We are hiring! Looking for a talented postdoc to work on clinical #DataScience @UCSF_BCHSI. If you know of folks with a background in any of the following please reach out! #PublicHealth #Epidemiology #MachineLearning #NLP #EHR bio #Stats #DeepLearning

It was a great conference! Looking forward to next year's program!

Congratulations @aryanayati @egnij and the team @UCSF and @Merck !

Congratulations @aryanayati and to the teams at @UCSF @UCIrvine and @GenentechEye! It was a great collaboration, and a nice way to show how large language models and #causalinference methods can help us practice evidence-based medicine

Congratulations @aryanayati, Göktuğ Önal, Mao-Yuan Chen, and Anshu Mukherjee! Really proud to be a part of this research team!

Check out https://t.co/iNCrJWEsmW and https://t.co/SIXF1WcAX5. Lots of opportunities to extend our methods beyond Crohn's disease to many other therapeutic areas:

My sincere thanks to co-authors @Doug_Arneson @armanmosenia and Vignesh @UCSF_BCHSI , and Shan @usfca. Thanks to @YODAProject, @VivliCenter, and to all trial participants whose data continues to advance medical research and IBD patient outcomes decades later.

We think there's something for everyone here: #ibd patients, medical researchers, data scientists, trialists, and people who believe in a future of personalized #PrecisionMedicine.

We found interesting differences between the patients recruited into trials and those seen in clinical practice. The subgroup of women over 50 were only 2% of the pooled population across 15 RCTs, yet they are actually 25% of IBD patients @UofCAHealth. RCT selection bias we think

We discover multiple subgroups of patients with Crohn's disease with different treatment responses. For example, women over the age of 50 w/ prior anti-TNF exposure have treatment responses that deviate from the majority trend, responding better to anti-IL12/23s than anti-TNFs.

Now we extend this concept to virtual cohorts of identical patients (#digitaltwins). Using nested models to separate placebo effects from drug effects, we performed multiple RCTs in silico to find the best treatments for each patient. We use data from 5703 patients, 15 RCTs.

In prior work, we developed a method for accurately predicting the results of head-to-head trials using historical data. We showed that we could correctly predict the results of a head-to-head trial in #crohnsdisease, using preexisting RCT data.

The ideal scenario would be to run many randomized trials on a hypothetical population of identical patients, studying pairs of drugs head-to-head, to see what works best for each patient. But this is clearly impossible!

Medical practice today is guided by studies that measure drug effects by looking at population averages. But in many cases, individual patient responses can deviate from the majority trend. How can we evolve from a world of one-size-fits-all medicine to #personalizedmedicine?

It is with a mix of emotions that I finally announce this work, my very last w/ @atulbutte. Atul believed in the power of open data and our ability to learn from the data we already have to improve healthcare now. This work is a tribute to this belief, made possible by his career

Thank you @microsoft research for supporting our work! We are committed to advancing the use of AI in medicine

Congratulations to @UCSF Duygu Tosun-Turgut @birtutamtuz Inducted into the 2025 Class of the @AIMBE College of Fellows! https:///press/tosun-turgut-COF-9122.pdf @UCSF_Ci2

One photo with: Milkyway, Zodical light, @Starlink satellites as streaks, stars as pin points, atmosphere on edge showing OH emission as burned umber (my favorite Crayon color), soon to rise sun, and cities at night as streaks. Taken two days ago from Dragon Crew 9 vehicle port

Flying over aurora; intensely green.

Truly grateful to the @KR_Foundation for their generosity. We are on a mission to advance artificial intelligence for IBD research and patient care. This award will go a long way towards helping us get there!

Thank you @UCJointCPH for the shoutout!