From a Finding to an Actionable Comment
A peer review identifies problems; a reviewer comment makes clear what should happen next. The difference matters in model-informed drug development (MIDD), where a missing analysis detail can affect whether a model result supports a dose, a subgroup claim, or a development decision. A vague comment may be accurate but leave the author unsure what evidence to supply and the reviewer unable to tell whether the concern has been addressed.
The aim is not to prescribe a preferred scientific conclusion. It is to connect a specific dossier claim to the evidence needed to assess it, then record what changed or why no change was made. Reviews of data provenance, model specification, estimation, diagnostics, validation, covariates, exposure-response, simulation, sensitivity, linked-model uncertainty, reproducibility, and cross-document consistency can all lead to comments. The comment-writing task is to state the relevant gap precisely, not to repeat the entire review.
A useful comment is evidence-focused and proportionate. It distinguishes what the dossier states from what the reviewer can verify. “The model is poorly documented” is a judgement without a defined target. “Section 4.2 reports the final clearance estimate but does not state the units or parameterization; please add both to the model description and confirm that the units in Table 7 are consistent” identifies a location, an observable omission, and a checkable remedy.
A Practical Structure for the Comment and Its Resolution
Use the following elements as a compact reasoning sequence. They do not need to appear as separate headings in every comment, but each should be clear enough that an author and a later reviewer can follow the issue.
Name the section, table, figure, model output, or simulation result and quote or accurately paraphrase the claim at issue. Include a page, section, or table identifier when available.
State what is absent, inconsistent, unsupported, or not reproducible. Separate the observed fact from an inference about its consequences.
Show how the gap limits interpretation of a specific result or decision. Avoid asserting that a result is invalid unless the evidence establishes that conclusion.
Specify the information, analysis, correction, or narrower wording needed. State what evidence would allow the concern to be checked.
After the response, identify the submitted evidence and document whether the original concern is resolved, partly resolved, or remains open, with a brief reason.
The justification for this sequence follows from what it takes to close a review issue. A claim depends on evidence and assumptions. If a relevant gap prevents a reviewer from checking that support, a remedy must produce information that addresses the gap; otherwise the original uncertainty remains. Therefore, a response is not sufficient merely because it is confident or extensive: it must be linked to the requested evidence and that evidence must be checkable against the claim. This reasoning also prevents overreach: if the gap concerns reporting, request clarification or supporting output before demanding a new analysis that may not be necessary.
A request should be specific without being needlessly prescriptive. “Provide the analysis dataset, executable code, and all exploratory outputs” may exceed what is needed to resolve a narrow discrepancy. Conversely, “please clarify” does not say what needs clarification. A well-scoped request names the minimum information needed to judge the claim, such as the analysis population, model equation, diagnostic output, simulation scenario, or versioned artifact.
Worked Example: A Covariate Claim Without Its Supporting Estimate
Hypothetical dossier. Section 5.1 states that body weight has no clinically relevant effect on exposure and that one dose is appropriate across the studied weight range. Table 9 reports a final population model, but does not show the weight effect estimate, its uncertainty, or a prediction of exposure across the weight range. The dossier gives a studied range of 48 to 112 kg.
Identify the claim and gap. The claim in Section 5.1 concerns the size and clinical relevance of the weight effect. The table’s omission of an estimate and uncertainty is observable; it does not, by itself, prove that the effect is important or that the dosing conclusion is wrong.
Connect the gap to the decision. Without an estimate and an interpretable exposure comparison across the stated range, the reader cannot assess how the model supports the conclusion that a single dose is appropriate. The concern is therefore about the evidentiary bridge between the model and the dosing claim, not simply about table completeness.
Reviewer comment. “Section 5.1 concludes that body weight has no clinically relevant effect and supports one dose across 48–112 kg. Table 9 does not report the weight effect estimate or its uncertainty, and the dossier does not show the predicted exposure contrast across this range. Please provide the estimated weight effect with its uncertainty and a model-based exposure comparison at representative values spanning the stated range, with the assumptions and interpretation stated. If these results do not support the current wording, please qualify the dosing claim.”
