AI-ASSISTED RWE STUDY DESIGN AND REPORTING: QUANTIFYING TIME SAVINGS AND SCIENTIFIC EQUIVALENCE ACROSS PROTOCOL, SAP, CODE, AND REPORT AND MANUSCRIPT GENERATION
Author(s)
Massoud Toussi, MBA, PhD, MD1, Motiur Rahman, MS, PhD2, Laurie Jean Lambert, MPH, PhD3, Gorana Capkun, PhD4, Ayad Ali, phd5, Andrew Cooper, PhD6, Mats Rosenlund, Phd7, Karl-Johan Myren, MSc8, François GAVINI, PhD9.
1Evidence Mastery, Lailly en Val, France, 2US Food and Drug Administration, Silver Spring, MD, USA, 3CADTH, Newington, ON, Canada, 4Merck KGaA, Allschwil, Switzerland, 5BeOne Medicines, San Carlos, CA, USA, 6Shionogi, London, United Kingdom, 7Daiichi Sankyo, Stokholm, Sweden, 8Alexion, Stockholm, Sweden, 9Takeda, Zurich, Switzerland.
1Evidence Mastery, Lailly en Val, France, 2US Food and Drug Administration, Silver Spring, MD, USA, 3CADTH, Newington, ON, Canada, 4Merck KGaA, Allschwil, Switzerland, 5BeOne Medicines, San Carlos, CA, USA, 6Shionogi, London, United Kingdom, 7Daiichi Sankyo, Stokholm, Sweden, 8Alexion, Stockholm, Sweden, 9Takeda, Zurich, Switzerland.
OBJECTIVES: Real-world evidence (RWE) study development involves labour-intensive document production across multiple stakeholders and weeks to months of calendar time. We quantified the time and resource savings achieved by EvidenceAi™ Suite, an AI-powered platform generating study concept sheet, protocol, statistical analysis plan (SAPs), statistical code, clinical study reports (CSRs), and study manuscript, and assessed scientific equivalence against expert human-generated drafts.
METHODS: A cross-sectional time-motion study was conducted among pharmacoepidemiologists and statisticians and RWE scientists across regulators, HTA bodies and pharmaceutical companies. Participants estimated standard calendar time and cumulative human effort required to produce documents equivalent to first drafts generated by ConceptAi™, ProtocolAi™, SAPAi™, CodeAi™, ReportAi™ and ManuscriptAi™ modules. Generated documents were reviewed by the participants and scored against domain-specific scientific rubrics. Time estimates were compared to real calender and human capital time used to generate the same documents by the participants teams.
RESULTS: The survey is ongoing. The preliminary results based on 15 responses provided from 12 organizations for 7 document types suggest that EvidenceAi™ Suite reduced median calendar time from weeks and months to minutes (calendar time saving: ~99%). Cumulative human effort was reduced from average 58.7 hours to minutes (human capital saving: ~98%). The users rated AI-generated documents as editorially and scientifically superior to humans in 93% of cases. Per Module analysis will be provided in the full presentation as the sample size is still small.
CONCLUSIONS: Purpose-built generative AI tools can deliver pharmacoepidemiological study documents of equivalent or superior scientific quality in a fraction of the time and resource investment of human-only workflows. Structured regulatory alignment embedded in AI generation architecture appears to drive both quality and efficiency gains. The human-in-the-loop principle along with these findings have direct implications for RWE study timelines, cost structures, and competitive evidence generation in HTA submissions.
METHODS: A cross-sectional time-motion study was conducted among pharmacoepidemiologists and statisticians and RWE scientists across regulators, HTA bodies and pharmaceutical companies. Participants estimated standard calendar time and cumulative human effort required to produce documents equivalent to first drafts generated by ConceptAi™, ProtocolAi™, SAPAi™, CodeAi™, ReportAi™ and ManuscriptAi™ modules. Generated documents were reviewed by the participants and scored against domain-specific scientific rubrics. Time estimates were compared to real calender and human capital time used to generate the same documents by the participants teams.
RESULTS: The survey is ongoing. The preliminary results based on 15 responses provided from 12 organizations for 7 document types suggest that EvidenceAi™ Suite reduced median calendar time from weeks and months to minutes (calendar time saving: ~99%). Cumulative human effort was reduced from average 58.7 hours to minutes (human capital saving: ~98%). The users rated AI-generated documents as editorially and scientifically superior to humans in 93% of cases. Per Module analysis will be provided in the full presentation as the sample size is still small.
CONCLUSIONS: Purpose-built generative AI tools can deliver pharmacoepidemiological study documents of equivalent or superior scientific quality in a fraction of the time and resource investment of human-only workflows. Structured regulatory alignment embedded in AI generation architecture appears to drive both quality and efficiency gains. The human-in-the-loop principle along with these findings have direct implications for RWE study timelines, cost structures, and competitive evidence generation in HTA submissions.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR114
Topic
Methodological & Statistical Research, Organizational Practices, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas