Signal Detection
Standard pharmacovigilance was designed for single-drug safety. TrialVyx was built for the polypharmacy reality.
When 40% of post-market adverse events involve three or more concurrent medications, pairwise ROR/PRR analysis leaves the majority of the drug-combination signal space structurally unexamined. That's not a query design problem — it's a method design problem.
The DDI Detection Gap
Why pairwise ROR/PRR misses multi-drug interactions
Reporting Odds Ratio (ROR) and Proportional Reporting Ratio (PRR) were developed for an era of simpler drug regimens. They examine the statistical relationship between a single drug and a single adverse event term — a method that works adequately for single-agent monitoring.
In polypharmacy — the clinical reality for most elderly and chronic disease patients — the interaction isn't between drug A and adverse event X. It's between the combination of drugs A, B, and C and the resulting adverse event profile that none of them would produce individually. Pairwise methods cannot detect this class of signal.
TrialVyx constructs N-drug interaction graphs — capturing triads, tetrads, and higher-order combinations — and propagates detection probability across the full drug-AE co-occurrence network.
Drug pairs only — N=2 combinations
All N-drug combinations including triads and tetrads
Methodology
Graph-based DDI analysis
The detection model constructs a network where indirect drug relationships become visible.
Node construction: Each unique drug (by INN) and each adverse event preferred term (MedDRA PT) becomes a node. Nodes accumulate report counts, temporal data, and seriousness weights from the ingested ICSR corpus.
Edge weighting: Edges between drug nodes encode co-prescription frequency — how often those drugs appear together in the same ICSR report. Edge weight increases with co-occurrence and is recency-discounted to reflect current prescribing patterns.
Propagation: Bayesian belief propagation flows detection evidence through the graph. A new adverse event report doesn't just update the direct drug-AE edge — it propagates probability updates to all connected drug combinations, revealing indirect interaction chains that pairwise analysis cannot detect.
Output: Signals are scored by IC (Information Component) value, trend velocity, and clinical severity. Multi-drug combinations scoring above the Investigate threshold surface as signal briefs for your PV team.
Signal Classes
Signal classes TrialVyx detects
Class examples from retrospective FAERS analysis. Drug names are therapeutic class descriptors, not specific compounds.
Co-prescription of statins with calcineurin inhibitors and PPIs creates a metabolic pathway interaction. Individual pairwise ROR for any two agents shows unremarkable values. The triad's combined CYP3A4 burden and QT-prolongation potential only becomes visible in graph analysis across concurrent prescriptions.
Detected via TrialVyx methodology in retrospective FAERS analysis. Class example only.Each agent has known individual bleeding risk. Standard monitoring tracks each pair. The triad's combined platelet inhibition and anticoagulation burden creates a signal that exceeds what any pairwise analysis would project — particularly concentrated in geriatric FAERS reporter populations with high co-prescription rates.
Detected via TrialVyx methodology in retrospective FAERS analysis. Class example only.Azole antifungals create significant CYP3A4 inhibition. When combined with certain HIV protease inhibitors and QT-extending cardiac agents, the metabolic interaction creates a cumulative drug exposure risk that standard monitoring of any individual pair fails to surface until adverse event accumulation is substantial.
Detected via TrialVyx methodology in retrospective FAERS analysis. Class example only.Multiple agents in oncology supportive care regimens carry hepatotoxicity signals individually below reporting thresholds. Combined administration in the FAERS record reveals a hepatotoxicity cluster — visible in graph analysis of the co-prescription network but not in any individual pairwise statistical calculation.
Detected via TrialVyx methodology in retrospective FAERS analysis. Class example only.Lead Time Comparison
Detection capability: TrialVyx vs. ROR/PRR/BCPNN
| Capability | TrialVyx | Standard ROR/PRR |
|---|---|---|
| Earliest detectable signal | 60–120 days before statistical threshold | At or after statistical significance threshold |
| Multi-drug (N ≥ 3) detection | Yes — N-drug graph analysis | No — pairwise analysis only |
| Regulatory-ready output | MedDRA-coded, audit trail included | Requires separate documentation step |
| Indirect interaction chains | Yes — Bayesian propagation across graph | No — direct association only |
| FAERS corpus coverage | Full corpus, all report types | Depends on query design |
Configure N-drug signal monitoring for your compound portfolio.
TrialVyx is configured per compound portfolio — we set the drug list, therapeutic area context, and detection thresholds specific to your post-market monitoring obligations. Request a methodology briefing to see how the graph model applies to your compounds.