I know usually because of rare diseases, people are usually interested to use more controls, but in case of an epidemic the number of disease is more and recruiting cases are feasible.
For example 2000 of cases and 1000 of control
Hi En-Tzu,
Just a point of clarification. The control group does not need to represent the general population of cases. Rather, the control group should represent the pool of patients from whence the cases are drawn. The Wacholder articles (1992) describe this in detail, and much more. In short, we would want as a control, anyone who would be selected (or eligible to be selected) as a case in our study if they had the disease. If some cases are not eligible to be selected for our study (e.g., they went to a different hospital or they had a different test for diagnosis), we want to be sure that our controls meet the same eligibility criteria. Representation of the general population is not criteria of control selection in most situations (unless we have population-based sampling of both groups).
Sure, valid estimates of the odds ratio may be obtained from any ratio of cases and controls, as long as you follow appropriate sample selection processes. Logistic regression and confounding adjustment would be applied in the usual manner.
Paul
Commonly the number of control is equal or more (2-3 times) than cases. Recruiting control is commonly easier than cases. But in case the number of control are limited to be recruited as long as clearly justified, it will be fine
As a general rule, the number of controls should, in theory, be at least twice as high. I imagine the sample corresponds to humans. Due to the difficulty of certain pathological conditions, 30 cases and 60 or 90 controls are often chosen, with all epidemiological, clinical and laboratory parameters well established and respecting the criteria of inclusion or exclusion, in relation to the intake of drugs, alcoholic beverages, Tobacco, basic diseases. However, there are few case studies, including only one, and they are also publishable. The important thing is that by guaranteeing a sample size of at least 30 cases and 30 human controls your conclusions can be generalizable.
Previous studies have shown that it is not efficient to have more than a 5:1 ratio. After that, there is diminishing returns from having a greater number of controls.
I agree with Paul Visintainer that you can have any ratio including a 1:0.5 or 2:1 as you proposed. The data analysis would be the same. I also agree with Adrian Esterman that a ratio more than 5:1 or in your case 1:0.2 or 5:1 is inefficient.
So long you apply correct calculation, it should be okay to have more Cases than Control..
Agree with above.
Generally, higher numbers of cases would produce more precise effect measures.
It will be ok to try different ratio between case and control (ex: 1:2, 1:4...etc.)
But the most important that need to be concerned is the selection of control group, it needs to represent the general population.
As colleagues mentioned above, normally controls are greater in number than cases. This is particularly useful if the number of cases is small. The increase in number of controls will improve the efficiency of statistical analysis. In the case of Sultan, there should be nor problem of using 2000 cases versus 1000 controls. However, it is still better to use at least 2000 controls.
Hi En-Tzu,
Just a point of clarification. The control group does not need to represent the general population of cases. Rather, the control group should represent the pool of patients from whence the cases are drawn. The Wacholder articles (1992) describe this in detail, and much more. In short, we would want as a control, anyone who would be selected (or eligible to be selected) as a case in our study if they had the disease. If some cases are not eligible to be selected for our study (e.g., they went to a different hospital or they had a different test for diagnosis), we want to be sure that our controls meet the same eligibility criteria. Representation of the general population is not criteria of control selection in most situations (unless we have population-based sampling of both groups).
Agree with above.
Generally, higher numbers of cases would produce more precise effect measures.
I think we can take more cases and less control since we do statistical analysis
Hi Sultan,
In a case control study, of key significance is to match each case to a control for comparison. However, having more cases than controls would imply that some cases would not be matched hence making this inappropriate. It is always best to have a matching ratio in favor of the controls.
On many occasions it is difficult for us to find the 30 patients that we consider a minimum requirement for human studies, if you do not have them, do your research the same. The results will be valid even if this theoretical or utopian number is not met ....
I have only come across one study where they used fewer controls than cases. As you may be aware, you can increase power up to a 4 to 1 match for controls to cases. From what I read, it is valid but you may have a hard time publishing
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology