Search PubMed/MEDLINE free, no key. For Embase, tick it and enter your organisation's Elsevier API key (+ institutional token if you're off-network) — Embase is subscription-gated, so without those it can't run. Results from all sources merge into one list, deduplicated on DOI. Each card shows highlighted drug/event/reporter hits and a validity flag. Found a real case? Hit "Open in ICSR: Literature case" to carry it across for full QC.
Two paths, same discipline:
Set Initial vs Follow-up before locking — Initial triggers a duplicate check against cases you've already processed on this device; Follow-up requires a reference to the original. Then Lock QC to unlock narrative generation. For E2B(R3) XML, fill in the sender/receiver/case IDs, and enter numeric MedDRA LLT/PT codes in the QC panel — without them the XML flags each gap with a comment and Argus will reject it. Always test-import into a sandbox first.
Upload one or more line listings (.csv/.xlsx — they merge). Map the drug, event, and optionally case ID and date columns. It computes PRR, ROR with 95% CI, and χ² per drug–event pair and flags Evans criteria (PRR≥2, χ²≥4, N≥3). These are screening aids only. Complete the GVP Module IX qualitative worksheet, set a recommendation, lock, then generate the report. Remember the stats only reflect the data you uploaded — a partial extract distorts them.
Pick the report type (PSUR/PBRER/DSUR/PADER/ACO/Other), upload the draft, and hit Detect errors. No key = structural check only (missing mandatory sections, leftover placeholders, suspiciously thin sections). With a key you also get a content review — factual/statistical inconsistencies, unsupported conclusions — all graded Critical/Major/Minor. If your template uses different headings, edit the expected-sections list rather than trusting a false "missing" flag. Accept/reject each finding, edit the suggested fix, then generate a redline.
Fill in the product/batch/complaint details, upload the source material (investigation reports, stability data, emails, photos), then generate. No key = scaffold only — the analytical sections stay as [Complete] placeholders, because that reasoning genuinely needs a toxicologist. With a key, the AI drafts them from your sources. Either way, every risk conclusion is yours to make.
NexaPV Assist — Intended Use Statement · Tool version · Statement generated
NexaPV Assist is a browser-based drafting and screening aid for pharmacovigilance professionals. It assists with literature screening, ICSR/E2B(R3) drafting, disproportionality screening (signal detection), aggregate report drafting and error-checking, RMP drafting, health hazard assessment report (HHAR) drafting, and provision of general regulatory-timeline reference information. It exists to reduce the manual effort of drafting and structural checking — it does not perform, replace, or certify any of the regulatory judgements involved in that work.
Qualified pharmacovigilance personnel (e.g. Clinical Safety Lead, QPPV/Deputy QPPV, PV physicians, signal management and aggregate-report staff) who are able to independently assess the medical, statistical, and regulatory correctness of the tool's output. It is not intended for use by anyone unable to independently evaluate its output before that output is relied upon.
As a custom-built, single-file browser application, this would typically fall under GAMP 5 Category 5 (custom application) if brought into a formal computer system validation (CSV) programme. Whether it requires formal validation depends on your organisation's own SOP for what constitutes a "GxP-relevant system" — in particular, whether a tool that only drafts content for a validated system of record (rather than acting as one) falls in or out of that definition. That determination is QA's to make, not the tool's; this statement is intended to give QA/IT the facts needed to make it, not to pre-empt it.
This tool is under active, ad hoc development. Any change to its logic (calculation methods, section templates, prompts, or scope) constitutes a new version and, if brought under formal validation, would require re-verification of at least the affected requirements before the new version replaces the prior one in routine use.
It's easy to read a notice titled "AI use" and assume everything here comes from a language model. It doesn't. Most of what this tool produces comes from three other sources, with AI as one optional layer on top:
Whichever combination produced a given output, every output is a draft proposal for a qualified person to review, correct, and approve — never a finished assessment, decision, or regulatory record. Nothing produced here is fit to be filed, submitted, or entered into a safety database without independent verification against the source data by an appropriately qualified reviewer.
Where AI drafting specifically is used, this reflects the human-centric, risk-based approach set out in the EMA's Reflection paper on the use of Artificial Intelligence in the medicinal product lifecycle (adopted September 2024), which expressly covers post-authorisation uses including adverse event report management and signal detection, and the risk-based credibility assessment framework in the FDA's draft guidance Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products (January 2025), together with the joint FDA–EMA Guiding Principles of Good AI Practice in Drug Development (January 2026).
This is an unvalidated development build. Specifically, none of the following has been done:
If you intend to use this for work supporting regulatory decisions, it must first be assessed, validated, and approved under your organisation's quality system and IT/InfoSec governance. Regulators may hold AI systems in this space to standards stricter than general industry practice.
Use of AI does not transfer, dilute, or share regulatory accountability. The QPPV, MAH, and the qualified individuals performing each activity remain fully responsible for the accuracy, completeness, and compliance of every record and submission — exactly as if no AI had been involved. "The AI produced it" is not a defensible position before any competent authority.
Where AI features are used, the text you submit is transmitted to Anthropic's API over the internet. Do not paste personal data, patient-identifiable information, or confidential material unless your organisation has approved that transfer and any required agreements and data-protection assessments (e.g. DPDP Act, GDPR) are in place. Local scan, OCR, and file parsing run in your browser; saved settings, credentials, and any persisted log live in this browser's local storage only, and are not secure storage.
© 2026 Dr Syed Mudasir Ahmad (docmuddie@gmail.com). All rights reserved. This software, including its structure, templates, prompts, and generated document formats, is the proprietary work of the author and is protected by copyright.
It is made available to the registered user named at sign-in, for that user's own use only. Copying, distributing, forwarding, sharing, uploading, sublicensing, reselling, publishing, or otherwise making this tool available to any third party — in whole or in part, modified or unmodified — without the prior written permission of the author is prohibited. Unauthorised reproduction or distribution constitutes copyright infringement and may give rise to civil and/or criminal liability under applicable copyright law. To request permission, or to enquire about a licence, contact docmuddie@gmail.com.
[tiab:~N]), which strongly favours papers about the co-formulated product over papers that merely mention both drugs in different contexts. Proximity applies to Title/Abstract only.{product} where the product name should be inserted; it's substituted (quoted) for each product in your list. The simple qualifier above is ignored while this is on.Section name | critical|major|minor