pharmacovigilance-signals
Drug safety signal detection — disproportionality analysis, case-series evaluation, and signal validation.
- License 许可
- MIT
- 最近更新
- 2026-11-07
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- pharmacovigilance-signals
- category
- scientific
- description
- Drug safety signal detection — disproportionality analysis, case-series evaluation, and signal validation.
03 SKILL.md 正文
Overview
pharmacovigilance-signals covers the detection and evaluation of adverse-drug-reaction signals from spontaneous reporting systems (FAERS, EudraVigilance, VigiBase): disproportionality statistics, case-series clinical review, and the validation workflow that separates real safety issues from reporting artifacts. Signal detection is hypothesis generation — every signal needs clinical and epidemiologic follow-up.
When to use
- Screening spontaneous-report databases for drug-event associations.
- Disproportionality analysis: PRR, ROR, IC (BCPNN), EBGM.
- Evaluating a signal: case review, temporality, dechallenge/rechallenge.
- Distinguishing true signals from notoriety, Weber, and masking effects.
- Signal prioritization for regulatory or clinical action.
- Designing post-signal pharmacoepidemiology studies.
Core concepts
- Spontaneous reporting. Voluntary reports of suspected adverse reactions — massive, global, but with no denominator, underreporting (typically >90% of events unreported), and reporting biases. Strength: detects rare, serious, unexpected events trials can't. Weakness: everything quantitative is fragile.
- Disproportionality statistics. PRR (proportional reporting ratio), ROR (reporting odds ratio), IC (information component, BCPNN), EBGM (empirical Bayes). They ask: is this drug-event pair reported more often than expected vs all other drugs/events? Thresholds (e.g. PRR ≥2, χ² ≥4, n ≥3) are screening filters, not proof. Shrinkage methods (IC, EBGM) handle small counts better.
- Reporting biases. Weber effect (reporting peaks after launch then declines); notoriety bias (media coverage spikes reports); stimulated reporting (regulatory actions prompt reports); confounding by indication (the disease, not the drug, causes the event); masking (a drug with many reports of one event suppresses disproportionality for others). Every signal must be examined for these before being believed.
- Case-series evaluation. The clinical core: review individual reports for temporality (plausible time-to-onset), dechallenge (event resolves on stopping), rechallenge (recurs on restarting — strong evidence, rarely available), dose relationship, alternative causes, and reporter quality. A handful of well-documented cases beats a thousand thin ones.
- Causality assessment. WHO-UMC or Naranjo scales structure individual-case judgment (certain/probable/possible/unlikely). They're guides, not algorithms — clinical judgment integrates the pattern.
- Signal prioritization. Seriousness, unexpectedness (not in label), strength of evidence, public-health impact, and preventability. Prioritize signals that are serious, unlabeled, and actionable — not just statistically striking.
- Validation. Disproportionality signals require confirmation: targeted pharmacoepidemiology (cohort/case-control in EHR/claims), literature review, mechanistic plausibility, and sometimes de novo studies. A signal is not a finding until validated.
- Data quality. Duplicate reports (same event reported by patient, doctor, manufacturer), missing data, coding variation (MedDRA granularity — group related preferred terms or miss the signal). Deduplicate and standardize before analysis.
Practical workflow
- Define the question. Drug(s), event(s) with MedDRA terms (group synonyms), time window, database.
- Clean. Deduplicate, standardize coding, handle missing fields transparently.
- Screen. Disproportionality statistics with shrinkage; stratify by age/sex/reporter where relevant; flag threshold-crossers.
- Triage for bias. Check Weber/notoriety patterns, indication confounding, masking — discard or down-weight artifactual signals.
- Clinical review. Case series: temporality, dechallenge/rechallenge, alternatives, WHO-UMC causality on key cases.
- Prioritize. Seriousness × unexpectedness × evidence strength × actionability.
- Validate and act. Pharmacoepi follow-up, literature, mechanistic review; communicate with appropriate uncertainty; monitor after action.
Example (R sketch):
# 2x2: a=drug+event, b=drug+other, c=other+event, d=other+other
PRR <- (a/(a+b)) / (c/(c+d)); ROR <- (a*d)/(b*c)
# use shrinkage methods (e.g. PhViD::BCPNN) for routine screening
Common pitfalls
- Treating disproportionality as incidence or risk (no denominator!).
- Ignoring notoriety/Weber effects (media-driven pseudo-signals).
- Confounding by indication mistaken for drug effect.
- MedDRA splitting hiding signals across synonymous terms.
- Duplicate reports inflating counts.
- Acting on unvalidated signals — or ignoring validated ones.
- Small-count signals without shrinkage (wild PRRs from n=2).
04 同仓库更多 Skills
web-scrape
Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open/fetch fails on JS-heavy pages, or when you need clean markdown/structured extraction from a site. Requires FIRECRAWL_API_KEY.
design-taste-frontend
tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML/CSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, "AI-made", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.
test-driven-development
Use when implementing any feature or bugfix, before writing implementation code
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes
requesting-code-review
Use when completing tasks, implementing major features, or before merging to verify work meets requirements
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).