School of Population Health

Automating systematic reviews

Clinical decision making relies on up to date summaries of research evidence. Producing these summaries, typically using a rigorous process called systematic review, is labour intensive and there is a huge backlog of such reviews because fo the rampant growth in research publications and limited human resource available for review. We aim to fast track this process using advanced computational systems, initially to help humans work faster, and ultimately to automate much or all of that process. Our work encompasses improvements in computer search technologies, text understanding, mining and text summarization methods.

Good clinical decisions are both informed and take little time to make. The voluminous information available to clinicians today means a long time is needed to locate and identify the relevant evidence for the decision. This project aims to help clinicians make better decisions by providing faster access to the right information.

Evidence based decision support for clinicians

Providing the right evidence to answer the right clinical question is increasingly challenging as both demands on clinicians increases and the evidence base is growing. With millions of trials, systematic reviews, guidelines and other scientific papers to choose from, clinicians need intelligent software to help get to the relevant evidence while attending to their clinical tasks, wherever they are. The QuickClinical meta-search engine simultaneously searches multiple evidence databases (such as PubMed , NHMRC Guidelines) and provides consolidated view of the best available evidence to the PC, tablet or mobile phone. Our work is already being translated into practice. Together with Therapeutic Guidelines Australia, The Centre for Research on Evidence Based Medicine at Bond University we create apps that support general practitioners, guideline developers and systematic reviewers.

Automation of systematic reviews

Clinical Systematic Reviews are robust summaries of the available evidence on a particular clinical question and are regarded as the best source of evidence. The process by which they are conducted is methodical and fairly well understood – but slow. It includes four main tasks:  information retrieval, evidence appraisal, information extraction and synthesis. Supporting the people who perform these steps with computational systems requires an informatics perspective.  We are developing tools to support these four tasks including sophisticated text-processing algorithms that simultaneously search multiple databases, intelligent agents that find and retrieve trials from the internet, artificially intelligent algorithms that extract and synthesise information found in the literature, and techniques to appraise the relevance and quality of clinical trials.

The long term aim of the project is to provide systematic reviews that are themselves software agents that provide summaries of evidence at the push of a button. Our vision was published in the British Medical Journal.

Project Members
image - 1323127848 Guy Tsafnat
Dr Guy Tsafnat
Casual Academic
image - Img 0587
Professor Enrico Coiera
Visiting Professor
Project Collaborators: External

Professor Paul Glasziou
Centre for Research on Evidence Based Practice, Bond University