C=L2 ×2 2 6 Score (weighted sum of differences) 5 7 A has the lower score: A gets probability 0.8, B 0.2. Random number 0.6137 is below 0.8 → the patient goes to arm A. The number is stored for the replay check.
For each factor: how far the arms would drift apart in this patient’s level if the patient went to A or to B. The arm with the lower score is preferred, but not certain, so the allocation cannot be guessed.

Randomising a patient

The patient page has a randomisation block on the right. The investigator presses Randomise, checks the factor values against the source documents, confirms with the password and gets the number (R0001, R0002…). A patient cannot be randomised twice: the number is final.

While the study is being set up, randomisation works in test mode from a separate test list: numbers TEST-0001. At the launch, test allocations are removed with the test patients and real numbering starts from one.

Randomisation on the patient page
The patient page: the randomisation number, the hidden arm, emergency unblinding and recording an error.

Blinding and emergency unblinding

Who sees what in a blinded study Investigator R0007 hidden number and date arm hidden Monitor, manager, DM R0007 hidden number, stratum, log arm and blocks hidden Unblinded statistician R0007 YE-101 arm, balance, list, full specification
The arm never goes into the audit trail, the letters or the blinded specification. An investigator who broke the code in an emergency sees the arm of that one patient afterwards.

When the treatment must be known to treat the patient now, the investigator presses Emergency unblinding, picks the reason, describes what happened, types the patient ID again and the password. Only they see the arm. Managers, data managers and the safety team are told at once that the code was broken, without the arm, and the patient’s open YouEDC Safety cases get a comment.

Errors

A randomisation error is not fixed by randomising again (as the ITT analysis guidance recommends). Press Record an error: wrong stratum, wrong patient, not eligible, or other. The note shows in the log and in the export (column RNDNOTE).

Checks and exports

  • Replay check of every allocation recomputes each allocation from the scheme’s secret seed, the list and the stored random numbers and compares with the log. The database refuses any change or deletion of log rows.
  • Why this arm (for unblinded members) shows the calculation for one number.
  • The randomisation log in CSV for SDTM DM (RANDNO, RANDDTC, STRAT_*, and for unblinded members ARMCD and ARM), the specification in DOCX (blinded or full with the seed), the allocation list, the balance by factor and the text for the publication by CONSORT 2010.
Why the patient got this arm
“Why this arm”: the score of each arm, the probabilities and the random number.

An example from the demo data

In the demo study of YE-101 against placebo, patients are allocated by minimisation on site, sex and HbA1c at screening (weight 2, cut point 8%). A data manager wrote the scheme and the unblinded physician approved it. Open Randomisation to see the log and the balance, and the page of any randomised patient.