Network meta-analysis (NMA) increasingly used in many areas of medical statistics, particularly health technology assessment. Therefore, NMA is a rapidly expanding methodological area of medical statistics, with many methodological challenges still to address.
The RSS conference session, ‘The latest developments in network meta-analysis’ was organised by Joy Leahy (National Centre for Pharmacoeconomics, Ireland) and Sarah Nevitt (University of Liverpool).
Despite her great efforts in organising the session and being scheduled to be the first speaker, Joy was unable to attend RSS Conference as she had welcomed a beautiful baby boy into the world just a couple of weeks earlier. Joy’s PhD supervisor Cathal Walsh (University of Limerick) stepped in and presented their work ‘Individual Patient Data (IPD) in Network Meta-Analysis: Is it worth the effort?’
Based on their simulation study, Cathal noted that while using IPD may produce more accurate results within NMA, for example allowing modelling of treatment covariate interactions, but IPD has a burden and a resource cost. Within a real example of a Hepatitis C NMA, Cathal showed that even with a small amount of IPD this greatly improves the ability to choose the correct NMA, but the treatment rankings do not change when using combinations of IPD and aggregate data. It would seem that in some settings, using IPD may be worth it for NMA.
The second speaker in the session was Maria Sudell from the University of Liverpool, speaking about NMA of joint longitudinal and time-to-event data which is ongoing work as part of her MRC Skills Development Fellowship. NMAs within a joint modelling framework are complex and considerations include covariance structures, consistency equations, use of study level random effects? Due to this complexity, IPD must be used – so this is definitely a context where the burden of retrieving IPD is worth it! Maria is currently implementing this developing methodology in R software and Stan – watch out for this coming soon!
The third speaker in the session was Howard Thom from the University of Bristol, speaking about Network Meta-Analysis of Disconnected Networks. Within a Health Technology Assessment context, disconnected networks may be unavoidable and current potential solutions may involve merging of nodes or pharmacological classes or using IPD – but these solutions are not always suitable. Howard described reference prediction or aggregate level matching, using a real example of a disconnected NMA in ischemic stroke, which may be potential solutions but depending on the context, results of the two methods may be very variable.
The session concluded with a panel discussion of the three speakers, led by session chair Sarah Nevitt. Questions ranged from specific methodological questions about NMA modelling aspects to broad questions regarding the application of NMAs to decision making processes and current gaps in the NMA evidence base.
All speakers, and many of the engaged audience members, shared their insightful thoughts on the current methodological field of NMA. A perfect end to the last parallel session of the RSS 2019 conference.