Artificial Intelligence-Powered Literature Reviews (AILRs): Streamlining Joint Clinical Assessment (JCA) Dossier Preparation for the Next Generation of Pharmaceutical Submissions
Author(s)
Gautamjeet Singh Mangat, MSc1, Astha Jain, MSc1, Sugandh Sharma, MSc2, Sangeeta Budhia, PhD3.
1Parexel, Mohali, India, 2Parexel, Chandigarh, India, 3Parexel, London, United Kingdom.
1Parexel, Mohali, India, 2Parexel, Chandigarh, India, 3Parexel, London, United Kingdom.
OBJECTIVES: The implementation of JCA has intensified time pressures for pharmaceuticals. Rather than relying solely on traditional, time-consuming literature review methods, sponsors should prioritize early planning and streamlined evidence collection for JCA. Artificial Intelligence (AI) can play a crucial role in this process, offering a means for early exploration and rapid generation of key data points. We conducted an AILR demonstrating its potential in this context.
METHODS: Our AILR consisted of three phases: 1) title/abstract screening (n=4,124 citations), where AI screened 70% of citations after being trained on 30% of human-reviewed citations; 2) full-text retrieval and screening (n=558 citations), utilizing automated features including bulk PDF downloading from open internet sources, automatic tagging, and AutoPRISMA, and 3) data extraction with critical appraisal (n=20 studies), using an LLM with standardized prompts. Expert human reviewers paralleled all phases, enabling direct comparison between the approaches.
RESULTS: AILR demonstrated an overall 35.5% improvement in time efficiency compared to the traditional approach. Phase 1 demonstrated a 60.0% efficiency gain, with AI reducing processing time from 70 to 28 hours, inclusive of a supplementary quality check implemented to mitigate AI's propensity for over-inclusive citation selection. Subsequent stages revealed differential efficiency improvements: Phase 2 achieved 19.0% reduction in time, while Phase 3 demonstrated a more pronounced 37.5% efficiency gain, aided by AI. Beyond time efficiency, AILR also demonstrated remarkable accuracy when compared to human reviewers, achieving 92.1% for first-stage screening, 91.3% for data extraction of three key outcomes, and 88.0% for critical appraisal.
CONCLUSIONS: This case study demonstrates the potential of AILRs in significantly reducing time and effort, particularly during the initial screening and data extraction phases. While there remains room for improvement, AILRs can facilitate early evidence assessment and streamline JCA dossier development, potentially enabling sponsors to meet the stringent 90-day turnaround requirement for dossier submissions more efficiently.
METHODS: Our AILR consisted of three phases: 1) title/abstract screening (n=4,124 citations), where AI screened 70% of citations after being trained on 30% of human-reviewed citations; 2) full-text retrieval and screening (n=558 citations), utilizing automated features including bulk PDF downloading from open internet sources, automatic tagging, and AutoPRISMA, and 3) data extraction with critical appraisal (n=20 studies), using an LLM with standardized prompts. Expert human reviewers paralleled all phases, enabling direct comparison between the approaches.
RESULTS: AILR demonstrated an overall 35.5% improvement in time efficiency compared to the traditional approach. Phase 1 demonstrated a 60.0% efficiency gain, with AI reducing processing time from 70 to 28 hours, inclusive of a supplementary quality check implemented to mitigate AI's propensity for over-inclusive citation selection. Subsequent stages revealed differential efficiency improvements: Phase 2 achieved 19.0% reduction in time, while Phase 3 demonstrated a more pronounced 37.5% efficiency gain, aided by AI. Beyond time efficiency, AILR also demonstrated remarkable accuracy when compared to human reviewers, achieving 92.1% for first-stage screening, 91.3% for data extraction of three key outcomes, and 88.0% for critical appraisal.
CONCLUSIONS: This case study demonstrates the potential of AILRs in significantly reducing time and effort, particularly during the initial screening and data extraction phases. While there remains room for improvement, AILRs can facilitate early evidence assessment and streamline JCA dossier development, potentially enabling sponsors to meet the stringent 90-day turnaround requirement for dossier submissions more efficiently.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR37
Topic
Health Technology Assessment, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Oncology