In-class assignment:

• Consider each of the following scenarios to identify and describe the problems (e.g., bias and confounding) that exist in the design and conduct of the study.

• Then propose an alternative study design that would answer the same research question but produce more valid results.

Scenario #1

Investigators want to know whether taking a new drug can reduce cholesterol levels. To test this hypothesis, they go to a local shopping mall and recruit 500 people to join their study. The investigators assign the first 250 people they find to the treatment group, give them a bottle of pills, and tell them to take one every day. The next 250 people form the control group and are not given any pills. The investigators ask everyone to come into the clinic in one month to have their cholesterol levels measured.

After one month, 150 of the treatment group come to the clinic for cholesterol measurements. 70 of them complain that the pills caused side effects after the first two days so they stopped using the pills. 100 of the control group also show up in the clinic for cholesterol measurement. 40 of these people say they'd heard good things about using the medication for cholesterol, so they asked their doctor for a prescription and started taking the pills. The investigators then compare the cholesterol levels in the treatment group to that of the control group and find no difference.

Scenario #2

Researchers are interested in determining whether alcohol intake is associated with risk of colon cancer. To evaluate this relationship, they identified 300 cases of colon cancer diagnosed at Baystate Medical Center between 1995 and 2001. As a control group, they identified 300 people diagnosed with liver cancer at Baystate during the same years. The researchers then hired two interviewers to conduct in-person interviews with the study subjects; interviewer 1 was assigned to interview the case group while interviewer 2 was assigned to the control group. 250 of the case subjects and 175 of the control subjects agreed to be interviewed. The interviewers asked each subject, "Do you ever drink beer?" and did not collect information on other factors. The researchers then classified subjects as alcohol drinkers or non-drinkers based on their response to this question. They calculated an odds ratio for the alcohol intake-colon cancer relationship of 0.5, suggesting that people who drink alcohol have a lower risk of colon cancer than non-drinkers.