Latent Class Model for Space Decompression Sickness Diagnosis Accuracy

• Activity Number: 353
• Type: Topic Contributed
• Date/Time: Wednesday, August 10, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Hsi-Guang Sung*+ and Alan H. Feiveson and Johnny Conkin
• Companies: USRA and Johnson Space Center and National Space Biomedical Research Institute
• Keywords: latent class model ; mixture of binomial distributions ; inter-rater agreement ; bends ; astronauts ; space medicine

Abstract

When astronauts perform work outside space vehicles, they risk decompression sickness (DCS) due to lower air pressure inside spacesuits as compared with the normal atmospheric pressure maintained in spacecraft. At the NASA Johnson Space Center, ground studies of breathing and exercise protocols are made in decompression chambers to develop a fast, yet reliable, way of preventing DCS in space. Due to concern for the safety of human test subjects, physicians often diagnose DCS and terminate tests when subjects complain of mild discomfort, even though actual DCS may not exist. In an attempt to estimate this conservative-driven bias, records of sensor data and subjective assessments of pain for 135 chamber tests were supplied to 15 experts in environmental physiology. The experts then studied the reports in a detached setting (i.e., their offices) and attempted to make an unbiased determination of whether actual DCS was present. These offline diagnoses were compared to actual test results, however no "gold standard" exists to verify true DCS. Nevertheless, we were able to use a latent class model to estimate the false positive rate for each of four symptomatic covariate patterns.