Most observational studies provide adjusted estimates in the form of OR (with 95% CI), however some of them use HR instead that includes the notion of time to event. Is there a way to transform one to the other?
26
If the time to event study is based on cumulative incidence (that is, the rate of persons experiencing the endpoint) then it's proably fine to consider it as an odds ratio. However, if it is based on incidence density (events per person-time) and includes recurrent events then the equivalence is not as close because risk factors may operate differently to produce the initial manifestation of a disease and its subsequent prognosis.
Hi Sofía, if you are doing a MA probbaly you have the raw data (n/N) so you can use these instead of HR, OR, RR...STATA allows you to do that using the "metan" library
The Cox model, other multivariate technique, allows , unlike the logistic regression which implies the same follow-up period for the subjects of the study, to consider participation time for each subject. If the subjects monitoring period is fixed (no censored data), the logistic regression model (results expressed par aOR with its 95% confidence interval and p) is as good as the Cox model .
Cox exponential model is often associated with survival studies with a different time of participation for subjects (censored data) . The results are expressed by hazard ratio (meaning of a relative risk as for the odds ratio) with its 95% confidence interval and p.
To use the Cox model, not only status (dependent variable) and covariates are introduced in the model, as a logistic regression, but also the participation time of each subject.
So we cant not transform HR to OR .
I have no much to add to the discussion. I agree with Ronán about the basics of the time to event study and the way that we have to use to estimate the relationships with potential risk factors, and the difference with the case-control studies. So, the transformation is not only not possible but even is not a good approach.
There are some effect sizes that are similar, if not exactly the same, and judgments are required as to whether it is acceptable to combine them. One example is odds ratios and risk ratios. When the outcome is rare, or in conducting a nested case control study then these are approximately equal and can readily be combined. As the event gets more common the two diverge and should not be combined. Other indices that are similar to risk ratios are hazard ratios and rate ratios.
Some researchers decide these are similar enough to combine; others do not. The judgment of the meta-analyst in the context of the aims of the meta-analysis will be required to make such decisions on a case by case basis.
I suggest to convert OR to risk ratio (RR) and with some consideration you can assess the similarity of HR and RR. HR deal with time and wouldn't cross-compare OR and HR. you can consider the type of study for concise selection of effect measures.
RR = odds ratio /1 − risk0 + risk0 × odds ratio
where risk0 is the risk of having a positive outcome in the control or unexposed group. Equation can be utilized for both the unadjusted and adjusted odds ratio.
If the rate ratio is the ratio of two rates calculated as cases/person-time, then that is actually a hazard ratio estimate as well.
When in doubt, it's probably best to split up the studies and look at each sub-group.
There are formula to convert RR and HR to OR.
You can find them in the appendix of this paper
Article The “Hispanic mortality paradox” revisited: Meta-analysis an...
some other useful links:
https://handbook-5- https://handbook-5-1.cochrane.org/chapter_12/12_5_4_4_computing_risk_ratio_from_an_odds_ratio_or.htm
https://stats.stackexchange.com/questions/130237/convert-hazards-ratio-to-odds-ratio
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis