I can only imagine the study that would need to be done to confirm this "90% accuracy" claim. In fact, I have to imagine it because I can pretty much guarantee that it hasn’t been done. https://twitter.com/TheNBACentral/status/1273667511788109829

Briefly, you’d have to track thousands of people’s metrics, using the data to make a prediction - “yes, covid-19”, or “no, healthy”, and then assess if you’re right over time. Then you’d assign them to batches of false negatives, false positives, true negatives, true positives,

True positive - Oura ring correctly predicts C-19.
True negative - Oura correctly predicts healthy
False negative - Oura says healthy, you get C-19
False positive - Oura says C-19, you stay healthy
Then calculate accuracy as (TP + TN)/(TP + FP + TN + FN).
Was this done? No chance

Here’s the ‘sleight of hand’ though - you just “guess” that most people will be negative based on known low prevalence of covid-19 (for example, 1%), and you’ll be right almost all the time. So your test has shitty PPV & sensitivity, but it’s “accurate”, so someone will buy it!