Corrections for Confounding and Censoring

Practice Questions

 

 


1. Early epidemiologic studies of AIDS in the US showed clearly that the disease occurred with elevated frequency among individuals of Haitian origin, homosexuals, and users of intravenous drugs. Substantial relative risks were found for each of these variables.
a) Do any of Haitian nationality, homosexuality, or intravenous drug use per se cause AIDS?
b) What are the confounding variables that generate the statistical associations of AIDS with
i) homosexuality
ii) intravenous drug use?

 

ANSWERS


In a multi-center cohort study comparing the effects of an experimental Drug A with that of a standard Drug B on mortality from a specific cardiovascular condition among both men and women, the relative risks of death from Drug A as compared to Drug B, for each gender, clinic, and gender within each clinic, are:

 

 

MALE 

FEMALE 

COMBINED 

CLINIC A 

0.5

0.7

1.1

CLINIC B

2.3

2.0

1.0

COMBINED

1.0

1.3

1.1

 

Answer the following two questions assuming that relatively small differences between the gender and clinic-specific relative risks are potentially explainable by chance.

 

2. With respect to the drug-mortality relationship controlling for Clinic, Gender is
a) neither a confounder nor an effect modifier.
b) a confounder but not an effect modifier.
c) an effect modifier but not a confounder.
d) both an effect modifier and a confounder.
e) irrelevant.

 

ANSWERS

3. With respect to the drug-mortality relationship controlling for Gender, Clinic is
a) neither a confounder nor an effect modifier.
b) a confounder but not an effect modifier.
c) an effect modifier but not a confounder.
d) both an effect modifier and a confounder.
e) irrelevant.

ANSWERS


4. In a quite large Chinese case control study of the relationship between nasopharyngeal cancer and consumption of salted fish, suppose the crude relative risk of at least weekly vs. rare salted fish consumption were 5.0. Also suppose that, when the data are stratified into occupational groups, the relative risks are 7.0 for a group consisting of farmers, fisherman, and other outdoor workers, 3.0 for factory workers, and 1.5 for white collar workers, and the Mantel-Haenszel occupation-adjusted odds-ratio was 3.
The most appropriate summary of the results of this study would be to give:

a) the three stratum-specific relative risks of 7.0, 3.0 and 1.5,, because there appears to be effect modification.
b) the three stratum-specific relative risks of 7.0, 3.0 and 1.5,, because there appears to be confounding.
c) the crude odds-ratio of 5.0, because it is the only number which summarizes exactly what was found in entire the study group.
d) the crude odds-ratio of 5.0, because the Mantel-Haenszel summary odds-ratio should not be used when there is substantial effect modification.
e) the Mantel-Haenszel summary odds-ratio of 3.5, because it adjusts for confounding by occupation and gives a fair representation of the magnitude of excess risk associated with salted fish consumption in each of the strata.

 

ANSWERS



5. The table below is similar in form to Table 3 of the exercise on the CASH study in the CDC exercise on oral contraceptives and ovarian cancer, but with the numbers modified. Which of the following accurately describes the data below?

 

Ever-use of oral contraceptives
and risk of ovarian cancer,by parity

 

Parity

Use of OC's

# Case-patients

# Controls

0

Ever user

10

50

Never user

25

200

>0

Ever user

160

200

Never user

60

120

 

a) Parity is not associated with risk of disease, hence can't be a confounder or modify any risk of oral contraceptive usage.
b) Parity is associated with risk of disease, but is does not appear to either confound or modify any risk associated with oral contraceptive usage.
c) Parity appears to both confound and modify the risk associated with oral contraceptive usage.
d) Parity appears to modify but not confound the risk associated with oral contraceptive usage.
e) Parity appears to confound but not modify the risk associated with oral contraceptive usage.

 

ANSWERS


6. By considering how the numbers in the table would change if the investigators had chosen to use twice the number of controls, and the additional controls were similar in oral contraceptive usage to those already selected:
a) explain why data from case-control studies are not suitable for the calculation of absolute risks of disease; and
b) show what would happen as the number of controls were changed if the computations for relative risk appropriate to a cohort study were mistakenly used to estimate relative risk, and compare this result to that from the odds-ratio estimate.

ANSWERS