Defining Eligible Cohorts Across Protocol Amendments

Why Eligibility Must Be Version-Specific

In a platform or umbrella trial, an amendment can change who may enter an arm while the trial is running. For example, an arm might broaden its permitted age range, revise a biomarker threshold, or remove an exclusion criterion. The amendment does more than change an operational checklist: it can change the population that the arm is designed to enroll. If eligibility is recorded without its protocol version, it may be impossible to tell which rules governed a participant’s entry decision.

The key task is to connect each eligibility decision to the protocol version in force at the relevant decision time. That preserves the distinction between people eligible under the earlier criteria and people eligible under the amended criteria. It also makes explicit whether a change has broadened, narrowed, or otherwise altered the intended cohort for an arm.

Definition: For arm $a$, protocol version $v$, and candidate $i$, define $E_{iav}=1$ if candidate $i$ satisfies every required inclusion criterion and none of the exclusion criteria for arm $a$ under version $v$; otherwise define $E_{iav}=0$. A version-specific eligible cohort is the set of candidates satisfying the criteria under a specified arm and version. The eligibility decision date is the prespecified date at which eligibility is assessed for entry, such as the date of randomization or enrollment.

The exact decision date must follow the protocol’s operational rule. The examples below use the date of randomization. A protocol could specify another entry milestone, but the milestone must be explicit and applied consistently. A candidate’s eligibility is assessed using the version applicable at that decision date; an amendment does not, by itself, rewrite the rules that governed an earlier decision.

This definition assumes that each criterion has an operational meaning, that the version and its effective date are known, and that the required information is available or its missing status is recorded. If required information is missing, the record should be marked unresolved pending the protocol’s specified procedure, rather than silently treated as either eligible or ineligible.

Representing the Change in Criteria

Let $C_v$ denote the set of candidates who satisfy the criteria for a particular arm under version $v$, evaluated at the same decision time and using the same candidate information. Comparing the sets makes the effect of an amendment concrete. Candidates in both sets remain eligible; candidates in only the amended set are newly eligible under the revised rules; candidates in only the earlier set no longer qualify under the revised rules.

\begin{aligned} \text{Eligible under both versions} &= C_{\text{old}} \cap C_{\text{new}},\\ \text{Newly eligible under the amendment} &= C_{\text{new}} \setminus C_{\text{old}},\\ \text{No longer eligible under the amendment} &= C_{\text{old}} \setminus C_{\text{new}}. \end{aligned}

This comparison is clearest when only one criterion changes and all other criteria stay fixed. If the sole change relaxes an age upper bound, then $C_{\text{old}}$ is a subset of $C_{\text{new}}$, assuming no other rule changes. If the sole change tightens a biomarker threshold, then $C_{\text{new}}$ is a subset of $C_{\text{old}}$. When several criteria change at once, neither set need contain the other: the amendment may make some candidates newly eligible and cause others to become ineligible.

Key result: The counts of newly eligible and no-longer-eligible candidates are not interchangeable with the total change in eligible counts. The difference in totals equals the number newly eligible minus the number no longer eligible: $$ |C_{\text{new}}|-|C_{\text{old}}| = |C_{\text{new}}\setminus C_{\text{old}}| - |C_{\text{old}}\setminus C_{\text{new}}|. $$

Worked Examples

Worked Example: Widening an Age Range

Suppose a hypothetical arm initially accepts ages 18 through 70, inclusive. An amendment changes the upper limit to 75, leaving every other criterion unchanged. Among 96 hypothetical candidates assessed using the same clinical information, 72 meet the original criteria, 9 fail the original rules only because they are aged 71–75, and 15 fail at least one unchanged criterion.

Under the original version, the eligible count is $72$. Under the amended version, the 9 candidates who failed only the age rule now qualify, while the 15 candidates who fail an unchanged rule remain ineligible. Thus:

|C_{\text{new}}|=72+9=81.

The amendment has made 9 of 96 candidates newly eligible and none no longer eligible. The increase is $81-72=9$ candidates, or $9/96=0.09375$, approximately 9.38 percentage points of this hypothetical candidate group. The conclusion relies on the stated check that the 9 candidates satisfy every other criterion and the 15 fail at least one unchanged criterion.

Worked Example: Raising a Biomarker Threshold

A hypothetical arm originally requires a biomarker value of at least 10 units. A revised version requires at least 20 units; all other criteria are unchanged. Among 30 candidates who satisfy all other criteria, 14 have values of at least 20 units, 11 have values from 10 up to but not including 20 units, and 5 have values below 10 units.

Under the original rule, the eligible candidates are those in the first two groups, giving $14+11=25$. Under the revised rule, only the first group qualifies, giving 14. The 11 candidates in the middle group are no longer eligible under the amended criterion; the 14 in the first group remain eligible, and the 5 in the last group fail both versions.

|C_{\text{new}}|-|C_{\text{old}}|=14-25=-11.

The eligible count decreases by 11, from 25 to 14. This calculation assumes that the biomarker values are measured and interpreted using the same assay definition in both versions. If the amendment also changes the assay or measurement window, the groups cannot be classified by threshold alone; those changes must be represented in the version-specific rules.

