Using software engineering principles to improve the completeness and efficiency of the systematic review ecosystem

Thesis by publication."Centre for Health Informatics, Australian Institute of Health Innovation, Faculty of Medicine and Health Sciences" -- title page.Bibliography: pages 80-104.Chapter 1. Introduction -- Chapter 2. Interoperability between trial registries and published results -- Chapter 3. Availability of clinical evidence and systematic review updates -- Chapter 4. Modelling the risk of conclusion change in systematic review updates -- Chapter 5. Discussion -- References -- Appendix.Systematic reviews are a critical component of evidence-based medicine because of their importance in clinical practice guidelines and decision making in practice, but it can be challenging to keep them up to date because of the rate at which new evidence is produced. Despite the number of systematic reviews published each year, new studies are incorporated into systematic reviews relatively slowly, which can delay the recognition of important safety issues. Guidelines on how and when systematic reviews should be updated appear to have little influence over how systematic reviews are updated in practice, suggesting that there may be benefit in developing tools to help systematic reviewers decide which reviews to update, avoiding redundancy and better targeting efforts where they are most needed.In this thesis I use software engineering principles to examine inefficiencies across the systematic review ecosystem, with a particular focus on the role of systematic review updates. First, I highlight the importance of improving data interoperability between trial registries and bibliographic databases. In a literature review I show that trial registries are often disconnected from the articles reporting their results. Second, I show that there is no clear evidence that systematic review updates are undertaken earlier following a signal of new evidence. Rather, systematic reviews often add no new evidence and rarely produce a change in conclusion. Third, I propose a new approach to help identify systematic reviews for which an update is warranted by modelling the risk of conclusion change in a curated set of published systematic review updates. To support the creation of a larger database for use in models of this type, I use a rule-based approach to automatically extract relevant information from published reviews and their updates. Finally, I make several recommendations about the need for an interoperable repository of structured systematic review information.Mode of access: World wide web1 online resource (xiii, 118 pages) graphs, table