Difference between revisions of "Criteria for confounding"
Bosmana fem (talk | contribs) (Created page with "=Criteria = Two conditions and two restrictions are necessary for a characteristic to be a confounding factor [1]: A confounding factor: # must be a proxy measure of a caus...") |
Bosmana fem (talk | contribs) m (→Criteria) |
||
| Line 2: | Line 2: | ||
Two conditions and two restrictions are necessary for a characteristic to be a confounding factor [1]: | Two conditions and two restrictions are necessary for a characteristic to be a confounding factor [1]: | ||
| − | + | A confounding factor: | |
# must be a proxy measure of a cause, in unexposed people | # must be a proxy measure of a cause, in unexposed people | ||
| Line 11: | Line 11: | ||
These four criteria must be verified whenever a characteristic is suspected of being a confounding factor. In the previous example the confounding factor (vaccination) is associated with both exposure (gender) and outcome (disease). Vaccination is not in any biological pathway between gender and disease and unvaccinated children have a higher risk of disease in both sexes. The two conditions and restrictions are met. The crude risk ratio was artificially increased by the unequal distribution of vaccinated among boys and girls and the fact that vaccination is a protective factor against disease. | These four criteria must be verified whenever a characteristic is suspected of being a confounding factor. In the previous example the confounding factor (vaccination) is associated with both exposure (gender) and outcome (disease). Vaccination is not in any biological pathway between gender and disease and unvaccinated children have a higher risk of disease in both sexes. The two conditions and restrictions are met. The crude risk ratio was artificially increased by the unequal distribution of vaccinated among boys and girls and the fact that vaccination is a protective factor against disease. | ||
| − | To numerically identify a confounding factor the measure of the crude effect is compared to a summary measure of the effect. This is a weighted measure taking into account the stratum specific value of the effect (i.e. the RR in each stratum), attributes a weight to each (based on the size of the sample). If the weighted measure of effect differs from the crude measure, then the characteristic on which we have stratified our analysis (vaccination in this example) may be a confounding factor. | + | To numerically identify a confounding factor the measure of the crude effect is compared to a summary measure of the effect. This is a weighted measure taking into account the stratum-specific value of the effect (i.e. the RR in each stratum), attributes a weight to each (based on the size of the sample). If the weighted measure of effect differs from the crude measure, then the characteristic on which we have stratified our analysis (vaccination in this example) may be a confounding factor. |
It must be considered, however, that there is no statistical method to compare a crude from a weighted measure of effect (a good rule of thumb: confounding may be considered if the difference between the two measures is more than 15-20%). | It must be considered, however, that there is no statistical method to compare a crude from a weighted measure of effect (a good rule of thumb: confounding may be considered if the difference between the two measures is more than 15-20%). | ||
Revision as of 20:04, 21 December 2022
Criteria
Two conditions and two restrictions are necessary for a characteristic to be a confounding factor [1]:
A confounding factor:
- must be a proxy measure of a cause, in unexposed people
- must be correlated (positively or negatively) with exposure in the study population. If the study population is stratified into exposed and unexposed groups, the confounding factor has a different distribution in the two groups
- must not be an intermediate step in the causal pathway between exposure and disease
- must not be an effect of the exposure
These four criteria must be verified whenever a characteristic is suspected of being a confounding factor. In the previous example the confounding factor (vaccination) is associated with both exposure (gender) and outcome (disease). Vaccination is not in any biological pathway between gender and disease and unvaccinated children have a higher risk of disease in both sexes. The two conditions and restrictions are met. The crude risk ratio was artificially increased by the unequal distribution of vaccinated among boys and girls and the fact that vaccination is a protective factor against disease.
To numerically identify a confounding factor the measure of the crude effect is compared to a summary measure of the effect. This is a weighted measure taking into account the stratum-specific value of the effect (i.e. the RR in each stratum), attributes a weight to each (based on the size of the sample). If the weighted measure of effect differs from the crude measure, then the characteristic on which we have stratified our analysis (vaccination in this example) may be a confounding factor.
It must be considered, however, that there is no statistical method to compare a crude from a weighted measure of effect (a good rule of thumb: confounding may be considered if the difference between the two measures is more than 15-20%).
