Theodore Pak
banner
tedpak.bsky.social
Theodore Pak
@tedpak.bsky.social
Assistant Professor @ucirvine.bsky.social. Former fellow @mgh-id.bsky.social, MGH Medicine resident, Sinai MD-PhD, mghwhitebook.app creator. Posts my own. Personal site: tedpak.com
Reposted by Theodore Pak
To be effective, data science agents need to be able to read plots reliably. @sara-altman.bsky.social and I wrote about some concerning findings on LLMs' ability to interpret plots when the content contradicts their expectations on the @posit.co blog.

posit.co/blog/introdu...
When plotting, LLMs see what they expect to see - Posit
Data science agents need to accurately read plots even when the content contradicts their expectations. Our testing shows today's LLMs still struggle here.
posit.co
November 13, 2025 at 3:07 PM
Reposted by Theodore Pak
Gave my 1st talk in NL: how to make multicenter data analysis with patient data actually work.

tl;dr: common data models (global interoperability) + LLMs (data harmonization, mapping, future-proofing) + federated learning (privacy)

Let me know what you think!

drive.google.com/file/d/1N3yk...
November 5, 2025 at 11:06 AM
Reposted by Theodore Pak
An #LLM extracted signs and symptoms for patients with sepsis with similar accuracy to humans, and identified 7 syndromes differentially correlating with multidrug-resistant infection risk and mortality. ja.ma/47onWVE
October 25, 2025 at 12:00 PM
Reposted by Theodore Pak
Maybe AI Knows Something That We Do Not!
🆕💫 Cohort study
LLM accurately extracted presenting signs & symptoms from admission notes that clustered into syndromes differentially correlated with infection sources, MDR infections, and mortality #IDsky
jamanetwork.com/journals/jam...
Syndromic Analysis of Sepsis Cohorts Using Large Language Models
This cohort study compares the performance of a large language model against manual medical record review for extracting presenting signs and symptoms data from hospital admission notes and uses these...
jamanetwork.com
October 24, 2025 at 7:23 PM
Reposted by Theodore Pak
My new piece in @medpagetoday.com

We Can Stop Deadly Hospital Outbreaks - So Why Aren't We?

Genomic surveillance can prevent outbreaks from spreading, save lives, and costs.

So what’s holding it back from being implemented in hospitals?

#IDSky

www.medpagetoday.com/opinion/seco...
Opinion | We Can Stop Deadly Hospital Outbreaks -- So Why Aren't We?
Genomic surveillance can help contain local and nationwide threats
www.medpagetoday.com
October 14, 2025 at 8:12 PM
#IDsky #Medsky 🧪🦠
It may be silly posting abt research in times like this but I have to share a small bit of joy with fellow ID heads. Our paper modeling the impact of colonization pressure on nosocomial pathogen acquisition just dropped on medRxiv! Only 3.5 years in the making.

www.medrxiv.org/content/10.1...

1/14
Hospital unit colonization pressure and nosocomial acquisition of drug susceptible and drug resistant pathogens
Background Hospitalized patients are at risk for developing hospital acquired infections. Active surveillance for bacterial colonization is effective at preventing infection but is resource-intensive ...
www.medrxiv.org
June 12, 2025 at 11:52 AM
Reposted by Theodore Pak
New preprint from our lab by Rangachar Srinivasa et al:

We WGS 4 major respiratory viruses in 3 hospitals (RSV, influenza, rhinovirus, HMPV) & found silent transmission happening under the radar

Genomics uncovered hidden spread, complex transmission chains #IDSky
📄: www.medrxiv.org/content/10.1...
April 27, 2025 at 8:55 PM
Reposted by Theodore Pak
How many hospitalized patients need care in the ICU? How many of them survive? What interventions do they receive there?

These are some of the questions I had two years ago while trying to write the "background" section for my masters thesis. But I couldn't find the answers... they did not exist.
April 14, 2025 at 9:41 PM
Reposted by Theodore Pak
The amount of green on this level of evidence guiding antifungal treatment duration demonstrates nicely the research knowledge gap in this field!

#IDSky #ClinMicro #AMR @drneilstone.bsky.social
March 15, 2025 at 11:19 AM
Reposted by Theodore Pak
ondering how this year’s measles outbreaks compare to previous years? This chart shows that 2025 (dark purple) is already about 4x higher than this time last year (light purple), which was itself a sharp increase over 2023. The trend is alarming. Prevention—including vaccination—is critical.
March 12, 2025 at 6:29 PM
Reposted by Theodore Pak
ID - Return to Basics:

This was the 1st reported case of an adult immunocompetent host who survived measles complicated by ARDS. A 36 yo female ED nurse, unvaccinated, who was exposed to a patient w active measles in October 1989. Ten days later, she developed measles
January 21, 2025 at 3:06 AM
Your AI can't see gorillas – A comparison of LLMs' ability to perform exploratory data analysis

chiraaggohel.com/posts/llms-e...
Your AI can’t see gorillas – Chiraag Gohel
A comparison of LLMs’ ability to perform exploratory data analysis
chiraaggohel.com
February 8, 2025 at 8:08 PM
Clever tricks with token sampling can guide a "reasoning" LLM like Deepseek's R1 (and the many new models it is already inspiring) into 'ing longer, and thereby producing more accurate output

There is a lot more juice to squeeze from these kinds of models.

timkellogg.me/blog/2025/02...
S1: The $6 R1 Competitor?
timkellogg.me
February 5, 2025 at 5:30 PM
Reposted by Theodore Pak
New paper in JAMA Network Open!

jamanetwork.com/journals/jam...

