ISS and ISE at AI Speed
The Integrated Summary of Safety (ISS) and Integrated Summary of Efficacy (ISE) are among the most time-consuming documents in NDA preparation. They require synthesizing data across multiple studies, identifying patterns across populations, and constructing a coherent narrative that supports the benefit-risk case. A well-resourced team typically spends three to four months on these two documents alone.
What makes ISS/ISE hard — and what doesn't
The genuinely hard part of ISS/ISE is the synthesis judgment: deciding which safety signals are clinically meaningful, how to frame the benefit-risk case for the target population, and how to anticipate and address FDA's likely concerns. This requires expertise.
The part that takes most of the time is not the hard part. It's:
- Pulling and formatting safety and efficacy data from multiple studies into consistent presentation
- Identifying all prior instances where a given adverse event or efficacy endpoint appeared across the development program
- Checking that every claim in the integrated summary has a cross-reference to a supporting study
- Ensuring terminology is consistent across all tables and narratives
- Drafting the initial narrative sections from structured data
This is exactly where AI operates at its highest value: large-scale, precision-dependent synthesis across complex document sets.
The AI-assisted ISS/ISE workflow
Step 1: Load all study reports, safety data listings, and TFLs into the AI context. Define the structure required by FDA guidance (21 CFR 314.50).
Step 2: Use AI to extract and tabulate safety data across studies — adverse events by system organ class, by severity, by relatedness — producing a consistent master safety table.
Step 3: Use AI to generate the first narrative draft for each ISS section from the structured data, flagging every claim with its source cross-reference.
Step 4: Expert medical review of the AI-drafted narrative, focusing on interpretation, framing, and benefit-risk judgment — not finding all the data.
Step 5: AI-assisted consistency review across the integrated summary and individual study reports.
The result: expert time is concentrated on expert work. The manual extraction, formatting, and cross-referencing — typically 60-70% of total ISS/ISE hours — moves to AI.