Person-Time in Network Meta-Analysis of Hazard Ratios

[Hans Messersmith]

Hello:

I have a question regarding the use of network to conduct a network meta-analysis of hazard ratios. In the primary reference: https://journals.sagepub.com/doi/10....867X1501500403

It states for hazard ratios the nvar must be person-time at risk. This is almost never reported in the publications of randomized trials. Can anyone provide a reference for how it can estimated or imputed?

I'm not the first to ask this question:
https://www.statalist.org/forums/for...r-hazard-ratio
https://www.statalist.org/forums/for...ed-effect-size

Hopefully I can get answers for them as well. :-)

Hans Messersmith
Clinical Guidelines Specialist
American Society of Clinical Oncology

[Hans Messersmith]

Well, I think I solved my own problem, so I will put the answer here for future folks that search for the same topic.

I am nearly certain, having tested it a bit, that the way to get at this is to skip the network setup procedure and use the network import procedure. It allows you to directly import effect sizes and their standard errors. See the help file for network import for more information on how the data should be set up. The effect size and standard error should be based on the log of the hazard ratio. This is a good reference for figuring out the standard error of a log hazard ratio in multiple situations, Tierney et al 2007: https://trialsjournal.biomedcentral....1745-6215-8-16

This seems to work like a charm!

[Hans Messersmith]

And I just received email confirmation from Dr. Ian White on that answer as well.