One of the ways results can be skewed in an observational/epidemiological study is through confounding, when a factor affects both the independent and dependent variables in a study. Put more plainly, the confounder is the extra or underlying factor that can affect the outcome even if it’s also affecting the exposure or reason for the exposure. Confounding by indication occurs when the reason (the medical condition or circumstances) that someone takes a treatment is also the reason for an observed effect, rather than the effect being from the treatment.
Deeper dive
For an example of confounding, consider observational studies that find that women who drink one or two glasses of wine during the second and/or third trimester of pregnancy have children with better outcomes – higher IQs, better academic achievement, better health, etc. – or at least no worse outcomes, than children born to mothers who drank no alcohol at all. However, a confounder here is the socioeconomic status and educational status of the women who tend to have a little wine: they are usually white, college-educated and in higher income brackets. Those women’s children are already more likely to have better outcomes than women from lower socioeconomic or educational backgrounds. Unless the researchers make adjustments to account for those factors, the results are not reliable.
Confounding by indication relates to the reason for the exposure and nearly always has to do with medication: why the medication was indicated. That is, could the reason a person took a certain medication be the reason for the outcomes observed instead of the medication itself? The easiest example is antidepressants and pregnancy. Thousands of studies look at pregnancy and infant outcomes when a mother takes antidepressants during pregnancy. Most find a very small increased absolute risk (about 1% to 3%) among women who take antidepressants. However, most of these studies compare women taking antidepressants to women not taking antidepressants regardless of mental state. Many of these studies did not consider why the antidepressants were indicated – usually a diagnosis of anxiety or depression – and whether that indication (the anxiety/depression) might contribute to poorer outcomes among those women’s children rather than the medication itself.
Some studies have compared women taking antidepressants to two groups of women not taking them: women with a mood/anxiety disorder and women without one. Most of these studies have found that women with a mood/anxiety disorder NOT taking antidepressants have similar pregnancy and infant outcomes as women who did take them. So is it the medication, the depression/anxiety or both that might contribute to a higher risk of certain outcomes? Or an entirely different factor such as a genetic predisposition for a mood/anxiety disorder that also affects an infant? When reading a study that looks at outcomes among people taking a certain medication (depression in women taking birth control, asthma in children exposed to acetaminophen in infancy, etc.), consider why the medication was indicated and whether that might affect outcomes.