A ternary model of decompression sickness in rats

• This study developed a ternary model of predicting probability of decompression sickness in rats.
• A dataset was compiled from 15 studies using 22 dive profiles and two strains of both sexes.
• Using ordinal logistic regression, model-fit was optimised by maximum log likelihood.
• The model predicted cases of DCS with 75% accuracy and 92.5% were within 95% confidence intervals.
• This model is reliable for predicting DCS within the range of parameters used to optimise the model.

BackgroundDecompression sickness (DCS) in rats is commonly modelled as a binary outcome. The present study aimed to develop a ternary model of predicting probability of DCS in rats, (as no-DCS, survivable-DCS or death), based upon the compression/decompression profile and physiological characteristics of each rat.MethodsA literature search identified dive profiles with outcomes no-DCS, survivable-DCS or death by DCS. Inclusion criteria were that at least one rat was represented in each DCS status, not treated with drugs or simulated ascent to altitude, that strain, sex, breathing gases and compression/decompression profile were described and that weight was reported. A dataset was compiled (n=1602 rats) from 15 studies using 22 dive profiles and two strains of both sexes. Inert gas pressures in five compartments were estimated. Using ordinal logistic regression, model-fit of the calibration dataset was optimised by maximum log likelihood. Two validation datasets assessed model robustness.ResultsIn the interpolation dataset the model predicted 10/15 cases of nDCS, 3/3 sDCS and 2/2 dDCS, totalling 15/20 (75% accuracy) and 18.5/20 (92.5%) were within 95% confidence intervals. Mean weight in the extrapolation dataset was more than 2 SD outside of the calibration dataset and the probability of each outcome was not predictable.DiscussionThis model is reliable for the prediction of DCS status providing the dive profile and rat characteristics are within the range of parameters used to optimise the model. The addition of data with a wider range of parameters should improve the applicability of the model.