Review Questions Section 9.1-9.4 (EPI)

Individuals at an elevated risk for a given medical condition are as likely to volunteer to take part in a RCT to prevent that condition as are individuals at moderate risk for that condition. This is an example of selection bias in a RCT.

False

A cross-sectional study was conducted to examine the relationship between regular exercise and obesity in children 8-12 years old. Based on prior findings from studies of adolescents, the investigators hypothesized that lack of regular exercise was also going to be associated with obesity in young children. They conducted a study with a sample of n=24 children in residing in Ohio. Once the data were analyzed, the investigators found that there was no statistically significant association between exercise habits and obesity (OR = 1.1, p = 0.11). The most likely explanation for the results obtained here is:

Random error (Feedback: The small sample size of only 24 children suggests that low power may have led to a type II error, especially since this association is likely to exist. A type II error is a type of random error.)

In order to understand how they could have obtained this result, the investigators compared the baseline data for those who remained in the study (completed all follow-up visits) and those withdrew from the study. This comparison revealed that women who had dropped out of the study were more likely to report having a close female relative who had breast cancer (60%) and being regular drinkers (84%) compared with women who remained in the study for the entire follow-up period (37% with family history of breast cancer and 59% regular drinkers).

Selection bias

Investigators conducting a prospective cohort study collected blood samples from participants at the beginning of the study and froze them for later analysis. After the cohort study was completed a nested case-control study approach was employed to understand whether blood levels of certain hormones (exposure) were associated with Alzheimer's disease (outcome). The previously stored blood samples were analyzed to quantify blood hormonal levels among cases and controls. After the nested case-control study analysis was completed, it was determined that the samples had degenerated during the storage period because of a failure to maintain consistent temperatures in the freezer. The above scenario is an example of information bias because:

The degenerated samples are likely to have resulted in some incorrect classification of exposure status in both the cases and controls

What is the definition of selection bias?

A type of bias that occurs when participants are recruited into a study in such a way that the study sample is not representative of the target population

In a case-control study examining the relationship between intravenous drug use and HIV infection, the investigators discovered after the study was completed that participants in the case group were more likely to underreport their drug use due to fears arising from previous convictions for drug-related violations. Underreporting was not a problem in the control group, however, since these participants were not receiving any medical care at the time of the study. The above scenario is an example of:

Information bias

In a large, hospital-based case-control study of smoking and lung cancer was conducted across 5 major cancer speciality hospitals in the US. Eligible case participants were individuals newly diagnosed with lung cancer. Eligible control participants were individuals newly diagnosed with noncancerous chronic pulmonary diseases (i.e. chronic bronchitis and emphysema). The investigators found that the association between smoking and lung cancer was relatively weak (OR = 1.1, p<0.05). This finding was surprising as a later case-control study of lung cancer and smoking found an OR = 3. Which of the following terms correctly identifies the direction of bias when comparing the measure of association obtained in the hospital-based case-control study compared with the measure of association reported in the later case control study ? (Select ALL that apply)

Negative bias Bias toward the null value

A cohort study was conducted to examine whether exposure to certain types of pesticides during early childhood increased the risk of developing neurological problems in later adulthood. The investigators obtained a CIR = 2.6 for the relationship between early childhood pesticide exposure and neurological problems in adulthood. However, upon re-analysis of the data, it appears that there is no relationship between early childhood pesticide exposure and neurological problems in adulthood (CIR = 1). Which of the following terms correctly identifies the direction of bias when comparing the original CIR=2.6 compared with the CIR=1.0 obtained when the data were re-analyzed? (Select ALL that apply)

Positive bias Bias toward the null value

The investigators of a prospective cohort study collected blood samples from the participants at the beginning of the study and froze them for later analysis. A subsequent nested case-control study was conducted to determine if blood levels of certain hormones (exposure) were associated with Alzheimer's disease (outcome). The previously stored blood samples were analyzed for the cases of Alzheimer's disease and the controls. It was later determined that the samples had degenerated during the storage period because of a failure to maintain consistent temperatures. Select the correct statement:

The misclassification is likely to represent nondifferential misclassification since presumably the samples for both cases and controls were affected uniformly.

If a study has good internal validity, what does that mean?

The study findings accurately reflect the true measure of association between the exposure and outcome.

If a study has good external validity, what does that mean?

The study findings are generalizable to the underlying or target population.

In a case-control study examining the relationship between intravenous drug use and HIV infection, the investigators discovered after the study was completed that participants in the case group were more likely to underreport their drug use due to fears arising from previous convictions for drug-related violations. Underreporting was not a problem in the control group, however, since these participants were not receiving any medical care at the time of the study. Information bias is likely to influence the results obtained in this study because:

The underreporting by the case group suggests that exposure may have been misclassified more often in the case group than in the control group; this is differential misclassification

In order to understand how they could have obtained this result, the investigators compared the baseline data for those who remained in the study (completed all follow-up visits) and those withdrew from the study. This comparison revealed that women who had dropped out of the study were more likely to report having a close female relative who had breast cancer (60%) and being regular drinkers (84%) compared with women who remained in the study for the entire follow-up period (37% with family history of breast cancer and 59% regular drinkers). How might these losses have influenced study results (select ALL that apply):

There was a low likelihood of selection bias due to loss to follow-up This study illustrates a high likelihood of selection bias due to loss to follow-up Those lost to follow-up tended to be women with a family history of breast cancer as well as regular drinkers. Family history of breast cancer is a very strong risk factor for breast cancer, so if they had been included in the study results, the findings may have changed (i.e., an association between alcohol consumption and breast cancer may have been found) It's unknown if the loss to follow-up was among drinkers or those with a family history of breast cancer; so it's unclear if they had been included in the study results, whether or not the findings may have changed

(True or False) The following is an example of selection bias: The participants who were enrolled in the study were the easiest to reach, and not representative of the participants who were more difficult to reach.

True

A researchers recruit cases and controls based on likelihood of exposure status. This is an example of selection bias in a case-control study.

True

An epidemiologist finds that individuals of one race or ethnicity are more likely to withdraw from a cohort study than those of any other race or ethnicity. This is an example of selection bias in a cohort study.

True

An investigator conducting a population based cross-sectional study finds that individuals who answered the landline-based telephone survey tended to be older than individuals who answered a cellphone-based telephone survey. The above example is an illustration of selection bias in a cross-sectional study.

True

In a case-control study examining the relationship between intravenous drug use and HIV infection, the investigators discovered after the study was completed that participants in the case group were more likely to underreport their drug use due to fears arising from previous convictions for drug-related violations. Underreporting was not a problem in the control group, however, since these participants were not receiving any medical care at the time of the study. The measure of association obtained in this study:

will tend to underestimate the true measure of association (be negatively biased) since the case group tended to underreport their drug use