I am doing a meta-analysis on prognostic factors predicting survival and want to extract hazard ratios as the effect measure. Some (very good) studies have reported the "relative risk derived by a Cox model" - I thought the Cox model is used to calculate a hazard ratio so I don't understand how the relative risk was calculated from this model? And is there a way of converting this relative risk back to hazard ratio?
A hazard ratio is pretty much the same as a relative risk. In practice you interpret the HR much like a RR. I think you can combine the HR and the RR in meta-analysis without blinking (providing the hazards were proportional, so just the one estimate summarizes the whole follow-up experience). It's not like an odds ratio where there are very much different estimates and you have to argue whether the OR is a good enough approximation or not. For what you need HR=RR.
Hi Stella
I absolutely agree with Darren. The HR can be interpreted as an RR. If you would do survival anaysis via parametric models instead of Cox such as Poisson models with log-Follow-up time as offset you get exactly the same number for the RR from the Poisson model as the HR from the Cox model. With Cox models you usually have to ensure proportional HR.
The problem is usually between RR and OR. An OR can only be interpreted as an RR in the case of low prevalent or indcident diseases because than they are pretty similar. But as soon as the prevalence or incidence increases the OR rises much faster than the RR.
http://link.springer.com/article/10.1007%2Fs00038-008-7068-3
I have seen systematic reviews refer to RRE - Relative Risk Estimate, as a pragmatic approach to combining similar RR measures (not OR), but indicating that they may not be identical.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
Also, holding time constant, say for events at day one post-op, in a Cox regression would provide a result equivalent to a risk ratio. This might be useful in situations where an odds ratio is not desirable from a logistic regression.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis