There could be a wide prior for hr estimated from the clinical trial but the clinical trial is an unbiased estimate of hr because patients were randomized to treatment arms. My own, observational data may have some unknown selection effects so I want to build a model that constrains hr to be consistent with the results of the clinical trial, and not let my own data influence hr. I think this idea is similar to the reasons discussed here http://statistik.wu-wien.ac.at/research/talks/resources/Plummer_WU_2015.pdf except that the data Z is not available and I only have access to the distribution for φ already estimated using Z by somebody else.