In human health, many queries can only be answered using observational studies; however, unlike experimental studies, observational studies are prone to several types of bias which may render the results invalid. The main types of bias experienced in observational studies include non-differential recall bias, publication bias, differential recall bias, loss to follow up, refusal to participate, interviewer bias and finally confounding. In this paper, I will discuss confounding as a type of bias and by using example explain how it can lead to invalidity of results.
Confounding
Confounding refers to the inclusion of the risk factor for a disease under study which is connectedto the exposure of interest although it does not form a part of the causal pathway between exposure and endpoint. This implies that if the confounder and the risk factor are associated and their association is not recorded at the beginning of the study, then the impact of the risk factor will be wrongly estimated thus making the results invalid.
Delegate your assignment to our experts and they will do the rest.
Example of a study
Smokers are highly associated with coffee drinking. Since smocking is a known causal factor for coronary heart disease (the end point), and most smokers take coffee often, a wrong impression can be formed whereby coffee drinking will be linked to coronary heart disease. Therefore, the researcher may get the wrong assumption and end up concluding that there is a correlation between coffee drinking and coronary heart disease; this is confounding bias.
Controlling confounding bias
Confounding in observational studies can be done in several ways. First, the researcher may divide the test group into strata that are defined by specific levels of the confounder, in the above example since smoking is the confounder, the divisions can include the non-smokers, heavy smokers and moderate smokers. The analysis will be done individually then each subgroup at a time. Finally, the results of the individual effects are compared to the stratified evaluation thereby comparing to the confounder. Most often, it is difficult for all probable confounders to be considered during matching. In such cases, regression models are used during observational studies.
Reference
Grimes D. A. and Schulz, K. F. (2002). Bias and Causal Associations in Observational Research. The Lancet; 359 (9302): 248-252
