From Model Uncertainty to an Agency Question
A model-based development discussion is most useful when it addresses a decision the sponsor must make. “Is the model acceptable?” may sound direct, but it leaves open what the model will be used for, which aspects need review, and what the sponsor would do with the answer. A focused question instead connects a defined uncertainty to a pending choice: for example, whether the available evidence supports using a model to select a dose for a planned trial, or whether another study is needed first.
The aim is not to make an agency choose the development strategy or to request a blanket endorsement of a model. It is to make the proposed use and the feedback sought clear enough that scientific advice can inform the sponsor’s next step. The question should be consistent with the selected interaction pathway and the decision’s timing. Advice and feedback are generally not the same as a binding regulatory decision.
The model’s context of use matters because a model’s adequacy is judged in relation to what it is intended to support. A model that may be informative for choosing doses for a trial does not automatically support a different use, such as replacing clinical evidence for another population. Earlier material on context of use and decision fitness helps distinguish the intended application from general claims about model quality.
Build the Question Around the Decision
Start by naming the decision in operational terms. “We need to decide how to use the model” is too vague. A useful statement identifies what action is pending, when it must be taken, and what options are genuinely available. Examples include selecting a dose regimen for a planned study, deciding whether a dedicated study is needed before a particular inference, or choosing between two trial designs.
Next, state what the model is proposed to do, for which population and conditions, and what evidence supports that use. Then identify the alternative that would be pursued if the proposed approach is not considered adequate. Finally, specify the feedback needed. It may concern a model assumption, the relevance of a validation set, a proposed sensitivity analysis, or the adequacy of the planned evidence for a defined use. A request such as “please comment on the model” does not tell the agency which of these matters is at issue.
Describe the action and its timing, such as selecting a regimen before finalizing a trial protocol.
State what the model will inform, in whom, under what conditions, and what evidence is available.
Describe the practical next option, such as collecting additional data or using a different trial design.
Identify the assumptions, analyses, or evidence gaps on which advice would help the sponsor act.
This sequence follows from the logic of a decision. If every plausible interpretation of a model uncertainty leads to the same action, resolving that uncertainty may not affect the pending choice. If credible assumptions lead to different actions, the uncertainty is decision-relevant. The agency question should then target the assumption or evidence that separates those actions, rather than invite an unbounded review of the entire modeling program.
For a simple threshold decision, let \(q\) be a model-derived quantity and \(c\) the threshold used to distinguish two actions. If the proposed action is selected when \(q \geq c\), then uncertainty about \(q\) matters to that choice when credible modeling assumptions place \(q\) on different sides of \(c\). The question should make clear what supports estimating \(q\), how the threshold is used, and what feedback is sought about that evidence or its assumptions. This is a framing aid, not a substitute for establishing that the threshold or model is scientifically appropriate.
Worked Example: Selecting a Dose for a Planned Trial
Hypothetical situation. A sponsor is preparing a phase II trial of a fictional oral medicine. A population PK/PD model predicts that a 20 mg once-daily regimen will achieve the prespecified pharmacodynamic target for most participants. The prediction depends partly on an exposure-response relationship estimated from a small early-phase data set. The sponsor must finalize the trial regimens in two months.
Decision and uncertainty. The pending decision is whether to include 20 mg once daily as the principal regimen, or to include a lower regimen and gather more dose-response information. The decision-relevant uncertainty is not simply whether the model “fits”; it is whether the exposure-response evidence and its uncertainty are adequate to use the predicted target attainment for this dose-selection purpose.
Proposed approach and alternative. The sponsor proposes to use the model, with prespecified sensitivity analyses for the exposure-response slope and between-participant variability, to support selection of the regimen. The alternative is a trial design that retains a lower dose arm to characterize dose response rather than relying on the model prediction alone.
Focused question. “For selecting regimens for the planned phase II trial, does the proposed population PK/PD analysis, including sensitivity analyses for the exposure-response slope and between-participant variability, provide an adequate basis to use predicted target attainment when choosing the 20 mg once-daily regimen? If not, which additional analyses or evidence would be most important before relying on that prediction?”
This question identifies the decision, model use, alternative, and requested feedback. It does not ask the agency to guarantee that 20 mg will be effective or to endorse the model for every future purpose. The stated timing also makes clear why the feedback is needed now.
Make the Request Answerable and Neutral
A focused question is not necessarily a question that can be answered with an unconditional yes or no. Agency feedback may be conditional on the proposed context of use, the evidence presented, or additional analyses. The question should therefore invite a scientifically informative response, including what would change the assessment, rather than imply that only one answer is acceptable.
A useful check is whether the briefing material gives the reader enough information to understand the question without guessing. At minimum, the supporting discussion should identify the decision, the intended context of use, the model’s role, the key evidence and limitations, the alternative strategy, and the specific feedback requested. The question itself should not carry an unsupported conclusion that belongs in the supporting rationale.
Worked Example: Deciding Whether to Conduct a Dedicated Study
Hypothetical situation. A sponsor is developing a fictional medicine cleared mainly by the kidneys. A physiologically based pharmacokinetic model predicts that exposure in people with moderate renal impairment may be 1.3 times that in people with normal renal function. The sponsor is considering whether to conduct a dedicated clinical pharmacology study before selecting a dosing recommendation for a later trial.
Decision and uncertainty. The pending decision is whether to conduct the dedicated study before making a dosing recommendation for the later trial. The important uncertainty is whether the model’s renal-function assumptions and available clinical PK data are sufficient for this specific use. The predicted ratio is not, by itself, a complete answer: its relevance depends on the model’s credibility and on how the proposed recommendation will be used.
