A large longitudinal study examined the association between obesity (defined as BMI >30 kg/m2) and risk of ischemic stroke in a cohort of patients. After assessing participant baseline characteristics, researchers followed 2,300 patients for a mean of 2.7 years and determined the incidence of major cardiovascular events. Within the entire cohort, the crude relative risk (RR) of ischemic stroke was 2.4 in obese patients as compared to nonobese patients, with a 95% confidence interval (CI) of 1.8-2.6. When the cohort patients were divided into those with diabetes mellitus and those without, among patients with diabetes, the RR of ischemic stroke was 1.1 in obese as compared to nonobese patients (CI, 0.9-1.6) whereas among patients without diabetes, the RR was 1.0 (CI, 0.7-1.5). Which of the following is the best interpretation of this study's findings?
Confounding bias results when an exposure-disease relationship is obscured by the effect of an extraneous factor associated with both the exposure and the disease. In this case, diabetes (confounder) is associated with both obesity (exposure) and stroke (disease). When a crude analysis is performed (without taking diabetes into account), there appears to be a significantly increased risk of stroke in obese individuals as compared to nonobese individuals in the cohort, as reflected by a crude relative risk (RR) of 2.4 with a 95% confidence interval (CI) that does not include the null value (1.0 for RRs). However, when the analysis is stratified by diabetes status, there is no statistically significant association as both 95% CIs cross 1.0, as expected in the presence of a confounding variable. Therefore, the crude RR >1 was due to confounding by diabetes. If there is a confounding variable and the analysis is stratified by this variable, there typically will be no significant difference in risk between the groups because the confounding effects have been removed. Confounding threatens the internal validity of a study.
The RRs reported are for stroke among obese compared to nonobese individuals, not for stroke among patients with diabetes compared to those without diabetes. These results do not imply that diabetes is not associated with ischemic stroke. Rather, they reflect that:
The corresponding RRs were ~1.0. Therefore, obesity does not independently increase the risk of stroke (Choice C). There may be a difference in the overall risk of stroke between the cohort patients with diabetes and those without diabetes; however, the RRs reported specifically compare obese to nonobese patients within those 2 groups.
(Choice B) In patients with diabetes, the RR of stroke in obese compared to nonobese individuals is 1.1, but the 95% CI includes 1.0. This means that the RR is not statistically significant.
(Choice D) The stratification in this analysis is based on diabetes, not obesity. Diabetes is the confounding variable.
(Choice E) Relative rates (as opposed to relative risks) involve incidence rates at a given point in time and are based on person-time calculations that are not given in this question.
Educational objective:
Confounding bias results when an exposure-disease relationship is obscured by the effect of an extraneous factor that is associated with both the exposure and the disease. If there is a confounding variable and the analysis is stratified by this variable, there will be no significant difference in risk between the groups because the confounding effects have been removed.