UPDATING SYSTEMATIC LITERATURE REVIEWS AND REPORTS USING A GENERATIVE AI-ENABLED FRAMEWORK
Author(s)
Ritesh Dubey, PharmD1, Nicola Waddell, HNC2, Rajdeep Kaur, PhD1, Gagandeep Kaur, M.Pharm1, Barinder Singh, RPh1, Mrinal Mayank, B.Tech1, Shubhram Pandey, MSc1;
1Pharmacoevidence, Mohali, India, 2Pharmacoevidence, London, United Kingdom
1Pharmacoevidence, Mohali, India, 2Pharmacoevidence, London, United Kingdom
OBJECTIVES: Periodic updates of systematic literature reviews (SLRs) are required to remain relevant for health economics and outcomes research (HEOR) decision-making. However, manual updating of SLRs is time- and resource-intensive. This study aimed to update existing literature reviews and reports using AI-assisted SLR and report generation, while preserving evidence traceability and ensuring appropriate human oversight.
METHODS: A previously conducted SLR and its report, originally developed in 2024, served as the base for this study (Kaur et al., 2025). An AI-assisted SLR was conducted, supported by automated data extraction tools. Newly identified studies were incorporated into an RAG-based multi-agent framework to update relevant sections of the SLR report, including humanistic burden, economic burden, and unmet needs. Updated report sections were evaluated by a human expert on a 5-point Likert scale to assess factual accuracy, completeness, and evidence traceability.
RESULTS: Over the two-year update period, five newly published studies were identified. Automated data extraction tables were generated using a human-in-the-loop approach, requiring approximately 10-20% human effort. These finalised extraction tables were subsequently used by the system to update the existing SLR report with new evidence. The framework generated revised narrative text, tables, and figures directly within the original manuscript structure without modifying previously validated content. Most updated sections were rated as “Strongly Agree” or “Agree,” reflecting high factual accuracy and clear linkage to source evidence. Minor terminology refinements were required in a limited number of cases. All updated tables and figures were appropriately aligned with the evidence. Compared with traditional manual updating workflows, the framework reduced update timelines by approximately 80-85%.
CONCLUSIONS: This study demonstrates that a RAG-based generative AI framework can efficiently update existing SLRs while maintaining transparency, traceability, and continuous human oversight, offering a scalable and compliant approach to support responsible AI adoption in HEOR evidence generation.
METHODS: A previously conducted SLR and its report, originally developed in 2024, served as the base for this study (Kaur et al., 2025). An AI-assisted SLR was conducted, supported by automated data extraction tools. Newly identified studies were incorporated into an RAG-based multi-agent framework to update relevant sections of the SLR report, including humanistic burden, economic burden, and unmet needs. Updated report sections were evaluated by a human expert on a 5-point Likert scale to assess factual accuracy, completeness, and evidence traceability.
RESULTS: Over the two-year update period, five newly published studies were identified. Automated data extraction tables were generated using a human-in-the-loop approach, requiring approximately 10-20% human effort. These finalised extraction tables were subsequently used by the system to update the existing SLR report with new evidence. The framework generated revised narrative text, tables, and figures directly within the original manuscript structure without modifying previously validated content. Most updated sections were rated as “Strongly Agree” or “Agree,” reflecting high factual accuracy and clear linkage to source evidence. Minor terminology refinements were required in a limited number of cases. All updated tables and figures were appropriately aligned with the evidence. Compared with traditional manual updating workflows, the framework reduced update timelines by approximately 80-85%.
CONCLUSIONS: This study demonstrates that a RAG-based generative AI framework can efficiently update existing SLRs while maintaining transparency, traceability, and continuous human oversight, offering a scalable and compliant approach to support responsible AI adoption in HEOR evidence generation.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR191
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas