Page 107 - Contributed Paper Session (CPS) - Volume 5

CPS1113 Madhu Mazumdar et al.

Instrumental variable analysis in healthcare

delivery science: Underutilized yet valuable
Madhu Mazumdar, Hsin-Hui (Vivien) Huang, Xiaobo Zhong, Jashvant Poeran
Institute for Healthcare Delivery Science, Department of Population Health Science & Policy,
Icahn School of Medicine at Mount Sinai, New York, NY

Abstract
Need for adjustment of confounders is necessary in answering queries of
comparative and association effect of treatment or policies with patient or
healthcare utilization outcomes. Various methods exist for confounder
adjustments. Three primary methods are 1) regression-based adjustment, 2)
propensity score-based adjustment, and 3) Instrumental Variable (IV) analysis.
Although conceptually superior due to the fact that only IV analysis adjust for
unmeasured confounders, it has remained underutilized in healthcare delivery
science research. Reason for underutilization include the fact that ‘instruments’
are difficult to formulate needing strict assumptions and the wrong perception
that the analysis is more complex than the other two methods. However, in
the era of easy availability of administrative claims-based databases and open
data sharing of national registries, formulation of IV has become easier. We
use the clinical question of whether there is increased risk for blood
transfusion after closed wound drainage are used for patients who have
undergone total shoulder arthroplasty to introduce the reader to an useful
administrative database (Premier Healthcare), compare the three statistical
methods discussed above, and provide codes and guidance for easy
implementation of IV analysis.

Keywords
Instrumental Variable; Administrative claims-based databases; Unmeasured
confounders; Closed wound drainage; Shoulder arthroplasty

1. Background
4
Unlike randomized clinical trials (RCTs) , observational studies must
acknowledge confounding; this can be addressed by multivariable approaches
2
such as regression modeling1 and propensity score analyses . These, however,
can only address known confounding factors, not unobserved confounders. In
contrast, IV analysis does address both known and unknown confounders, a
major advantage.

Instrumental Variable Analysis
3 5
The basic principle behind IV analysis , is choosing an IV to represent a
mechanism for assigning treatment to patients. It should be closely associated

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