Page 108 - Contributed Paper Session (CPS) - Volume 5

CPS1113 Madhu Mazumdar et al.
with the treatment/intervention of interest but not the outcome. The concept
is presented in Figure 1, where IV represents the IV, T the treatment, U the
measured and unmeasured confounders, and Y the outcome. The principal
(and most challenging) task is to choose an IV that fulfills these assumptions:
1. IV is positively associated with T
2. IV is independent of U

IV is associated with Y only through T
Here, there are two paths from T to Y:
1. T→Y (the direct treatment effect on outcome)
2. T←U→Y (the effect contributed by confounders).

The second path indicates the potential bias involved in estimating a
treatment effect when confounding is present. For example, if we adjust for U
in a regression model, the T←U→Y path is eliminated; in this case, we can
obtain an unbiased estimate of T on Y. However, in observational studies, U
may not be available. An IV is independent of Y and the effect of IV on Y can
only be estimated through the effect of IV on T; therefore, the effect of IV on
Y represents the effect of T on Y. Indeed, as IV should also be independent of
U, there is only one viable path from IV to Y: IV→T→Y.
Conceptually, an IV analysis resembles the so-called ‘intention-to-treat’
approach, where patients randomized to a treatment group are analyzed
regardless of actual treatment received.
Types of Instrumental Variables: Although several types of IVs exist,
among those most applicable to observational studies are the following:
▪ Distance to (specialty) providers is a commonly used IV, since it may be

highly related to the chance of getting a specific treatment. In an observational
study on the type of anesthesia used in hip fracture repair surgery, the IV was
patients’ distance to hospitals that either specialize in regional anesthesia (RA)
6
or general anesthesia . Someone who lives closer to a hospital with a high RA
utilization rate was assumed to be more likely to receive RA, compared to
someone who lives closer to a hospital where RA is avoided or not offered.
▪ Preference-based IVs are suitable if treatment patterns (e.g. preferences
for surgical approaches or interventions) are recognizable at the hospital or
physician level. One observational study chose as their IV the surgeons’
preference for managing the posterior cruciate ligament in total knee
arthroplasty (either minimally or posterior-stabilized) to investigate effects on
7
time to revision surgery . Patients treated by surgeons with a specific preferred
surgical approach were assumed to be more likely to have that procedure.

97 | I S I W S C 2 0 1 9