A Latent Class Analysis to Identify Subgroups of Heart Failure Under Missingness And/Or Uncertainty in the Indicator Variables

• Activity Number: 536 - Contributed Poster Presentations: Section on Statistics in Imaging
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Wendimagegn Alemayehu* and Cynthia M Westerhout and Jason R Dyck and Todd Anderson and Justin A Ezekowitz
• Companies: University of Alberta and University of Alberta and University of Alberta and University of Calgary and University of Alberta
• Keywords: latent class analysis; missing data; measurement error ; propensity score; multiple imputation

Abstract

A latent class analysis (LCA) method was applied to identify diagnosis subgroups of heart failure patients sharing similar clinical profiles characterized by baseline characteristics, sign and symptoms at presentation, biomarkers and imaging markers. The imaging data were generated from two different diagnostic tests namely the echocardiography and the cardiovascular magnetic resonance (CMR). The CMR is generally assumed to provide more accurate measurement of cardiac functions and volumes compared to the echocardiography. However, in our data, the CMR test was not performed for all patients and thus was missing for a large number (up to 50% in some groups) of patients. Further, the missing data process was dependent of the underlying patient group and possibly of the unobserved missing values. We apply and evaluate a number of alternative LCA models to address the problems of imprecise and/or incomplete measures of the indicator variables. We evaluate the performance of the methods using the sensitivity and specificity to correctly classify patients in reference to diagnoses groups that were centrally adjudicated by expert clinicians.