REPORTING AI-ASSISTED METHODS IN PROSPERO-REGISTERED ONCOLOGY SLR PROTOCOLS: TRENDS AND GAPS
Author(s)
Hoda Fotovvat, PhD1, Grace Fox, PhD2, Bianca Jackson, PhD3, Ella Jones, BS2;
1Open Health Group, Strategic Market Access, New York, NY, USA, 2OPEN Health, Strategic Market Access, New York, NY, USA, 3OPEN Health, Strategic Market Access, Hingham, NY, USA
1Open Health Group, Strategic Market Access, New York, NY, USA, 2OPEN Health, Strategic Market Access, New York, NY, USA, 3OPEN Health, Strategic Market Access, Hingham, NY, USA
OBJECTIVES: Artificial intelligence (AI) is increasingly used in systematic literature reviews (SLRs) to support timely health technology assessment. As rapid evidence generation is especially critical in oncology, we assessed PROSPERO-registered oncology SLR protocols to quantify AI use and describe its reporting to provide insights for researchers and decision-makers.
METHODS: We searched PROSPERO for SLR protocols using the keyword “cancer” and included records registered 2020 to 2025. Protocols were classified as ongoing, completed, or discontinued. Ongoing and completed protocols were filtered for references to AI or AI platforms (e.g., Nested Knowledge or Covidence). For records mentioning AI, we extracted the named tools and applications within the review process (e.g., screening).
RESULTS: The search identified 53,005 oncology-related PROSPERO protocols, with annual registrations rising from 5,677 in 2020 to 16,355 in 2025. At registration, 90.3% were ongoing and 9.4% completed. Among 52,867 completed or ongoing protocols, 374 unique records mentioned AI use. These protocols contained 402 tool-specific references, indicating that more than one AI platform was reported. Rayyan (49%) and Covidence (37.6%) were most cited; ChatGPT (9%), Distiller (3.5%), and Nested Knowledge (1%) were less common. AI mentions increased from 13 in 2020 to 174 in 2025. Reported AI use was largely confined to screening and, less frequently, data extraction; no protocols described AI integration across the full review workflow.
CONCLUSIONS: SLR protocol registrations in PROSPERO increased more than threefold between 2020 and 2025 in oncology. AI‑assisted methods were rarely reported (<1%) and inconsistently described, suggesting limited uptake as well as underreporting. This underscores the need for improved registry fields and standards to gauge AI’s value, ensure human oversight, and deliver timely, high‑quality evidence for decision‑making. Our findings are limited by reliance on self‑reported text and by the restriction to oncology protocols.
METHODS: We searched PROSPERO for SLR protocols using the keyword “cancer” and included records registered 2020 to 2025. Protocols were classified as ongoing, completed, or discontinued. Ongoing and completed protocols were filtered for references to AI or AI platforms (e.g., Nested Knowledge or Covidence). For records mentioning AI, we extracted the named tools and applications within the review process (e.g., screening).
RESULTS: The search identified 53,005 oncology-related PROSPERO protocols, with annual registrations rising from 5,677 in 2020 to 16,355 in 2025. At registration, 90.3% were ongoing and 9.4% completed. Among 52,867 completed or ongoing protocols, 374 unique records mentioned AI use. These protocols contained 402 tool-specific references, indicating that more than one AI platform was reported. Rayyan (49%) and Covidence (37.6%) were most cited; ChatGPT (9%), Distiller (3.5%), and Nested Knowledge (1%) were less common. AI mentions increased from 13 in 2020 to 174 in 2025. Reported AI use was largely confined to screening and, less frequently, data extraction; no protocols described AI integration across the full review workflow.
CONCLUSIONS: SLR protocol registrations in PROSPERO increased more than threefold between 2020 and 2025 in oncology. AI‑assisted methods were rarely reported (<1%) and inconsistently described, suggesting limited uptake as well as underreporting. This underscores the need for improved registry fields and standards to gauge AI’s value, ensure human oversight, and deliver timely, high‑quality evidence for decision‑making. Our findings are limited by reliance on self‑reported text and by the restriction to oncology protocols.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
SA42
Topic
Study Approaches
Disease
SDC: Oncology