The Effect of Multiple Imputation of Routine Pathology Variables on Laboratory Diagnosis of Hepatitis C Infection
Date
2022-05-04
Authors
Menon, N.
Lidbury, B. A.
Richardson, Alice
Journal Title
Journal ISSN
Volume Title
Publisher
arXiv
Abstract
Pathology tests are central to modern healthcare in terms of diagnosis and
patient management. Aggregated pathology results provide opportunities for
research into fundamental and applied questions in health and medicine, but
data analytic challenges appear since test profiles vary between medical
practitioners, resulting in missing data. In this study we provide an
analytical investigation of the laboratory diagnosis of Hepatitis C (HCV)
infection and focus on how to maximize the predictive value of routine
pathology data. We recommend using the Influx - Outflux measures to help
construct the imputation model when using multiple imputation.
Data from 14,320 community-patients aged 15 - 100 years were accessed via ACT
Pathology (The Canberra Hospital, Australia). Influx and Outflux were
calculated to identify which variables were potentially powerful predictors of
missing values. Available Case analysis and Multiple Imputation were used to
accommodate missing values in the dataset. Logistic regression model and
stepwise selection method were used for analysing the imputed datasets. The
predictive power of all methods was compared.
The predictive power of the models on multiply imputed data was similar to
the power of the models based on complete data. The advantage of multiply
imputed data was that it allowed for the inclusion of all the completed
variables in the logistic models, thus identifying a broader selection of test
results that could lead to the enhanced laboratory prediction of HCV.
Multiple imputation is an important statistical resource allowing all
individuals in a study to contribute whatever data they have supplied to the
analysis. MI in combination with the values of Influx and Outflux identifies
potential predictors of HepC infection. Variables age, gender and alanine
aminotransferase have been shown to be strong laboratory predictors of HCV
infection.
Description
Keywords
Hepatitis, logistic regression, multiple imputation
Citation
Collections
Source
Type
Manuscript
Book Title
Entity type
Access Statement
Open Access
License Rights
Creative Commons Attribution 4.0 International License
Restricted until
Downloads
File
Description