OBJECTIVES: Systematic literature reviews (SLRs) are routinely required as part of health technology appraisal (HTA) processes in many countries. However, the traditional manual approach to SLRs is becoming ever more time-intensive given the growth rate of published clinical literature. Natural language processing (NLP) is a form of artificial intelligence (AI) which can reduce this burden; industry and HTA agencies must align on the appropriate way to use this technology to harness efficiencies, whilst maintaining the scientific rigour and transparency of the SLR.

METHODS: A novel NLP-supported SLR update was piloted with the UK HTA agency, the National Institute of Health and Care Excellence (NICE), as part of an HTA appraisal in the advanced ovarian cancer setting. The methodology was developed after discussion with NICE and was designed to balance the need for improved efficiency, maintained accuracy, and sufficient transparency to enable the evidence assessment group (EAG) to provide a thorough critique. In this initial pilot the use of NLP was limited to search query creation and matched entity highlighting during document retrieval; final retrieved documents were then screened manually. A previous retrospective validation exercise showcased the robustness and precision of this approach as compared to a fully manual SLR.

RESULTS: The methodology and results presented for the NLP-driven SLR were validated by the EAG and considered appropriate for HTA decision making.

CONCLUSIONS: This signals that HTA agencies are increasingly willing to explore novel AI-based solutions to improve efficiency, particularly for time-intensive and low-risk components such as SLRs, and where human oversight is retained. There remains an opportunity to explore additional NLP functionality applied to SLRs, such as topic modelling, information extraction, and summarisation. There also remains a need to collaborate with other global HTA agencies to develop broader consensus on the use of these approaches.