Hello!
I am working on meta-analysis which will examine survival differences between two groups of patients undergoing lung transplantation. I use RevMan 5.4 to generate forest plots of hazard ratios which I calculate based on the information in included studies. I use formulas provided in paper by JF Tierney (doi:10.1186/1745-6215-8-16). In one case, hazard ratio was 0,61 with 95% CI 0,52-0,7 while log rank test was p=0,4 which is not significant. Is that possible that HR is
this looks like an individual study and graphically it looks like there may be some distributional problems .
usually a non parametric test, like log rank test will be more likely to give a statistically significant result than a parametric test when distributional non conformation is present but a parametric test will be more sensitive if distributional assumptions are met.
whilst your results may be possible, Ithink you should check your calculations.
also see attached reference
Article On Confidence Intervals for the Hazard Ratio in Randomized C...
It seems that there was survial difference before 1500 days. On this regard, maybe you could try to perform Landmark analysis and recalculate the P value from 0 day to this point.
It's possible to have difference between both tests. Other tests for survival analysis such as Wilcoxon test assess different parts of the survival curve so they can yield different results. But I recommend to check if HR is for univariable vs multivariable analysis.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python