Why Lung-MAP S1400 Is a Case Study in Protocol Adaptation
A platform trial is not a single, unchanging comparison. It is an organized structure for evaluating treatments over time, often by directing eligible participants to different substudies. That structure can continue to serve its purpose even when a particular treatment arm closes. The central design question is therefore not simply whether an arm succeeded, but how the protocol can respond to evidence without losing the coherence of its comparisons.
Lung-MAP, also known as SWOG S1400, was developed as a master protocol for advanced squamous non-small-cell lung cancer. Its biomarker-directed approach used genomic screening to assign participants to substudies testing therapies matched to particular molecular alterations. Early substudies included targeted-treatment comparisons such as AZD4547 for tumors with FGFR alterations, taselisib for tumors with relevant PI3K-pathway alterations, and palbociclib for tumors with relevant cell-cycle alterations. The exact eligibility rules and treatment details were defined within each substudy protocol; a shared screening framework did not make the substudies identical.
The treatment landscape changed rapidly during the program. New evidence and approvals, particularly in immunotherapy, affected which options were clinically relevant and what constituted an appropriate comparator. Some targeted approaches also failed to show enough promise to justify continued testing. Lung-MAP responded through protocol amendments, including closing or changing substudies and adding or revising treatment options. Over time, the platform’s screening approach also broadened beyond its original focus on squamous disease. The case illustrates adaptation as ongoing governance: evidence, available therapies, eligibility, and operational feasibility all matter.
What Can Prompt an Adaptation?
A trial may change because of accumulating results within an arm, external evidence, or practical constraints. These reasons should be separated because they support different conclusions. A prespecified interim futility rule asks whether the arm’s observed evidence makes continued enrollment insufficiently promising under the trial’s decision plan. An external change in standard of care asks whether the original comparison remains clinically appropriate. Poor recruitment or a treatment’s unavailability can make an arm infeasible even when its scientific question remains interesting.
These are not interchangeable explanations. If an arm closes under a prespecified futility boundary, the decision is evidence about that arm under its design assumptions; it does not establish that every treatment in the same class is ineffective. If an arm closes because a comparator is no longer appropriate, that is a design and clinical-context decision, not a direct finding that the experimental treatment lacks activity. A careful report states the actual reason and the evidence supporting it.
Protocol adaptation should also distinguish planned decisions from ad hoc changes. A protocol may specify in advance which interim data will be reviewed, who reviews them, and what criteria could lead to continuation or closure. New evidence outside the protocol may instead require a formal amendment and review. Either route needs a documented decision trail: what changed, when it changed, which participants were affected, and why the revised design remains interpretable.
Worked Decisions Under Explicit Assumptions
A numerical stopping calculation is meaningful only after the endpoint, decision rule, and probability model have been stated. The following examples are hypothetical illustrations of reasoning, not Lung-MAP rules or results. In particular, the probabilities and thresholds below are not reported trial findings.
Worked Example: Probability of Reaching a Hypothetical Response Threshold
Suppose a hypothetical substudy uses tumor response as an early activity measure. Its illustrative continuation rule requires at least three responses among the next four evaluable participants. At a response probability of \(p=0.25\), calculate the probability of meeting that rule. Assume each participant’s response is independent, the response probability is the same for all four participants, and response is recorded without error.
Let \(X\) be the number of responses among those four participants. Under the assumptions, \(X\sim\operatorname{Binomial}(4,0.25)\). Continuing requires \(X\geq3\), so
The two terms are \(4(0.015625)(0.75)=0.046875\) and \(0.00390625\), respectively. Thus
Under this assumed response rate and rule, the chance of reaching the threshold is about 5.08%. This calculation alone does not determine whether to close an arm: the rule’s clinical rationale, timing, other evidence, and protocol governance must also be considered. It does show how a prespecified threshold translates into a probability under a stated model.
Worked Example: Separating Futility From a Changed Comparator
Consider two hypothetical reasons for stopping enrollment in a biomarker-matched substudy. In Scenario A, an interim review finds that the arm crosses its prespecified futility boundary. In Scenario B, a new treatment becomes the accepted standard, making the original control no longer clinically suitable, while the experimental arm has not crossed its futility boundary. What should the closure records say?
