Decompression sickness predictive models for unsafe human exposure.

Decompression sickness (DCS) incidence prediction models have achieved useful predictive success under conditions of routine Navy diving. However, extrapolation into higher-risk exposures, e.g., emergency conditions, has been a problem. We have assembled a calibration data set of 3,300 single exposures with 200 DCS cases emphasizing high-incidence data from the U.S. Navy compilation of manned diving trials. We also evaluated a variant of the older linear-exponential risk model family where the instantaneous risk is defined as the relative supersaturation squared. Goodness of fit was assessed by maximum likelihood, by comparison of categories of observed and predicted cases in three ways (component data set, depth-time group, and risk level), and by reproduction of data dose-response trends. Four models fit the data well. Two had the old risk definition, and two had the new. With each risk definition, a satisfactory set of parameters was found differing mainly in treatment of gas kinetics in the fastest compartment. Multimodel inferences were made with a combination of the four models weighted using the Akaike Information Criterion. The combined model is recommended for use in emergency preparations where compressed-air exposures may lead to a 40% or higher incidence of DCS.