How to Train Your Models for Safety

We consciously omit race and ethnicity from our model so that they're hopefully not biased. We remove things like names and anything else that might add skew in there. It's also just not statistically significant. You should always update your models because if you trained it two years ago and then there's a pandemic, new terminology, new ICD and CPT codes need to go into the model. So you actually want to retrain things every few months with the fresher data that you have. And that's the only thing that makes me feel comfortable as a CEO and as the head of product that this is safe.