This is a well-researched and well-written article. I have three comments.

1. If you are interested, please check out my pub Brain Labs (https://medium.com/brain-labs). This is just the sort of article I love to publish.

2. I have some relevant experience. I worked on a SaaS web application targeted at physicians to better interpret test results. One of the company founders is an ER doctor who noticed how often physicians misinterpret test results. The problem is that overwhelmed and undertrained doctors have a lot of trouble factoring the significance of false negatives and false positives into their test results, and as a result may misdiagnose. Most of the doctors we showed the app to agreed this is a problem.

3. As mentioned, you did a lot of thorough research that shows a lack of rigor in coming up with the number 800,000. That said, I think you undercut your own push for rigor in your conclusion by writing "the true number of people who are killed or disabled by misdiagnosis in the United States is likely to be orders of magnitude lower than 800,000 a year." I'll conservatively two orders of magnitude since you used a plural. That would be 8,000. That's a huge difference from 800,000. Without corroborating data, it feels like a dart throw. If throwing darts is legitimate, I would venture that 8,000 feels way way too low to me.