support@nirnasa.com +91 83097 56689 Mon – Fri: 10:00am – 6:00pm IST
24/7 support for all services
Signal detection · Evaluation · Risk action

See emerging safety signals sooner, and act on evidence

Nirnāśā Signal Detection Management runs proven statistical methods over FAERS and your other data sources. It ranks every signal with a priority score you can see into, and takes each one through a governed GVP Module IX lifecycle, from detection to a monitored risk action.

4Disproportionality methods
500Drug–event pairs per bulk scan
10GVP IX lifecycle stages
7Specialised AI agents
Detection methods

The statistics regulators know, in one engine

From classic disproportionality to Bayesian shrinkage and sequential surveillance, you can use the method that suits the question and show your working.

PRR

Proportional Reporting Ratio

How often an event is reported with a drug, compared with all other drugs, with a 95% confidence interval.

ROR

Reporting Odds Ratio

The odds of the drug–event pair against the background, with a 95% CI. A lower bound above 1.0 flags a signal.

IC

Information Component

A Bayesian measure in the WHO VigiBase style that stays stable when case counts are small.

EBGM

Empirical Bayes (MGPS)

EBGM and EB05 from a DuMouchel multi-item gamma Poisson shrinker, to reduce false positives.

TreeScan

Hierarchical scan

Looks for signals across the MedDRA hierarchy (SOC → HLGT → HLT → PT), so related terms that are each too weak on their own still surface together.

MaxSPRT

Sequential testing

Maximised sequential probability ratio testing to catch reporting that is speeding up over time, as FDA Sentinel does.

CUSUM

Poisson CUSUM

Cumulative-sum surveillance that alerts when reporting of a drug–event pair shifts upward.

FAERS analysis in a few clicks

Query the FDA Adverse Event Reporting System through openFDA for any drug and up to five events, or screen hundreds of pairs in one bulk run.

  • Drug + event dashboard. PRR and ROR with 95% CI, top 10 suspect drugs and co-reported events.
  • Demographics. Distributions by sex, age, reporter type and country.
  • Bulk detection. Up to 500 pairs for a single drug or event, filtered to ROR lower 95% CI > 1.0 and sortable by cases, PRR or ROR.
  • Reliable at scale. Rate limiting with automatic back-off keeps long scans running.
  • Exports. CSV line listings and full-page PDF.
Drug X
5 events selected
Query run
Hepatic failure
PRR 3.6 · ROR 3.8 (2.9–5.0)
Signal
Nausea
ROR 1.1 (0.9–1.3)
No signal
Demographics
62% female · median 54 y
Profile
Line listing
CSV · PDF
Exported

Evidence-led signal management, with the discipline regulators expect.

Multi-source evidence

See where your sources agree

In line with EU Implementing Regulation 2025/1466, signals are checked across all the data you hold. For each drug–event pair you see how many sources agree, which source flagged it first, and a combined EB05.

  • FAERS via openFDA, connected live.
  • EudraVigilance (EVDAS), VigiBase and FDA Sentinel, connected with your organisation's credentials.
  • VAERS and Embase through file import.
  • Your own case data and literature, including live PubMed search with built-in PV filters.
Discuss your data sources
Concordance · Drug X / hepatotoxicityDemo view
FAERSEB05 2.6Flagged first
EudraVigilanceEB05 2.1Agrees
Internal casesEB05 1.4Agrees
Literature3 reportsSupporting
Combined EB052.43 of 3 statistical sources agree
Transparent prioritisation

A priority score you can explain to an inspector

Every signal gets a score from 0 to 100, built from weighted, published factors. You can see exactly how much each factor added, and a reviewer can override the result with a recorded reason.

EB05 strengthSeriousnessFatal casesIME / DMEReporting trendVulnerable groupsUnlisted eventCase count
Priority score breakdown81 / 100
EB05 strength
+18
Seriousness
+16
IME / DME
+14
Reporting trend
+12
Unlisted event
+10
Fatal cases
+7
Case count
+4

Human override available · reason and user recorded

GVP Module IX lifecycle

A governed path for every signal

Signals move forward one stage at a time, and gates stop a signal closing without evidence. Every move back needs a reason, and every transition is audited.

  1. Detected
  2. Triage
  3. Validation
  4. Confirmation
  5. Prioritisation
  6. Assessment
  7. Recommendation
  8. Risk action
  9. Effectiveness
  10. Closed

Signal committee

Meetings with quorum rules, recorded votes and automatically generated minutes.

Risk actions

Label updates, communications and other measures recorded against the signal.

Effectiveness gate

A signal can't close until the effectiveness of its risk action has been assessed.

Watchlists

Watchlisted pairs are re-checked automatically as new data arrives.

Human-governed AI

Seven specialised agents, one reviewer in charge

The AI triage copilot drafts an assessment based only on the signal's own metrics, with citations and a capped confidence score. A person must confirm it, and every triage is audited. Without an AI key, a rules-based version keeps working offline.

Ingestion

FAERS/openFDA, EHR extracts, PubMed and manual entry.

Detection

PRR, ROR and IC runs, including rate-limited bulk scans.

Causality

Scoring against the WHO-UMC criteria.

Lifecycle

Stage management with EMA-aligned gates.

Reporting

PRAC summaries, PSUR/PBRER signal sections and line listings.

Demographics

Breakdowns by sex, age, reporter and country.

Normalisation

NLP entity extraction and de-duplication.

Orchestration

Run agents one at a time, as a pipeline or in parallel.

Compliance

Every decision signed and traceable

EMA GVP Module IX

A signal management lifecycle with stage gates and documented decisions.

EU IR 2025/1466

Evidence across all available data sources, with concordance shown.

21 CFR Part 11

Password re-entry to validate, close, escalate or reject a signal.

Hash-chained audit trail

A tamper-evident record of every metric run, transition and override.

AI signal radar detecting a drug-event signal with PRR, ROR and EB05 values
Nirnāśā AI · human-governed
AI inside

An AI copilot for signal triage

The triage copilot drafts an assessment from the signal’s own metrics, with citations and a capped confidence score. A reviewer confirms every triage.

AI triage copilot

Draft assessments grounded only in the signal’s metrics, with citations.

Seven specialised agents

Intake, detection, causality, lifecycle, reporting, demographics and normalisation.

Flexible orchestration

Agents run one at a time, as a pipeline or in parallel.

Works offline too

Without an AI key, a rules-based version keeps triage running.

How Nirnāśā AI works

Find the signal before it finds you

See disproportionality, TreeScan and the GVP IX lifecycle running on live FAERS data for your products.

Support available now

24/7 on-call technical support for every service

Our support team is available around the clock for all Nirnāśā services, so your safety operations never stop.

  • PV Intake
  • ICSR Suite
  • Aggregate Manager
  • Signal Management
  • PSMF
  • QMS
  • Voice Connector