Confounding by Indication

Confounding by Indication is a type of confounding in observational studies that makes a treatment appear to look harmful just because they are only given to people with a condition. It is also referred to as treatment selection bias.

For example, almost all those with AIDS in western countries that die are on antiretroviral drugs. That could wrongly be interpreted as the drugs were the cause of the death (confounding by indication). It is a common claim by HIV deniers that the drugs cause the deaths.

The Encyclopedia of Public Health defines this as:

Confounding by indication is a bias that occurs when the drug of interest is selectively used or not used by those who developed the outcome of interest; a causal relationship between the drug and its effect on disease outcome cannot be established. This situation frequently occurs when, within a study group, some of the patients with the given illness have other health problems requiring the prescription of several available treatments. It is not always possible to control for this type of confounding, although randomized control trials are considered as a method of choice for the assessment of drug-disease relationships in these circumstances.

This type of confounding is addressed by and is not a problem in prospective randomized controlled trials.

Podiatry Relevant Example:
A relevant example was this publication that claimed the foot orthotics were a risk factor for a running injury. However, those with a foot injury are more likely to be wearing a foot orthotic as an intervention and also those with a foot orthotic are more likely to have had a previous injury which is also a risk factor, so was the risk due to the previous injury or the intervention (foot orthotics)? ie confounding by indication. Not withstanding that, the study allegedly did not actually find that anyway. See these two analyses:
Foot orthotics associated with increased injury risk in runners (Podiatry Arena)
Do foot orthotics increase the risk for running injury? (Running Research Junkie)

External Links:
How confounding by indication can indicate problems in observational research on the effectiveness of an intervention! (explanation of the concept using erectile dysfunction as the example)