One-Liners, Part One

I've fallen behind and I can't get up!

Having been overtaken by “life” over the past few weeks, there’s a whole host of neglected articles – here are a handful of one-liners from the first of the 20+ I’ve not had a chance to fully summarize:

Maternal paracetamol (acetaminophen) use during pregnancy and risk of autism spectrum disorder and attention deficit/hyperactivity disorder in offspring: umbrella review of systematic reviews

From an academic standpoint, it’s always valid to explore associations between exposures and outcomes. But, it’s professionally dishonest to promote a single, outlier review absent the greater context of multiple other reviews. This “umbrella review of systematic reviews” collates the acetaminophen/paracetamol link to autism and ADHD and finds generally low quality evidence that fails to demonstrate a clear link.

Assessing deep learning artificial intelligence support for detecting elbow fractures in the pediatric emergency department

Pediatric elbow fractures can be challenging. In this French emergency department, sensitivity of clinicians alone was only 77.3%. Adding AI increased this to 98.9% – but specificity dropped from 88.3% to 63.5%. A classic trade-off needing either 1) improved technology, or 2) an appropriate pathway to mitigate the false-positives.

Can radiology requisition quality reflect clinical reasoning? Insights from a RI-RADS evaluation of emergency CT referrals

This article only crossed my feed because it mentioned “emergency CT referrals” and their high level of inappropriateness. Turns out, it includes all referrals from across clinics and non-emergency services, and “inappropriateness” is primarily a factor of poor descriptiveness the narrative request, not the test itself.

Evaluating human-in-the-loop strategies for artificial intelligence-enabled translation of patient discharge instructions: a multidisciplinary analysis

It’s not rocket science. AI can translate discharge instructions into other languages, but <gasp>, they’re not as good as professional translators. However, human + AI translation was the best.

Peer perceptions of clinicians using generative AI in medical decision-making

In this qualitative evaluation, surveyed physician participants disapproved of other clinicians who depended upon generative AI to support their decision-making. This was partially ameliorated if said colleagues were reported to make their own diagnostic decisions and used LLMs only for “verification”.

Peripheral Vasopressor Use in Early Sepsis-Induced Hypotension

This is a simple retrospective analysis of CLOVERS, which was a study looking at liberal versus restrictive fluid resuscitation in septic shock. In this analysis, it did not appear to have any negative association with survival to have received vasopressors through peripheral access, rather than central.