Bias and Confounding Question 43

Question

Because of an unusually high occurrence of endogenous anxiety syndrome among undergraduate students at University X, a case-control study was carried out to determine whether taking an introductory statistics course (a 0,1 exposure variable, where 1 = exposed and 0 = unexposed) might be a cause. The study involved 57 students diagnosed with the anxiety syndrome, and these ‘cases’ were compared with a sample of 750 ‘normal’ controls. Consider the following 2 x 2 tables that stratify for previous history of mental disorder (MD):

• 	MD = 1				MD = 0
  	E = 1	E = 0			E = 1	E = 0
  D = 1	25	15			2	15
  D = 0	50	50			50	600

What is the estimated measure of effect that describes the E-D relationship that ignores the control of the variable MD (i.e., what is the estimated crude effect for these data)?

b. What are the estimated effect measures for each MD group? Again, show your calculations. 

c. Based on your calculations for either or both of the above questions, should MD be controlled because there is meaningful interaction? Explain. 

d. Based on your calculations for parts a and b, should the variable MD be controlled because it is a confounder? (Assume that the a priori conditions for confounding are already satisfied for the variable MD.) Explain your answer with appropriate information and logic. 

e. What is the estimated odds ratio that describes the association of MD with exposure? What does this odds ratio say about the distribution of previous history of mental disorder (i.e., MD) when comparing students taking introductory statistics with those not taking introductory statistics? Does this result support the conclusion that MD is a confounder in these data?  [Hint: To answer this question, you need to consider the following alternative way to define confounding: If the measure of effect is an odds ratio, a covariate (e.g., MD) is a confounder provided it is associated with both the exposure (i.e., E) and the disease (i.e., D) separately, i.e., ORMD,E|D=0 is meaningfully different from 1 and ORMD, D|E=0 is meaningfully different from 1.

Link to Bias & Confounding Answers

Source: 250B Problem Set