03-03-2018 05:52 PM
I am wondering if it is possible to compare two separate hazard ratios obtained from Cox models i.e. not a nested model, and obtain a p-value for comparison.
I came across a recent high impact journal publication where the authors state in their statistical methods: "Relative risks were expressed as hazard ratios with 95% confidence intervals derived from Cox models; hazard ratios for a positive FFRCT versus positive CTA were compared with the chi-square test." Link to paper: http://www.massgeneral.org/imaging/assets/pdfs/cardiac-mrpetct/lu-promise-ffrct-jacc-cvi-2017.pdf
How could this be done? My understanding is that chi-squared test is a global test, like the log-rank test; My take is that authors have done a 2x2 contingency table and obtained a p-value, but can you then extrapolate from this a comparison of hazard ratios? I did not think so because this is missing the time-to-event component.
Is this a valid method to perform?