Table 3

Hypothetical example in which the effect of G is estimated without bias even though the control disease is caused by E

In this hypothetical example, E+ triples the risk of CVD but G has no effect. If one used CVD controls with lung cancer cases, the odds ratio for E on lung cancer is biased by a factor of one-third; the odds ratio would be 5 instead of 15. But the effect of G is 3 both when proper controls and when CVD controls are used.

Proper controls100100100100
Improper (CVD) controls100300100300
Cases (lung cancer)10015003004500
Odds ratio: cases vs. proper controlsReference15345
Odds ratio: cases vs. improper controlsReference5315
Odds ratio: improper vs. proper controlsReference313