Bias in Epidemiologic Research
For the following study descriptions, describe potential bias (es) most likely to be present. Indicate how differential bias would affect the measure of association (would it be underestimated, overestimated or unchanged):
1. A cohort study was designed to look at the association of pesticides and the occurrence of childhood leukemia. Children living in rural areas were followed and leukemia rates were compared between those children living in areas with high and low pesticide exposure. Pesticide exposure was considered high if it was over 20 "units" and low if it was under 20 "units." The method used to quantify pesticide exposure was accurate to within +/- 12 units.
The potential bias that can occur in this situation would be a selection bias, which is normally introduced in a retrospective study because exposure are been measured indirectly, is self-reported, and could be recalled differentially the parents of children who are not sick, versus those that are sick. Bias can also be introduced if some of the group of individuals refuse to respond to the study survey or are loss to follow (Belson, Kingsley, and Holmes, 2007). This differential loss follow-up could bias the relative risk measure of association of pesticides and the occurrence of childhood leukemia to either underestimate or overestimate the result. Thus, loss to follow is problematic because it could reduce the power of the study to detect the associations that are truly present (Aschengrau, and Seage III, 2008, p.268).
2. Persons diagnosed as HIV positive or negative were interviewed about their number of lifetime sexual partners. The interviews occurred after they were told their HIV status. Indicate how the bias would affect the measure of association (would it be underestimated, overestimated or unchanged)
This situation would produce a differential recall bias between persons diagnosed as HIV positive and HIV negative. Recall bias is very likely because both groups have differential level of accuracy, which suggests that the extent of inaccurate recall could be related to the characteristics of the HIV exposure and the respondents. Interviewing technique, such as the design of questionnaires that is under the control of the investigator and the respondent's motivation, also attribute to recall bias. Differential recall bias could bias the odds ratio measure of association to either underestimate or overestimate the result, thus, leading to false assumption of an association between exposure and disease (Aschengrau, and Seage III, 2008).
3. A cohort study looked at the association of exposure to PCBs in the workplace and the occurrence of cancer over 20 years. At the 10-year follow-up, 15% of the original participants could not be located to assess their disease status.
Healthy worker effect, which is another form selection bias "that occurs in two special types of cohort studies; proportional mortality ratio (PMR) and standardized mortality ratio (SMR) studies" (p.270), which occurs because the general population consists of both healthy and ill people, is selected for comparison. The loss follow-up could bias the relative risk measure of association of PCBs in the workplace and the occurrence of cancer to either underestimate or overestimate the result. Due to the 15% of the original participants that could not be located to assess their disease status (Aschengrau, and Seage III, 2008, and Ibrahim, Alexander, Shy, and Farr, 1999).
4. In a clinical trial of a new drug, the intervention group received the medicine and outcomes were assessed by a doctor working for the pharmaceutical company. The placebo group received the placebo and outcomes were assessed by a doctor from the local community hospital.
This is an interviewer bias because the doctor working for the pharmaceutical company and the doctor from the local community hospital are aware of the treatment status of the case and control group. Interviewer bias could bias the odds ratio measure of association to either underestimate or overestimate the result intervention group and placebo group (Aschengrau, and Seage III, 2008).
schengrau, A., and Seage III, G. R. (2008). Essentials of Epidemiology in Public Health [second edition]. Jones and Bartlett Publishers, LLC
Belson, M., Kingsley, B. and Holmes, A., (2007). Risk Factors for Acute Leukemia in Children: A Review. Retrieved October 8, 2009, from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1817663
Ibrahim, M., Alexander, L., Shy, C., and Farr, S., (1999). Eric Notebook: Selection Bias. Retrieved October 8, 2009, from http://eric.unc.edu/notebooks/issue8/eric_notebook_8.pdf
NMOR, T. A