Reading an Access Plan as a Set of Specific Permissions
An adaptation access plan describes which interim information may be available to which trial roles, under what masking conditions, and for what purpose. Reading it well requires more than noticing that a plan says “blinded” or names a committee. The key question is item-specific: what information can this recipient see, in what form, and what decision is that information meant to support?
This tutorial focuses on interpreting a plan that has been written. It does not establish that the plan was followed in practice, nor does it determine whether the adaptation rule itself is statistically sound. Those are separate questions. A well-interpreted plan makes the intended route of information legible enough to recognize missing permissions, unnecessary disclosures, and mismatches between information and decision.
The “recipient” should be read as a specific role or group, not as a vague organizational label. “The sponsor” may be too broad if only an independent statistician is authorized to see a result, while sponsor personnel responsible for trial conduct are not. Similarly, “the committee” may be ambiguous if it does not distinguish the members who receive comparative results from staff who prepare meeting materials.
Masking status is also relational. A treatment code may conceal the treatment identity from a recipient, but it does not necessarily conceal the existence or direction of an arm difference. Conversely, an aggregate pooled estimate may be available without revealing separate treatment-group results. A plan should be read by asking what this recipient can infer from this particular item, not by assigning one permanent label such as “blinded” to an entire person or organization.
Trace Each Item From Its Form to Its Intended Use
A useful reading sequence is to locate the result, identify its recipient and presentation, then connect it to the authorized decision. A decision can be a formal adaptation, such as changing the sample size or selecting a treatment arm, or a restricted monitoring purpose. The stated use matters: information that is appropriate for one purpose is not automatically authorized for another.
Distinguish, for example, a pooled estimate from arm-specific estimates, an arm-coded contrast from a treatment-named contrast, or a participant-level record from a summary.
Determine which role receives the item directly and whether the plan allows it to be passed to anyone else.
Ask whether treatment identities, the direction of comparative results, or both remain concealed. Do not treat a code label alone as proof that comparative information is masked.
Check that the recipient has the information needed for the specified adaptation and that the plan does not imply broader use without permission.
This sequence is a reading aid, not a claim that a plan is complete merely because its rows can be parsed. If an item, recipient, masking status, or permitted use is missing, record the ambiguity rather than supplying an assumption. In particular, “as needed” does not identify who decides that information is needed or what additional information may then be released.
Worked Example: Pooled Variability for Sample-Size Re-Estimation
Hypothetical plan excerpt: At the interim review, an independent statistician receives the pooled estimate of outcome variability, without treatment-specific means or treatment identities. The statistician calculates a revised sample size using the prespecified rule. The sponsor’s trial-management team receives only the resulting total target sample size and does not receive arm-specific outcomes.
Item and recipient. The first item is the pooled variability estimate; its recipient is the independent statistician. The later item is the revised total sample-size target; its recipient is the sponsor’s trial-management team.
Form and masking. The statistician receives a pooled estimate, not separate means by treatment group. The plan therefore withholds the arm-specific outcome summaries needed to inspect the observed treatment contrast. It does not, by itself, prove that the statistician is unaware of every treatment identity: that depends on the data and coding arrangements. The sponsor receives only a target number, not the interim treatment outcomes, so the stated disclosure does not reveal an arm contrast.
Intended decision and numerical check. Suppose, hypothetically, that the rule targets a two-sided comparison with \(\alpha=0.05\), 90% power, equal allocation, a clinically relevant difference of 6 outcome units, and a pooled standard deviation estimate of 12.4 units. Under the usual normal-approximation calculation for two independent groups with common standard deviation, the required number per group is approximately
Rounding up gives 90 participants per group, or 180 total. The illustration assumes independent observations, equal allocation, a common standard deviation represented by the pooled estimate, and the stated normal-approximation design rule. The access-plan interpretation is that the statistician receives information sufficient for this variability-based calculation, while the sponsor receives the decision output rather than the comparative outcomes. The plan would still need to specify any cap, rounding convention, or other constraint if those are part of the actual rule.
When a Decision Requires Comparative Information
Some adaptation decisions depend on distinguishing treatment groups. For those decisions, a plan may authorize a designated independent role or committee to see arm-specific or comparative results while limiting disclosure to trial personnel who could influence conduct. The information can be treatment-named or arm-coded. These are different presentations: arm coding conceals which treatment corresponds to a code, but the recipient may still see which coded arm has the more favorable result.
Worked Example: Arm-Coded Results for Treatment Selection
Hypothetical plan excerpt: An independent statistician receives interim response rates for arms A, B, and C, with treatment names replaced by codes X, Y, and Z. A designated data monitoring committee receives the same coded rates and may recommend dropping an arm under the prespecified selection rule. The sponsor’s trial-management team receives no arm-specific rates.
Suppose the hypothetical interim response rates are 0.42 for X, 0.35 for Y, and 0.28 for Z. The plan gives the statistician and committee comparative information: the coded rates can be ranked, and the committee can consider the stated selection rule. Treatment identities remain masked if the code key is withheld from those recipients. The sponsor’s trial-management team, by contrast, receives no such comparison under this excerpt.
The interpretation should not be “the committee is blinded, so it cannot distinguish the arms.” It can distinguish X, Y, and Z by their interim results, even if it does not know which named treatments those codes represent. Nor should the intended action be inferred from the rates alone: the plan authorizes consideration of dropping an arm only as specified by the selection rule, not an unrestricted choice of any action. This example assumes response rates are calculated on the same defined interim population and that the plan’s rule and code-key arrangements are as stated.
