Not really. I could spend time and energy building filters that matched current known patterns, but those filters probably won't be very good. Suppose instead of a stroke it had been a gun accident. I'm not confident I could successfully anticipate and build a filter good enough to recognize that it wasn't a political post about the NRA and gun rights.

Ultimately, that line of thought ends with me developing my own algorithmic feed, which I don't really want to do and isn't scalable anyway.