Even 14-days post-293 enrollees, a Cox model would use all the data. It's the standard for "time to event" analyses - here the event is death.

A Cox model with only one observation date simplifies to a hazard ratio test.

Some of last 25% of pats contribute only to the 14 day data, some to 14 and 28 days, and a few to all three. The model refers to the too-early-to-collect data as "censored."

Basic calc:

P(alive @ 42)=
(# alive @ 14 / # started trial & have 14 day data) *
(# alive @ 28 / # alive @ 14 & have 28 day data) *
(# alive @ 42 / # alive @ 28 & have 42 day data)