The authors found that these apparently impressive ANNs were poorly generalizable (i.e., the performance was much worse on a new validation set compared to the training set).
Compare the red vs. green ROC curves. The performance drops from an AUC of 0.99 to 0.7! Yikes! 2/
There’s a reason for this: They used one dataset for all their positive images and a separate dataset for all their negative images.
This is risky for confounding because the model could pick up on any number of differences in CXRs that aren’t clinically meaningful.
3/
Turns out the neural networks weren't actually looking at the chest in the CXR; they were ‘cheating’ by looking at laterality markers and other parts of the images that were different between the two datasets.
Look at the salience maps (pixels the model deems most important): 4/
The AI ‘learned’ that specific laterality markers used in one hospital go with the COVID19+ cases & a different marker used in a different hospital go with the COVID19- cases
This marker cheating phenomenon was also described in PNA diagnosis in 2018 arxiv.org/abs/1807.00431 5/
3 things we do:
🤔be skeptical: ROC of 0.99 for detecting COVID on CXR is simply too good to be true
💾use better data: use a training set that has both + & - examples, validate with another dataset
👩⚕️medical experts should be involved in developing AI models to spot these biases
• • •
Missing some Tweet in this thread? You can try to
force a refresh
If you intubate you need to read the #PREOXI trial!
-n=1301 people requiring intubation in ED/ ICU were randomized to preoxygenation with oxygen mask vs non-invasive ventilation (NIV)
-NIV HALVED the risk of hypoxemia: 9 vs 18%
-NIV reduced mortality: 0.2% vs 1.1%
#CCR24
🧵 1/
Hypoxemia (SpO2 <85%) occurs in 10-20% of ED & ICU intubations.
1-2% of intubations performed in ED/ICU result in cardiac arrest!
This is an exceptionally dangerous procedure and preoxygenation is essential to keep patients safe.
But what’s the *BEST* way to preoxygenate? 2/
Most people use a non-rebreather oxygen mask, but because of its loose fit it often delivers much less than 100% FiO2.
NIV (“BiPAP”) delivers a higher FiO2 because of its tight fit. It also delivers PEEP & achieves a higher mean airway pressure which is theoretically helpful! 3/
Results from #PROTECTION presented #CCR24 & published @NEJM.
- DB RCT of amino acid infusion vs placebo in n=3511 people undergoing cardiac surgery w/ bypass.
- Reduced incidence of AKI (26.9% vs 31.7% NNT=20) & need for RRT (1.4% vs 1.9% NNT=200)
Potential game changer!
🧵 1/
I work in a busy CVICU & I often see AKI following cardiac surgery.
Despite risk stratification & hemodynamic optimization, AKI remains one of the most common complications after cardiac surgery with bypass.
Even a modest reduction in AKI/CRRT would be great for my patients. 2/
During cardiac surgery w/ bypass, renal blood flow (RBF) is reduced dramatically. This causes injury, especially in susceptible individuals.
But what if we could use physiology to protect the kidneys?
Renal blood vessels dilate after a high protein meal increasing RBF & GFR! 3/
77 yo with respiratory distress, RR 30, SpO2 80% on non-rebreather at 15 lpm
CXR & TTE are unrevealing
pH 7.58 / PaCO2 24 / PaO2 >500 / HCO3 22
MetHb 0% CarboxyHb 0%
The ABG looks like this:
The answer is sulfhemoglobinemia.
Sulfhemoglobinemia is a *permanently* modified hemoglobin associated with exposure to TMP/SMX, dapsone, phenazopyridine, & other amino & nitro compounds.
It has an altered oxy-hemoglobin dissociation curve.
2/
Sulfhemoglobinemia is easily confused with methemoglobinemia. Both have very dark colored blood & present with cyanosis. Diagnosis typically requires a specialized lab.
Spoiler: you may have heard that SulfHb is green. It isn’t really. You’re thinking of Vulcans’ blood.
Damn. Under Trump the White House Medical Unit was a pill-mill. Thousands of ambien & provigil per month.
Worse, for a clinic that doesn’t typically do procedures w/ moderate sedation they sure are they ordering prodigious quantities of morphine, fentanyl, versed, & ketamine…?
Honestly, this reminds me of Norman Ohler’s Blitzed.
The AG report was largely concerned with the enormous cost of prescribing these non-genetic meds.
It’s worth pointing out that dispensing prescription meds without documentation is malpractice. In the case of controlled substances it’s also likely a crime.