[Evidence synthesis 2.0: how artificial intelligence is making systematic reviews more efficient.] / La sintesi delle evidenze 2.0: come l'intelligenza artificiale sta rendendo più efficienti le revisioni sistematiche.

Systematic reviews (SRs) are essential tools for synthesising the available scientific evidence on a given topic, and in some healthcare fields they represent the core for public health decisions according to the principles of evidence-based medicine. However, keeping up to date with the volume of scientific production is not always easy given the estimated annual increase in scientific publications of 4.10%. Indeed, SRs take a long time, with an average time of eleven months from design to submission to a scientific journal; to make more efficient this process and timely achieve evidence collection, systems such as living systematic reviews and artificial intelligence tools have been developed for the automation of SRs. These tools can be divided into three categories: visualisation tools, active learning tools and automated tools with Natural Language Processing (NLP). Nlp makes it possible to reduce the time spent and human error, for example, in the screening of primary studies; there are already many tools that apply to all stages of a SR, currently the most widely used are those with "human-in-the-loop" where the reviewer is involved in the various steps to verify the goodness of the work performed by the model. At this time of transition in SRs, new approaches are emerging and are increasingly appreciated by the community of reviewers; leaving some more basic but also error-prone tasks to machine learning tools can increase the efficiency of the reviewer and the overall quality of the review itself.