Endpoint changes based solely on increasing event counts, which therefore shouldn't introduce bias, are roughly the same as an increase in N.

Example: changing PE from "alive" to "alive and not intubated," if recommended only using unblinded event counts data, should be OK based on my reading.