In Scenario A, the record should identify the interim data cut, the prespecified boundary, the result relative to that boundary, and the resulting recommendation. The conclusion is that continued testing did not meet the arm’s defined criterion for promise. It is not a conclusion that all treatments targeting that biomarker fail.
In Scenario B, the record should identify the external clinical change and explain why the original comparison is no longer suitable. It should not label the closure as a futility finding when the futility criterion was not met. A revised comparison might be considered, but only through a reviewed amendment that addresses eligibility, control selection, and interpretation.
The distinction matters because the scientific evidence differs. Scenario A is a decision based on the arm’s interim evidence; Scenario B is a decision about the continued relevance of the design. Both may justify stopping enrollment, but they support different statements about what was learned.
Justifying a Closure Rule and Preserving Interpretation
A simple futility rule can be understood by comparing the probability of future success under assumptions about the treatment’s activity. Suppose a decision rule says that an arm continues only if future observations can plausibly meet a prespecified target. The calculation conditions on the accrued data and models the remaining data. This is sometimes called conditional power when the target is a final statistical test, but the general logic also applies to simpler response-count rules.
For the binomial illustration above, the derivation is direct. The event of reaching at least three responses in four participants consists of two mutually exclusive events, exactly three and exactly four responses. The binomial probability for exactly \(k\) responses is \(\binom{4}{k}p^k(1-p)^{4-k}\). Adding the two disjoint probabilities gives the probability of meeting the rule. The calculation is valid only if the assumed common response probability and independence are reasonable for the participants being projected. If response probabilities differ substantially across biomarker groups or patient characteristics, this simple model may not represent the future accrual.
For a platform such as Lung-MAP, closure also has consequences beyond the arithmetic of one decision boundary. The protocol must specify whether participants who have already enrolled continue follow-up, how data from the closed arm will be reported, and whether the screening network redirects newly identified participants. A closed arm’s existing data remain part of the scientific record; closure does not erase them. Nor does it automatically permit pooling its participants with a later arm using a different treatment, eligibility definition, or comparator.
When protocols change, the version of the design is part of the evidence. Investigators need to track which version governed each participant’s screening and assignment, the dates of amendments, and the rationale for each change. If an eligibility rule expands, for example, the enrolled population may differ across periods. If the control changes, comparisons across periods may no longer estimate the same treatment contrast. Analysis and reporting must reflect those differences rather than treating the platform as though it had operated under one fixed protocol throughout.
State whether the proposed change follows a prespecified interim rule, external evidence, a safety concern, or feasibility constraints.
Confirm the relevant review process, data cut, and protocol language before stopping or modifying enrollment.
Specify which participants, eligibility criteria, treatments, comparators, and follow-up procedures are affected.
Version the protocol and analysis plan, document the rationale, and report results according to the design actually experienced by participants.
Worked Example: Platform Continuity After One Arm Closes
Suppose a hypothetical platform has three active biomarker-directed substudies, each enrolling about 12 eligible participants per month across the screening network. One substudy closes, while the other two remain open. Assume for this illustration that screening capacity is unchanged, all three arms had received an equal share of eligible participants before closure, and participants formerly eligible for the closed arm are not eligible for either remaining arm.
Before closure, the three arms collectively enroll approximately \(3\times12=36\) participants per month under the stated simplification. After closure, the two continuing arms together enroll approximately \(2\times12=24\) participants per month. The reduction is \(36-24=12\) participants per month, or \(12/36=0.333\), about 33.3% of the former total. This is not evidence that the platform has failed: screening and enrollment continue for the two remaining questions. But it does show that platform continuity does not guarantee unchanged recruitment volume or preserved access for the closed biomarker group.
The assumptions are deliberately strong. If participants who would have matched the closed arm can now enter a different substudy, or if screening capacity and recruitment patterns change, the projected volume differs. The operational response must therefore be based on actual eligibility and allocation rules, not on the arithmetic alone. The example demonstrates why arm closure requires communication and updated procedures throughout the screening network.