Skip to main content
PVgenix
Platform · Signal detection

Quantitative signal detection across the lifecycle

Run established disproportionality methods over your full safety database, manage masking, stratification and thresholds, and move every signal of disproportionate reporting through a configurable, fully audit-trailed review workflow — aligned to GVP Module IX.

PRR Proportional Reporting RatioROR Reporting Odds RatioBCPNN Information ComponentEBGM Empirical Bayes
How the statistics work

Every method starts from one 2×2 contingency table

Disproportionality analysis compares the observed reporting of a drug–event pair against what would be expected if there were no association. PVgenix computes all four established statistics from the same counts, so reviewers see a consistent, cross-checked read on each pair.

Contingency countsVelmora · hepatic injury

Reports in the database, cross-tabulated by drug and event. Cell a — reports naming both — drives the signal.

Event of interesthepatic injury
All other eventsnot the event
Drug of interestVelmora
a
48
observed
b
1,902
reports
All other drugscomparator
c
380
reports
d
96,500
reports
Expected reports for this pair: ~8.5 — observed 48 is 5.7× higher.
PRR
Proportional Reporting RatioReporting fraction for the drug vs. all other drugs.[a/(a+b)] ÷ [c/(c+d)]
6.28
≥ 2.0 · χ² 31
ROR
Reporting Odds RatioOdds of the event with the drug vs. without it.(a·d) ÷ (b·c)
6.41
95% CI 4.7–8.7
BCPNN
Information Component (IC025)Bayesian shrinkage — stable at low counts.log₂(observed ÷ expected)
1.92
IC025 > 0
EBGM
Empirical Bayes (EB05)Gamma-Poisson shrinkage (MGPS), 5th percentile.shrunk observed ÷ expected
3.81
EB05 ≥ 2.0
Signal of Disproportionate Reporting flaggedAll four statistics cross their configured thresholds — the pair is queued for validation, not auto-confirmed.
N = 48 reportsfirst seen 12 days ago
4
Methods in parallel
PRR · ROR · BCPNN · EBGM
5
Stratification dimensions
age · sex · region · seriousness · time
30
Day review cadence
configurable per product family
100
% audit-trailed
every disposition versioned
Interpreting signals

Case volume shapes statistical signal detection

Disproportionality methods compare observed reporting against expected patterns, so their results are most meaningful where case volume supports them. PVgenix surfaces strength and stability together, and applies masking and subgroup analysis so genuine signals are not hidden by high-volume reporters.

Masking detection

Identify drug–event pairs whose disproportionality is suppressed by a dominant co-reported product, and re-run the analysis with the masking term removed.

Stratified & subgroup analysis

Stratify by age, sex, region, seriousness and reporting period to control confounding and expose signals concentrated in a subpopulation.

Configurable thresholds

Set the signalling rules per method — PRR ≥ 2 with χ² ≥ 4 and N ≥ 3, EB05 ≥ 2, IC025 > 0 — tuned per product family and reviewed on a fixed cadence.

Disproportionality landscape3 signals above threshold
case count →RORsignal thresholdVelmora — hepatic injuryOncoval — neutropeniaCardizen — QT prolongation
PRRRORBCPNN / ICEBGMmasking offstratify: all
Signal management lifecycle

From statistical flag to documented disposition

A quantitative flag is the start, not the conclusion. Every signal moves through a structured, role-gated workflow where qualified personnel validate, prioritize, assess, and decide — with a complete audit trail from detection to closure, aligned to GVP Module IX.

1

Detection

Disproportionality runs across the safety database, with masking and subgroup analysis surfacing candidate pairs.

PRR · ROR · EBGM
2

Validation

Confirm the flag is genuine — check data quality, duplicates, confounding and known associations before escalating.

de-duplication
3

Prioritization

Triage by seriousness, strength, novelty and public-health impact to focus review on what matters most.

risk-ranked
4

Assessment

Medical evaluation of causality, biological plausibility, literature and mechanism, captured as a structured record.

medical review
5

Recommendation

Decide and record the action — label change, RMP update, further analysis or no action — with rationale.

action linkage
6

Tracking

Signals carry a status through their lifecycle with a clear audit trail from detection to documented resolution.

full audit trail
Detection · configurationactive run

Methods & thresholds

PRR Proportional Reporting Ratio≥ 2.0 · χ² ≥ 4 · N ≥ 3
ROR Reporting Odds RatioCI lower ≥ 1
IC Information ComponentIC025 > 0
EBGM Empirical BayesEB05 ≥ 2.0

Masking & stratification

Masking detection
Stratified analysis
agesexregionseriousnessperiod

Review workflow

Run cadenceevery 30 days
Disposition statesNew → Validated → Closed
Audit trail
Thresholds, masking & workflows

You set the rules — the platform runs them, every cycle

Detection criteria are configuration, not code. Tune the signalling rule for each method, decide which subgroups to stratify by, switch masking on for high-volume databases, and define how often the analysis re-runs — all per product family and all version-controlled.

Per-method, per-product rules

Different thresholds for vaccines, biologics and small molecules — each method tuned independently and stored as a versioned configuration.

Scheduled, repeatable runs

Analyses re-run on a fixed cadence against the latest data, so emerging signals are caught between aggregate reporting cycles.

Defensible audit trail

Every threshold change, run and disposition is captured for inspection — who changed what, when, and why.

Signal-detection statistics are screening tools, not proof of causality — they prioritize where qualified review is needed. Scope and thresholds are defined per client agreement.

See quantitative signal detection in action

Request a demo to see disproportionality methods, configurable thresholds, masking, and workflow-driven signal management running on your safety data.