Why an Agreed Strategy May Need Revisiting
An agreed model strategy gives a development team a working basis for generating evidence and making decisions. It does not guarantee that every later analysis, assumption, or use will remain within the scope discussed. New data may contradict a key prediction; an assumption may become implausible in the population of interest; or a planned decision may become more consequential than it was when the strategy was discussed. The practical question is not simply whether anything has changed, but whether the change could affect the decision the model is intended to support.
Renewed consultation is most useful when the change creates a material uncertainty about the model’s role, evidence requirements, or interpretation in a regulatory decision. A minor update that leaves the intended use and decision unchanged may be documented and handled within the existing plan. Conversely, when new evidence undermines an assumption that supports a dose, trial-design, or extrapolation decision, continuing without discussion can leave an important gap between the agreed strategy and the strategy now being used.
An earlier discussion or feedback record should be read in context: identify the question asked, the assumptions and evidence presented, the intended use, and any conditions attached to the feedback. Agency feedback is not a blanket endorsement of later analyses that differ in population, data, model structure, or decision use. The relevant comparison is between the strategy that was actually discussed and the proposed strategy now under consideration.
A Decision-Focused Review
A disciplined review separates four questions: what changed, how the change affects the model, whether it changes the decision, and whether the unresolved uncertainty is important enough to seek advice. This prevents two opposite errors: treating every new observation as a reason to reopen the strategy, and treating an earlier agreement as a reason to ignore consequential new evidence.
Record the model’s intended use, target population and conditions, key assumptions, planned analyses, decision to be informed, and the scope of agency feedback. Distinguish explicit feedback from the team’s interpretation of it.
State what new data, analysis, operational change, or scientific information is available. Identify which assumptions are now challenged and whether the change is observed evidence, a revised hypothesis, or a practical constraint.
Explain how the change could affect model performance, prediction uncertainty, applicability, or the decision itself. Compare the existing and revised strategies using the same decision context and decision criteria.
Decide whether to document and proceed, conduct additional analyses, revise the strategy, or seek renewed agency input. Specify what evidence or answer would resolve the uncertainty and when it is needed.
This sequence is a reasoning aid, not a regulatory rule or numerical threshold. Its central justification is straightforward: consultation has value when it can influence a consequential choice. If the change cannot affect the intended use, evidence plan, or decision, renewed discussion may add little. If plausible interpretations of the new evidence lead to different development actions, timely advice may prevent resources being committed to a strategy that no longer fits the evidence.
Worked Examples
Worked Example: New Clinical Data Challenge a Dose Prediction
Hypothetical scenario. A team discussed using a population PK model to support a dose for adults with moderate renal impairment. The strategy assumed that renal function had no clinically meaningful effect on exposure after accounting for a specified covariate relationship. Later, a small study reports dose-normalized exposure ratios of 1.00 in the reference group and 1.35 in the impairment group. For this illustration, suppose the team’s prespecified review range is 0.80 to 1.25. The new point estimate is above that range, but the study is small and its uncertainty interval is wide.
What changed? There is direct new evidence relevant to an assumption central to the intended use. The point estimate, 1.35, exceeds the upper review boundary of 1.25 by \(1.35-1.25=0.10\). This comparison is a screening result, not proof that the true exposure ratio is outside the range: the uncertainty interval and study limitations must also be examined.
Decision impact. If the higher exposure is credible, the proposed dose may need adjustment or additional evaluation. If it is largely sampling variation, the existing recommendation may remain appropriate. Because these interpretations could lead to different dosing actions, the change is decision-relevant. The team should assess study quality, model fit and predictive performance, and sensitivity to alternative renal-function relationships before treating the original strategy as unchanged.
Next step. Prepare a focused consultation that presents the original assumption, the new estimate and its uncertainty, the analyses performed or planned, and the potential dosing consequences. Ask whether the proposed evidence package is adequate to support the dose recommendation across the intended renal-function range. Do not present the small study’s point estimate alone as establishing a required dose change.
Worked Example: A Changed Assumption May Not Change the Decision
Hypothetical scenario. A model-based trial simulation was discussed using an assumed treatment effect of 20 percentage points. A new pilot analysis supports a smaller effect of 16 percentage points. Under the team’s stipulated simulation results, a trial with 100 participants per arm has an 80% probability of meeting its success criterion at a 16-point effect, while the agreed design criterion is at least 75%. At a 20-point effect, the corresponding probability was 90%.
Quantify the change. The effect assumption falls by \(20-16=4\) percentage points. The estimated probability of meeting the criterion falls by \(90\%-80\%=10\) percentage points. However, the revised result remains \(80\%-75\%=5\) percentage points above the stipulated design criterion.
Interpretation and response. The new assumption affects predicted performance, but under these hypothetical results it does not cross the stated decision criterion. That does not make the change irrelevant: the team should verify that the simulation assumptions and uncertainty have been appropriately examined. If the result is robust and the intended use and decision remain the same, documenting the update and its sensitivity analysis may be proportionate. Renewed consultation becomes more compelling if plausible effect values near 16 points move the trial below the criterion, or if the agency’s earlier feedback was explicitly conditional on the 20-point assumption.
Next step. Report both scenarios, the 75% criterion, and the sensitivity of the design decision to the effect assumption. If advice is sought, frame the question around whether the revised simulation evidence is sufficient for the design decision, rather than asking generally whether the model is acceptable.
