Evidence Thread AI
Independent research in Medical Affairs

Exploring AI for evidence generation.

Evidence Thread AI studies how AI agent workflows can help Medical Affairs teams find and synthesize evidence, identify gaps, and formulate questions for future research.

Explore case studies ↓

Research directions

Each case study starts with a specific Medical Affairs task and asks how to evaluate the AI workflow against human review.

Current pilot

Literature screening

Evaluate how AI agents apply eligibility criteria and flag records that need a reviewer’s judgment.

Potential future work

Evidence synthesis

Explore how AI can organize findings by population, intervention, and outcome while preserving links to the original studies.

Potential future work

Evidence gap assessment

Explore whether AI can help identify unanswered questions and inform priorities for new studies.

Potential future work

Knowledge networks and GraphRAG

Connect studies, populations, interventions, and outcomes in a knowledge graph. Explore whether GraphRAG (graph-based retrieval-augmented generation) can use those connections to answer questions across studies and point reviewers to the publications behind each answer.

First case study · Feasibility pilot

AI-assisted abstract screening

The first case study focuses on title and abstract screening for a systematic literature review. The workflow will apply prespecified PICO criteria (population, intervention, comparator, and outcomes) to propose screening decisions for human review.

A small set of PubMed abstracts was screened and reviewed, and reviewed records were loaded into a prototype evidence graph. The next evaluation will compare AI screening proposals with independent human decisions and track missed records, uncertain cases, and rationales that need closer review.

Illustrative workflow · Abstract screening
  1. Source abstractTitle, abstract, and PICO criteria
  2. AI agent 1Assesses eligibility
    AI agent 2Independently assesses eligibility
  3. Screening proposalCombines assessments and flags uncertainty
  4. Human reviewChecks the source and makes the decision

Publications

Conference abstract in preparation

AI-assisted abstract screening for systematic literature reviews

The abstract will describe the feasibility pilot for AI-assisted screening using prespecified PICO criteria.

Insights

Perspectives on AI in Medical Affairs, from lessons in evidence review to developments in the wider field.

What should AI agents have to show before we trust an abstract screening decision?

AI can help screen a large set of abstracts, but the quality of the workflow starts with how those records were found. When an AI assistant searches the open web, it can surface press releases, reviews, and other secondary sources alongside original research. Even an accurate summary of a press release offers a different basis for screening than the original study abstract. Start with a documented search of bibliographic databases and preserve the source record for each decision.

Screening introduces a different risk. A model may infer that a study enrolled the population of interest or measured a particular outcome when the abstract never says so. That inference can sound convincing. Ask the model to assess each eligibility criterion against the supplied text, cite the passage it relied on, and flag missing information. At the abstract stage, an unclear criterion can be a reason to advance a record to full-text review.

One prompt can ask a model to screen and check its answer, but that self-check is part of the same assessment and can repeat the same error. Separate the initial decision from verification. A second assessment can compare the proposed decision and supporting passages with the original abstract and eligibility criteria. Disagreement is a signal for review, and agreement can still hide a shared mistake.

Humans should define the protocol, resolve ambiguous cases, and remain accountable for what gets excluded. The goal is to focus their time on disagreements, missing information, and borderline cases. Before relying on that approach, compare the workflow with independent human screening and measure how often it would exclude a record a reviewer would advance. If performance supports selective review, continue auditing a sample of routine exclusions. The workload benefit depends on how much review can be safely avoided and how much time verification adds.

Team

Stefan Kolata, PhD

Founder & Research Lead

Global Medical Affairs leader with 15+ years of combined experience in scientific research and biopharma, spanning medical strategy, global launch readiness, evidence generation, and lifecycle planning.

More about Stefan

I combine scientific expertise with strategic thinking, cross-functional leadership, and experience navigating complex global development alliances. I earned my PhD at Rutgers University, with a concentration in behavioral neuroscience.

I champion responsible AI adoption for evidence synthesis and medical communications in regulated environments. I received an internal AI Innovation Award for work on secure multi-agent verification workflows and served on an enterprise AI group supporting governance and global AI training.

I’ve been invited to speak at international life sciences and biotechnology conferences on Medical Affairs, scientific publications, and compliant AI innovation.

About the project

Evidence Thread AI is an independent research project. Across case studies, the project examines how to keep AI-assisted work traceable to source evidence and verifiable by scientific reviewers.