Greetings all. If you have two groups (one exposed and not exposed) using an unmatched case-control study design, do you need to match all other variables? Could this be avoided by just performing unconditional logistic regression with matches every variables?
62
First: it's a cohort study. A case-control design has two groups based on the occurrence or non-occurrence of the disease. A cohort study is based on exposure and non-exposure.
Second, unconditional logistic regression with matched data produces estimates of the odds ratio that are biased towards 1 - in other words, towards no effect.
Why not control for confounders by using multivariable models?
I take it that you really are doing a case -control (comparison) study that is selecting on the outcome and that the exposure (which has not been used in the selection) is two groups (this differs from what Ronan infers as you have not made it clear what you are actually doing!).
GIven that; you cannot study prevalence as you have chosen the number of controls and cases and not nature! Eg if it is 1 case to 1comparison, the prevalence is 50% by design -t cannot be anything else,
The reason for matching is bracket off the effect of these other variables giving you more power to detect effects in the exposure of real interest, Doing this by including extra variables in the model requires you have full interactions to get an equivalent control and the models can be very unwieldy. Note if you match on a variable you then cannot study its effect as you have effectively removed that. There are also dangers of overmatching if you have a case control design - see
Removal of radiation dose response effects: an example of over-matching BMJ. 2002 Aug 10; 325(7359): 327–330. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1123834/
Have a look at this for effective matching strategies
http://gking.harvard.edu/matchit
It is not applicable to calculate prevalence rate in a case-control design.
Thanks guys for all your responses. After a careful look, I realized that my study wasn't really a case control but a cohort study involving two groups. They were selected randomly and comparisons were made between these two groups.
Martín : An excellent reference – thank you. I am always impressed that people who answer RG questions can back up their suggestions with must-read publications.
the matching varible must be connected to the disease and the exposure,otherwise matching based on the varible not of interest can pulldown the efficiency of the study.Second ,the matching varible should be few as possible to prevent the result of overmatching and or=1 .Third ,matching does not mean improve efficiency.It can only be of use in study of rare disease or long latency.Here,a eminent book ,Modern Epidermiology,stated the answers very well.I hope it can help you !
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis