Page 111 - Contributed Paper Session (CPS) - Volume 5

CPS1113 Madhu Mazumdar et al.
or registries) often lack information on important confounders. This gap may
be addressed by IV analysis. Here, IV analysis provided results comparable to
multivariable regression and propensity score analysis. However, since the IV
method also addresses unmeasured confounders, it is conceptually superior.
IV analysis is not commonly used in healthcare delivery research.
Researchers may believe that identifying an IV that fulfills all three
assumptions is difficult, or that the method is overly complicated. Both
obstacles can be overcome by early collaboration with statistical collaborators.
Moreover, statistician-collaborators can assist in examining IV assumptions via
formal statistical tests and sensitivity analyses, discussing each step of the
analysis with the study team. Indeed, our 2-stage analysis is no more complex
than propensity score analyses. In conclusion, IV analysis may complement
other analytic approaches for observational studies and thereby increase the
overall value of such studies.
For this retrospective cohort study data from the Premier Healthcare
Database10 (Premier Healthcare Solutions, Inc., Charlotte, NC) was used. This
database contains administrative claims data on approximately 20-25% of US
hospital discharges. Records include International Classification of Disease-9th
revision (ICD-9) codes, Current Procedural Terminology (CPT) codes, and
complete inpatient billing items. Preparing analytic dataset with patient as unit
using these codes is a valid way to perform healthcare delivery research
projects.

Table 1. Baseline characteristics of patients in the Instrumental Variable-
derived cohort
Drain Use
Yes (n=21,218) No (n= 83,898)
n % n % Standardized
difference**
PATIENT DEMOGRAPHICS
Median Age* 70 (63, 76) 70 (63, 76) 0.0071
Gender 0.0056
Male 9,075 42.77 35,650 42.49
Female 12,143 57.23 48,248 57.51
Race / Ethnicity 0.0554
White 17,426 82.13 70,469 83.99
Black 975 4.60 3,579 4.27
Hispanic 89 0.42 432 0.51
Other 2,728 12.86 9,418 11.23
HEALTHCARE-RELATED
Insurance Type 0.0435
Commercial 5,022 23.67 19,514 23.26
Medicaid 449 2.12 2,039 2.43
Medicare 14,554 68.59 58,259 69.44
Uninsured 118 0.56 331 0.39
Unknown 1,075 5.07 3,755 4.48

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