Pharmacovigilance signal detection and assessment, with every number traced to its query
Every potential safety signal gets its evidence assembled in days, each number traced to its query.
See one case, screen by screen ↓The weekly disproportionality run is the easy part. It screens every drug–event pair and hands back a list of combinations above threshold. Then the safety scientist has to decide, for each one, whether there is a new potential causal association — and to do that, someone has to assemble the evidence.
That means pulling the cases behind the pair and reading every narrative for time to onset, dechallenge, rechallenge and confounders; re-running the statistics; searching the literature for published cases and a background rate; checking sections 4.4 and 4.8 of the reference safety information; and writing it all into a memo where every figure can be checked. Done by hand, that takes weeks per signal, while the validation clock runs.
Estimated split for one potential signal, by hand and with the solution, in days.
Five specialist agents score, gather and write up the evidence; a safety scientist validates and the safety committee decides.
The Monday detection run flags Veltrimab, an IL-23 p19 antibody for plaque psoriasis and psoriatic arthritis, with interstitial lung disease. Here is what happens next, screen by screen, in the working solution.
The weekly run on Monday Oct 5 at 02:00: 1,043,112 reports screened, 6,412 drug–event pairs scored, 9 above threshold. At the top of her list sits Veltrimab × Interstitial lung disease — “ROR 3.4 and rising · 17 cases, 5 new · not in the label”. Lena Ortiz, the signal analyst, assigned it to her at 07:40.
“3 of the 9 pairs are new this week; 2 are not explained by the label.”
The board plots each combination’s reporting odds ratio with its 95% confidence interval against a threshold of 2. Interstitial lung disease sits at 3.4 (2.1–5.5) and is rising, with priority High, stage Detected and 29 days left on its validation clock, due Nov 4.
ROR 3.4 (95% CI 2.1–5.5), PRR 3.4 with χ² 25.8, EBGM 3.0 with EB05 2.1, and 17 cases, 5 of them since the last run. The quarterly ROR has climbed from 1.1 in Q3 2025 to 3.4 in Q3 2026. Exposure is 38,400 patient-years since launch in March 2024, against a background rate of 1.6 per 10,000 patient-years.
“Observed 17 against about 6 expected from the background rate — and spontaneous reporting usually under-counts.”
Maya clicks the 3.4. The exact query opens: data lock 2026-09-30, MedDRA 27.1, the narrow interstitial lung disease SMQ, suspect products only — with the inputs of the two-by-two table (a 17, b 4,283, c 1,211, d 1,037,601, expected 5.1). She can copy it or re-run it now.
“This number was produced by the query below. It re-runs on every detection run and the result is stored with the signal.”
The assembler pulled the 17 cases with the narrow search and read every narrative. Serious 15 of 17, one fatal. Median time to onset 85 days, the largest group between 61 and 90 days. Positive dechallenge in 9, a positive rechallenge in study VLT-302. Confounders counted, not hidden: methotrexate in 6 cases, smoking history in 5.
Each row opens its narrative with the key sentences highlighted. The fatal case, VLT-2026-005804, is a 73-year-old man on methotrexate 15 mg weekly, with a 40 pack-year smoking history and known COPD, admitted with respiratory failure 401 days after starting Veltrimab. Causality: unassessable.
“The reporting physician could not separate the contribution of methotrexate, smoking and Veltrimab.”
Maya asks for the memo. Each step shows what it found: the statistics with 9 queries stored; 17 cases with onset, dechallenge and confounders; 2 relevant articles and 1 background-rate study; and the label check, flagged amber — not in sections 4.4 or 4.8, unexpected. The memo writer pulls it together.
Summary, label check, case series, literature and plausibility, observed against expected, and proposed actions: follow-up questionnaires for the 6 cases missing CT findings, notify regulators of the validated signal by Nov 4, include it in the periodic report with data lock Dec 31, 2026, and ask the committee to consider it as an important potential risk in the risk management plan.
“Unexpected. Interstitial lung disease, pneumonitis and pulmonary fibrosis are not in sections 4.4 or 4.8 of the reference safety information.”
Three choices: validated and proceed to assessment, refuted and close, or monitor and check again next run. Maya validates, with the reason on the record — serious, unexpected, a consistent time to onset and 9 positive dechallenges; methotrexate and smoking do not explain all cases — and her electronic signature. The clock moves to regulator notification, due Nov 4.
Meaning of the signature: “I reviewed the evidence and validate this signal.”
She sends the signal to the safety committee on Wed Oct 14, 2026 at 14:00. The package goes with it: the memo with its label check, the line listing of 17 cases, the statistics with the query behind every number, the literature and background rate, and the signal history. Dr. Omar Haddad, Dana Okafor and Sam Patel decide.
The same signal, seen by the five people who move it from detection to decision.
Five specialist agents score, gather and write up the evidence; a safety scientist validates and the safety committee decides.
Computes PRR, ROR and empirical Bayes scores for every drug–event pair each week, flags pairs that meet the PRR rule or the Bayesian rule, and stores the query behind each number.
