Can someone explain to me - assuming they have enough data - why not train different models explicitly for each group / subgroup you want to model? You even could then just take the top N% of each group by score, effectively guaranteeing equal treatment for each group. Why would this not work?
Regeneron is among the most science and data focused biopharma and has a long, long history in genetics. They know how critical privacy is and will ensure the data are used to advance human health.
Potentially an even better home for the data than in the company, since now will not have pressure of quarterly reports.
Founders can make a "competitive" salary, if they have existing experience for the actual job of being a founder (i.e., previous executive leadership experience).
For many, being a founder is not just "taking a risk on an unproven idea" it's also a career change from being a solo IC or technical leader to being an executive, with very, very different requirements for success, so the risk here is compounded.
Said in the George Senior voice: And thats why you don’t use a non-profit to do world critical work: politics will always beat true value at a non-profit.
1. Most of the data cited stops around 2010, ignoring the huge progress that checkpoint inhibitors and cell therapies have made in many cancers.
2. Mixed up issues of treatment progress and treatment access. Treatment progress is a fact, but can be missed if patients don’t have access.
Access in the US is essentially a political issue - states with Medicaid expansion, for example, have better overall access than those that don’t. Also, in general, changing Medical practice is slow, even once a therapy is approved.
The narrative that we have only poor treatment options that don’t advance misses the true progress that has been made, and let’s those responsible, ie insurance companies and politicians off the hook.