8 ms·

Google will succeed at healthcare only after medical decisions can be reduced to A/B tests. That is to say, never. After wasting many millions on failures lik

Google will succeed at healthcare only after medical decisions can be reduced to A/B tests. That is to say, never. After wasting many millions on failures like Calico, Google hasn't learned that healthcare is a service that's continuous, custom, and complex. And since the "customer's" health and life are at stake, maintaining health is a damned sight more important and difficult to do sufficiently well than Google's current business: spam. Tech companies like Google simply don't have the DNA or the persistence to succeed in science-based markets driven by R&D and heavy regulation. They want faster profits driven by higher multipliers and lower standards for accuracy than you can get in healthcare, where accuracy really counts and every case is a little bit different. In medicine it really matters if you get the right answer for your A/B test, otherwise someone may die or they may sue you. That's why so little of healthcare has been automated, and why that's likely to continue.

dekhn 4y ago
Why do you think Calico is a failure? I think the jury is still out on that one.
- randcraw 4y ago
  I have a little experience trying to work with Google in the R&D space at a large pharma. Google and other external R&D agents consistently underestimated the difficulty in getting access to sufficient data from us to employ large scale ML. That data often exists within a pharma's clinical trials, but it's acquired and stored in a way that makes it inaccessible to subsequent mining. This is a problem for dataminers like me within pharmas too. Most of our clinical trial data is locked up contractually within contract research organizations (CROs) who are responsible for taking specific measurements of it as part of a trial. From what I've seen, dataminers within the company can gain access to only selective bits of that data (one or two features, perhaps) from all the patients in a trial, but never to all the data on each patient, and never data from the trial that spans multiple CROs. This 'federation' of clinical trial data collection serves the needs of the trial design (and satisfies HIPPA regs), but it makes mining the data or asking new questions of it almost impossible. Too little data is available (and never longitudinal longer-term outcomes, since trial durations are short) to allow mining of more than short-term signals, like immediate response to treatment. Pharmas are unlikely to change that practice, in part because it's easier to federate the analyses and allow the raw data to remain in distributed silos. But federation is also a good way to discourage asking creative yet inessential questions that might later prove awkward or invite lawsuits, since litigation is an omnipresent worry in US healthcare. How companies like Calico/Google hope to navigate such minefields is a question they have yet to answer successfully.
  - dekhn 4y ago
    You didn't answer the question of why Calico is a failure. They have very little to do with Google. Calico is its own independent institute that does exactly what Art and Larry wanted when they set it up- produce world-class research into longevity. I think you mean Verily, which works closely with Google on health care projects. (note: I worked at google over a decade and work for pharma now. I helped establish Google's move into health care and even launched products intended for pharma users and set up the ethical guidelines to work with private data like clinical records. Unfortunately, Google's Research leadership chose to take their health efforts in a completely different direction so now I focus on stuff that will actually make a difference for pharma).
mateo1 4y ago
>Google will succeed at healthcare only after medical decisions can be reduced to A/B tests. That is to say, never. Never, or literally tomorrow if they take over. "Personalised medicine" decisions made with "AI" will provide a perfectly unaccountable black box to blame when (if ever) the general public realises that vulnerable and poor people have been getting the B end of the A/B testing. I agree with the rest of the post but god have mercy on us if these companies take over healthcare. I never thought I'd be living in such a totalitarian state in my life.

dekhn 4y ago

Why do you think Calico is a failure? I think the jury is still out on that one.

randcraw 4y ago
I have a little experience trying to work with Google in the R&D space at a large pharma. Google and other external R&D agents consistently underestimated the difficulty in getting access to sufficient data from us to employ large scale ML. That data often exists within a pharma's clinical trials, but it's acquired and stored in a way that makes it inaccessible to subsequent mining. This is a problem for dataminers like me within pharmas too. Most of our clinical trial data is locked up contractually within contract research organizations (CROs) who are responsible for taking specific measurements of it as part of a trial. From what I've seen, dataminers within the company can gain access to only selective bits of that data (one or two features, perhaps) from all the patients in a trial, but never to all the data on each patient, and never data from the trial that spans multiple CROs. This 'federation' of clinical trial data collection serves the needs of the trial design (and satisfies HIPPA regs), but it makes mining the data or asking new questions of it almost impossible. Too little data is available (and never longitudinal longer-term outcomes, since trial durations are short) to allow mining of more than short-term signals, like immediate response to treatment. Pharmas are unlikely to change that practice, in part because it's easier to federate the analyses and allow the raw data to remain in distributed silos. But federation is also a good way to discourage asking creative yet inessential questions that might later prove awkward or invite lawsuits, since litigation is an omnipresent worry in US healthcare. How companies like Calico/Google hope to navigate such minefields is a question they have yet to answer successfully.
- dekhn 4y ago
  You didn't answer the question of why Calico is a failure. They have very little to do with Google. Calico is its own independent institute that does exactly what Art and Larry wanted when they set it up- produce world-class research into longevity. I think you mean Verily, which works closely with Google on health care projects. (note: I worked at google over a decade and work for pharma now. I helped establish Google's move into health care and even launched products intended for pharma users and set up the ethical guidelines to work with private data like clinical records. Unfortunately, Google's Research leadership chose to take their health efforts in a completely different direction so now I focus on stuff that will actually make a difference for pharma).

mateo1 4y ago

>Google will succeed at healthcare only after medical decisions can be reduced to A/B tests. That is to say, never. Never, or literally tomorrow if they take over. "Personalised medicine" decisions made with "AI" will provide a perfectly unaccountable black box to blame when (if ever) the general public realises that vulnerable and poor people have been getting the B end of the A/B testing. I agree with the rest of the post but god have mercy on us if these companies take over healthcare. I never thought I'd be living in such a totalitarian state in my life.