Drug-safety triage with defensible causal bounds.
For drug-safety directors at top-20 pharma and safety leads at major CROs. Adverse-event signal triage with Rosenbaum-Γ bounds on every estimate, MedDRA preferred terms, a dual-attestation reviewer state machine, and a 21 CFR Part 11 ALCOA+ audit ledger that pharma uses inside their own regulated workflow.
FAERS-disproportionality misses signals it can't quantify the fragility of.
Standard PV triage gives you a PRR. It doesn't tell you how much unmeasured confounding flips it.
Disproportionality scores (PRR, ROR, IC) are point estimates. Reviewers spend hours deciding whether a signal is real, an artifact of co-reporting, or a known indication-bias. By the time the PSUR submission goes out, the methodology footnote is a paragraph of qualitative caveats.
"Identifying and addressing the presence of confounding and other forms of bias is critical… planned sensitivity analyses to assess the robustness of study findings." — FDA Non-Interventional Studies draft guidance, March 21, 2024.
Rosenbound delivers the sensitivity analysis as a default output. Every signal arrives with the Rosenbaum-Γ bound that quantifies how strong an unmeasured confounder would have to be to flip it. Reviewers move from "this looks suspicious" to "Γ = 1.06, very sensitive — escalate" in seconds.
Eight features the incumbents don't bundle.
Each feature is in production today and visible to design partners at first login. Try them on a synthetic cohort at rosenbound.ai.
Γ-bound per signal
Every adverse-event signal in the reviewer queue carries its Rosenbaum-Γ sensitivity bound. Sort by fragility. Triage by signal AND by how robust the evidence is.
MedDRA preferred terms + SOC class
Version-pinned MedDRA browser with verbatim-to-PT auto-suggest. Signals carry the Preferred Term, System Organ Class, and the LLT match where applicable. WHODrug roadmap.
Dual-attestation reviewer state machine
State flow: queued → triaged → in-review → quality-review → escalated/dismissed → signed-off. High-severity sign-offs require two independent reviewers. Role-based filters for "My queue", "Team queue", "All signals".
Provenance per signal
Methodology and provenance block on every signal-detail page distinguishes corpus-level model AUROC from per-signal estimates. No silent overclaiming — reviewers see exactly which population a metric refers to.
21 CFR Part 11 ALCOA+ audit ledger
Every state transition committed to a SHA-256 chained ledger. Attributable, Legible, Contemporaneous, Original, Accurate — built in, not bolted on. Read-only export to pharma audit teams.
Cognitive Validation Report
The system refuses to ingest incoherent data and explains why — temporal vetoes (treatment-before-diagnosis), ontological vetoes (dose-above-FDA-max), causal-acyclicity vetoes, biological vetoes. Bad data gets blocked at intake, not silently confidence-graded.
E2B(R3) export, MedWatch 3500A first
FDA MedWatch 3500A export today; full ICH E2B(R3) ICSR XML on the 90-day roadmap. Bidirectional import in the same release window. CIOMS I as international add-on.
WorkOS SSO + RBAC
AuthKit/WorkOS with SAML and OIDC. Email + Google + Microsoft + GitHub + Apple SSO out of the box. Three default roles (admin, reviewer, viewer); custom roles via the founder Slack.
Try the synthetic PV cohort at rosenbound.ai
Upload a sample CSV, watch the Cognitive Validation Report block bad signals, slide the Γ-curve on a flagged signal, sign it off, see it materialize in the audit ledger. No login. rosenbound.ai →
Built to ship in 2026 under the right exemption.
✓ 21st Century Cures Act § 3060 (non-device CDS)
Pharmacovigilance triage satisfies all four exemption criteria: for a healthcare professional, advisory not directing, transparent algorithm with documented provenance, independent-review-of-basis enabled via the Γ-curve and methodology block. No 510(k) required for 2026 ship.
✓ 21 CFR Part 11 ALCOA+ by design
Audit module satisfies § 11.10(a)–(k) without operator configuration. SHA-256 chained ledger, monotonic timestamps, fsync + multi-writer protection, external-verifiable. Pharma uses our outputs inside Argus/Vault Safety/LifeSphere workflows without bolt-on audit infrastructure.
✓ FDA RWE Framework alignment
Sensitivity-bounded reporting is exactly what the FDA RWE Framework (final Aug 2023) + the EHR/Claims Data guidance (final Jul 2024) + the Non-Interventional Studies draft (Mar 2024) are now asking for. Rosenbound provides it as the default output.
✓ Complementary to your safety system of record
Rosenbound does not replace Argus, Vault Safety, or LifeSphere. We sit beside them as the methodology + audit-defensibility layer. Signal-evaluation sections feed your PSUR/PBRER/DSUR/PADER directly; we don't author the whole report.
Watch the platform walkthrough at rosenbound.ai — three moments: the Cognitive Validation Report refusing incoherent data, the live Γ-bound sensitivity visualization, and the reproducibility certificate generated on every study. The full platform is open to Design Partners.
pip install rosenbound
— Official Python SDK for programmatic access: cohort upload, sensitivity-bounded study runs, and reproducibility certificate retrieval. Apache 2.0; Pydantic v2 typed; py.typed for IDE autocomplete + mypy. Platform access gated by Bearer token + RBAC + tenant scoping — the SDK is open, the audit substrate is not.
Run Rosenbound on synthetic cohorts. Tell us what breaks.
Rosenbound is selecting a small number of design partners across pharma drug-safety teams, CRO methodology groups, and academic RWE units. A Design Partnership costs nothing and moves no data: the platform provides production-shape synthetic cohorts; your team runs the full workflow — intake, pentagon, Γ-bounds, sign-off, certificate — in the live environment and tells us where it falls short of what a regulator would accept.