Thread by @kareem_carr

Speaking as a black statistician, I don't think we can completely eliminate using race in medical decisions if we want to make the best decisions for each patient given our current state of technology. Gene testing for specific ancestry would be better but we aren't there yet.

Leaving out race can have a lot of unintended consequences. Algorithms might default to the standard of care for the largest group. This means treating everybody as if they are white which would be problematic in many cases.

Alternatively, algorithms may relearn race from the data. They will use family history, geography and other demographic variables to guess the race. This can be tricky to detect if you aren't looking for it. "Racist" algorithms get released to the public all the time.

I think the best solution is to use race to make decisions in medicine but not in a *racist* way. In other words, use it *only* when it's the best way to improve people's standard of care. This is hard because it requires something lots of people aren't good at. Nuance.

In general, I would suggest using race but following careful guidelines like:
1. Using better ways (like genetic testing) when available

2. Only using race for the benefit of the patient not the institution (for instance, providing less compensation to black patients based on racial assumptions isn't about helping them, it's about saving the institution money.)

3. Auditing decision making systems that use race to make sure that their decision-making quality isn't disproportionately bad for marginalized racial groups.

4. We must also be careful not to fall into the logical trap of assuming that making decisions based on race means the races are fundamentally different. It just means it's a piece of data that was (perhaps) helpful in that situation where we had limited knowledge.

Humans often make simplifying assumptions to make decision making easier, and this often works, but that doesn't mean these assumptions are the truth.