LOGISTIC REGRESSION AND DECOMPRESSION SICKNESS

Rubicon Research Repository/Manakin Repository

LOGISTIC REGRESSION AND DECOMPRESSION SICKNESS

Show full item record


Title: LOGISTIC REGRESSION AND DECOMPRESSION SICKNESS
Author: Southerland, DG
Abstract: Decompression sickness is a multi-faceted illness believed to be caused by the formation of gas bubbles in the tissues as a result of gas supersaturation. It may result in debilitating illness or death and so is an event to be avoided. This paper introduces some simple terminology and concepts of empirical statistical modeling to allow future application of these techniques to more complex, physiologically-based models to predict decompression sickness in multi-level and multiple inert gas diving. The primary focus is on binary logistic regression and how it might be used in an empirical model to predict decompression sickness. The method is applied in a crude meta-analysis on a pooled database of single stage compressed air dives from the literature. The working hypothesis for the analysis was that depth and time of dive are associated with the occurrence of decompression sickness in the dataset. A logistic model was created in which both depth and time were found to be statistically significant predictors of decompression sickness. Preliminary validation of the model was performed using bootstrap techniques for the variables chosen and the parameter estimates obtained. The model still requires validation using a new dataset and also limited usefulness because it is limited to single stage air dives under controlled conditions, which are relatively uncommon among divers in the field. Practical use of logistic regression for multi-stage diving will require some method to collapse complex depth-time profiles to fewer predictor variables.
Description: MS Thesis
URI: http://archive.rubicon-foundation.org/3433
Date: 1992

Files in this item

Files Size Format View
SoutherlandMS.pdf 283.5Mb PDF View/Open

This item appears in the following Collection(s)

Show full item record

Browse

My Account