10
TPP attributes traced
7
Chains that carry weight
3
Complete but not carrying
3
Backward orphans
Contents
1. The Chain Is Not Requirement to Test Case
Traceability in a development program runs from the commercial promise to the words a sales representative may say. Five links, and the program's entire argument lives in whether they hold.
| Link | What it fixes | When it is set |
|---|---|---|
| TPP attribute | What the product must be able to claim to be commercially viable | Gate 0, revised by governance only |
| Protocol endpoint | What will be measured, in whom, over what period | Before Phase 3 opens. ⚠ Effectively irreversible once participants are enrolled. |
| SAP analysis | How it will be tested, and in what order | ⚠ Signed before database lock. The testing hierarchy is fixed here. |
| CSR section | Where the result is reported | After lock, from the locked data |
| Label claim | What may actually be said | Negotiated with the agency at review — and determined by the four links above |
The distinctive property of this chain is that the last link cannot be repaired by working
harder on the others.
In a systems project, a requirement that fails its test gets a fix and a retest. Here, if the analysis at hierarchy position 4 does not reach significance, there is no retest. The participants are enrolled, the data are locked, the analysis was pre-specified, and the claim is simply not available.
Which makes traceability in this setting a design-time discipline rather than a verification one. By the time you can check whether the chain held, every decision that determined it is years old.
In a systems project, a requirement that fails its test gets a fix and a retest. Here, if the analysis at hierarchy position 4 does not reach significance, there is no retest. The participants are enrolled, the data are locked, the analysis was pre-specified, and the claim is simply not available.
Which makes traceability in this setting a design-time discipline rather than a verification one. By the time you can check whether the chain held, every decision that determined it is years old.
2. The Matrix
| Ref | TPP attribute | Protocol endpoint | SAP analysis | CSR section | Label target | Weight |
|---|---|---|---|---|---|---|
| T-01 | Mean weight reduction ≥15% at week 68 | Co-primary endpoint, both pivotals | Primary analysis, hierarchy position 1 | §11 Efficacy | §14 Clinical Studies | carries |
| T-02 | Responder rate ≥80% achieving ≥5% reduction | Co-primary endpoint, both pivotals | Primary analysis, hierarchy position 2 | §11 Efficacy | §14 Clinical Studies | carries |
| T-03 | Proportion achieving ≥10% reduction | Key secondary | Hierarchy position 3 | §11 Efficacy | §14 Clinical Studies | carries |
| T-04 | GI-attributed discontinuation ≤4% | Key secondary, vs active comparator | Hierarchy position 4 | §11 Efficacy and §12 Safety | ⚠ Comparative claim SOUGHT — not granted unless positions 1–3 all succeed | complete, does not carry |
| T-05 | Once-weekly dosing, maintenance within 20 weeks | Fixed by protocol design | Descriptive; exposure summary | §9 Investigational plan | §2 Dosage and Administration | carries |
| T-06 | No unexpected serious safety signal | Safety monitoring throughout; DMC review | Pooled safety analysis | §12 Safety | §5 Warnings and §6 Adverse Reactions | carries |
| T-07 | Waist circumference reduction | Secondary | Hierarchy position 5 | §11 Efficacy | ⚠ Below the likely cut — supportive text only | complete, does not carry |
| T-08 | Patient-reported physical function | Secondary | Hierarchy position 6 | §11 Efficacy | ⚠ Below the likely cut — supportive text only | complete, does not carry |
| T-09 | Cardiovascular safety adequate for the indication | CV sub-study (CR-02), 640 participants | Separate analysis, not in the hierarchy | Standalone CSR | §5 Warnings — and pre-empts a post-marketing requirement | carries |
| T-10 | Chronic weight management indication as sought | Enrolled population and inclusion criteria | Population definitions, blind data review | §10 Study participants | §1 Indications and Usage | carries |
7 of 10 chains carry weight. The other 3 are structurally complete — every link present, every analysis pre-specified, every result reported — and still cannot support a claim.
