I stumbled over it in E.D. Thalmann's paper where he introduces the linear-exponential kinetics that are kind of USN standard today (VVAL18, ...):
http://www.diverbelow.it/attachments/article/131/Thalmann et alii. Improved probabilistic decompression model risk prediction using linear-exponential kinetics.pdf
where he writes on p.286 "All models severely underpredicted the DCS incidence for Air and O2 Decompression and Surface Decompression dives. When O2 fractions near 100% are used, all models rapidly drop the tissue PN2 levels and risk rapidly dissipates. The O2 itself is assumed not to contribute to risk in these models."
Another paper from NMRI is: E.C. Parker et al: "Probabilistic models of the role of oxygen in human decompression sickness"
ARTICLES | Journal of Applied Physiology
There the say, previous models severely underpredicted DCS risk when 100% O2 was used for decompression or surface decompression. They had to include two effects to improve the models: (a) high pPO2 causes vasoconstriction and reduces off-gassing, (b) oxygen that's not bound to hemoglobin can bubble just like any inert gas.
This paper also has references to earlier research on oxygen bends.