OBJECTIVES: SLRs are fundamental to evidence-based decision making and often require screening thousands of articles. The increasing scientific data raises the need to incorporate artificial intelligence and machine learning (AI/ML) to analyse vast volume of data. Although there are many AI/ML tools that facilitate SLR process, there is lack of awareness in conducting AI/ML-enabled SLRs. The objective of current review was to analyse the trends and understand gaps of using AI/ML tools in published SLRs.

METHODS: Targeted searches were conducted on Embase^®, MEDLINE^® and Cochrane databases using OVID SP^® on 07 Jun 2023. Search terms used were AI/ML, deep learning, SLRs and meta-analysis and specific AI/ML-enabled databases. No restrictions were applied on indication, geography and timeframe. Hand searches were also supplemented.

RESULTS: Thirty-seven SLRs used AI/ML-assisted databases such as Abstrackr^® (11 SLRs), Rayyan^® (10 SLRs), Covidence^® (10 SLRs) and DistillerSR^® (five SLRs) between 2015-23. Majority of identified AI/ML-enabled SLRs were conducted in United States and United Kingdom and used for title/abstract screening (35 SLRs), followed by full-text screening (nine SLRs) and data extraction (four SLRs).

The availability of AI/ML tools in conducting SLRs increased over the years, from one in 2015 to 11 in 2022 (5 by Mid-2023). Although availability has increased, AI/ML tool was used in only 37 SLRs between 2015-23 (4 by Mid-2023). Reasons of limited use of AI/ML tools in SLRs were: 1) Validity and reliability of most of AI/ML tools currently not established, 2) costs and training required and 3) lack of guidance on acceptability of automated SLRs by health technology assessment (HTA) bodies.

CONCLUSIONS: Availability of AI/ML tools and their use in published SLR has increased in past and might further increase with recent development in AI/ML tools like GPT-3. However, guidance from HTA bodies and increased experience of these tools is needed for growth of AI/ML in SLR.