Increasing the number of controls per case increases the statistical power of the study , but this effect is negliible for more than 4 controls per case so ,other things being equal, cost considerations in obtaining data suggest 1: 4 is optimal.

In studies where cost of obtaining data is irrelevant, eg in using already obtained data ,use of more than a 1: 4 ratio would increase power somewhat and should be used. .

I recommend you to use more control groups rather than more controls in single group. Grimes and Schulz described in perfectly in Article Epidemiology 2: Compared to what? Finding controls for case-...

Thank you so much Aleksandar P. Medarevic for your valuable answer which could mostly applied at the design stage. Now, the problem is the data was already collected therefore, what care & measurement shall I do during the analysis so that precise result can be obtained. You know 1:4 ratio lacks statistical power to detect the difference.

Thank you again

Desta, I am not sure that I am able to help you, but there is solution, certainly. Can you describe what you have done so far?

Beyond 1:3 it is futile to take more controls, straight away I can not explain statistical background about the ratio, 1:1 or 1:2 is good enough and a maximum of 1:3 but not more than that it would increase the cost of your study and at the same time findings are no better. Please refer to standard epidemiology text books viz. one sucy by Charles Hennekens

Thank You Aleksandar P. Medarevic for your time.

What I did is that now the data was collected with 1 case to 4 controls ratio. Right now, I was conducting analysis and writing the whale paper. However, my supervisors & friends suggested that a 1:4 ratio in case-control study is lack potency (power) to detect the difference between cases & controls observed. Since the data was already collected & run by now, what statistical measure can I apply (compensate) so that I can find good precision in detecting the actual difference between cases & controls observed?

Thank you!

Desta,

I recommend you sincerely to follow your hypothesis. After that, you should check you date to find appropriate statistical methods, by which you test your hypothesis. Don't be disappointed if you don't find connection between independent and dependent variable. Gordis's Epidemiology gives you illustrative examples about CS studies, as well as answers to most common questions.

What is your topic about?

Informative articles are:

Article Analysis of matched case-control studies

http://researchonline.lshtm.ac.uk/333606/ (click Download)

I feel it is difficult to provide a clear solution unless you provide the criterion of the sample size you calculated initially. Ideally you should have the odds ratio, prevalence of the disease in unexposed(controls), power (1-beta) and level of significance (alpha), one or two sided test and ratio (case:control) ; these are the criteria to calculate the sample size you change the ratio scenario and look for the sample size and proceed for data analysis accordingly

Ravi Pitani, good answer.

i herewith attach a sample size calculation , where i kept different ratios and you have got different sample sizes you can omit the extra samples or go a head with the existing sample size of yours, by the way the ratio should have decided before the recruitment or start of the study NOT NOW after the data collection. goodluck with your analysis

Case control study is a study design for rare diseases. So in order to get requisite number of study subjects we have to strive very hard. So increasing the number of controls is the solution. But it has been found that if we increase the controls beyond 1 : 4, there is no gain in statistical power of the study. So going beyond 4 is futile....

Please feel free to ask any further question

Dear Desta

As long as you have done the data collection phase of your study and the case to control ratio was 1:4, you start data analysis and you need not to worry a lot about this ratio. As one colleague pointed out, in case control studies, the controls are equal, double, triple or more of the cases . We need this particularly if the cases are few in number to improve statistical efficiency.

With best regards