The problem with instrumental variables

When using real-world data, researchers must always deal with a key issue: selection bias.  To get around this bias, many health care researchers use an instrumental variable that can predict the explanatory variable of interest (e.g., receipt of a specific treatment) but is not correlated with patient outcomes (e.g., mortality). A commonly used IV is patient distance from a provider; distance from a provider would certainly affect receipt of specific services that provider specializes in, but it may not be correlated with outcomes.

An interesting editorial by Soumerai and Koppel (2016), note the following, however:

Like any cross-sectional analysis, IV analysis relies on the absence of any unmeasured patient and health system confounders (e.g., socioeconomic status, health status, and other lifesaving treatments, such as medications) that may provide an alternative explanation for the relationship between the IV and the patients survival. This assumption is the Achilles heel of IV studies. Most administrative data lack important variables correlated with survival (e.g., urban/rural status, or receipt of other lifesaving treatments), or they measure them poorly (e.g., race), representing a violation of the IV assumptions.

Sure there are problems with IV, but are there better alternatives?  The authors hold up the Angrist, Chen, and Frandsen (2010) paper as a prime example of high-quality IVs.  In this paper, the Vietnam draft lottery was used as a source of real-world randomization.

The authors also claim that IV is a cross-sectional technique and thus is subject to bias due to the potential for pre-existing trends.  Instead, the authors recommend using a formal interrupted time series design, as they claim that it is more resistant to bias.




Leave a Reply

Your email address will not be published. Required fields are marked *