The IDSA released their guidelines for treatment of uncomplicated UTI back in 2011. We looked at a large claims database formatted into the OMOP common data model to see whether the recommendations still hold

tl;dr They do.

(1/7)
Use of Machine Learning to Assess the Management of Uncomplicated UTI
This cohort study examines the effectiveness and adverse event profile for first-line antibiotics, fluoroquinolones, and oral β-lactams for treating uncomplicated urinary tract infection (UTI) in cont...
jamanetwork.com
February 1, 2025 at 3:28 PM
Reposted by Theodore Pak
The removal of #HIV- and #LGBTQ-related resources from the websites of CDC and other health agencies is deeply concerning.

Access to this information is crucial for ID and HIV health care professionals.

Our full statement: https://buff.ly/42AO8LI
January 31, 2025 at 10:27 PM
Reposted by Theodore Pak
It is simply not enough for universities to say, “Our grants & research offices are looking into how research will be affected by the latest directive.”

University leaders must actively and very publicly make the case for the kind of work many of us do. The silence on this front is deafening.
January 31, 2025 at 3:16 AM
Reposted by Theodore Pak
It appears that the CDC has removed its guidelines for contraception and STIs. www.cdc.gov/contraceptio...
January 31, 2025 at 9:25 PM
Reposted by Theodore Pak
The entire website for the NIH Office of Research on Women's Health (ORWH) is very nearly stripped bare. This is so, so devastating. orwh.od.nih.gov/research/fun...
orwh.od.nih.gov
January 31, 2025 at 6:25 PM
Reposted by Theodore Pak
It's one thing to take down pages talking about diversity in science.

It's another thing entirely to take down research databases that are used for observational studies.

That is exactly what it seems just happened at the CDC.

www.statnews.com/2025/01/31/c... www.statnews.com/2025/01/31/c...
CDC removes data on sexual orientation, gender identity from website
Data from an expansive federal survey on youth behavioral habits, including their sexual orientation and gender identity, have been removed from the website of the CDC.
www.statnews.com
January 31, 2025 at 6:53 PM
Reposted by Theodore Pak
🚨If you use public data products from the CDC or other US federal agencies 🚨

Download them NOW

[This is how I’m spending my Friday]
A colleague asked whether I thought the government would continue the public release of its data products. The data from CDC health monitoring surveys, decades of data on mortality, crime data, climate data of a variety of types that go back for years, etc.
1/
January 31, 2025 at 1:02 PM
Reposted by Theodore Pak
I'm just here to say for the 2nd week in a row, the CDC has not released its morbidly and mortality weekly report on schedule. Everyone who provides healthcare to patients in the US is less informed and less up to date as a direct result. This impacts negatively at the population level.
Every week, I check the CDC's Mortality and Morbidity Weekly Report. This week for the first time in its 60 years it was not published on schedule. American healthcare workers all over the country are less informed about health crises that affect their patients. CDC led reports matter a 🧵
January 30, 2025 at 5:04 PM
Reposted by Theodore Pak
The communications pause for federal health agencies meant that the CDC couldn't publish the MMWR this week -- the first time in the publication's 60+ year history this has happened.

Unhappy about this! Getting good data published on infectious threats is critical, especially as H5N1 looms. #IDSky
Let's Hope the MMWR Resumes Publication Sooner Rather Than Later
To us specialists in Infectious Diseases, there are certain verities we hold near and dear to our hearts: Antibiotics are miracle drugs, but the bugs will become resistant if we don’t use them respons...
blogs.jwatch.org
January 25, 2025 at 10:42 AM
Board certified!

Glad that I'm now officially qualified to wear sunglasses while ranting about doxycycline, the definition of "FUO," and Streptococcus taxonomy controversies 😎
December 19, 2024 at 6:38 PM
Reposted by Theodore Pak
New: Universal masking&testing reduce hospital infections, so does staff masking

Stopping universal masking&testing led to 25% increase in respiratory viral infections
Restarting staff masking led to 33% decrease

8 out of 100 (random sample) died in the hospital
jamanetwork.com/journals/jam...
Testing and Masking Policies and Hospital-Onset Respiratory Viral Infections
This cohort study examines the ratio between hospital- and community-onset respiratory viral infections at different levels of testing and masking from 2020 to 2024.
jamanetwork.com
December 2, 2024 at 9:52 PM
Reposted by Theodore Pak
short thread on this important paper I reviewed today:
www.acpjournals.org/doi/10.7326/...
Morbidity and Mortality of Hospital-Onset SARS-CoV-2 Infections Due to Omicron Versus Prior Variants
1/
November 29, 2024 at 10:18 PM