GVP-VI Guidelines and biologit MLM-AI

Updated: Sep 19

Naveen Basar and Bruno Ohana.


The Guidelines on good pharmacovigilance practices - module VI is a document issued by the European Medicines Agency outlining practices expected by marketing authorization holders on the collection, management, and submission of individual reports of suspected adverse reactions.


In particular, GVP-VI's Appendix 2 presents guidelines for conducting medical literature monitoring for pharmacovigilance that ensure compliant and auditable processes.


biologit MLM-AI is a literature monitoring tool designed from the ground up pharmacovigilance teams' productivity and compliance needs in mind. In this article, we show how we meet GVP-VI guidelines and offer a single, easy-to-use, and compliant tool for pharmacovigilance literature screening processes.



Let’s review the sections of GVP-VI Appendix 2 relevant to medical literature monitoring and how they map to biologit MLM-AI features:


LITERATURE SEARCH REQUIREMENTS


VI.App.2.1. When to start and stop searching in the medical literature

Literature searches should be conducted for all products with a marketing authorisation, irrespective of commercial status. It would therefore be expected that literature searching would start on submission of a marketing authorisation application and continue while the authorisation is active.

In MLM-AI, all searches and screening results are timestamped for traceability. Searches can run retrospectively (or previous search results uploaded) ensuring coverage from the date of application.


Periodic searches are also run on a schedule on a specified day zero (daily or on a weekday) ensuring no gaps.

VI.App.2.2. Where to look

Medline, Embase and Excerpta Medica are often used for the purpose of identifying ICSRs. These databases have broad medical subject coverage. Other recognised appropriate systems may be used. The database providers can advise on the sources of records, the currency of the data, and the nature of database inclusions. It is best practice to have selected one or more databases appropriate to a specific product.

biologit’s scientific literature database aggregates three large scale databases of broad medical coverage and global reach: PubMed (including Medline), Crossref, and the Directory of Open Access Journals (DOAJ).


They are complemented by regional databases and users can select the ones that suit their needs. Together they provide a powerful integrated scientific literature repository from which to search for events globally.


📖 LEARN MORE: Read about how open access repositories enrich and complement results for adverse event searches in this study.


MLM-AI also supports uploading search results from any database which customers are licensed for, such as Embase, EBSCO, etc. Finally, all search results from biologit MLM-AI are automatically de-duplicated.


SEARCH CONSTRUCTION


Many sections in GVP-VI provide guidance on how to correctly build database searches. In general searching broadly and with high recall is desirable.


(VI.App.2.3.2. Search construction) When constructing a search for pharmacovigilance, the highest recall for a search would be to enter the medicinal product name and active substance name (in all their variants) only
(VI.App.2.3.3. Selection of product terms) Searches should be performed to find records for active substances and not for brand names only This can also include excipients or adjuvants that may have a pharmacological effect
(VI.App.2.3.4. Selection of search terms) As described previously, there is no acceptable loss of recall when searching published literature for pharmacovigilance. The use of search terms (free text or use of indexing) to construct more precise searches may assist in managing the output.

Database queries are greatly simplified in MLM-AI: to set up a new product simply enter the product name which automatically expands into known brand names and synonyms (the final list can be updated).


This ensures a high recall search and minimizes the risk of missing relevant articles caused by complex and error-prone queries:



MLM-AI uses AI predictions to rank and filter articles according to the user’s desired workflow. The level of automation and QC is configurable and all results are fully auditable. This ensures the best of both worlds: simple, high-recall searches with productivity gains from AI.


📖 LEARN MORE: Learn more about productive screening workflows with AI-based tags.


VI.App.2.3.5. Limits to a search

The use of limits that reduce the search result to only those published in the English language is generally not acceptable.

MLM-AI searches articles in any language and performs automatic language detection so users can filter articles and apply the most appropriate workflow for non-English content.


Limits applied to patient types, or other aspects of an article, for example, human, would need to be justified in the context of the purpose of a search.
[…] ICSRs may be presented within review or study and such records may not be indexed as "case-reports", resulting in their omission for preparation of periodic safety update reports from search results limited by publication type.

biologit's unique AI tags built from models fine-tuned for pharmacovigilance lets users quickly prioritize or filter by the presence of patients (case reports), special situations, animal studies, etc.


Note that tags do not impose a limit to search results: users have various productivity options for screening such as batch-screening, sample QC or automatically screening of potentially non-relevant articles.

Limits can be applied to produce results for date ranges, for example, weekly searches can be obtained by specifying the start and end date for the records to be retrieved.

MLM-AI supports the scheduling of searches according to a pre-determined day zero. Users can also submit ad-hoc searches for any period. All search events are recorded in the audit log.

The search should also retrieve all records added in that period, and not just those initially entered or published during the specified period (so that records that have been updated or retrospectively added are retrieved). This should be checked with the database provider if it is not clear.

MLM-AI addresses this guideline by supporting searches on both the publication date and the article “entry” date into the scientific database. This is the default behavior and users do not need any additional configuration to run searches this way.


SCREENING PROCESS AND OUTPUTS


VI.App.2.4. Record keeping

It is always good practice to retain a record of the search construction, the database used and the date the search was run. In addition, it may be useful to retain the results of the search for an appropriate period of time, particularly in the event of zero results. If decision making is documented on the results, it is particularly important to retain this information

In MLM-AI all actions are recorded in a permanent audit log, along with the date, time, and user who triggered the action. Audit log information can later be retrieved via reports.

Screening results are permanently logged and fully visible to the team facilitating QC and audits. All screening decisions can be exported for reporting purposes.


📖 LEARN MORE: See how MLM-AI traceability and audit features enable CFR-11 level compliance.

VI.App.2.5. Outputs

Consistent with the requirement to provide the full citation for an article and to identify relevant publications, the title, citation, and abstract (if available) should always be retrieved and reviewed.

Full citation details, including any screening decision and other relevant article metadata, can be exported in easily consumable formats (excel, CSV) or machine-readable formats (XML) for later processing.

VI.App.2.6. Review and selection of articles

It is recommended that quality control checks are performed on a sample of literature reviews / selection of articles to check the primary reviewer is identifying the relevant articles.

biologit MLM-AI supports effective quality check processes with sample QC, batch review, and a permanent audit log. All screening decisions can be exported.

All articles for search results that are likely to be relevant to pharmacovigilance requirements should be obtained, as they may contain valid ICSRs or relevant safety information.
[…] Outputs from searches may contain enough information to be a valid ICSR, in which case the article should be ordered.

The MLM-AI screening workflow supports integration with browser-based article procurement tools (buy buttons) and secure upload of full text PDFs or translations as attachments ensuring all information remains centralized and accessible to the entire team.

The review should also be used as the basis for collating articles for the periodic safety update report production, therefore relevant studies with no ICSRs should also be identified, as well as those reports of events that do not qualify for submission as ICSR.


The screening workflow also allows marking articles relevant for safety reports and other non-ICSR workflows. Users can optionally select articles for aggregate reporting, safety studies, customizable special situations, and customizable exclusion criteria.

VI.App.2.7. Day zero

day zero is the date on which an organisation becomes aware of a publication containing the minimum information for an ICSR to qualify for submission. For articles that have been ordered as a result of literature search results, day zero is the date when the minimum information for an ICSR to be valid is available.

biologit MLM-AI provides date & time stamps when search results become ready indicating. Article uploads and screening decision actions are also timestamped to indicate day zero.

VI.App.2.8. Duplicates

It is expected that ICSRs are checked in the organisation database to identify literature articles that have already been submitted.

Search results are automatically de-duplicated in MLM-AI. When a previously screened article is identified the duplicate is linked with the corresponding review decision.

VI.App.2.9. Contracting out literature search services

[…] The nature of third-party arrangements for literature searching can range from access to a particular database interface only (access to a technology) to full literature searching, review and ICSRs submission (using the professional pharmacovigilance services of another organisation).

MLM-AI supports multiple teams working together and various forms of integration:

VI.App.2.10. Electronic submission of copies of articles on suspected adverse reactions published in the medical literature

Upon request of the Agency, the marketing authorisation holder that transmitted the initial report shall provide a copy of the relevant article taking into account copyright restrictions, and a full translation of that article into English.

MLM-AI can retain copies of obtained articles and translations by attaching them to search results. These can later be retrieved to facilitate this requirement.


CONCLUSIONS

In this article we reviewed what are the EMA's GVP-VI requirements for pharmacovigilance medical literature monitoring. biologit MLM-AI was built from the ground up with pharmacovigilance professionals embedding best practices and compliance requirements into its feature set.


ABOUT BIOLOGIT MLM-AI

biologit MLM-AI is a complete literature screening platform built for pharmacovigilance teams. Its flexible workflow, unified scientific database, and unique AI productivity features deliver fast, inexpensive, and fully traceable results for any screening needs.