iiinova — where your digital self keeps evolving

Reposted by Andrew Admon

Jack Iwashyna @iwashyna.bsky.social · 09/06/2025

#NotEvenTrying New in @sccmcriticalcare.bsky.social Crit Care Med, @hauschildt.bsky.social + @ajadmon.bsky.social review 120 ML or AI studies predicting sepsis, sepsis-related outcomes or sepsis treatment in adult populations NONE reported formal fairness metrics pubmed.ncbi.nlm.nih.gov/40488579/

Abstract text:
Consideration of Sociodemographics in Machine Learning-Driven Sepsis Risk Prediction
OBJECTIVES: Use of machine learning (ML) and artificial intelligence (AI) in prediction of sepsis and related outcomes is growing. Guidelines call for explicit reporting of study data demographics and stratified performance analyses to assess potential sociodemographic bias. We assessed reporting of sociodemographic data and other considerations, such as use of stratified analyses or use of so-called “fairness metrics", among AI and ML models in sepsis.
DATA SOURCES: PubMed identified systematic and narrative reviews from which studies were extracted using PubMed and Google Scholar.
STUDY SELECTION: Studies were extracted from selected review articles published between January 1, 2023, and June 30, 2024, and related to sepsis, riskprediction, and ML; we extracted studies predicting sepsis, sepsis-related outcomes, or sepsis treatment in adult populations.
DATA EXTRACTION: Data were extracted by two reviewers using predefined forms, and included study type, outcome of interest, setting, dataset used, reporting of sample sociodemographics, inclusion of sociodemographics as predictors, stratification by sociodemographics or assessment of fairness metrics, and reporting a lack of sociodemographic considerations as a limitation.
DATA SYNTHESIS: Thirteen of 96 review studies (14%) met inclusion criteria: 6 systematic reviews and 7 narrative reviews. 120 of 170 studies (71%) extracted from these review articles were included in our review. 99 of 120 studies (83%) reported a measure of geography or where data was collected. Eighty (67%) reported sex/gender, 24 (20%) reported race/ethnicity, and 4 (3%) reported other sociodemographics. Only three stratified performance results (2%) by sociodemographics; none reported formal fairness metrics. Beyond a lack of geographic heterogeneity (39/120, 33%), few studies reported a lack of sociodemographic consideration as a limitation

172

Andrew Admon @ajadmon.bsky.social · 25/12/2024

I’m disappointed that I never got a set of these when I left the house!

120

Reposted by Andrew Admon

Gabriel Bosslet @gbosslet.bsky.social · 16/12/2024

#PCCMSky Update We are now 2 starter packs deep! Repost so the noobs know who the cool kids are CCM only folks definitely welcome Reply if you are not on one and should be (need to see your creds in bio) go.bsky.app/2gcvXuy go.bsky.app/88wg6oG

3168

Reposted by Andrew Admon

Başak Çoruh @basakcoruh.bsky.social · 17/12/2024

1/ Great advice for med students & residents on the DON’Ts for a personal statement: tinyurl.com/yjewsknc ❌ Don’t write >1 page ❌ Don’t write before you think ❌ Don’t write about a momentous clinical experience ❌ Don’t focus only on medical experiences ❌ Don’t explain the specialty

tinyurl.com

What NOT to Write in Your Personal Statement

Succeeding in residency takes grit, determination, compassion, and industrial doses of caffeine. For program directors, figuring out which applicants are most likely to navigate the transition from pe...

173

Andrew Admon @ajadmon.bsky.social · 16/12/2024

I’ll preorder the coffee table book

020

Andrew Admon @ajadmon.bsky.social · 16/12/2024

Maggie made a checklist for her birthday party (with some spelling help :-)). This is going to be an epic fifth birthday bash!

160

Reposted by Andrew Admon

Julia M. Rohrer @dingdingpeng.the100.ci · 13/12/2024

You know what’s the real disappointment here? We never went all the way. We never had a logit vs probit showdown.

19795

Andrew Admon @ajadmon.bsky.social · 14/12/2024

I tried to add to your total but was already following you!

120

Reposted by Andrew Admon

Ann Leckie @annleckie.com · 12/12/2024

Do not confuse "interesting thoughts about art" for ironclad instructions about how to make art.

311520

Andrew Admon @ajadmon.bsky.social · 11/12/2024

Wait until we tell them about the benefits of a third stent

010

Reposted by Andrew Admon

Julia M. Rohrer @dingdingpeng.the100.ci · 11/12/2024

I think they turned it into a movie later, Regression Runner

0131

Andrew Admon @ajadmon.bsky.social · 11/12/2024

That’s also why I like target trial emulation- it helps identify study design decisions that would have been clearly weird or impossible in an RCT (“we’ll enroll people today and start them on treatment sometime in the past six months!”) but sometimes happen anyways in observational studies

120

Andrew Admon @ajadmon.bsky.social · 11/12/2024

Love it, but more importantly strong choice of colors. I'd add a line between Asthma and death too, but the measured relationship is being distorted by the open path via the "Pneumonia and hospitalized" since they're conditioning on it

110

Andrew Admon @ajadmon.bsky.social · 11/12/2024

Conditioning on hospitalization for pneumonia induces a false (inverse) association between asthma and other severe risk factors/complications that get you admitted with pneumonia and makes it look like pneumonia patients do well.

220

Andrew Admon @ajadmon.bsky.social · 11/12/2024

Their inclusion criteria included both pneumonia and hospitalization, so that's the 'selection node' or 'collider' (the box at the top). Asthma is one reason for hospitalization with pneumonia (e.g., it's a risk factor for poorer outcomes, so the ED might admit). Septic shock is another.

