CLIScan™: The Future of Literature Screening for Pharmacovigilance

In this blog, you’ll discover:

1. What CLIScan™ is and how it streamlines clinical and pharmacovigilance literature screening.
2. The powerful features designed to improve screening efficiency and accuracy.
3. How CLIScan™ compares with other literature screening tools and why it stands out.

Introduction

Literature screening is a crucial and important aspects in pharmacovigilance processing, that would never miss any reports that contain any drug safety information.

Literature volume is outpacing manual screening capacity, and GVP Module VI still demands weekly reviews. So here we discuss in this article we explore and introduce our powerful AI/ML modelling CLIScan – Clinical Literature Scan.

Why Is PV Literature Volume Outpacing Manual Screening Capacity?

Every day, thousands of new biomedical articles are published across databases like PubMed, Embase, local journals, and conference proceedings. However, the number of PV professionals available to manually screen these publications has not increased at the same rate. As a result, the volume of literature is increasing faster than people can review it manually.

1. PubMed alone holds more than 40 million citations and adds over 1 million new biomedical articles every year (NIH/NCBI, 2024).
2. Add Embase, which covers 98% of MEDLINE journals plus another 3,500-plus titles (Elsevier)

There are several literature database available out there day-to-day 1000’s of articles published in it

A small to mid-size PV team can face thousands of new abstracts a month across these databases. Manual abstract screening runs roughly 1-2 minutes per article, and full-text review takes 15-20 minutes per article

Important takeaways Features:

Why Does AI-Native Active Learning model is important?

Peer-reviewed research shows active-learning models can cut the number of publications needing full screening by 63.9% to 91.7%, while still identifying 95% of relevant records (Systematic Reviews journal / BioMed Central, 2023).

1. Boolean and keyword searches are static: they don’t improve as a reviewer works, and they can’t tell the difference between a term used in a relevant clinical context versus an irrelevant one.
2. Active learning is dynamic. Our active learning model with our own data shows up to 83-95% workload reduction while still surfacing about 95% of relevant articles because the model retrains on every decision a reviewer makes.

Active-learning models reduce full-text screening workload by 63.9% to 91.7% while still surfacing 95% of relevant records, according to a 2023 peer-reviewed study in Systematic Reviews (BioMed Central), the underlying methodology CLIScan’s personalized reviewer models build on.

AI/ML with our tool

Our AI/ML model has capable of compete with all other models available where the other tools can have it.

1. Naive Bayes or Logistic Regression
2. TF-IDF feature extraction
3. Uncertainty/max/random query strategy
4. Class balancing

Every Include, Exclude, or Maybe decision retrains that individual’s model, so the ranking gets sharper the more a reviewer works through a project. For brand-new projects with zero labeled decisions yet, CLIScan falls back to a hand-weighted PV-term dictionary combined with TF-IDF scoring, so screening can start immediately instead of waiting on a cold-start model.

Advanced de-duplication

Available Strategies

1. DOI Only:
   Matches articles exclusively by their Digital Object Identifier (DOI). This is the most precise method with virtually zero false positives, since DOIs are globally unique identifiers assigned to each published article.
2. Conservative
   Combines exact DOI matching with fuzzy title comparison using a 90%+ similarity threshold. Catches duplicates even when DOIs are missing, as long as titles are nearly identical.
3. Aggressive
   Uses a lower similarity threshold (85%+) and also compares author lists. Catches duplicates with minor title variations, different formatting, or partial author overlaps across databases.
4. Content Similarity( Advanced)
   Uses Levenshtein distance (edit distance) algorithm to measure character-level similarity between titles and optionally abstracts. Detects duplicates even with typos, formatting differences, special characters, or encoding issues.

CLIScan Workflow Builder — Advanced Node Configuration

Visual drag-drop canvas, no code. Node = single configurable step, connected by arrows.

What is node workflow builder

same node type reusable multiple times in one workflow — e.g., two Deduplication Nodes (exact early, semantic late).

For example in the same workflow you can implement multiple stages of de-duplication before it go for screening.

Advantages of node workflow builder

• No-code flexibility — pharmacovigilance logic built visually, adapts per project without engineering
• Node reuse — same node type placed twice for staged filtering (cheap exact pass before expensive semantic pass) = efficiency + accuracy (98%)
• Branch logic — Decision/Gate/Switch nodes route serious/IME cases to stricter review chains automatically, routine cases through lighter path — faster triage, no manual sorting
• Auto-routing based on logic no manual intervention required
• Full traceability — every article’s path (nodes passed, decisions, who/when) logged immutably → 21 CFR Part 11 audit trail
• Versioning — editing live workflow creates new version, old versions stay linked to articles processed under them — audit continuity guaranteed

How Does CLIScan Actually Compare to other Legacy Tools?

Comparison table

Here’s how the three approaches stack up on the capabilities that matter most for a regulated PV workflow.

CLIScan That Adapts to Your Process, Not the Other Way Around

Most literature screening tools force you into their fixed sequence. But CLIScan with advanced features in whatever order and configuration your process actually requires.

the tool is inbuilt and developed with a advice of experienced candidates within Pharmacoviglance and clicnical research, who had have the things in mind as with thier mind it developed in firsthand. It is none other like other companies a randoom tech guy with an idea.

Efficiency is the obvious win, but the deeper value is traceability. Every article’s path through the workflow — which nodes it passed, what decision was made at each step, who made it and when — is logged immutably. That’s not a nice-to-have; it’s what makes the pipeline defensible under 21 CFR Part 11.

And because editing a live workflow creates a new version rather than overwriting the old one, you can always reconstruct exactly which workflow configuration was active when any given article was screened. Your process improves over time without breaking your audit trail.

Closing thoughts

CLIScan doesn’t just automate literature screening — it lets you design the screening logic that fits your team’s actual risk tolerance and regulatory obligations, then enforces it consistently and audibly on every article that comes through. Fewer duplicate reviews, faster routing for serious cases, and a complete record of every decision along the way.

That’s what “enhancing the workflow” should mean.