Why Traceability Matters
A development report may contain sound data, a well-executed model analysis, and a reasonable decision, yet still leave the central claim unsupported. The problem is often not a missing result but a missing link: the report does not show how the data and analysis justify the result, or how that result supports the claim and decision. Traceability makes those links visible for review.
For example, a report might state that a proposed regimen is appropriate because simulated exposures are “similar.” A reviewer needs to know which data informed the model, what comparison was made, what range counts as similar, and how the result bears on the regimen decision. A traceable argument answers those questions without asking the reader to infer the reasoning from scattered tables and prose.
Building a Claim-to-Evidence Matrix
Start with a claim that is specific enough to review. Then record the evidence chain in separate columns. The exact layout can vary, but the matrix should distinguish what was observed, what was analyzed, what the model reported, what is being claimed, and what decision follows. Keeping these items separate prevents a decision from being mistaken for evidence or a model output from being treated as the claim itself.
| Matrix field | Question for the reviewer |
|---|---|
| Development claim | What statement is being made, and for which population, regimen, or development question? |
| Supporting data | Which observations or data summaries are cited as evidence? |
| Analysis | What model-based analysis connects those data to the reported result? |
| Model result | What estimate, prediction, comparison, or uncertainty summary is reported? |
| Interpretive bridge | Why does this result support the claim, and what conditions or thresholds are used? |
| Decision | What development action is informed by the claim? |
A useful matrix gives enough detail to locate the relevant evidence in the report, such as a table, figure, analysis section, or data summary. It should not merely repeat a conclusion in several columns. If the claim says “the regimen maintains target exposure,” the result column should identify the exposure measure and its value; the bridge should state the target and how the result was compared with it.
A compact notation can clarify the logic. Let \(D\) denote the supporting data, \(A\) the analysis, \(R\) the reported model result, \(C\) the claim, and \(J\) the development decision. The intended chain is:
This notation does not say that each arrow is automatic. It emphasizes that each transition requires an explanation. A result can be accurately reported but irrelevant to the claim; a relevant claim can still be insufficient to justify the decision. The matrix is an audit of the reasoning, not a method for proving that the model is correct.
A Short Justification: Why the Links Must Be Checked Separately
Suppose a decision is justified by a claim, and that claim is said to follow from a model result. For the decision to be supported by the evidence, the result must support the claim and the claim must support the decision. Likewise, the result is only evidence for the claim if its analysis is connected to the relevant data and the interpretation is warranted. Thus, a chain with a broken link cannot be repaired just by making another link more persuasive.
This is a logical justification, not a numerical formula for combining evidence. In particular, the links are not probabilities that can be multiplied unless a separate probabilistic model and its assumptions justify such a calculation. The practical consequence is simple: review every arrow, not only the final model estimate.
Worked Example: Testing a Claim About Exposure in Renal Impairment
Hypothetical scenario. A development team proposes that the standard dose can be used in patients with moderate renal impairment without adjustment. The report summarizes a model-based comparison of steady-state exposure with that in patients with normal renal function. For this example, the team defines a geometric mean exposure ratio from 0.80 to 1.25 as the range that would support its stated “similar exposure” rationale.
Data and analysis. The report identifies concentration-time observations from 18 participants with moderate renal impairment and 20 with normal renal function. A population PK analysis is used to estimate steady-state AUC for each group under the same dose. The matrix should point to the data summary and the analysis result rather than treating the participant counts alone as evidence of similarity.
Model result. The estimated geometric mean ratio, moderate impairment divided by normal renal function, is 1.18, with a 90% confidence interval from 1.04 to 1.34. The point estimate is within the stated range, but the interval extends above 1.25.
Traceability assessment. The data-to-analysis link is described, and the result-to-threshold comparison is explicit. However, if the claim requires evidence that the ratio is within 0.80 to 1.25, the reported interval does not establish that: it includes values above the upper bound. The point estimate alone does not remove this uncertainty. The matrix should mark the “no dose adjustment” claim as not fully supported by this exposure-similarity rationale, rather than silently converting a point estimate into a conclusion about the entire plausible range.
Decision implication. The development team may still consider other evidence when deciding on a dose, but the matrix should not imply that this particular model result proves the stated exposure claim. The inferential gap is the leap from a point estimate inside the range to a claim that exposure is adequately similar, despite uncertainty extending beyond the range.
Reviewing a Matrix for Missing Bridges
The most useful review question is not only “Is there a model result?” but also “What must be true for this result to support this claim?” Check that the report states the comparison, the relevant target or criterion, and the interpretation connecting them. A result about average exposure, for instance, does not by itself establish a claim about the proportion of individuals who attain a target. Those are different quantities, and the bridge between them must be supplied by an appropriate analysis and interpretation.
The matrix also helps distinguish three common situations:
- Supported link: the report identifies the evidence and explains why it supports the next step.
- Unclear link: the report may contain relevant evidence, but the reader cannot tell which result supports the claim or how it was interpreted.
- Unsupported link: the result does not establish the claim as stated, or a necessary comparison or premise is absent.
A gap should be recorded as a gap, not concealed by stronger wording. A reviewer can then ask whether the report needs a narrower claim, an additional analysis, a clearer decision criterion, or a qualification describing what the result does and does not establish. The matrix does not determine which remedy is appropriate; it makes the choice visible.
