73:poster Living Systematic Reviews (LSR) and Prospective Meta Analysis (PMA): a call-of-duty for Bayesian analysis. (28th April 2022)

• Record Type: Journal Article
• Authors: Tanna, Gian Luca Di; Sunjaya, Anthony Paulo; Santos, Joseph Alvin; Bhaumik, Soumyadeep; Grant, Robert
• Is Part Of: BMJ global health. Volume 7(2022)Supplement 2
• Page Start: A14
• Page End: A15
• Publication Date: 2022-04-28
• Subjects: World health -- Periodicals; 362.105
• DOI: 10.1136/bmjgh-2022-ISPH.40
• Languages: English
• ISSNs: 2059-7908

Abstract

Abstract : Background: The recent Covid-19 pandemic has accelerated the use of LSRs and PMAs, viewed as the 'next generation systematic reviews and meta-analyses'. LSRs and PMAs are prospective designs that can reduce the problems of traditional retrospective meta-analyses (MA) such as selective outcome reporting and publication bias, missing data, etc., and thus offer a better option for incorporating and generating new evidence. Objectives: We propose the Bayesian approach as a method for analysing LSRs and PMAs. Bayesian Meta Analysis (BMA) is particularly appealing - actually, natural - for these designs as it clearly reflects the process of learning, defined as new evidence coming to update the previous knowledge, that is intrinsic to LSRs and PMAs. Methods: Results pooled in the previous update of the LSR, or derived from the studies already known in the PMA, can be used to provide an objective/historical prior distribution. The combination of this information with the accumulated results (conditioning on these) provides the posterior probability distribution that can be used as the prior in the next iteration of the LSR/PMA (yesterday's posterior becomes tomorrow's prior). Results: We will show an example of BMA on a LSR of the association between Covid-19 and asthmatic patients and give practical suggestions for its use. Discussion: Without relying on asymptomatic normality assumptions, BMA is suitable as it is a coherent and flexible framework that, in comparison with frequentist MAs, allows a better assessment of the between-study variance and overcomes some common issues as dealing with missing data and publication bias.