Pulls the cases behind a signal and extracts time to onset, dechallenge, rechallenge, seriousness, outcome, co-suspect and concomitant medicines and risk factors from each narrative.
Screens the weekly literature for each product, flags individual case reports, and finds background rates and published cases for an open signal.
Compares the signal’s terms with sections 4.4 and 4.8 of the current core data sheet, answers listed or not listed, and quotes the nearest entry.
Writes the signal assessment memo from the other agents’ outputs — summary, label check, case series, literature and plausibility, observed against expected, proposed actions.
Validates, then the committee decides. The agents propose; a named person decides.
Ask about any signal, case or deadline — or tell it what to change. Answers point to their sources.
Why is Veltrimab × interstitial lung disease rising?
Five new cases arrived since the last run, four of them serious. The reporting odds ratio moved from 2.9 to 3.4 (95% CI 2.1–5.5) and has risen for four quarters. Median onset is 85 days with 9 positive dechallenges, and the event is not in the label. Methotrexate and smoking explain some cases but not all.
Show only signals that are not in the label
Filtered the signal board to the 8 signals where the event is not in the reference safety information. Three of the open ones are high priority: interstitial lung disease, acute kidney injury and DRESS.
Add a rule: always review any lung event for Veltrimab
Added Lung events on Veltrimab to Always review in Settings. From the next run, every report coding to the interstitial lung disease search goes to Maya Chen, whatever the statistics say. The change is in the audit trail.
Which deadlines fall in the next 14 days?
Uveitis assessment (Veltrimab) on Oct 12, owner Maya Chen; the safety committee on Oct 14 with DRESS and hyponatraemia; and, just outside, the acute kidney injury assessment (Corventa) on Oct 21, owner Sam Patel.
13 screens from the working solution, on its sample data. Pick one to see it large.
Thresholds, clocks and approvers are settings. A threshold change applies from the next run.
| Setting | Default | Choose from |
|---|---|---|
| PRR rule — reporting ratio | PRR ≥ 2.0, with chi-square ≥ 4 | 1.0 to 5.0, in steps of 0.5 |
| PRR rule — minimum number of cases | 3 | 1 to 10; fewer cases are tracked but not flagged |
| Bayesian rule — EB05 ≥ 2 | On | On · Off |
| Detection run | Weekly · Monday 02:00 | Daily · 02:00 · Weekly · Monday 02:00 · Every two weeks · Monday 02:00 |
| Validation clock | 30 days | 10 to 60 days, in steps of 5 |
| Emerging safety issue — tell regulators within | 3 working days of validation | 1 to 5 working days |
| Who validates or refutes a signal | Maya Chen, safety scientist for the product | Maya Chen · Lena Ortiz · Sam Patel |
| Who notifies regulators | Dana Okafor, QPPV | Dana Okafor · Dr. Omar Haddad |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample safety data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · PRR, ROR, EBGM = standard disproportionality scores; EB05 = the lower 5th percentile of the empirical Bayes estimate. Products, cases and people named on this page are from the working solution’s sample data; the people are characters in it.
Software that takes a potential safety signal from detection through validation and assessment to a decision, with every step recorded. Signal Management adds agents that compute the disproportionality scores, assemble the case series, search the literature, check the label and draft the assessment memo.
Each detection run builds the two-by-two table for every product and event and computes PRR with chi-square, ROR with its 95% interval and EB05. A pair is flagged when the PRR rule (PRR at least 2, chi-square at least 4, at least 3 cases) or the Bayesian rule (EB05 at least 2) holds. Signals from literature, clinical studies, regulators or partners can also be logged and go through the same agents.
Yes. Every number on the board, in the statistics and in the memo opens the exact query that produced it, with its inputs, the data lock and the MedDRA version. You can copy the query or re-run it.
No. The agents propose; people decide. The safety scientist validates, refutes or keeps a signal under monitoring with a signed reason, the medical safety lead signs the memo, the safety committee decides the outcome and the QPPV notifies regulators.
No. The agents have read-only access to the case database, and every agent step is on the signal’s history.
The label check agent compares the signal’s terms with sections 4.4 and 4.8 of the core data sheet in force at data lock, answers listed or not listed, and quotes the nearest entry.
Yes. The PRR ratio and minimum cases, the Bayesian rule, the run schedule, the always-review rules, who validates and who notifies, the validation clock and the emerging safety issue clock are all settings.
The Agentic Solution Engine builds and deploys it from your requirements — your signal procedure, memo template, products and reference safety information — and it goes live once every quality gate has passed. We will walk you through it on your own past signals first.
Every side-effect report becomes a cited case in minutes, filed with days to spare.
1-min filmSafety LiteratureEvery journal article screened with a written reason, and reportable cases at intake in about a day.
1-min filmPeriodic Safety ReportsEvery periodic safety report is fully drafted within two weeks of data lock, each number traced to its query.
1-min filmWe’ll run Signal Management on a sample of your own past signals.