Ziad Obermeyer

@oziadias

Physician + prof . Machine learning, health policy, emergencies, espresso.

Berkeley, CA
Vrijeme pridruživanja: ožujak 2013.

Tweetovi

Blokirali ste korisnika/cu @oziadias

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @oziadias

  1. Prikvačeni tweet
    31. lis 2019.

    We'd like to help fix the bias we found in algorithms used for population health. And we'd love help from a few friends to spread the word! 📢

    Poništi
  2. proslijedio/la je Tweet
    27. sij

    Last night in the ED, wondering which of three patients to pick up, I pondered whether they could be auto assigned, or their cc blinded. And the next morning, I am treated to an excellent related analysis by oziadias

    Poništi
  3. 26. sij

    And while she is Twitter-less, I should say that this work was led and carried out entirely by my fabulous colleague CJ Chang

    Prikaži ovu nit
    Poništi
  4. 26. sij

    Fact 4: These differences across docs are far less pronounced on the overnight shift. There's often only 1 doc working overnight—in which case, you don’t get to choose your patients.

    Prikaži ovu nit
    Poništi
  5. 26. sij

    But some docs like to see chest pain; other like to see toe pain. Those patients won’t cost the same, even if there were *no* practice variation across docs. Adjusting for that, the rankings change a lot. E.g., 77% of docs get reassigned to a new cost quintile.

    Prikaži ovu nit
    Poništi
  6. 26. sij

    Fact 3: A lot of what we consider “practice variation” across docs may be variation in *preferences*. Take cost variation as an example. Let’s first rank all the docs in our sample by their average cost per patient (in quintiles).

    Prikaži ovu nit
    Poništi
  7. 26. sij

    Fact 2: Docs’ preferences change over the course of a clinical shift. Towards the end of their shift, they prefer to take simpler, more straightforward cases.

    Prikaži ovu nit
    Poništi
  8. 26. sij

    Interlude: Emergency docs who let patients sit on the board without signing up, I’M LOOKING AT YOU. We all know who you are.

    Prikaži ovu nit
    Poništi
  9. 26. sij

    Fact 1: Some emergency docs like to see older/sicker patients, others like to see younger/simpler patients. This happens because there’s often >1 doc working in the ER. So when a patient pops up on the board, docs can choose to sign up... or wait for a colleague to sign up.

    Prikaži ovu nit
    Poništi
  10. 26. sij

    You can’t pick your doctor in the ER. But your doctor can pick you. Two implications: 1) A lot of "practice variation" may be *preference* variation 2) Don't assume "as-if-random assignment" to docs Thread on a new, short paper

    Prikaži ovu nit
    Poništi
  11. proslijedio/la je Tweet
    24. sij
    Poništi
  12. proslijedio/la je Tweet
    23. sij

    Cross sectional study finds emergency physicians prefer to see different types of patients, accounting for a substantial part of practice variation

    Poništi
  13. proslijedio/la je Tweet
    19. sij

    Prediction: getting mass spectrometry to the standardized, streamlined place where DNA sequencing is right now would have at least 10X the effect that sequencing has had. It is such a cool piece of technology with endless applications

    Prikaži ovu nit
    Poništi
  14. proslijedio/la je Tweet
    13. sij

    Shameful lack, non-compliance of reporting of clinical trials by academic medical centers by comparison with industry striking

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    16. sij

    Building on great work by , racial bias is one of many biases unintentionally baked into risk scores. how, why, and a way forward.

    Poništi
  16. proslijedio/la je Tweet
    14. sij

    Jenna Wiens, , & I have a short commentary just out discussing the terrific et al. piece on bias in medical AI. Using proxies to train AI can result in NASTY bias problems; getting labels right is key.

    Poništi
  17. proslijedio/la je Tweet
    14. sij

    Where are all the complaints of headaches and nausea?? Chinese Restaurant Syndrome my ass.

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    14. sij
    Poništi
  19. proslijedio/la je Tweet
    12. sij

    Interesting new JEBO paper by Lin and exploits variation in sleep due to daylight savings time, and finds that extra sleep reduces hospitalizations by ~8% for up to a week afterwards.

    Prikaži ovu nit
    Poništi
  20. 11. sij

    Even better idea: Be the kind of patient who gets your surgery scheduled for a Monday.

    Poništi
  21. proslijedio/la je Tweet

    You: This one untested supplement will restore balance to my body Your body:

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·