3. A Complete Chain That Does Not Carry
T-04 is the one that matters, and it is worth walking end to end because every link is sound.
| Link | T-04 — GI-attributed discontinuation | Status |
|---|---|---|
| TPP attribute | ≤4% target, ≤7% minimum acceptable. The differentiator. | ✔ Present since v1.0 |
| Protocol endpoint | Key secondary, measured against an active comparator | ✔ Collected in both pivotals |
| SAP analysis | Pre-specified, hierarchy position 4 of 6 | ✔ Signed before lock |
| CSR section | Reported in §11 and §12 | ✔ Will be reported whatever the result |
| Label claim | ⚠ Comparative tolerability claim in §14 | Only if positions 1–3 all succeed first |
The chain is complete and the claim may still not exist, because the testing hierarchy is a
gate that sits between the analysis and the label.
Testing stops at the first failure. If any of the three positions above T-04 misses significance, testing stops there and everything below becomes descriptive — the number is still reported in §6 Adverse Reactions, and it simply may not be stated as a comparison anywhere.
A traceability matrix that showed only link presence would mark T-04 fully traced and green. That is why this matrix carries a weight column: traced and load-bearing are different properties, and only the second one is worth anything.
Testing stops at the first failure. If any of the three positions above T-04 misses significance, testing stops there and everything below becomes descriptive — the number is still reported in §6 Adverse Reactions, and it simply may not be stated as a comparison anywhere.
A traceability matrix that showed only link presence would mark T-04 fully traced and green. That is why this matrix carries a weight column: traced and load-bearing are different properties, and only the second one is worth anything.
T-07 and T-08 fail the same way and matter less — they were always supportive rather than differentiating. The difference is that nobody built a commercial case on waist circumference. The business case's share assumption rests on T-04, which makes it the single most consequential row in the matrix and the only one anybody should be losing sleep over.
4. Both Directions
Forward traceability asks whether every requirement is evidenced. Backward asks whether every piece of evidence serves a requirement. Programs check the first and skip the second.
| Direction | Found | What it means |
|---|---|---|
| Forward — a TPP attribute with no analysis that tests it | 0 | Every attribute in the profile maps to at least one pre-specified analysis. ⚠ This is the direction most programs check, and it is the easier one. |
| Backward — an analysis serving no TPP attribute | 2 | ⚠ Two exploratory analyses in the SAP trace to no profile attribute. Both are legitimate science; neither can support a claim, and reporting them alongside pre-specified analyses is how a CSR loses a reviewer's trust. |
| A claim sought with no analysis positioned to deliver it | 0 | The comparative claim IS positioned — at hierarchy position 4. The link exists; it simply may not carry weight. |
| A protocol endpoint collected but never analyzed | 1 | ⚠ One exploratory biomarker is collected and has no analysis in the SAP. Collecting data nobody will analyze imposes burden on participants for no evidentiary return. |
The backward direction found three things the forward direction cannot see.
Two exploratory analyses in the statistical analysis plan trace to no profile attribute, and one biomarker is collected with no analysis planned at all. None is misconduct — all three are legitimate scientific curiosity that survived into a regulated document.
But the collected-and-never-analyzed biomarker is a real cost: it imposes a blood draw on 2,480 participants for no evidentiary return. And the two orphan analyses carry a subtler risk — reported alongside pre-specified analyses in a CSR, they invite a reviewer to ask which of the others were also decided after the fact.
Two exploratory analyses in the statistical analysis plan trace to no profile attribute, and one biomarker is collected with no analysis planned at all. None is misconduct — all three are legitimate scientific curiosity that survived into a regulated document.
But the collected-and-never-analyzed biomarker is a real cost: it imposes a blood draw on 2,480 participants for no evidentiary return. And the two orphan analyses carry a subtler risk — reported alongside pre-specified analyses in a CSR, they invite a reviewer to ask which of the others were also decided after the fact.
The fact base asserts that backward orphans must be greater than zero, on the reasoning that a matrix reporting none has almost certainly only been walked in one direction. Any sufficiently large program accumulates analyses nobody can trace to a requirement; finding zero is evidence about the search, not about the program.
5. Using It for Change Impact
The matrix earns its keep when somebody proposes a change, because it answers the question a change request cannot answer about itself: what else does this touch?
| Change | What the matrix showed | What it cost to know |
|---|---|---|
| CR-01 — add an Asian-population cohort to Phase 2 | Touches T-10 only. The cohort enlarges the population supporting the indication; no endpoint, analysis or claim changes. | A contained change. Approved without re-opening the SAP. |
| CR-02 — add a cardiovascular sub-study | ⚠ Creates a NEW chain (T-09) end to end — a new endpoint, a separate analysis, a standalone CSR and a §5 label target — without touching any existing chain. | Why it could be added mid-program at all. A change that had altered an existing analysis would have required re-opening the hierarchy. |
| A protocol amendment adding one assessment | Touches no chain — and touches three work packages and 2,480 participants. | ⚠ The matrix says it is harmless and the WBS says it is expensive. Both are right. |
| Re-positioning T-04 higher in the hierarchy | Touches T-01, T-02 and T-03 — every chain above it — because testing order is shared state. | The reason it was never done. A local change with program-wide consequence. |
That final row is the whole argument for maintaining a matrix at all.
“Move the tolerability endpoint up the hierarchy” sounds like a change to one endpoint. The matrix shows it is a change to four chains, because the testing hierarchy is shared state — every position depends on every position above it.
That is not obvious from the change request, from the protocol, or from the SAP read on its own. It is obvious from a matrix, and it is the kind of thing a program discovers expensively when nobody has one.
“Move the tolerability endpoint up the hierarchy” sounds like a change to one endpoint. The matrix shows it is a change to four chains, because the testing hierarchy is shared state — every position depends on every position above it.
That is not obvious from the change request, from the protocol, or from the SAP read on its own. It is obvious from a matrix, and it is the kind of thing a program discovers expensively when nobody has one.
The third row is the useful counter-example. A change can be trivial in traceability terms and severe in cost terms — an added assessment moves no claim and moves three work packages. A matrix is one input to a change decision, not the decision. Read alone it would have waved through the single most expensive category of protocol amendment this program can receive.
6. What Traceability Would Have Caught Earlier
| If the matrix had been built at | It would have shown | Actionable? |
|---|---|---|
| Gate 3 — before the Phase 3 protocol was fixed | ⚠ That the differentiating attribute sat below three endpoints it depended on, and that no head-to-head trial was planned to support it | Yes. The endpoint could have been powered differently, or a separate comparative study scoped. |
| Before the SAP was signed | That hierarchy position 4 made the commercial case conditional on three prior successes | ⚠ Partially. Position could still move, at the cost of risking the efficacy claim — which is why it was placed there. |
| After database lock | The same thing, definitively | No. By then it is a reporting fact. |
| At the label negotiation | That the claim was never available | No. And this is when most programs discover it. |
The matrix is most valuable at exactly the point it feels most premature.
At Gate 3 there are no results, no CSR sections and no label — three of the five columns are empty. It looks like an exercise in filling in boxes for a future that has not happened.
And that is the only moment when the answer it produces can still change anything. Built at Gate 3, this matrix would have shown a program whose commercial differentiation depended on an endpoint positioned to protect the approval rather than to deliver the claim — which is a correct choice, and one worth making consciously rather than discovering.
At Gate 3 there are no results, no CSR sections and no label — three of the five columns are empty. It looks like an exercise in filling in boxes for a future that has not happened.
And that is the only moment when the answer it produces can still change anything. Built at Gate 3, this matrix would have shown a program whose commercial differentiation depended on an endpoint positioned to protect the approval rather than to deliver the claim — which is a correct choice, and one worth making consciously rather than discovering.
7. Borrowed From Systems Delivery
| Systems traceability | Here |
|---|---|
| Requirement → design → build → test case → result | TPP → endpoint → analysis → CSR → label claim |
| A failed test produces a defect and a retest | ⚠ A failed analysis produces nothing. There is no retest; the participants are enrolled and the data are locked. |
| Coverage is the headline metric | ⚠ Coverage is necessary and nearly meaningless. Weight is the metric — whether the chain can support a claim. |
| Requirements can be re-baselined mid-build | The TPP can be revised by governance; the protocol cannot, once participants are enrolled under it. |
| Bidirectional checks are standard practice | Rarely done. ⚠ The backward walk is where the orphans are, in both disciplines. |
The discipline transfers; the assumption underneath it does not.
Systems traceability assumes iteration — that a broken link can be found, fixed and re-verified. That assumption is what makes coverage a useful metric there, and it is false here.
Which is why the weight column exists. A matrix imported unmodified from a systems methodology would report this program 100% traced and would have nothing to say about the one row that decides whether the business case survives.
Systems traceability assumes iteration — that a broken link can be found, fixed and re-verified. That assumption is what makes coverage a useful metric there, and it is false here.
Which is why the weight column exists. A matrix imported unmodified from a systems methodology would report this program 100% traced and would have nothing to say about the one row that decides whether the business case survives.