Tetranomial decompression sickness model using serious, mild, marginal, and non-event outcomes

Abstract

Abstract Decompression sickness (DCS) is a condition resulting from reductions in ambient pressure, causing inert gas bubbles in tissues. This work focuses on hyperbaric exposures, specifically DCS resulting from underwater diving. Signs and symptoms of DCS can range from mild skin rashes and joint pain to serious neurological and cardiological malfunction, and even death. Marginal DCS is defined as symptoms associated with DCS that resolve spontaneously without recompression treatment. There are two categories of decompression modeling used to mitigate risk of DCS: deterministic and probabilistic; neither address DCS symptom severity. Symptom severity is important to U.S. Navy dive planning, as the Navy has different limits for the number allowable cases of mild-symptom DCS and more severe-symptom DCS for a given dive. In this work, a probabilistic model for predicting the tetranomial outcomes of serious, mild, marginal, and no DCS was developed, analyzed, and compared with trinomial and trinomial marginal models from our previous works. Six variants of exponential-exponential (EE1) and linear-exponential (LE1) models were calibrated with 3322 air and N2–O2 dive exposures detailed in the BIG292 empirical human dive trial data set. Two methods of symptom severity splitting were compared. The log likelihood difference test indicated the LE1 model using a previously-disclosed Type A/B splitting provided the best fit to the empirical dive data of all tetranomial models tested in this work. When comparing this tetranomial model to our previous trinomial and trinomial marginal models using the Pearson chi-squared statistic, we find that the tetranomial and trinomial marginal models’ predictions of marginal DCS are not aligned well with the incidence of marginal DCS in the data. Both the trinomial marginal model in our previous work and tetranomial model presented here are unable to accurately replicate the occurrence of marginal DCS events observed in the BIG292 dataset. These marginal DCS events may hinder model fit during calibration. We recommend the use of the trinomial model from our previous work, which simultaneously predicts the probability of mild, serious, and no DCS.