The item’s intended use should be explicit enough to distinguish decision support from general awareness. A result authorized for a committee’s treatment-selection recommendation does not automatically authorize its release to recruitment staff, investigators, or sponsor personnel. The existence of a recipient permission does not create an onward permission.
Worked Example: A Restricted Futility Summary
Hypothetical plan excerpt: An independent statistician calculates a prespecified futility statistic at an interim analysis. The statistician and a designated monitoring committee receive the statistic and its interpretation under the trial rule. The committee may recommend stopping for futility. The sponsor receives only the committee’s recommendation and the operational information needed to implement it, not the statistic or arm-specific efficacy estimates.
Here the item is not raw participant data or a full set of arm-specific results; it is a specified statistic plus its interpretation. Its recipients are the statistician and designated committee. The intended decision is a futility recommendation under the stated rule. The sponsor is authorized to receive the resulting recommendation, but not the underlying comparative estimates described as restricted.
This plan excerpt is interpretable only if “the statistic and its interpretation” are sufficiently defined elsewhere in the plan: for example, the statistic’s scope and what result corresponds to the futility boundary. If the excerpt does not define those terms, a reader should flag the omission rather than assume that the committee receives a particular confidence interval, p-value, or treatment contrast. The plan’s wording identifies a boundary between the decision group and operational recipients; it does not establish that the boundary was maintained in practice.
Why the Information-to-Decision Link Matters
A plan can be read as a chain: a specified interim item is provided to an authorized recipient, who uses it for a specified decision. The chain is coherent only if the item is sufficient for that decision and the permission is no broader than intended. This gives a simple justification for checking each link rather than relying on labels such as “independent” or “blinded.”
The implication runs from decision to information: first ask what information the decision rule requires, then check that the plan gives that information to the role responsible for applying the rule. If the specified decision requires comparing arms but the recipient receives only a pooled summary, the described access is insufficient unless the plan identifies another source of the comparison. If the plan provides arm-specific results to a person who has no stated role in the decision, the access is not justified by that stated purpose alone.
This check is logical, not a statistical guarantee. A plan may specify the correct information for an adaptation while the rule has other methodological limitations. Conversely, a sound statistical rule cannot be applied as specified if the authorized decision-maker lacks the required input. Access interpretation and statistical validity are related but distinct.
Comparison, Ambiguities, and Failure Modes
Plans can describe access in different ways. A role-based description names who receives which item; an item-based description starts from a result and names its recipients; a broad organizational statement may name only an institution or committee. The first two can both be interpretable when all relevant details are present. Broad labels are harder to interpret because they can conceal differences among people within an organization or among the information each person receives.
A plan may also distinguish direct access from a summary or recommendation passed onward. These are not equivalent permissions. In the futility example, receiving a recommendation is not the same as receiving the underlying statistic. Likewise, a coded table is not equivalent to a pooled table: the coded table preserves comparisons across arms even when treatment names are concealed.
Common warning signs include undefined recipients (“the study team”), underspecified items (“interim data”), unqualified masking claims (“blinded results”), or purposes with no clear decision (“for review”). Check whether the plan explains who holds a code key, whether recipients may forward information, and what information is withheld from roles whose conduct should not be influenced by comparative findings. If an answer is absent, that is an ambiguity to report, not a license to guess.
Correction: Coding can conceal treatment identity while leaving the relative performance of coded arms visible. State separately whether the recipient can see a comparison and whether the recipient can map each code to a named treatment.
A second mistake is to assume that a permission to receive information includes permission to share it. Read onward disclosure as a separate access question. A third is to treat a written plan as evidence of actual access. A plan states intended or authorized access; confirming what individuals actually received requires records or other evidence beyond the plan itself.
Finally, access descriptions do not remove the need to consider whether an adaptation rule has appropriate operating characteristics, whether outcomes are sufficiently mature, or whether the trial’s conduct could be influenced by disclosures. Those issues are not resolved simply by naming an independent recipient. The plan’s contribution is narrower: it defines who is permitted to see what, and for what stated purpose.
Check Your Understanding
For each item, identify the recipient, masking status, and intended use. Solutions follow the questions.
- A plan gives an independent statistician a pooled standard deviation but no treatment-specific means. What can you conclude about the information made available, and what should you avoid concluding about treatment identity?
- A committee sees response rates for coded arms X and Y, but not the code key. Is the committee able to compare the arms? What remains masked?
- A sponsor receives a recommendation to stop for futility but not the interim statistic. Does this mean the sponsor received the same information as the committee?
- A plan says “the study team may review interim findings.” Name two details needed to interpret this permission.
- A plan authorizes an independent statistician to view treatment-specific results. Does that establish that the statistician actually viewed them, or that the sponsor did not? Explain.
Solutions. (1) The statistician receives pooled variability information and not the stated treatment-specific means. This supports a variability-based use, but the excerpt alone does not establish whether treatment identities are known through other information or coding arrangements. (2) Yes. The committee can compare X with Y if both rates are shown; the code-to-treatment mapping remains masked if the key is withheld. (3) No. A recommendation is a decision output, not the underlying statistic or all information available to the committee. (4) At minimum, specify which roles count as “the study team,” what interim items are included and in what form, and what decisions or uses are permitted; onward disclosure may also need clarification. (5) No. Authorization describes permitted access, not proof of access in practice, and it says nothing by itself about sponsor access. Actual access and any additional permissions require separate evidence.