Implementing machine learning in an evidence synthesis group: Recommendations based on a three-year implementation process
Ames, Heather Melanie R; Borge, Tiril Cecilie; Bergsund, Hans Bugge; Jardim, Patricia Sofia Jacobsen
Research report
Published version
Permanent lenke
https://hdl.handle.net/11250/3154873Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
Sammendrag
The evidence synthesis process is a labour- and resource intensive process but using machine learning (ML) is one way to expedite the evidence synthesis process without compromising quality. Therefore, in 2020, the Cluster for Reviews and Health Technology Assessments (HTV) at the Norwegian Institute of Public Health (NIPH) established a dedicated ML team to implement ML in evidence synthesis processes in HTV. The main aim was to enhance evidence synthesis practices by combining human intelligence with ML to optimize workflow changes throughout the evidence synthesis process. This report provides recommendations on how to carry out implementation of ML functions in ML-naïve evidence synthesis groups, based on the experiences implementing ML at HTV. It offers "best practice" suggestions, rooted in our reflections on implementation, aiming to assist other ML naïve groups or institutions in implementing ML functions in the evidence synthesis process. The guide is adaptable to different organizational goals and objectives, while providing insights applicable to implementation of various ML tools and functions. The report is structured into three main sections corresponding to different phases of implementation: pre-implementation, implementation, and sustainment/evaluation. We use the EPIS framework throughout the document as a tool to explain the different implementation phases and important aspects to consider in each phase. Each section concludes with a "Take home message" based on our implementation experiences, summarized as practical tips on important aspects that we believe are important to consider in the implementation process.