A HYBRID AI-ASSISTED SCREENING FRAMEWORK FOR SYSTEMATIC LITERATURE REVIEWS IN HER2-POSITIVE METASTATIC BREAST CANCER

Author(s)

Alex Bates, PhD1, Sam Mettam, MSc2, Susan McCutcheon, PhD3, Barinder Singh, RPh4, Shivom Prajapati, MPharm5, Sumeet Attri, MPharm5, Sukriti Sharma, MSc5.
1Jazz Pharmaceuticals, London, United Kingdom, 2Jazz Pharmaceuticals, Bramley, United Kingdom, 3SCM22 Ltd, London, United Kingdom, 4Pharmacoevidence, London, United Kingdom, 5Pharmacoevidence, Mohali, India.
OBJECTIVES: Recent position statements from health technology assessment (HTA) bodies, including NICE (UK) and CDA (Canada), emphasize responsible use of artificial intelligence (AI) for evidence generation, aided by appropriate human oversight. Building on the first NICE‑accepted AI‑assisted systematic literature review (SLR), this study utilizes a hybrid AI-human framework to identify outcomes beyond those conventionally predefined in treatment‑ or outcome‑focused SLRs.
METHODS: A literature search was conducted across EMBASE®, MEDLINE®, CENTRAL, and the Cochrane Database of Systematic Reviews (CDSR) from January 2019 to December 2025. Using the AI‑assisted SLR methodology described by Makhija et al., the Pharmacoevidence AI‑SLR tool was applied to an oncology use case in human epidermal growth factor receptor 2 (HER2)-positive metastatic breast cancer (mBC). Outcomes included: trastuzumab deruxtecan (T-DXd) discontinuation, efficacy, safety, tolerability, and subsequent therapy patterns post T-DXd. The AI system served as an independent second reviewer for screening and data extraction.
RESULTS: A total of 2,082 citations were screened using the hybrid framework. A pilot sample of 100 citations was screened to optimize prompts. Title and abstract screening performance was high (accuracy: 96.5%, precision: 87.8%). An accuracy of ~98% was attained for both full-text screening and data extraction of study outcomes. Overall, the hybrid approach markedly improved efficiency, delivering twice the speed of traditional human review while maintaining accuracy above the ideal 95% benchmark, compared with less than 90% for conventional review. These gains translated into an estimated 50% reduction in operational costs and time.
CONCLUSIONS: This study demonstrates that AI‑assisted SLR with a human-in-the-loop, can be efficiently integrated into SLR workflows. Unlike traditional SLRs, which typically focus on a single treatment, this study examined a broader range of therapies used post T-DXd. Agreement rates across first- and second-stage screening outperformed the traditional two-human review (reaching the near-ideal 95% benchmark), with significant efficiency gains and scalability for future HTA submissions.

Conference/Value in Health Info

2026-11, ISPOR Europe 2026, Vienna, Austria

Value in Health, Volume 29, Issue 12S

Code

P17

Topic

Clinical Outcomes, Methodological & Statistical Research, Study Approaches

Topic Subcategory

Artificial Intelligence, Machine Learning, Predictive Analytics

Disease

No Additional Disease & Conditions/Specialized Treatment Areas, Oncology

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×