Advanced Signal Management: how to build a signal dossier ready for PRAC and inspections

Many companies “do signal detection.” Very few companies can actually demonstrate how a signal was:

• identified,
• validated,
• prioritized,
• evaluated,
• and closed, or converted into action.

And during an inspection, the difference between the two is huge.

The Master Guide explains that signal management includes aggregate analyses and disproportionality methods, referencing statistical approaches, and that the QPPV must closely follow the process and bring it to the internal committee.

Here we go one step further: how to make the process not just “existing,” but “defensible.”

The myth to dismantle (contrarian insight): “A signal is a number”

This is a classic shortcut.

Myth: “If the metric does not cross the threshold, it is not a signal.”
Risk: you miss clinically relevant signals, especially underreported events, rare events, or signals in special populations.

Related myth: “We will discuss it in the next PSUR.”
Risk: governance starts to look reactive and slow.

In an audit, what really matters is not the threshold. What matters is the traceability of the decision.

Technical language (LSI terms) that an auditor or expert expects to hear

Here are the key terms that are new compared with your previously published articles:

• EVDAS (EudraVigilance Data Analysis System)
• SMQ (Standardised MedDRA Queries)
• MedDRA hierarchy: SOC / HLGT / HLT / PT / LLT
• IME list (Important Medical Events)
• ROR (Reporting Odds Ratio)
• EBGM (Empirical Bayes Geometric Mean)
• Information Component (IC)
• CCSI / CCDS (Company Core Safety Information / Data Sheet)
• Reference Safety Information (RSI)
• signal validation and signal prioritization
• case-series analysis
• dechallenge/rechallenge
• time-to-onset distribution
• confounding by indication
• signal tracking log

The “audit-proof” process in 6 steps, without reinventing GVP

Step 1 — Detection (screening): define how you search

Strong companies avoid “gut-feel screening” and formalize three elements:

• Data source: internal safety database, EVDAS, literature, real-world evidence
• Standardized queries: use of SMQs and consistent PT clusters (MedDRA)
• Clinical filter: IME list plus seriousness/preventability criteria

Typical mistake: screening based on 2–3 PTs “from memory.”
Result: not reproducible.

Step 2 — Triage: distinguish noise from priority signals

This step requires a simple but documented matrix.

Examples of prioritization criteria:

• clinical seriousness (death, life-threatening, disability)
• vulnerable populations (pediatrics, pregnancy, renal impairment)
• novelty versus RSI/CCSI
• biological plausibility
• temporal trend and clustering, such as sudden peaks
• risk of medication error or device-related issues, when applicable

Step 3 — Validation: when you truly call it a “signal”

Signal validation does not mean confirming causality. It means declaring that the information deserves structured evaluation.

In audits, the most common problem is this: the signal was discussed “verbally,” but there is:

• no meeting minute,
• no written rationale,
• no owner,
• no due date.

Step 4 — Evaluation: how you perform the clinical assessment, and what you document

This is where the signal becomes a dossier.

Practical elements that make the assessment robust:

• Case-series analysis: review cases as a series, not as isolated records
• Duplicate handling: “same patient, multiple reporters”
• Time-to-onset distribution: compatible or incompatible patterns
• Dechallenge/rechallenge: when present, extremely valuable, even if rare
• Confounding by indication: underlying disease versus drug effect
• Data quality: missingness, follow-up obtained, estimated exposure
• Comparison with RSI/CCSI for expectedness

Step 5 — Decision: who decides, where, and with which outputs

A “well-managed” signal has a clear output:

• closure, with rationale and conditions, for example “monitor for 3 cycles”
• or action, such as label change, RMP update, PASS, or DHPC, where applicable

Step 6 — Action & follow-up: the part that is often missing

If you decide on an action, the auditor wants to see that:

• the action was implemented, not just proposed
• you checked effectiveness, or at least progress
• the signal remained traceable until closure

The operational core: the “Signal Tracking Log” as a high-ROI artifact

If I had to recommend a single high-value artifact, it would be this: a single, controlled signal tracking log.

Fields that make the log truly defensible:

• Signal ID, unique, plus opening date
• source: EVDAS, internal, literature, affiliate
• trigger: SMQ/PT, IME, cluster, case series
• numerical snapshot: total cases, serious cases, fatal cases, temporal trend
• metrics: ROR / EBGM / IC, if used, plus the method
• comparison with RSI/CCSI for expectedness
• clinical summary in 5–10 lines, not a novel
• decision plus rationale
• actions plus owner plus due dates
• links to minutes and attachments, such as Safety Committee records
• status: open / monitoring / closed, plus closure date

Table: what an inspector wants to see in a signal dossier

“30-second” mini-checklist to test your maturity

If I pick a signal that was closed 6 months ago, can you immediately retrieve:

• the query that generated it, SMQ/PT
• the case series and inclusion/exclusion criteria
• the rationale for closure or for the action taken
• the committee minutes and approvals
• the follow-up, meaning what you monitored afterwards

If one of these answers is “no,” the risk is not only audit failure. It is also loss of organizational memory.

Box: remember this

A signal is not only detection. It is governance.

Your objective is not to prove that “there are no signals.” Your objective is to prove that when signals emerge, you have a process that produces consistent, traceable, patient-oriented decisions.

Conclusion

Advanced signal management does not require statistical magic. It requires documentation discipline, a shared language, MedDRA and SMQ, and a decision log that allows anyone, whether in an audit, PRAC, or acting as QPPV deputy, to reconstruct what happened without interpretation gaps.