Alyssa Pradhan

Our team wore blue for World Antimicrobial Awareness Week 2025💙

Research into how resistance emerges, spreads and can be prevented is vital. Our group works to understand these processes and develop better tools to detect, monitor and combat antimicrobial resistance.

#WAAW @oxfordbrc.bsky.social

When helpfulness backfires: LLMs and the risk of false medical information due to sycophantic behavior - npj Digital Medicine npj Digital Medicine - When helpfulness backfires: LLMs and the risk of false medical information due to sycophantic behavior

An overemphasis on helpfulness makes LLMs vulnerable.
Research shows models will comply with illogical medical requests, generating false information. This sycophantic tendency can be corrected with specific prompting and fine-tuning. #MedSky #MedAI #MLSky

Advisory Committee on Immunization Practices at a Crossroads

Viewpoint: The sudden removal of all Advisory Committee on Immunization Practices (ACIP) members and changes to #COVID19 vaccine policy have critically weakened the US vaccine program.

#MedSky

https://ja.ma/43MpHvd

"All the data in the world will not make a difference, unless there are people ready to use it."

— Atul Butte, MD, PHD, FACMI
2024 Morris F. Collen Award Winner

I like these papers where the risk from subtle patterns we pick up on clinically can be somewhat quantified:

pubmed.ncbi.nlm.nih.gov/40345937/

#MedicalAI #MedSky #MLSky #DigitalHealth #AIinHealthcare

Can #LLMs introspect? Can they do maths? We're back (again) busting myths about #LLMs and #AIinmedicine! Listen now: podcasts.apple.com/au/podcast/m...
#MedicalAI #MedSky #FOAMed #digitalhealth

Texas Banned Abortion. Then Sepsis Rates Soared. ProPublica’s first-of-its-kind analysis is the most detailed look yet into a rise in life-threatening complications for women experiencing pregnancy loss under Texas’ abortion ban.

NEW: Pregnancy became far more dangerous in Texas after the state banned abortion, according to ProPublica’s first-of-its-kind analysis, which found the sepsis rate for women hospitalized as they miscarried in the 2nd trimester shot up by more than 50%.

https://propub.li/43487Tj

@isabelhanson.bsky.social

Would love to collaborate on one if we can't find one!

A Cross-Sectional Study of GPT-4–Based Plain Language Translation of Clinical Notes to Improve Patient Comprehension of Disease Course and Management The 21st Century Cures Act provides patients with access to their clinical notes, but most patients, particularly those with limited health literacy, have difficulty understanding and utilizing the...

#GPT-4 generated plain language notes improved comprehension of discharge notes, particularly in Black and Hispanic patients and those with decreased #healthliteracy. Very cool to see the practical effect of this work
#MedicalAI #digitalhealth #MedSky #AI

ai.nejm.org/doi/full/10....

Interesting discussion with Hoffman, ft. some concerning sidestepping about regulation, conflating any guardrails with stifling innovation. Also interesting to see the anxiety around #DeepSeek immediately leading to justification / apology for tech billionaires..
#MLSky #digitalhealth #AI

This is gorgeous 😍

How Can We Diagnose and Treat Bias in Large Language Models for Clinical Decision-Making? Recent advancements in Large Language Models (LLMs) have positioned them as powerful tools for clinical decision-making, with rapidly expanding applications in healthcare. However, concerns about bias...

I really like the detail in this paper! Exploring the bias in #LLM responses through counterfactuals, SHAP, embeddings (oh my) and looking at intersections of gender x ethnicity 😍
TLDR: #Algorithmicbias runs deep
#MedSky #MLSky #digitalhealth #healthequity
arxiv.org/abs/2410.16574

Large Language Models lack essential metacognition for reliable medical reasoning - Nature Communications Large Language Models demonstrate expert-level accuracy in medical exams, supporting their potential inclusion in healthcare settings. Here, authors reveal that their metacognitive abilities are under...