220

Andrew Admon @ajadmon.bsky.social · 11/12/2024

Oooh, I’m going to use that with my 4 year old- “I need to see more efficacy data before I keep doing that”

100

Reposted by Andrew Admon

ATS Blue Journal @atsblueeditor.bsky.social · 05/12/2024

📣 The AJRCCM invites you to submit your cutting-edge research in pulmonary and critical care medicine @atscommunity.bsky.social #medsky SUBMIT HERE: tinyurl.com/3hezurez

065

Reposted by Andrew Admon

Jack Iwashyna @iwashyna.bsky.social · 06/12/2024

last night, Hallie Prescott had her NEJM review on sepsis released: www.nejm.org/doi/full/10.... Today I had the pleasure of giving some remarks on her installation as the Toews Family Legacy Professor at the University of Michigan

nejm.org

Sepsis and Septic Shock | NEJM

Sepsis is a life-threatening, heterogeneous disorder involving a dysregulated immune response to infection. Research to identify subtypes and targeted therapies for host dysregulation is ongoing.

36214

Andrew Admon @ajadmon.bsky.social · 04/12/2024

Figures with manually added references…

000

Reposted by Andrew Admon

American Thoracic Society (ATS) @atscommunity.bsky.social · 03/12/2024

Hello, Bluesky! We've finally started building out our starter pack for our ATS member community. If you're a current or former member, let us know and we will add you. go.bsky.app/GLzJtqT #medsky #lungsky #pedsky

225220

Andrew Admon @ajadmon.bsky.social · 01/12/2024

8-year-old Beau. He’s a good boy and loves his little brother.

160

Andrew Admon @ajadmon.bsky.social · 30/11/2024

We’d never let you anyways

010

Andrew Admon @ajadmon.bsky.social · 30/11/2024

Who’s got it better than us?!

000

Andrew Admon @ajadmon.bsky.social · 30/11/2024

You’re as cool as the other side of the pillow

120

Reposted by Andrew Admon

Gabriel Bosslet @gbosslet.bsky.social · 25/11/2024

I have been a big proponent of 15 minutes of writing per day- especially for busy clinical academicians as laid out in the book "How to write a lot" (a great and quick read) We suffer from the idea that we need chunks of time to write So with this 🧵 I will put my money where my mouth is 1/

1314027

Andrew Admon @ajadmon.bsky.social · 28/11/2024

I haven’t been able to get @nealstephenson.bsky.social’s Anathem out of my head, and am already looking forward to re-reading. And if that isn’t enough, it has DAGs!

Cover for book “Anathem” by Neal Stephenson, featuring a person in a tunic walking out of a gate beneath an analemma.

030

Andrew Admon @ajadmon.bsky.social · 27/11/2024

Thanks for making this so easy!

100

Andrew Admon @ajadmon.bsky.social · 26/11/2024

Of course! I owed you for the number of times on rounds we learn from your posts! Re: the estimator in an IV analysis, the time-outcome relationship is a solid option with few assumptions. But It's not the only (or even most patient-relevant) estimator. More at doi.org/10.1136/bmj-2023-078093

020

Andrew Admon @ajadmon.bsky.social · 26/11/2024

In the meantime, thanks again for all of the great stuff you post online and now here. It's done a lot to move all of our care forward.

130

Andrew Admon @ajadmon.bsky.social · 26/11/2024

More broadly, I share your healthy skepticism of observational comparative effectiveness studies and would love a well-powered RCT with extended follow-up and good separation between the anaerobe-depleting and anaerobe-preserving abx.

130

Andrew Admon @ajadmon.bsky.social · 26/11/2024

0.5% absolute increased mortality. If that was the right analysis all along, I think that's still clinically meaningful given the numbers of people treated, absent a specific indication to go with zosyn over cefepime (e.g., concern for intrabdominal infection).

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

Extra: Other criticism describes possible of collider stratification b/c we adj for flagyl. This is reasonable! But, we made this choice because our target comparison was anti-anaerobe vs anaerobe-perserving abx. Reassuringly, the same analysis w/o flagyl was in the same direction, but smaller w/...

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

5. Inconsistent with Cochrane: It'd be good to dive into each of the RCTs directly, but at first glance of the cefepime vs. pip-tazo studies, those confidence intervals look a little wide to draw any clear conclusions.

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

4. Inconsistent with ACORN: ACORN was an important study but a lot of people in the "cefepime" arm got a dose of either flagyl (47%) or zosyn (19%). And so if the effect is due to anaerobe depletion, it didn't really look at that.

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

3. Confounding by indication: Definitely a possibility! But the instrumental variable analysis (with or without flagyl- see below) helps protect against this.

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

2. Mortality differences emerge late: I agree, and I get why this feels weird for typical ICU interventions. It feels plausible for effects of anaerobe depletion, though, which may work by impacting secondary infection. While this is based on lab/animal data, I agree that this is a bit handwavy

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

1. Raw data is negative: The raw data wasn't negative (bottom of table 1, or just calculate the proportion who died who in each antibiotic group). It's still worth pursuing the instrumental variable analysis due to confounding by indication, but it's reassuring that the results are compatible.

120

Andrew Admon @ajadmon.bsky.social · 26/11/2024

Thanks for sharing this, and thanks for all that you do for our field, especially bringing solid meded to Bluesky! Your work gives a lot of us a great reason to join. I was involved in the original paper and while I'm (always) open to being wrong I don't think it's for the reasons here. ...

120