Pulmonary Critical Care, data-science, machine-learning, and AI in critical care, health services research, University of Michigan
Joined August 2013
- Tweets468
- Following74
- Followers1.1K
- Likes184
Michael Sjoding retweeted
.@AnnalsATS study making national news: “Latino patients with respiratory illnesses are 5 times more likely to be oversedated”
nbcnews.com/news/latino/lati…
Michael Sjoding retweeted
A recent @JAMA_current study co-authored by MIDAS affiliate faculty member @umichmedicine's Dr. Michael Sjoding cautions the use of AI tools with baked-in systemic bias to help diagnose patients. Read more from @axios. axios.com/2023/12/20/ai-bias…
JAMA produced a video on our just published study!
Although standard AI models improve diagnostic accuracy, systematically biased AI models reduced diagnostic accuracy, and commonly used image-based AI model explanations did not mitigate this harmful effect.
ja.ma/3ROAuPf
Michael Sjoding retweeted
In #AI for medicine, we often hear, "Oh, this tool just assists clinicians - they decide care"
In @JAMA_current, @SarahJabbour_ et al's randomized study finds potential harm from decision support w assistive AI
We cover "Automation Bias" in an editorial
jamanetwork.com/journals/jam…
Michael Sjoding retweeted
Of 4 collaboration strategies to deploy #AI for the diagnosis of ARDS from chest XR, the most accurate is to allow AI review the XR first & defer to the physician if uncertain. 79% of cases were decided by AI - significantly reducing physician workload.
nature.com/articles/s41746-0…
On this day when a mentee has a great paper accepted, I'm grateful to have learned how to generously and effectively respond to manuscript reviewers from @ColinRCooke @iwashyna and now can model this with my mentees
Michael Sjoding retweeted
I've been accused of "relentlessly" covering the pulse oximeter issue. Guilty as charged. I find this issue critical. It's a microcosm of the painful, historical, scientific & nuanced ways race interests with medicine. Here's a🧵of our coverage. statnews.com/2022/11/01/puls…
Michael Sjoding retweeted
Ahead of an 11/1 @US_FDA mtg on pulse oximeters, #UMPrecisionHealth members @tsvalley, @iwashyna, @msjoding authored a piece in @ATSBlueEditor proposing actions that FDA, senators, clinicians, manufacturers can take to ensure they work accurately for all linkedin.com/feed/update/urn…
Proposal #4: Leverage widely available data to ensure real world safety in relevant subgroups
- We need more real world, real patient performance testing in clinical settings, not just data from laboratory on healthy patients
Proposal #3: Power pulse oximeter tests to detect small but clinically important differences between racial groups
- Small performance differences translate into significant population disparities, hidden hypoxia rates of 2% versus 12% for example!
Proposal #2: Align pulse oximeter evaluation with use in clinical practice
-ARMS (accuracy root mean square) doesn’t tell us what we need to know at the bedside. Rates of “occult” or “hidden” hypoxemia more relevant
Proposal #1: Require device manufacturers to report pulse oximeter performance in patient subgroups defined by relevant attributes
-This should be a mandatory labeling requirement!
As Nov 1 FDA meeting on PulseOx approaches, @tsvalley, @iwashyna and I wrote editorial in @ATSBlueEditor discussing actions @US_FDA, @ATS, @SCCM, @SenWarren, @SenBooker, clinicians, manufactures could all take to ensure they work accurately for everyone atsjournals.org/doi/abs/10.1…
Michael Sjoding retweeted
NEW: @msjoding & colleagues have spent 2+ years looking at the accuracy of #pulseox devices in hospital patients with darker skin, and finding major issues.
Read our new brief summarizing their work, in advance of an @US_FDA hearing on the topic this fall: ihpi.umich.edu/news/pulse-ox…
ALT Infographic about pulse oximeters in hospitalized Black patients. Click the link in the tweet to get to the brief that describes the findings in full.
Michael Sjoding retweeted
1/ Excited to announce our new work at @MLHC2022! Many patient risk-stratification ML models are trained using lab test results as labels. But testing disparities are known to exist across patient groups. (w/ @msjoding, Jenna Wiens) @MichiganAI (arxiv.org/abs/2208.01127)
Michael Sjoding retweeted
Hi, here’s my (@shanestorks) next Michigan AI #StudentTakeover Spotlight with Trenton Chang (@chang_trenton)! We discussed equity in healthcare and diagnostic testing, the role of AI, and overcoming imposter syndrome. #AcademicChatter
youtube.com/watch?v=KpnmyCj3…
Distillation of complex figure/nuanced thread on pulse ox accuracy/reliability. If pulse ox and ABG are way off, additional pulse ox readings likely unreliable for all. If pulse ox and ABG align, additional pulse ox readings likely accurate for White but less so for Black pts.
This quoted post is unavailable.