Proposed approach and alternative. The sponsor proposes to use the model, supported by available clinical PK data and sensitivity analyses for renal clearance, to inform whether dose adjustment should be evaluated in the later trial. The alternative is to conduct a dedicated study in participants with renal impairment before finalizing that recommendation.
Focused question. “For deciding whether a dedicated renal impairment study is needed before selecting a dosing approach for the later trial, is the proposed PBPK analysis, supported by the available clinical PK data and sensitivity analyses for renal clearance, an adequate basis for that decision? If not, what additional clinical or modeling evidence would be most informative?”
The question seeks feedback on the model’s adequacy for a defined development decision. It does not treat the predicted 1.3-fold exposure ratio as a regulatory conclusion or imply that a model-based assessment would necessarily replace a clinical study.
Compare Broad, Leading, and Decision-Focused Questions
| Question style | Example | Why it helps or hinders |
|---|---|---|
| Broad | “Is our model acceptable?” | Does not specify use, decision, alternative, or the aspect on which feedback is needed. |
| Leading | “Do you agree that our model proves the proposed regimen is appropriate?” | Overstates what a model can establish and encourages a simple endorsement rather than targeted scientific feedback. |
| Decision-focused | “For selecting regimens for this trial, is the model and planned sensitivity analysis adequate to inform the dose choice, or is further dose-response evidence needed?” | Links the proposed approach and alternative to a concrete decision and asks about the evidence needed for that use. |
A broad question can be appropriate as an opening discussion prompt, but it is weak as the only request when a particular action is pending. A leading question is also problematic: it presents the sponsor’s preferred conclusion as though it were established and can obscure the uncertainty that advice should address. A decision-focused question is more informative because it presents the proposal and alternative without disguising the sponsor’s position or asking the agency to make the sponsor’s decision.
Check the Framing Before Submission
A question can be clear in wording yet still be poorly framed if its scope exceeds the evidence or if the decision is not genuinely pending. Review it against these checks:
- Decision: Is there a specific action, decision-maker, and relevant timing?
- Use: Is the model’s proposed role limited to a defined population, condition, and purpose?
- Alternatives: Is there a realistic option if the proposed model-based approach is insufficient?
- Feedback: Does the request identify the assumption, analysis, or evidence gap at issue?
- Evidence: Does the supporting material show why the model could inform this decision and where uncertainty remains?
- Scope: Does the question avoid implying that advice on one use constitutes endorsement for another?
A practical diagnostic is to ask what action would follow from each plausible response. If the sponsor cannot describe a next step for a response that the approach is adequate, one that it is inadequate, or one that it is adequate only after specified work, the question may not be connected tightly enough to the decision. Another warning sign is a request for general agreement without naming the model’s intended use or the evidence that is open to discussion.
The framing also has limits. A well-written question cannot compensate for a poorly supported model, an unclear context of use, or evidence that omits important assumptions. Nor does wording guarantee that an agency can resolve every uncertainty in one interaction. Some issues may need to be separated into a primary decision question and a small number of supporting questions. Keep those supporting questions subordinate: a long list of unrelated requests can make it harder to identify the decision that matters most.
Worked Example: Choosing a Pediatric Trial Regimen
Hypothetical situation. A sponsor has a model linking weight, clearance, and exposure in adults, together with limited PK data from adolescents. The sponsor proposes a weight-based regimen for a planned adolescent trial. A fixed regimen with additional PK sampling is a feasible alternative. The sponsor’s immediate decision is which regimen to specify in the protocol.
Decision-focused question. “For selecting the initial regimen for the planned adolescent trial, is the proposed weight-based approach, supported by the adult model and the available adolescent PK data, sufficiently justified? In particular, are the proposed assumptions about weight-related clearance and the planned PK sampling adequate for this use, or should the protocol use a fixed regimen with additional dose-ranging information?”
The question names the immediate protocol decision, explains the proposed model-based approach, presents a practical alternative, and focuses the requested feedback on two relevant areas. It does not claim that the adult model is automatically transferable to adolescents or ask for an open-ended endorsement of pediatric use in general. If the sponsor also wants advice on a later labeling application, that is a distinct use and should not be implied by this trial-regimen question.
Check Your Understanding
For each item, identify the framing issue or state a more decision-focused version.
- A sponsor asks, “Is our model good enough?” The decision is whether to use a model to select a dose for a planned trial. What is missing from the question?
- A model-based exposure estimate is favorable, but plausible sensitivity analyses could change the selected regimen. Why should the question address those analyses?
- A sponsor asks whether a model can “replace all further studies.” Give one reason this request is too broad.
- Write a focused question for a hypothetical decision about whether a model is adequate to select one of two trial regimens, where further dose-response data are the alternative.
Solutions.
1. The question omits the particular dose-selection decision, the model’s proposed role and context of use, the alternative to relying on it, and the specific feedback sought. It should ask whether the evidence supports using the model for that defined trial decision and what additional work would be important if it does not.
2. The sensitivity analyses address uncertainty that may change the action. If credible assumptions lead to different regimen choices, feedback on those assumptions or analyses could affect the decision; a favorable central estimate alone would not describe that decision-relevant uncertainty.
3. “All further studies” combines potentially different populations, decisions, and contexts of use into one request. It also treats the model as a universal substitute without specifying which evidence gap or development decision is under consideration.
4. One suitable version is: “For selecting between regimen A and regimen B for the planned trial, is the proposed model, supported by the available data and the specified sensitivity analyses, an adequate basis for the choice? If not, what additional dose-response evidence would be most important before making that selection?” The wording identifies the decision, proposed approach, alternative evidence strategy, and requested feedback.