This is an interesting question but does the methodology address whether an LLM is "uncertain"? It would be interesting to see how the LLM reports of confidence in an answer correspond to probability of the token for that answer... #MLSky #medsky #healthcareAI
www.nature.com/articles/s41...

Ever wondered what it’s like being stuck talking to me at a party? Look no further.
#MedSky #MLSky #algorithmicbias #healthcareAI

"Replacement of just 0.001% of training tokens with medical misinformation results in harmful models more likely to propagate medical errors.... Corrupted models match the performance of their corruption-free counterparts on open-source benchmarks routinely used to evaluate medical LLMs."

If RFK Jr. is confirmed. @donttreadonkids.bsky.social

I'd love to be included if possible! Thanks ☺️

Specifically, at what point do you *not* run the model that works really well for the majority population, but performs very poorly in the minority...

The focus on close tracking of, and transparency about performance across subgroups is on point though

I agree with the call for more and better data and think there are cases where using protected attributes can improve diagnosis. But am not sure that the diagnostic nature of pathology / radiology mean that significant variation in performance across subpopulations is acceptable.

Ethical Use of Artificial Intelligence in Medical Diagnostics Demands a Focus on Accuracy, Not Fairness Artificial intelligence (AI) promises to be a transformative technology for medicine and health care. As such, there is an increasing interest in ensuring its ethical use. In this perspective, we c...

Interested to see what people's thoughts are regarding this #NEJMAI perspective 🧵
#MedSky #MLSky #algorithmicbias #healthcareAI #healthequity
ai.nejm.org/doi/full/10....

Great to see "TRIPOD+AI and DECIDE-AI are the highest quality guidelines with respect to rigor of development and the involvement of the stakeholders" 😄 in

academic.oup.com/jamiaopen/ar...

Get TRIPOD+AI from www.bmj.com/content/385/...

#MLSky #StatsSky #MedSky

Check out our new paper "Tackling Algorithmic Bias and Promoting Transparency in Health Datasets: The STANDING Together Consensus Recommendations" jointly published in NEJM AI and The Lancet Digital Health, led by @jaldmn.bsky.social @xiaoliu.bsky.social

The importance of microbiology reference laboratories and adequate funding for infectious disease surveillance Microbiology reference laboratories perform a crucial role within public health systems. This role was especially evident during the COVID-19 pandemic. In this Viewpoint, we emphasise the importance of microbiology reference laboratories and highlight the types of digital data and expertise they provide, which benefit national and international public health. We also highlight the value of surveillance initiatives among collaborative international partners, who work together to share, analyse, and interpret data, and then disseminate their findings in a timely manner.

I agree with this Lancet Digital Health viewpoint that data harmonisation and funding for microbiology reference labs is a public health issue. Investment in the expertise to standardise and analyse this data in real time is also needed
@thelancet.bsky.social
www.thelancet.com/journals/lan...

Move over Spotify wrapped, it's time for 49 minutes of us chatting about the trends in #healthcareAI for 2024!
#MLSky #MedSky #MedEd #Podcast

Haematological setpoints are a stable and patient-specific deep phenotype - Nature Complete blood count indices are tightly regulated around setpoints for decades in healthy adults, and represent a deep phenotype providing opportunities for investigating differential disease risks, ...

A cool illustration of the richness of routine laboratory data. In all the hype of #healthcareAI, this use to unveil patterns we can't perceive is particularly powerful. #MedSky #MLSky
www.nature.com/articles/s41...

Nice paper - tricky case. Patient w COVID-19 isolated for 7 months, impacting rehab.

Rules and interpretation of tests were too rigid for lung Tx patient, long after IPC threat had passed.

Authors @priyagargiddoc.bsky.social @alyssapradhan.bsky.social et al offer some solutions!

Tagging 🩺🖥️

LLM-based agentic systems in medicine and healthcare - Nature Machine Intelligence Large language model-based agentic systems can process input information, plan and decide, recall and reflect, interact and collaborate, leverage various tools and act. This opens up a wealth of oppor...

The opportunities, challenges and outlook for LLM-based agents in medicine and healthcare—our paper published today

nature.com/articles/s42...