Read September 2026

From Protocol to Analysis Plan: SAPAI

My takeaway

Problem

Manual writing of Statistical Analysis Plan for clinical trial study

Solution

This paper introduces a LLM pipeline to successfully generate a SAP from protocol document with 78-81% of accuracy. They predominantly did a prompt engineering technique by generating each section from protocol. Important thing is their Validation and scoring framework to validate claude, gpt, gemini against their pilot SAPs. It involves 0-4 scoring technique whereas each rank’s requirement varies. For example ‘0’ indicates when a section is not extracted at all to ‘1’ indicates when a section is extracted with major inaccuracy and so on.

clinical-trials
sap