Question

1. The NHS investigators were interested in evaluating associations between certain characteristics such as hair color and tanning ability and risk of melanoma (Weinstock MA et al., Am J Epidmiol 1991; 133:240-245). They identified 34 incident cases of melanoma and 234 controls randomly sampled from the disease-free members of the cohort. Both cases and controls reported their hair color and tanning ability at baseline (prior to the date of melanoma diagnosis for cases) and again several years later (after the date of melanoma diagnosis for cases). The results are summarized on the next page:

Exposure information from baseline questionnaire: Exposure information from post- melanoma questionnaire: Hair Color Cases Controls Hair Color Cases Controls Red or blond (exposed) Brown or black (unexposed) Red or blond (exposed) Brown or black (unexposed) 37 197 220 234 268 48 52 216 234 268 193 34 34 Tanning AbilityCases Controls Tanning Ability Cases Controls None light (exposed) Medium / dark (unexposed) None / light (exposed) 79 155 234 268 15 92 Medium/ dark (unexposed)19157176 34 25 180 34 234 268 Calculate the odds ratios [(a x d)/(b x c)] for each of the four 2X2 tables above. a. b. Do you see any evidence for information bias? If so, which type of information bias? c. What might explain the pattern of results shown above?

Answer 1

a) odds ratio for each of the tables:

(a/b)/(c/d) = ad/bc

for table no1

(11*197)/(37*23) = 2.55

for table no 2

(11*193)/(23*41) = 2.25

for table of 3

(9*155)/(25*155) = 0.36

for table no 4

(15*157)/(77*19) = 1.61

b & c) for table 1 and 2, results are coherent but, for tables 3 and 4 the odds ratio values are less and greater than 1. respectively. One says that it's protective cause while the other says it's harmful. So there is an information bias here.

It's a detection bias.

The cause is because the exposed group is monitored more closely and the hair color of red, brown or black has a more significant effect rather than having a light or dark hair color