A Network-Based Algorithm for Clustering Multivariate Repeated Measures DataThe National Aeronautics and Space Administration (NASA) Astronaut Corps is a unique occupational cohort for which vast amounts of measures data have been collected repeatedly in research or operational studies pre-, in-, and post-flight, as well as during multiple clinical care visits. In exploratory analyses aimed at generating hypotheses regarding physiological changes associated with spaceflight exposure, such as impaired vision, it is of interest to identify anomalies and trends across these expansive datasets. Multivariate clustering algorithms for repeated measures data may help parse the data to identify homogeneous groups of astronauts that have higher risks for a particular physiological change. However, available clustering methods may not be able to accommodate the complex data structures found in NASA data, since the methods often rely on strict model assumptions, require equally-spaced and balanced assessment times, cannot accommodate missing data or differing time scales across variables, and cannot process continuous and discrete data simultaneously. To fill this gap, we propose a network-based, multivariate clustering algorithm for repeated measures data that can be tailored to fit various research settings. Using simulated data, we demonstrate how our method can be used to identify patterns in complex data structures found in practice.
Document ID
20170006954
Acquisition Source
Johnson Space Center
Document Type
Presentation
Authors
Koslovsky, Matthew (Wyle Labs., Inc. Houston, TX, United States)
Arellano, John (Wyle Labs., Inc. Houston, TX, United States)
Schaefer, Caroline (Wyle Labs., Inc. Houston, TX, United States)
Feiveson, Alan (NASA Johnson Space Center Houston, TX, United States)
Young, Millennia (NASA Johnson Space Center Houston, TX, United States)
Lee, Stuart (Wyle Labs., Inc. Houston, TX, United States)