Hypothetical response and disposition. The authors add the covariate estimate and uncertainty interval, provide predicted exposures at 48, 80, and 112 kg, and explain the model’s covariate parameterization. The predictions show a modest increase with weight, while the authors’ stated exposure-based decision criterion remains met. The reviewer checks that the added output uses the stated model and range. The comment can be marked resolved if the evidence supports the qualified claim; it should not be marked resolved merely because the requested numbers were added if their interpretation still contradicts the conclusion.
Make the Requested Remedy Testable
A remedy is verifiable when another reviewer can identify what changed and assess whether it addresses the stated gap. Depending on the issue, useful evidence may be a revised sentence, a corrected table, a model equation, a diagnostic plot, a sensitivity analysis, a simulation summary, or a reproducible computational artifact. “We have clarified this” is difficult to assess unless the response points to the changed location and the clarification is visible.
Keep the request aligned with the type of problem. A missing definition may need a reporting correction; a numerical discrepancy may need reconciliation across outputs; an unsupported claim may require additional evidence or narrower wording. A reviewer should not automatically demand a new model run whenever a result is unclear. First ask whether the existing analyses can answer the question and whether their methods and outputs are sufficiently documented.
Worked Example: An Unspecified Simulation Scenario
Hypothetical dossier. A simulation summary reports that a proposed dose achieves a target response in 87% of simulated participants. The methods section describes the model and number of replicates but does not state the distribution of baseline covariates used to generate the participants. The dossier uses the 87% result to support a dosing recommendation.
Why the omission matters. The reported percentage is conditional on the simulated population and scenario. If the covariate distribution is not stated, a reader cannot judge which population the operating characteristic represents or reproduce the stated scenario. The omission does not establish that 87% is incorrect.
Reviewer comment. “The simulation summary in Section 6.3 reports 87% target attainment and uses this result to support the proposed dose, but the simulation methods do not specify the baseline covariate distributions used to generate participants. Please report the distributions and parameter values used for each simulated covariate, and identify whether they match the intended population. If they do not, please provide results for the intended population or qualify the scope of the recommendation.”
Hypothetical response and disposition. The authors add a table of covariate distributions and parameter values, identify the target population, and provide an additional simulation for that population. The reviewer checks that the new scenario is described, that its reported result corresponds to that scenario, and that the recommendation is consistent with the result. If the added simulation changes the attainment rate materially, the original result may be correctly reported yet insufficient to support the original recommendation; the response should then revise the recommendation or explain the remaining limitation. The issue is closed only when the claim and its stated population are aligned with the evidence.
Comments should also distinguish required correction from a scientifically reasonable alternative. If more than one remedy could address the gap, state the evidentiary objective and allow the authors to propose a suitable approach. For example, a concern about uncertainty propagation might be addressed by providing a missing uncertainty component, demonstrating why it is negligible, or qualifying the downstream conclusion. The reviewer can judge the response against the objective without dictating a single method in advance.
Documenting Responses and Closure
A useful response record has four parts: the comment identifier, the author’s reply, the specific change or evidence location, and the reviewer’s disposition with rationale. For an unresolved issue, state what remains missing and what evidence would close it. This record makes the review traceable and prevents a revised sentence from being mistaken for a corrected analysis.
Use “resolved” when the requested evidence or correction is present and addresses the concern. Use “partly resolved” when some elements are supplied but a material part remains unaddressed. Use “unresolved” when the response does not supply the requested evidence, when the evidence cannot be checked, or when it reveals that the claim remains unsupported. A claim can also be revised or withdrawn; in that case, the reviewer should check that the revised dossier no longer relies on the original unsupported wording elsewhere.
Worked Example: A Response That Does Not Reconcile a Discrepancy
Hypothetical dossier. Table 4 reports 312 participants in the analysis set, while the model-results section says 298 participants contributed to the final fit. The report does not explain the difference. The authors reply, “The model used the appropriate patients.”
Why the reply is insufficient. It does not identify which participants were excluded, give the analysis count for the final fit, or reconcile the two reported counts. The assertion that the participants were appropriate does not let a reviewer verify the population underlying the model result.
