Automatize or perish Towards userled instant and at

Automatize or perish: Towards user-led, instant and (at least) semi-automated systematic reviews 1 Torres , 1 Rana , 2 Stark , 2 Hagmann , 2 Knahl Mercedes Torres Aakash Benjamin Sebastian Constanze 2 2 2 3 Stefanie Polzmacher , Annabelle Wolff , Selin Nur , Clive E Adams School of Computer Science 1, Cochrane Schizophrenia Group 3, University of Nottingham, UK 2 University of Applied Sciences, Ulm, Germany , Background The reviewer’s hand-crafted masterpiece • Current systematic reviewing, from protocol generation to review publication, is, largely, manual. • This process is slow, expensive, repetitive, inaccurate and lacks transparency. • A framework to procure, curate, and synthesise tailored evidence when needed has to be developed. Karl Benz’s hand-crafted masterpiece The idea Cochrane’s mass manufacturing “You can have any review as long as it’s systematic” Writing the protocol Finding evidence Dissemination Data extraction Write up Henry Ford’s masterpiece of mass manufacturing “You can have any colour as long as it’s black” Results of extensive search for relevant work Question setting Update • We propose a short-medium term rejuvenation of the current framework for systematic reviewing employing existing technology but tailoring and developing it for this purpose. This approach would combine Biomedical Informatics, Document Engineering, Machine Learning, Human-Computer Interaction and end-user needs to automatically and instantly: • Relate to only relevant best evidence. • Optimize PDF and HTML documents to make data extraction more accurate. • Extract accurate quantitative and qualitative data from randomized trials efficiently using methods such as Deep Learning, Support Vector Machines or ensemble classifiers. • Create an online repository with all the data extracted from the optimized documents. • Present data tailored to users’ previously-defined priorities/values and specifications. Synthesis Activities in the reviewing cycle with already developed systems of automation • Much relevant expertise and software exists. • Aspects of this work are ongoing in many areas relevant to health care. • Current approaches, however, are entrepreneurial, idiosyncratic and disjointed. It is only a short matter of time before teams unite to ensure the conveyer-belt of best current evidence runs swiftly. Mercedes Benz’s current mass manufacturing The future Soon a new framework for systematic reviewing will: • Support instant, efficient and accurate best evidence generation. • Save time, costs and labour. • Present data to all users via a thoroughly-tested user-friendly online repository. • Allow users to access personalised data according to needs. Conclusions • • • The production line of systematic reviews related to health should never aim to mass produce a perfect cloned product These types of reviews will always involve peoples’ values and judgements However, the time between question-setting and interpretation of synthesised best evidence can be seconds not months or years Much effort has already been expended in this process and, with leadership, co-ordination and investment, this investment could be amalgamated This would produce a new system where repetitive processing involving high-precision drudgery is left to machines and the personal value-linked consideration of what to ask and what data mean left to people. Contact: Dr Mercedes Torres (psxmt 3@nottingham. ac. uk) School of Computer Science, University of Nottingham, Jubilee Campus, Wollaton Road, Nottingham, NG 8 1 BB, UK.