Worked Example: Applying the Effective Version at Entry

In a hypothetical trial, version 1 is effective through June 30 and permits ages 18–70. Version 2 becomes effective July 1 and permits ages 18–75. The protocol specifies that the version in force on the randomization date governs eligibility. One candidate aged 73 is screened on June 28 and is not randomized until July 3. A second candidate aged 68 is randomized on June 29.

For the 73-year-old, the screening date is not the governing date under the stated rule. At randomization on July 3, version 2 is in force, and age 73 meets its age criterion. The candidate is eligible on age under version 2, subject to meeting every other criterion. For the 68-year-old, randomization occurs on June 29, so version 1 governs; age 68 meets the age criterion under that version.

The record should retain both dates, the version applied at randomization, and the criterion-level decision. It should not label the first candidate ineligible merely because the initial screen preceded the amendment. This conclusion follows from the example’s explicit effective-date rule; a different protocol rule would require applying that rule instead.

A Practical Documentation Process

A useful record makes it possible to reconstruct both the decision and the cohort definition later. The record should distinguish the protocol’s written criteria from the information used to assess them and from the final eligibility status. A concise process is:

Identify the governing version.
Record the arm, eligibility decision date, protocol version, and effective date. Apply the protocol’s stated rule for which date determines the version.

Evaluate each criterion.
Record criterion-level pass, fail, or unresolved status, with the relevant source and measurement date where appropriate. Do not infer a pass from missing information.

Assign the version-specific decision.
Classify the candidate as eligible only if all required inclusion criteria are met and no exclusion criterion applies under that version.

Summarize the amendment’s effect.
Compare the old and new rules to identify unchanged, newly eligible, and no-longer-eligible groups. Keep these categories distinct from the count actually enrolled.

A version history should state the amendment identifier, approval and effective dates, arm affected, exact old and new wording, and whether any other criteria changed at the same time. For each candidate, retain the version applied, decision date, criterion-level results, and any documented exception or unresolved item. If the protocol authorizes re-screening or reassessment, record that as a new decision under the applicable rules; do not overwrite the earlier decision.

This documentation also distinguishes three different quantities: the number who meet a version’s criteria, the number screened under that version, and the number who enter the arm. They may differ because some candidates are not screened, some eligible candidates do not enroll, and some screened candidates have unresolved information. Reporting an enrollment count as though it were the size of the eligible cohort obscures these distinctions.

Diagnostics, Alternatives, and Limitations

Several checks can reveal versioning errors. Look for eligibility records without a protocol version or decision date; decisions made under a version not yet effective; criterion values outside the permitted measurement window; missing values coded as passes; and abrupt changes in enrollment that are not explained by the amendment. Recalculate a sample of records from the criterion-level data and the versioned rules. When feasible, check that the category counts reconcile: candidates eligible under both versions, newly eligible, and no longer eligible should account for the relevant comparison set without double-counting.

A tempting alternative is to apply the latest criteria to every participant and store only the resulting eligibility flag. This may simplify a current roster, but it erases which rules actually governed earlier entry decisions and can reclassify participants retrospectively. Another alternative is to retain only a single “ever eligible” flag. That can be useful for a limited operational summary, but it cannot show which version made a candidate eligible or how the target cohort changed. Version-specific records preserve those distinctions.

Eligibility criteria define the population an arm is permitted to enroll under a given version; they do not alone establish who was approached, who enrolled, or what population is represented in the resulting data. A documented change in criteria therefore supports a clear description of the intended cohort, but does not show that the enrolled participants match that cohort. Where multiple criteria change together, the set comparison can identify the net groups only if candidate-level data are available and each rule is operationally specified.

Common pitfall: “The age range was widened, so everyone enrolled before the amendment is eligible under the new rule.” This is incorrect: a broader rule may make additional people eligible, but it does not establish that previously enrolled participants meet every criterion in the new version, nor does it change the version that governed their original entry. Preserve the original decision and, if a new assessment is required and authorized, record it separately under the new version.

Check Your Understanding

Answer each question using the version-specific rules described in the tutorial. Brief solutions follow the questions.

An age limit changes from 18–65 to 18–70, with every other criterion unchanged. Among 80 candidates, 50 meet the old rules, 7 fail only because their ages are 66–70, and 23 fail another criterion. What is the new eligible count?
A biomarker threshold changes from at least 5 units to at least 8 units. Of 40 candidates who meet all other criteria, 12 have values below 5, 10 have values from 5 to below 8, and 18 have values of at least 8. How many become ineligible under the new rule?
A candidate is assessed for eligibility on the day before an amendment takes effect but randomized the day after. The protocol says the randomization-date version governs. Which version applies, and what information should be retained?
Why is it not generally valid to apply the newest criteria retrospectively and replace earlier eligibility decisions?

Solutions. 1. The new count is $50+7=57$; the 7 candidates meet every unchanged criterion, while the 23 do not. 2. The 10 candidates in the 5-to-below-8 group become ineligible; the 18 at or above 8 remain eligible, and the 12 below 5 failed both versions. 3. The version effective on the randomization date applies. Retain the screening and randomization dates, governing version, and criterion-level results. 4. Retrospective replacement obscures which rules governed the original decision and may assign participants a status they did not have under the applicable version; retain the original decision and record any authorized reassessment separately.