For fatigue or steady load?
I have found that under steady load the keyway doesn't make as much difference as you would expect.
And for fatigue it is a lot worse.

= = = = = = = = = = = = = = = = = = = =
P.E. Metallurgy, Plymouth Tube

Although withdrawn, ASME B106.1M provides a good methodology - still endorsed by CEMA for pulley shaft sizing.

The case is actually overtorsion, so no fatigue was present. I'm looking at calculating what torque was required to break the shaft, and didn't know whether the keyway would decrease the allowable torque slightly...

This is more just for my interest, I'm not actually doing an official report or anything...

Keys are a miserable method for transferring torque between a shaft and something like a pulley/coupling/etc. A key is suitable for indexing the shaft and a mating component at assembly. But it is much better to use some type of connection that provides sufficient clamped friction to prevent any relative sliding at the contact interface. Even a tiny amount of relative motion at the contact interface will usually produce fretting.

There are excellent textbooks on such design whether the shaft is either under static or cyclic loads. One reference is " Design of Machine Elements" authored by Spotts which has several examples and problems. Mechanical Engineering handbooks such as Kent or Mark have also copious information on the subject matter. Libraries tend to have theses references.

In the cases that I worked with the load was not transferred via the keyway, but there was a full length keyway for locating other components.
We torsion tested all of the time, and found that a 1.375" shaft with a shallow (0.090" deep x 0.187" wide) was basically a solid shaft.

= = = = = = = = = = = = = = = = = = = =
P.E. Metallurgy, Plymouth Tube

This depends on the design. Is the flange shrink-fitted? in that case the key is only a safety (or should be only a safety) and not carry load. Even so, a key is a risk with regards to fatigue, not mush so with regards to overload.