Why Preferences and Fair Access Matter
An allocation rule can be statistically well specified and still raise ethical concerns for participants. A person may strongly prefer one treatment because of its route, schedule, or perceived acceptability. Another may be unable to reach the clinic needed to receive a treatment. These issues matter to the ethical case for adaptive allocation, but they do not by themselves show that one treatment is better or that its assignment probability should be increased.
The earlier material on clinical equipoise and on why equal allocation can be ethically attractive provides relevant background: allocation decisions are made under uncertainty, and fairness is one consideration among several. Here the focus is narrower. We ask what preferences tell us, what fair access requires, and how to distinguish a fair opportunity to participate from a particular allocation ratio.
Preferences are ethically relevant because trial participation involves real people, not interchangeable units. They can help investigators understand which treatment features matter to participants and whether the trial's options are acceptable. But preferences vary: some people have a strong preference, some have a mild one, and some are indifferent. A preference expressed by people who enroll may not describe those who decline or cannot access the trial.
Fairness also has more than one dimension. A useful assessment distinguishes fair access to the trial, fair opportunity to receive each treatment among participants for whom it is appropriate, and fair treatment of people who face practical barriers. Equal assignment probabilities can address one part of this assessment, but cannot alone ensure that people have an equal chance to enroll or can actually receive the assigned treatment.
Preferences Are Relevant, but Not Decisive
A preference can matter to the ethical assessment in at least three ways. It may reveal a feature of treatment burden that investigators should take seriously; it may affect whether a participant is willing to enter or remain in a randomized trial; or it may expose a mismatch between the trial options and participants' values. These are reasons to measure and consider preferences carefully. They do not turn a preference survey into a comparative-effectiveness study.
In particular, an aggregate preference share should not automatically become an allocation ratio. If more people prefer Arm A than Arm B, that fact alone does not show that assignment to A produces more benefit, that all participants share the preference, or that those assigned to B are treated unfairly. A preference may concern convenience rather than clinical outcomes. Conversely, equal assignment odds do not erase a substantial burden associated with one treatment or a barrier that prevents some groups from participating.
The ethical question is therefore not simply “Which arm is preferred?” It is whether the trial has understood whose preferences are represented, what those preferences concern, and how the allocation procedure treats participants with different preferences. Any change to assignment probabilities needs its own justification; preference information is context for that judgment, not a substitute for it.
Worked Example: A Preference Survey Does Not Set the Allocation Ratio
Suppose, hypothetically, that 80 people eligible for a two-arm trial are asked about treatment preference before enrollment. Of those, 48 prefer Arm A and 32 prefer Arm B. Assume all 80 are willing to be assigned to either arm, both arms are suitable for them, and the trial uses equal randomization.
The observed preference proportions are:
With equal assignment probabilities, each enrolled participant has probability \(0.5\) of assignment to either arm. Across 80 assignments, the expected arm counts are:
These are expected counts, not guaranteed counts. The preference proportions, 60.0% and 40.0%, describe this hypothetical group; they do not imply that a 60:40 allocation is ethically required. The stated assumptions matter: if some people are unwilling to accept one arm, or if an arm is unsuitable for them, the trial's approach to those participants needs separate consideration.
Fair Access Before and During Assignment
Fair access begins before randomization. Eligibility rules, recruitment locations, visit schedules, transportation demands, language access, and other trial procedures can determine who is realistically able to participate. An allocation rule cannot correct an access barrier that has already kept a group out of the trial. Nor does identical randomization among enrolled participants establish that the enrolled group fairly represents the people affected by the research question.
During assignment, fairness concerns the opportunity each participant has to receive each suitable arm. The opportunity may be equal, or it may differ under a justified allocation rule. To make the distinction precise, let \(r_{ik}\) denote participant \(i\)'s probability of assignment to arm \(k\) at the time of assignment. Let \(I_{ik}\) equal 1 if participant \(i\) is assigned to arm \(k\), and 0 otherwise. Then \(E(I_{ik})=r_{ik}\). Summing across participants gives:
This follows from the linearity of expectation; it does not require assignments to be independent. The result clarifies what an allocation rule implies for expected arm counts. It does not establish that the probabilities are fair, that access is equitable, or that participants' preferences have been addressed. Those are ethical judgments informed by the trial's context.
Worked Example: A Treatment Burden Creates an Access Question
Consider a hypothetical trial comparing a daily tablet (Arm A) with an infusion requiring weekly clinic visits (Arm B). Of 120 people considered for the trial, 80 live near the clinic and 40 live farther away. Assume all 80 nearby people and 10 of the farther-away people can manage the weekly visits; the other 30 cannot. Assume both treatments are otherwise suitable for those able to receive them.
If the trial requires every participant to be able to receive either arm, 90 people meet that practical condition: \(80+10=90\). Under equal randomization among them, the expected counts are:
The arithmetic checks the expected allocation among the 90 participants; it does not resolve the larger fairness issue. Thirty people were not included because weekly travel was infeasible. If travel burden is related to location or resources, the trial may have limited access for that group. Assigning those people to the infusion despite their inability to attend would not make access fair; it could make treatment receipt impossible. Investigators would need to consider whether the access barrier can be reduced and whether the trial's eligibility and recruitment choices are justified for its intended population.