Reviewer follow-up. “Thank you for the response. The count difference remains unexplained. Please identify the analysis set used for the final fit, report the number excluded relative to the 312 participants in Table 4, and give the reason for each exclusion category. Please also correct or cross-reference the dossier counts so the modeled population is clear.”
Hypothetical resolution. The authors provide a reconciliation showing that 14 participants lacked the required concentration records for the specified analysis, update the model-results text to 298, and cross-reference the exclusion description. The reviewer checks that the totals reconcile (\(312 - 14 = 298\)) and that the reason matches the stated analysis requirements. The issue is resolved if the reconciliation and affected text are consistent. If the response only changes 312 to 298 without documenting the excluded participants or reason, the count discrepancy is corrected but the evidentiary question is only partly resolved.
Prioritization, Diagnostics, and Common Pitfalls
Not every comment has the same decision importance. Prioritize issues that could materially alter interpretation, the target population, dose selection, or the strength of a development conclusion. A typographical error with no effect on interpretation still merits correction when it could mislead, but it should not be framed as equivalent to an unsupported dosing claim. Priority is about the consequence of the gap, not the forcefulness of the wording.
Before sending a comment, check that it can be acted on and later assessed. Ask: Is the claim located? Is the gap observable rather than speculative? Does the stated consequence follow from that gap? Does the remedy address it? Is there a clear criterion for closure? If any answer is no, revise the comment. When evidence is unavailable or the analysis is outside the dossier’s scope, a proportionate resolution may be to state the limitation and narrow the claim rather than imply that an unperformed analysis has been completed.
A common mistake is to write, “The subgroup result is not credible; redo the analysis.” This leaps from a concern to a conclusion, offers no evidence for the judgement, and does not say which result or feature is problematic. A better comment names the subgroup estimate and relevant uncertainty or diagnostic gap, explains why it limits the stated subgroup interpretation, and requests the missing evidence or a suitably qualified claim. The reviewer should not treat a non-significant result as proof of no effect, or an association as proof of causation; the requested remedy must preserve the limits of the underlying evidence.
| Weak approach | Why it fails | More actionable approach |
|---|---|---|
| “Clarify the model.” | No component or desired clarification is identified. | Name the missing specification, its location, and the information needed to evaluate it. |
| “The conclusion is wrong.” | States a verdict without showing the evidence gap or decision relevance. | Identify the result that does not support the wording and request evidence or narrower wording. |
| “Please rerun everything.” | Is broad and may require work unrelated to the concern. | Request the specific analysis or artifact needed to resolve the stated uncertainty. |
| “Addressed.” | Leaves no record of what changed or why closure is justified. | Record the response location, check performed, and reason for the disposition. |
There are limits to what comment writing can achieve. A precise request cannot compensate for inadequate underlying data, an unidentifiable model, or an analysis that was not conducted. Nor should a reviewer demand certainty where the evidence supports only a qualified conclusion. The appropriate outcome may be a transparent limitation, a narrower claim, or an open concern. The value of the process is that the dossier’s evidentiary status and its implications for the decision are made explicit.
Check Your Understanding
For each item, identify the key improvement or disposition. Suggested answers follow the questions.
- A report claims that a model is predictive, but gives no validation results. What should a reviewer request, and why?
- An author replies to a request for units by saying, “Units are standard.” Is the concern resolved? Explain.
- A reviewer notes that a dosing conclusion is supported only for adults, while the dossier wording says “all patients.” State a proportionate comment objective.
- A corrected table resolves a numeric mismatch, but the text still cites the old value. How should the reviewer disposition the response?
Solutions. (1) Locate the predictive claim and request the validation evidence relevant to that claim, with enough description to identify the evaluated data or predictions; without it, the reader cannot assess the stated predictive support. (2) No. The response does not identify the units or permit verification; request that the units be explicitly stated at the relevant parameter or result. (3) Ask the authors to provide evidence supporting the broader population or narrow the wording to the population actually supported; do not imply that adult evidence establishes the claim for all patients. (4) Mark the issue partly resolved or unresolved, depending on whether the text remains consequential: the table correction alone does not make the dossier internally consistent, so request correction of the text and recheck affected references.