In essence it does not make any difference to performance, there will be some, but it will not be significant. And it depends on the load.
The first one may have a minimal performance problem because the index trees may be deeper and may have 1 or 2 more levels to analyze before arriving at the desired item, but this is usually paltry.
The second tends for more information in the memory since everything is separated and not used often, so the cache starts working against and not in favor, and there will be occasional losses.
Which is better you will not know in advance or know your toric scenario, only the actual use will indicate what works best, and may change according to a change of pattern, until tests can fail because it involves a lot of variable, but you will hardly see big difference.
Do according to the real need, according to requirements, and worry less about the architecture, just leave it easy to change if one day needs it, but it will not be because of the performance. If you have performance problems it is because you have many other things that are much more wrong, for example not having an index where you need it, or searching unnecessarily.
All doctors doing CRUD will not tingle because it will be 99.99% of the time with the screen stopped without any operation in the database. And what would all the doctors be? All of Brazil at the same time? Then it may be a problem and it will require heavy engineering. All from a clinic? It has almost zero impact. Is it something important for a hospital? Better hire an experienced engineer to take care of this.