Thus, equal assignment odds among participants who can receive both treatments and fair access to the trial are related but distinct questions. The example does not show that equal randomization is wrong; it shows why its fairness cannot be judged from the allocation ratio alone.
Contrasting Allocation Approaches
Three broad approaches help organize the ethical comparison. They are not interchangeable solutions. The appropriate approach depends on the trial's treatments, participant preferences, eligibility constraints, and reasons for using a particular allocation rule.
| Approach | What it can address | What it cannot establish on its own |
|---|---|---|
| Equal randomization | Gives each suitable arm the same assignment probability at the relevant assignment point. | That participants had equal access to enroll, or that treatment burdens and preferences are fairly handled. |
| Preference-sensitive arrangements | May take strong preferences seriously, for example by assessing willingness to be randomized or considering a nonrandomized option where appropriate. | That preferences are accurate measures of treatment benefit, or that every preference can be accommodated within a randomized comparison. |
| Response-adaptive allocation | Changes assignment probabilities over time using accumulating trial responses, under a specified rule. | That participants' preferences are represented, that access barriers have been corrected, or that changing probabilities is fair merely because the rule is response-based. |
A preference-sensitive arrangement should not be confused with response-adaptive randomization. The former is organized around what participants say they prefer or are willing to accept; the latter changes allocation probabilities in response to accumulating outcome information. A trial might consider both preferences and responses in its ethical review, but one does not automatically justify the other.
Worked Example: Changing Assignment Probabilities Raises a Fairness Question
Suppose, hypothetically, that 40 participants enter a two-arm trial in two successive groups of 20. All are eligible for both treatments, and each says they are willing to receive either treatment, although 24 of the 40 state a preference for Arm A. For the first group, each arm has assignment probability 0.5. For the second group, assume the trial's response-adaptive rule sets the probability of Arm A to 0.7 and Arm B to 0.3.
The expected numbers assigned to each arm are calculated group by group. For Arm A:
For Arm B:
The expected total is \(24+16=40\), matching the number of participants. These expected counts do not guarantee that exactly 24 participants receive Arm A. Nor do they show that the 24 participants who prefer A will be the 24 assigned to A: stated preferences do not determine random assignments. The ethical review should ask whether the unequal probabilities are adequately justified and whether participants with different preferences have a fair opportunity to take part. The arithmetic alone cannot answer either question.
Diagnostics, Limitations, and a Common Mistake
An ethical review of preferences and access should look for warning signs rather than relying on an allocation ratio as a summary of fairness. Useful questions include:
- Were preferences sought from a group that includes, or is reasonably informative about, the people the trial aims to reach?
- Does the stated preference concern likely benefit, treatment burden, convenience, or another feature? Are those being treated as distinct?
- Can every enrolled participant realistically receive every arm to which they may be assigned?
- Do recruitment, visit, or communication requirements create avoidable barriers for particular groups?
- If assignment probabilities differ, is the reason clear, relevant to the ethical question, and assessed separately from preference shares?
These checks have limits. Preference reports may change after people learn more about the treatments, and a survey of willing participants may omit people who declined. Practical access can also differ within a group and change over time. No simple calculation can settle these issues, and equal allocation does not guarantee that all relevant burdens are equal.
Common mistake: “Most participants prefer Arm A, so a fair trial should assign more people to Arm A.” This conclusion treats a preference share as if it were evidence of greater benefit and as if all participants shared the same preference. The correction is to identify what the preference represents, whose preferences were measured, and whether the allocation rule has an independent justification. Preference information can inform ethical deliberation, but it does not by itself prescribe assignment probabilities.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- In a hypothetical group of 50 participants, 35 prefer Arm A and 15 prefer Arm B. Does this fact alone justify assigning more participants to Arm A? Why or why not?
- A trial gives each enrolled participant a 0.5 probability of assignment to either arm. Name one reason this does not establish fair access to the trial.
- In a response-adaptive trial, 10 participants have assignment probability 0.6 for Arm A and 10 have probability 0.4. What is the expected number assigned to Arm A across these participants?
- Why might excluding people who cannot attend frequent visits raise an ethical question even if equal randomization is used among those enrolled?
Solutions. (1) No. The preference proportions are \(35/50=0.700\) for A and \(15/50=0.300\) for B, but they do not establish comparative benefit or show that every participant prefers A. The trial should assess what the preferences mean and justify its allocation rule separately. (2) Equal assignment odds apply to people who have enrolled; recruitment, eligibility, travel, or other practical barriers may have prevented some people from participating. (3) By summing assignment probabilities, \(E(N_A)=10(0.6)+10(0.4)=6+4=10\). This is an expected count, not a guarantee of exactly 10 assignments to A. (4) The visit requirement may prevent a group from accessing the trial or treatment, so equal odds among enrolled participants do not resolve whether the trial's procedures create an avoidable barrier.