What is the optimal case/control ratio in a case-control study? Why do most textbooks or monographs suggest that it is more than 1? Can it be less than 1 (what are the drawbacks?)? Thank you.
Best Answer
As @EpiGrad says – there is no optimal ratio since otherwise everyone would use it. I suggest you address the issue by looking at the cost of a control versus the cost of a case.
Cases
The basis for a case-control study is that you want to study rare outcomes (cancer, re-operations etc). By being rare your problem is that finding these patient is the major cost.
Controls
Controls are basically anyone without the disease and therefore you have an abundance of these. Finding 10 more controls is usually not so difficult.
Statistics
What you want to see is something where you have a difference between the two studied sample like in the case below:
If you think you'll end up in a situation where you can't see the difference you need to increase your number of patients. In other words you have this situation:
That you want to change by recruiting more patients in one group into this one:
The statistics is very straightforward you gain most power by having groups of equal size. Since your usually in a situation where you can't find more patients in the rare outcome group you want to increase the number of patients in the control group. The central limit theorem gives that the with of the normal curve is given by this simple equation:
$SE = frac{SD}{sqrt{n}}$
- SE = standard error (the standard deviation of the sampling distribution of the mean)
- SD = standard deviation of your sample
- n = number of patients in your sample
As you can see, the effect on the width of the curve each studied person has, decreases as defined by the $sqrt{n}$. This gives that the optimal ratio is where you get most out of the time and effort you spend recruiting patients/controls.
What's vital in case-control studies is that you have to put just as much effort into the controls as you do with the patients. For instance you can't interview the interesting cases yourself while sending a student to talk to the controls. Identifying the correct source population can also be rather challenging.
Similar Posts:
- Solved – the correct method to calculate the Z score for the mean of a variable between two groups
- Solved – Intercept bias in logistic case-control regression: which is the reason
- Solved – Can every retrospective cohort study be transformed into a case control study
- Solved – How to separate groups using PCA
- Solved – How to separate groups using PCA