03 · What You Need to Know
How Predictable Allocation Threatens Randomized Trials
Why Is Random Sequence Generation Not Enough?
Random sequence generation determines the treatment assignments using an appropriate chance mechanism. Allocation concealment prevents people who can influence enrollment from knowing or predicting those assignments before participants are irrevocably entered into the randomized comparison.
These safeguards address different parts of the process.
A computer can generate a perfectly valid random sequence, but if the resulting list is visible to recruiters, the assignments may influence which participants are enrolled and when.
For example, a recruiter who believes that a particularly promising participant should receive the experimental intervention might delay enrollment after learning that the next assignment is to the control condition.
Such behavior can create systematic baseline differences between treatment groups. It may occur intentionally, but subtler changes in eligibility interpretation or recruitment timing can also arise without deliberate misconduct.
The purpose of allocation concealment is to prevent assignment information from influencing those decisions.
How Can Fixed Block Randomization Become Predictable?
Permuted-block randomization maintains specified treatment allocation numbers within each completed block.
In a two-arm trial with equal allocation and blocks of four, every completed block contains two assignments to intervention A and two to intervention B.
The sequence within each block is randomized. Possible arrangements include AABB, ABAB, ABBA, BAAB, BABA, and BBAA.
However, if a recruiter knows the block size, the beginning of each block, and the assignments already released, the remaining possibilities become progressively restricted.
For instance, if the first two assignments in a block are A and A, the final two must be B and B.
At that point, the next assignment is fully predictable despite the block having been generated randomly.
Predictability is not limited to the final assignment in a block. Knowledge of earlier allocations can change the probabilities of subsequent assignments before the block is completed.
How Can Assignment Probabilities Change Within a Block?
Consider a randomly permuted block of four containing two A assignments and two B assignments. Suppose the first participant has been assigned to A.
Three assignments remain: one A and two B. Conditional on the information available, the probability that the next assignment is B is therefore two-thirds rather than one-half.
This calculation illustrates why an allocation scheme can be random at generation but partially predictable during implementation.
It does not mean that every blocked trial is compromised. The risk depends on what information recruiters can access or infer.
Why Are Small Fixed Blocks Especially Vulnerable?
Small fixed blocks maintain close numerical balance during recruitment, but their limited number of possible arrangements can make future assignments easier to anticipate.
The risk is particularly relevant in open-label trials where recruiters may observe previous assignments and know that a fixed block size is being used.
For example, in a block of two with equal allocation, knowing the first assignment reveals the second assignment immediately if the block boundaries are known.
Larger blocks generally offer more possible sequences, although predictability may still arise as a block approaches completion.
The appropriate block size should therefore balance numerical allocation requirements with the risk that assignment information could be inferred.
The broader comparison of simple, block, and stratified randomization helps explain why allocation balance and unpredictability must be considered together.
Do Randomly Varying Block Sizes Prevent Prediction?
Randomly varying block sizes can reduce predictability by making it harder for recruiters to identify block boundaries and determine how many assignments remain in a particular block.
For example, a trial might use concealed block sizes of four and six, with the size of each block selected through a prespecified random procedure.
A recruiter observing previous assignments would then have less certainty about whether a block is approaching completion.
However, varying block sizes does not guarantee unpredictability. If the allocation sequence or block-size schedule is exposed, future assignments may still be known.
Even without direct access, information about the set of possible block sizes and previous assignments can sometimes alter the probability of the next treatment.
Randomly varying blocks should therefore be viewed as one safeguard rather than a substitute for allocation concealment.
Can Stratified Randomization Also Become Predictable?
Yes. Stratified permuted-block randomization maintains separate allocation sequences within predefined participant categories.
If small fixed blocks are used within each stratum, recruiters who know the stratum and previous assignments may be able to infer upcoming allocations.
This can be particularly concerning when only a few participants enter each stratum and recruitment personnel have detailed knowledge of the allocation procedure.
For example, a trial may stratify students by baseline achievement. If a recruiter knows that the next lower-achievement student must receive intervention A to complete a known block, that information could influence enrollment decisions.
Concealment must therefore protect the allocation sequence within each stratum, not merely the overall trial list.
Excessive stratification can also complicate allocation administration, although sparse strata and predictable assignments are distinct problems.
What Other Allocation Methods Can Be Predicted?
Predictability can arise from procedures other than permuted blocks.
Alternating assignments between intervention and control are entirely predictable and do not constitute genuine random sequence generation.
Assigning treatments according to birth dates, medical record numbers, student identification numbers, or days of the week similarly allows treatment allocation to be anticipated. Such procedures are commonly classified as quasi-random rather than genuinely randomized.
Deterministic minimization algorithms may also make assignments predictable when the relevant balancing variables and allocation history are known.
Introducing an appropriate random component can reduce predictability in covariate-adaptive procedures.
However, the random component must be implemented according to the prespecified design, and future assignments must remain protected from recruitment personnel.
How Can an Exposed Allocation List Compromise the Trial?
One of the most straightforward concealment failures occurs when the full randomization sequence is accessible before participants are enrolled.
The sequence might be stored in a shared spreadsheet, printed in an unsecured location, or included in a protocol appendix available to recruitment staff.
In such circumstances, no statistical sophistication is required to predict the next assignment. The recruiter can simply read it.
This is why random sequence generation and allocation administration should be treated as separate operational functions.
Access to the complete sequence should be limited to appropriately authorized personnel or systems, and treatment assignments should be released only after enrollment requirements have been completed.
Can Sealed Envelopes Prevent Predictability?
Sequentially numbered, opaque, sealed envelopes can provide allocation concealment when prepared and administered correctly.
However, the method is vulnerable if envelopes are insufficiently opaque, can be opened and resealed, or are selected out of sequence.
For example, a recruiter might be able to determine an assignment by holding an inadequately opaque envelope against a light source.
Even when envelopes are genuinely sealed, staff may compromise concealment by opening several in advance or failing to document the order in which they were used.
Appropriate safeguards include independent preparation, sequential numbering, secure storage, tamper-evident sealing, and opening only after eligibility and enrollment are finalized.
Merely reporting that sealed envelopes were used is not enough to establish adequate concealment.
How Does Centralized Randomization Reduce the Risk?
Centralized randomization separates enrollment from access to the complete allocation sequence.
Recruitment personnel submit the necessary eligibility and stratification information to a secure system or independent allocation service. The assignment is released only after the required enrollment steps are completed.
A properly designed system can prevent recruiters from viewing future allocations or manipulating the order in which assignments are released.
It may also maintain audit records showing when enrollment was finalized, when assignments were generated or released, and whether any unusual access occurred.
However, electronic systems are not automatically secure. Inadequate access controls, administrator privileges, or software implementation errors can still compromise allocation concealment.
The system should be tested before recruitment and monitored for departures from the intended procedure.
How Does Predictability Create Selection Bias?
Selection bias can arise when knowledge of upcoming treatment assignments influences which participants enter the trial or when they are enrolled.
Suppose a researcher knows that the next assignment is to an experimental treatment and believes that a particular participant is especially likely to benefit.
The researcher might expedite that participant's enrollment, delay another participant, or interpret an ambiguous eligibility criterion differently.
As a result, the distribution of prognostic characteristics across treatment groups may reflect recruitment decisions rather than the intended chance mechanism alone.
Such bias can undermine causal inference because treatment assignment is no longer independent of participant selection.
The problem does not require deliberate fraud. Expectations about treatments can influence decisions subtly, especially when eligibility involves professional judgment.
This is one reason adequate concealment should be maintained even when investigators believe recruitment staff will act impartially.
Does Predictability Always Mean Selection Bias Occurred?
No. A predictable assignment creates an opportunity for selection bias, but it does not prove that anyone manipulated enrollment.
The actual risk depends on whether people who could influence recruitment had access to the relevant information and whether enrollment decisions could be affected.
For example, an allocation list may be accessible to a statistician who has no involvement in participant recruitment. That access does not necessarily compromise concealment if the statistician protects the information and assignments are released securely.
Conversely, even partial predictability may be consequential when recruiters can delay or accelerate enrollment.
Risk-of-bias assessment should therefore consider both the allocation procedure and the opportunities for foreknowledge to influence recruitment.
Can Baseline Characteristics Reveal a Concealment Problem?
Unexpected baseline differences may raise concerns about the randomization process, particularly when accompanied by evidence that assignments were accessible or predictable.
However, baseline imbalance alone does not establish selection bias. Proper randomization can produce chance differences.
Similarly, well-balanced baseline characteristics do not prove that concealment was adequate.
The Cochrane RoB 2 framework evaluates sequence generation, allocation concealment, and baseline differences together when assessing bias arising from randomization.
Researchers should therefore investigate allocation logs, enrollment timing, eligibility records, and system access rather than relying on baseline significance tests.
The distinction between chance baseline imbalance and evidence of compromised randomization is essential for interpreting suspicious allocation patterns.
What Should Researchers Do if Predictability Is Discovered?
If an ongoing trial reveals that future assignments can be predicted, investigators should promptly protect the remaining allocation process and document the issue.
The research team should determine when the vulnerability began, who could access the information, whether enrollment decisions may have been influenced, and which participants or sites were potentially affected.
Appropriate trial governance, statistical, and ethics personnel should assess the consequences and any necessary corrective action.
Depending on the circumstances, corrective measures may include securing the allocation system, revising access permissions, changing future block-generation procedures through an approved amendment, or conducting sensitivity analyses.
Researchers should not silently regenerate assignments for participants already randomized or conceal the incident from the final report.
Any changes should preserve participant safety and be documented sufficiently to support a transparent assessment of trial integrity.
