B1838
Title: Analysis of ordinal longitudinal data under case-control sampling: Studying mortality in critically ill patients
Authors: Chiara Di Gravio - Imperial College London (United Kingdom) [presenting]
Ran Tao - Vanderbilt University Medical Center (United States)
Jonathan Schildcrout - Vanderbilt University (United States)
Abstract: The CLOVERS trial compared the effect of two fluid resuscitation strategies on mortality in patients with sepsis. At recruitment, blood samples of each participant were collected and stored, and, over the course of 14 days, information on their daily health status was recorded. In particular, on any given day, participants could have been in one of four ordered outcome states: discharge, hospital, intensive care unit, or death. It is common in clinical trials, such as CLOVERS, to store blood samples at recruitment and analyze them at a later date to retrospectively obtain information on a new exposure. However, high costs reduce the number of samples that researchers can analyze. The experience in setting up a secondary analysis of the CLOVERS trial is discussed where interest was in the association between glycocalyx degradation and mortality, but glycocalyx degradation was not collected in the main trial and, due to budget constraints, could be collected for a third of the patients originally enrolled. First, efficient study designs are discussed that can be implemented in scenarios where an ordinal longitudinal outcome and a set of covariates are available to everyone, but information on a novel exposure needs to be collected. Then, a semiparametric likelihood approach is presented to estimate the parameters. Finally, the finite sampling operating characteristics of the proposed approach are examined and results from the CLOVERS trial are presented.
