Using a data base of 2,383 air and nitrogen-oxygen dives resulting in
131 cases of decompression sickness (DCS), risk functions were develop
ed for a set of probabilistic decompression models according to surviv
al analysis techniques. Parameters were optimized using the method of
maximum likelihood. Gas kinetics were either traditional exponential u
ptake and elimination, or an exponential uptake followed by linear eli
mination (LE kinetics) when calculated supersaturation was excessive.
Risk functions either used the calculated relative gas supersaturation
directly, or a delayed risk using a time integral of prior supersatur
ation. The most successful model (considering both incidence and time
of onset of DCS) used supersaturation risk, and LE kinetics (in only 1
of 3 parallel compartments). Several methods of explicitly incorporat
ing metabolic gases in physiologically plausible functions were usuall
y found in lumped threshold terms and did not explicitly affect the ov
erall data fit. The role of physiologic fidelity vs. empirical data fi
tting ability in accounting for model success is discussed.