Worked Example: Auditing a Long-Acting Formulation Claim
Hypothetical scenario. A team claims that a long-acting formulation will maintain concentrations above a prespecified pharmacologic target throughout the dosing interval for at least 95% of patients. The report describes simulations for 1,000 virtual patients. In 930 simulations, the minimum concentration over the interval exceeds the target.
Calculation. The simulated proportion is:
Traceability assessment. The simulation result is directly relevant to the claim, but it does not support the stated 95% threshold: the estimated proportion is 93.0%, which is 2.0 percentage points below it. The matrix also needs to make clear whether the claim is based on this simulation estimate alone or on an additional uncertainty analysis. Without that bridge, it would be incorrect to report that the 95% claim has been demonstrated.
Decision implication. If the development decision requires at least 95% of patients to remain above target, this result does not meet that criterion as stated. The decision could be reconsidered using additional evidence or a revised regimen, but the report should not present the 93.0% simulation result as if it met a 95% requirement.
Comparing Traceability Approaches
A matrix is especially useful when several claims draw on different analyses or when a decision depends on a sequence of model-based results. It is not the only way to communicate evidence. A well-organized narrative can explain the same chain, but prose can make omissions harder to spot. A result-only summary is concise, yet it can leave the reader unable to determine what evidence supports a claim or what decision criterion was applied.
| Approach | Strength | Common limitation |
|---|---|---|
| Claim-to-evidence matrix | Makes claim, result, interpretation, and decision links visible side by side. | Can become a checklist with vague entries unless each cell points to substantive evidence. |
| Narrative rationale | Can explain context and qualifications in a readable sequence. | Links may be dispersed across sections or implied rather than stated. |
| Result-only summary | Efficient for communicating key estimates. | Does not, by itself, show why a result supports a claim or decision. |
These approaches can complement one another: the matrix can serve as a review map while the narrative explains the reasoning in full. Neither format substitutes for evidence. A complete-looking table cannot make an unsupported inference sound, and a polished narrative cannot compensate for a missing result.
Diagnostics, Failure Modes, and Limitations
Several warning signs indicate that traceability is weak. A claim may have no identifiable supporting result; a result may be listed without the data or analysis that produced it; or the report may change the quantity being discussed between the result and claim. Another warning sign is a decision that appears in the conclusion but is not tied to a stated criterion or rationale. These are reporting and reasoning concerns; finding them does not by itself establish that the analysis is invalid.
A matrix also has limits. It can organize the stated argument but cannot certify the quality of the underlying data, model, or implementation. Reviewers may need separate checks of those components. In this tutorial, the focus is whether the development claim can be followed through its supporting evidence and reasoning—not a full technical evaluation of every component of the analysis.
Worked Example: Separating a Model Result From a Safety Claim
Hypothetical scenario. A report uses a model to predict exposure under a proposed coadministration regimen. The estimated exposure ratio is 1.42 relative to the reference regimen, and the report concludes, “The combination is safe.” The model analysis is described, but the report provides no safety endpoint, exposure-response result, or prespecified safety criterion connecting the predicted ratio to the safety conclusion.
Traceability assessment. The matrix can record the exposure result as a model output and the safety statement as the claim. The critical bridge is missing: an exposure ratio, by itself, does not establish that safety is acceptable. The claim is therefore not traceable to the result as presented, even if the exposure prediction is accurately reported.
Correction. The report should either narrow its statement to what the result supports, such as “the model predicts a 1.42 exposure ratio under the specified regimen,” or provide relevant evidence and reasoning that support the safety claim. The decision may be described separately, with its rationale, rather than being used as proof of the claim.
Common Mistake
Wrong reasoning: “The model result is in the report, and the development decision was made, so the claim is supported.” This confuses the presence of a result and a decision with a justified connection between them.
Correction: Ask what data produced the result, how the analysis connects those data to the result, what interpretation supports the claim, and why that claim informs the decision. If one link is missing, mark it explicitly. A decision can be reasonable for broader reasons, but those reasons should not be attributed to a model result that does not establish them.
Check Your Understanding
For each question, identify the traceability issue or the appropriate conclusion.
- A model predicts a 90% probability of meeting an exposure target, but the claim says that at least 95% of patients will meet it. Is the claim supported by that result?
- A report lists a model estimate but does not identify the supporting data. Which link in the matrix is unclear?
- Why does a decision to use a regimen not, by itself, show that a model-based claim supporting the regimen is valid?
- A hypothetical simulation has 760 of 800 virtual patients above target. Calculate the simulated proportion and state whether it meets a stated requirement of at least 95%.
- A report predicts an exposure ratio of 1.20 and claims there is no safety concern, but gives no safety criterion or supporting safety analysis. What should the matrix show?
Solutions. (1) No. A 90% result is below the 95% threshold; the stated claim is not supported by that result. (2) The data-to-analysis or data-to-result link is unclear because the report does not identify what evidence produced the estimate. (3) A decision is an action informed by evidence, not evidence itself; the report must explain how the result supports the claim and how the claim supports the decision. (4) \(\hat{p}=760/800=0.950=95.0\%\), so the point estimate meets the stated threshold exactly. This calculation alone does not establish additional claims about uncertainty. (5) Record the ratio as the model result and mark the result-to-safety-claim bridge as unsupported or missing; an exposure ratio alone does not establish the safety conclusion.