When I run the Cox regression analysis, I obtain HR with values of 20 and 15 with large CI. Is this acceptable. I've divided a variable into quartiles. The baseline category has fewer events and the 2,3 and 4 category when compared to the baseline provides results with HR of 20 and 15.
The question is not if the results are acceptable (to be published) but if the study was well-designed and the data is trustworthy. If the latter is the case, any result should be acceptable to be published. The large CI only tells you that your data does not allow you a very precise estimate of a parameter in a model, but it also very clearly shows that your data suggests some quite large value. This should be more than enough to demonstrate that the factor your research is about is at least very "interesting", possibly relevant.
CIs and p-values do not indicate anything about the quality of the study or the data. These things must be thoroughly described and explained in the methods section. Nobody is able to interpret a statistical result if he/she does not study the methods. This is often overlooked. People focus on catchy numbers and (often) falsely think that they would be able to understand what the figures mean. Just and only noting a result (p-value, estimate, confidence interval) means nothing unless one one knows the details of the study design.
I have seen such large HRs before in published articles so I am assuming they are acceptable but just think about a few things. Is the analysis crude or adjusted? How wide are the confidence intervals though? If previous studies have look at what you are studying or similar - what were their HRs - very large too? Based on clinical or whatever experience - is large HRs possible? Answers to some of these will help you decide it it is acceptable or not - e.g if clinically it is possible or similar studies found large HRs then i will be confident.
Because you have large CI which could be due to small sample size, I will just be cautious about making bold recommendations based on the results but that is the result you got so you can't change it. If it is crude, run a KM curve and see how close or wide the lines are to each other which will give you some indication that one is far better than the others.
On the interpretation, I don't what the outcome is but say you have 2 treatments (A and B) and the HR for B is 20 - you can just state in the adjusted cox regression model (or unadjusted), treatment B was associated with increased mortality (aHR=20.0; 95% CI = xxx, yyy). Something like this will work.......I hope this helps. Thanks!!
That is very wide, it is probably due to low sample size but just report it.
The question is not if the results are acceptable (to be published) but if the study was well-designed and the data is trustworthy. If the latter is the case, any result should be acceptable to be published. The large CI only tells you that your data does not allow you a very precise estimate of a parameter in a model, but it also very clearly shows that your data suggests some quite large value. This should be more than enough to demonstrate that the factor your research is about is at least very "interesting", possibly relevant.
CIs and p-values do not indicate anything about the quality of the study or the data. These things must be thoroughly described and explained in the methods section. Nobody is able to interpret a statistical result if he/she does not study the methods. This is often overlooked. People focus on catchy numbers and (often) falsely think that they would be able to understand what the figures mean. Just and only noting a result (p-value, estimate, confidence interval) means nothing unless one one knows the details of the study design.
- Badges
- Science topic