What and why?
- is a type of observational study.
- can be used to find risk factors, for example in an outbreak
- cheap and fast
- is often claimed to be inferior to prospective cohort study, but many epidemiologists are disagree on this notion.
How?
- define the cases (for example using CDC criteria or other valid criteria)
- choose controls
- do not match too much since you might miss the effect that you want to find. Patients in the same ward without infection might be a good control group.
- for every case a maximum of 4 controls is enough. Adding more controls will not change the effect much.
- calculate the effect size (odds ratio) of the factors that you might think the risk of the outbreak (you can do this by the hand using 2 x 2 table or statistic program such as SPSS and SAS).
- try to check whether the odds ratio is still significant when other possible confounders are included in the models.
- try not to look at p-value (yes.. I know it is against what you have been teached), but actually you can have p-values for every effect size. But, look at the effect size itself and its confidence interval (CI). 1.1 is less convincing than 2.5 for example. Moreover, a very broad CI can raise an eyebrow, and this is caused by a small sample.
(and when we are talking about small sample, I read a very nice book entitled 'Thinking, fast and slow' by Daniel Kahnemann and it is interesting to know the phenomenon of small numbers. When the population is small, extremes result (as well one that point to positive as well the other that point to negative side) can be expected).
- is a type of observational study.
- can be used to find risk factors, for example in an outbreak
- cheap and fast
- is often claimed to be inferior to prospective cohort study, but many epidemiologists are disagree on this notion.
How?
- define the cases (for example using CDC criteria or other valid criteria)
- choose controls
- do not match too much since you might miss the effect that you want to find. Patients in the same ward without infection might be a good control group.
- for every case a maximum of 4 controls is enough. Adding more controls will not change the effect much.- calculate the effect size (odds ratio) of the factors that you might think the risk of the outbreak (you can do this by the hand using 2 x 2 table or statistic program such as SPSS and SAS).
- try to check whether the odds ratio is still significant when other possible confounders are included in the models.
- try not to look at p-value (yes.. I know it is against what you have been teached), but actually you can have p-values for every effect size. But, look at the effect size itself and its confidence interval (CI). 1.1 is less convincing than 2.5 for example. Moreover, a very broad CI can raise an eyebrow, and this is caused by a small sample.
(and when we are talking about small sample, I read a very nice book entitled 'Thinking, fast and slow' by Daniel Kahnemann and it is interesting to know the phenomenon of small numbers. When the population is small, extremes result (as well one that point to positive as well the other that point to negative side) can be expected).
Geen opmerkingen:
Een reactie posten