>>Return to Tell Me About Statistics!
Last time(vol.11), we explained the difference between the risk and odds ratio (OR). The risk ratio has immediate clinical utility, but the OR may seem less clinically useful than the risk ratio. However, in clinical research, the OR is often used. Why is OR used in clinical research?
In this article, we explain what can be learned from the OR.
What are cohort and case-control studies?
How are ORs used in clinical research?
Typical clinical research include “cohort studies” and “case-control studies.” When analyzing the latter study’s data, risk ratios cannot be used. This is where the OR is used.
First, I would like to explain what is a “cohort study” and a “case-control study.”
Clinical research can be divided into “prospective” and “retrospective,” cohort studies are “prospective,” and case-control studies are “retrospective.” For example, suppose we want to examine the association between the presence or absence of smoking and the occurrence of stroke in patients with hypertension.
[Cohort study]
We will randomly select 1,000 people with hypertension who have never had a stroke, investigate whether they have ever smoked, and then follow them over the next five years to see if they have had a stroke.
Such studies are referred to as cohort studies. This study is called a “prospective” study because it looks into the future five years from now.
[Case-control studies]
A total of 1,000 patients with hypertension who have had a stroke and 1,000 randomly selected stroke-free patients with hypertension will be surveyed for smoking habits.
Such studies are called case-control studies. This study is called a “retrospective” study because it is retrospective in nature.
Let us look at the difference between these two studies in terms of cause and effect.
In the above example, the presence or absence of smoking is the cause and stroke onset is the result. A cohort study examines future outcomes (whether a stroke has occurred), and a case-control study examines past causes (whether someone has smoked).
We now explain that when analyzing data collected in a case-control study, the risk ratio cannot be used, but the OR can.
Case-control studies use odds ratios
We refer to the data in the table as data collected in a case-control study (retrospective study).
[Table] Examples of case-control studies on the relationship between stroke and smoking status
In this table we see that, smokers have a 77% risk of having a stroke. Based on this data, can we generalize that smokers with hypertension have an approximately 80% risk of having a stroke?
This example is set up with data collected in a case-control study.
To examine whether smoking is a factor in the onset of stroke in a case-control study, as a case, we collect data on people who have had strokes and whether or not they smoked. Additionally, as a control, we collect data on people who have not had a stroke were collected and examined to see if they smoked.
Essentially, the available data are “what percentage of people with hypertension who have a stroke smoked” and “what percentage of people with hypertension who have not had a stroke smoked.”
Therefore, from case-control studies, we cannot obtain data on either “what percentage of smokers had a stroke” (numerator of the division when calculating the risk ratio) or “what percentage of non-smokers had a stroke” (denominator of the division).
The risk of developing a stroke in all participants is 50% and depends on their sampling (extraction). It would be a big mistake to calculate the risk ratios using risks that are dependent on sampling. Therefore, in case-control studies, ORs can be used to determine whether there is an influencing factor.
In this table, the OR is greater than 1.0, indicating that the presence or absence of smoking is an influencing factor in the incidence of stroke in patients with hypertension. However, it should not be interpreted as “the OR indicates that smokers have a 4.0 times greater risk of having a stroke than non-smokers.”
Moreover, if the incidence of a disease is extremely rare, the risk ratio can be approximated using the OR.
>>Return to Tell Me About Statistics!
Comments are closed