Worked Example: A Changed Population Alters Model Applicability
Hypothetical scenario. A strategy was discussed for predicting exposure in adults using a model supported by data from adults aged 18 to 75 years. The development plan now proposes applying the same model to adolescents aged 12 to 17 years. The team’s preliminary analysis predicts a median exposure of 100 units in adults and 82 units in adolescents at the same nominal dose. For this illustration, suppose the team uses 90 to 110 units as an internal target range for comparing these predictions.
Calculate the comparison. The adolescent prediction is \(100-82=18\) units below the adult prediction. Relative to the adult prediction, this is \(18/100=0.18\), or an 18% decrease. The adolescent prediction of 82 units is also \(90-82=8\) units below the lower edge of the stipulated internal target range.
What the calculation does and does not show. It flags a potentially important difference under the model; it does not establish that adolescent patients will have lower exposure by exactly 18%, nor that the target range is a regulatory acceptance boundary. The adult evidence may not establish predictive performance in adolescents, and the assumptions behind the age-related prediction require evaluation.
Next step. The intended population has changed, and the prediction could affect dose selection. The team should assess the evidence supporting the age-related assumptions and the uncertainty in the prediction, then consider seeking advice on the proposed extrapolation and any additional evidence needed. The consultation should make clear that the population expansion is the change, show the 100- and 82-unit predictions with their limitations, and ask a decision-linked question about the evidence package for the adolescent dose.
Preparing the Explanation and Request
A useful explanation is concise enough to expose the decision logic but detailed enough for the agency to evaluate the evidence. It should not hide uncertainty behind a single preferred model result. A clear account generally includes:
- Agreed baseline: the original model purpose, population, assumptions, decision, and relevant scope of feedback.
- Change and evidence: what became available or changed, when it occurred, and how reliable and applicable it is.
- Model consequence: which assumptions, predictions, diagnostics, or uncertainty assessments are affected, and which remain unchanged.
- Decision consequence: what action the original and revised strategies would support, including whether the action changes under plausible alternative assumptions.
- Proposed response: analyses, data collection, strategy revisions, or consultation requested, together with timing and the development decision they inform.
The consultation question should state the choice on which advice is sought. For example: “Given the new renal-impairment data and the resulting uncertainty in exposure, is the proposed sensitivity analysis and confirmatory evidence sufficient to support the planned dose recommendation?” This is more actionable than “Does the agency agree with our model?” The team should distinguish the question requiring advice from background information and should avoid implying that prior feedback has already resolved a materially different use.
Diagnostics, Alternatives, and Limitations
Before deciding that consultation is warranted—or that it is unnecessary—check whether the apparent change is robust and relevant. Useful checks include whether the new data are from the intended population and conditions; whether the data quality and analysis are fit for the question; whether model diagnostics reveal systematic misfit; whether predictions are sensitive to the challenged assumption; and whether uncertainty spans different decisions. Verify that comparisons use the same endpoint, population, time horizon, and decision criterion. A numerical shift can be misleading when these quantities are not comparable.
There are several proportionate responses. The team may document an update and proceed when the change is minor and does not alter the decision or intended use. It may first run additional analyses when the key uncertainty is addressable with available data. It may revise the model strategy when evidence shows that the original approach is no longer suitable. Renewed consultation is especially useful when the choice between these responses is consequential, when the evidence supports materially different interpretations, or when the changed use extends beyond the scope previously discussed.
These options are not mutually exclusive: a team may perform sensitivity analyses and then consult on the remaining uncertainty. Consultation also has limits. It cannot make weak evidence strong, replace appropriate model evaluation, or guarantee a particular regulatory outcome. The team remains responsible for transparently describing limitations and for supporting its development decisions. The appropriate interaction pathway and timing depend on the question and development context; the purpose here is to identify when the strategy merits reconsideration, not to prescribe a particular meeting mechanism.
- Underreacting: continuing unchanged because the strategy was previously discussed, despite evidence that challenges a decision-critical assumption.
- Overreacting: reopening consultation for every numerical update without showing how the change could affect a decision.
- Overstating certainty: treating a point estimate or a model prediction as a confirmed clinical effect, without accounting for uncertainty and applicability.
- Asking too broadly: requesting general agreement with a model rather than identifying the specific decision, uncertainty, and proposed next step.
Check Your Understanding
Use the stated information in each question. Suggested solutions follow the questions.
- A new dataset shifts a model prediction by 3%, but the intended use, decision, and sensitivity analyses are unchanged. What should the team establish before deciding whether to consult again?
- Earlier feedback addressed adults, but the proposed strategy now includes adolescents. Why is this more than a routine data update?
- A revised analysis changes the probability of trial success from 78% to 72%. The team’s stated design criterion is at least 75%. What is the decision-relevant consequence?
- Name four elements that should appear in a focused explanation of a strategy revision.
Solutions. (1) Verify that the new dataset is reliable and relevant, that the prediction comparison is like-for-like, and that plausible uncertainty does not change the decision or expose a limitation in the original strategy. If those checks support a minor, decision-neutral update, documentation and additional analysis may be sufficient; the 3% shift alone does not determine the response. (2) The intended population has expanded beyond the scope discussed. The team must assess whether the model’s assumptions and evidence support use in adolescents and whether the prediction informs a different decision. (3) The revised probability is \(72\%-75\%=-3\) percentage points relative to the criterion, so it falls below the team’s stated design threshold. The strategy may therefore support a different trial-design action; the team should examine uncertainty and consider whether revised analyses or renewed advice are needed. (4) Suitable elements include the agreed baseline and its scope, the new evidence or changed assumption, the effect on model predictions and uncertainty, the consequence for the decision, and the proposed analyses or specific question for advice.