Discussion Section 9: Selection and Information Bias

PART A

The "Truth"

Suppose an investigator is able to conduct a case-control study in a reference population (complete population) of 10,000 individuals. In this study, the investigator can correctly identify all individuals who have disease as cases and all individuals who do not have disease as controls. In addition, the investigator is able to correctly ascertain exposure status for all individuals. The data for this complete population are presented below in Table 1:

Table 1: Exposure status for case-control

study of a full population

	Cases	Controls
Exposed	500	1800
Unexposed	500	7200
	1000	9000

1. Calculate the exposure odds for both groups and the exposure odds ratio given the above data.

The "Reality" we hope for....

Since conducting a case-control study of 10,000 individuals is challenging for a number of reasons, suppose that the investigator is able to conduct a case control of 1,000 individuals. The investigator determines that a 50% sample of cases and a 10% sample of controls should be recruited and enrolled from the reference population (as shown in Table 1). Again, let us assume that the investigator is able to correctly identify those who have the disease as cases and those who do not have the disease as controls and is also able to correctly ascertain exposure status for cases and controls. Finally, let's assume that recruitment of cases and controls is conducted such that selection into the study is not affected by exposure status.

1. What would the 2x2 table for this sample look like?

Table 2: 50% sample of cases and a 10% sample of controls

from the reference population shown in Table 1

	Cases	Controls
Exposed	60%	40%
Unexposed	10%	90%

1. Calculate the exposure odds and the exposure odds ratio for this sample.

1. Is this OR different from the one calculated in question 1? Please explain why or why not.

The "Reality" we may ACTUALLY GET ...

Now suppose again that the investigator seeks to recruit a 50% sample of cases and a 10% sample of controls are chosen from the reference population shown in Table 1. However, in this scenario, suppose that selection of cases occurs in such a way that 60% of exposed cases are enrolled and 40% of unexposed cases are enrolled in the study while recruitment of controls does not differ by exposure status.

1. What would the 2x2 table for this sample look like?

Table 3: Case/Control study where selection is not

independent of exposure status

	Cases	Controls
Exposed
Unexposed

Table 3: Case/Control study where selection is not

independent of exposure status

	Cases	Controls
Exposed
Unexposed

1. Calculate the exposure odds and the exposure odds ratio for this sample.

1. Is this OR different from the one calculated in question 1?

1. What type of bias could have led to this result?

1. What is the direction of the biased OR calculated in q6 compared to the OR calculated in q1 and q2?

PART B

To study alcohol consumption and hypertension in Hispanic/Latino(a) men and women ages35years old, investigators conducted a cross-sectional study in several public and private clinic and hospital-based sites across the US. Information on alcohol use was collected via responses to a self-administered health check-up questionnaire completed in waiting rooms prior to a health visit. Blood pressure measurements were taken via trained nurses when patients were in the exam room. Participants were categorized as either heavy alcohol users (3 drinks/day) or not heavy alcohol users (<3 drinks/day). The following are the data collected from this study:

Table 4: Alcohol consumption & hypertension in a sample of Hispanic/Latino(a)

adults (n=24,000)

	Hypertension	No Hypertension	TOTAL
3 drinks/day	3,120	7,296	10,416
<3 drinks/day	1,680	11,904	13,584
TOTAL	4,800	19,200	24,000

1. What is the OR for heavy alcohol use and hypertension in this sample?
2. Instead of the above, suppose that the study was conducted as follows: Investigators collected information on alcohol use via responses to a nurse administered health check-up questionnaire completed in the exam room prior to the health examination. Blood pressure measurements were taken by the same nurse who administered the health questionnaire.

1. First, could there be any problems with the measurement of alcohol consumption during this health check-up questionnaire?

1. What type of bias is a "reporting bias" or "socially desirable responding" broadly known as?

1. Assume that individuals who were heavy drinkers under-reported heavy alcohol use during the nurse-administered survey. In fact, it turned out 20% of individuals who were heavy alcohol users (3 drinks/day) reported consuming<3 drinks/day, in other words they reported being not heavy alcohol users. Compared to Table 4, what would your 2x2 table look like now?

Table 5: Alcohol consumption & hypertension in a sample of Hispanic/Latino(a)

adults assuming 20% under-report heavy alcohol use (n=24,000)

	Hypertension	No Hypertension	TOTAL
3 drinks/day
<3 drinks/day
TOTAL			24,000

Table 5: Alcohol consumption & hypertension in a sample of Hispanic/Latino(a) adults assuming 20% under-report heavy alcohol use

	Hypertension	No Hypertension	TOTAL
3 drinks/day
<3 drinks/day
TOTAL

1. What type of information bias is this?

1. What is the OR in this scenario?

1. What is the direction of bias of the OR calculated in 3e compared with the OR calculated in Part B/q1?

1. Now suppose that the following happened:

Investigators collected information on alcohol use via responses to a nurse administered health check-up questionnaire completed in the exam room prior to the health examination. For any individual who reported consuming3 drinks/day, the nurse would then take 2 blood pressure readings in order to make sure the reading was correct. If one reading was higher than the other, then the nurse would record the higher BP measurement in the survey. As a result of this, 5% of individuals who reported heavy drinking were categorized as being hypertensive even though they were not hypertensive. No mistakes in classifying hypertension status were made among individuals who reported consuming <3 drinks/day.

1. What type of bias is described here?

1. Compared to Table 4, what would your 2x2 table look like now?

Table 6: Alcohol consumption & hypertension in a sample of Hispanic/Latino(a) adults assuming 5% of heavy drinkers misclassified as hypertensive (n=24,000)

	Hypertension	No Hypertension	TOTAL
3 drinks/day
<3 drinks/day
TOTAL			24,000

6. Calculate the exposure odds and the exposure odds ratio for this sample.

7. Is this OR different from the one calculated in question 1?

8. What is the OR in this scenario?

9. What is the direction of bias of the OR calculated in 3c compared with the OR calculated in Part B/q1?