Girish Nadkarni's Avatar

Girish Nadkarni

@girishnadkarni

Immigrant. Professor. Data Scientist. Father.Son.Friend. Opinions Own. RT≠endorse

2,075
Followers
1,646
Following
43
Posts
10.10.2023
Joined
Posts Following

Latest posts by Girish Nadkarni @girishnadkarni

Go Eagles Go.

#eagles

10.02.2025 02:22 👍 1 🔁 0 💬 0 📌 0

The economic chaos that will occur across the United States on Monday is something I hope is being well covered by journalists. Hospitals and Universities employ large sectors of communities.

08.02.2025 02:27 👍 80 🔁 23 💬 5 📌 1
Thieme E-Books & E-Journals -

Can carefully developed electronic alerts help with management of #AKI? Yes, but developing an effective alert is both an art and science. Read about our experience in #ACI www.thieme-connect.com/products/ejo...
@girishnadkarni.bsky.social #informatics

01.01.2025 19:41 👍 5 🔁 1 💬 0 📌 0

Happy New Year! 2025 = 45² — we’re celebrating a perfect square year. Cheers to fresh starts, big dreams, happiness and contentment!

01.01.2025 13:48 👍 8 🔁 1 💬 0 📌 0

For all of mankind

11.12.2024 03:29 👍 0 🔁 0 💬 0 📌 0

Also has a blue suit though.

11.12.2024 02:26 👍 0 🔁 0 💬 1 📌 0

Superman gets beaten up way too much.

11.12.2024 02:21 👍 0 🔁 0 💬 1 📌 0

Ok I dont like Superman and Lois anymore

11.12.2024 02:14 👍 2 🔁 0 💬 1 📌 0

I have tried starbucks instant coffee and its a revelation.

10.12.2024 20:48 👍 1 🔁 0 💬 0 📌 0

Had so much fun discussing how #AI is shaping medicine and sharing the work from our lab at #apexpathshala @akforg.bsky.social @girishnadkarni.bsky.social

08.12.2024 06:03 👍 10 🔁 2 💬 0 📌 0

How have I not watched Superman and Lois till now

01.12.2024 02:51 👍 3 🔁 0 💬 0 📌 0
Preview
Extracting International Classification of Diseases Codes from Clinical Documentation using Large Language Models - PubMed Current LLMs have poor performance in extraction of ICD-10-CM codes from inpatient notes when compared against the human coder.

Are #LLMs as good as human coders in extracting ICD codes? Of 6 LLMs we tested, GPT-4 achieved highest agreement with human coder but still only at 15.2%. Read our full study in #ACI at - pubmed.ncbi.nlm.nih.gov/39608761/ @girishnadkarni.bsky.social

30.11.2024 18:00 👍 6 🔁 1 💬 0 📌 0
Preview
A Primer on Reinforcement Learning in Medicine for Clinicians npj Digital Medicine - A Primer on Reinforcement Learning in Medicine for Clinicians

Excited to share our review article, which provides a step-by-step comprehensive overview of reinforcement learning tailored specifically for clinicians, presented in a clear and non-technical manner - www.nature.com/articles/s41... @girishnadkarni.bsky.social #AI #RL

27.11.2024 02:47 👍 6 🔁 3 💬 1 📌 0
Participate – SAIL: Symposium on Artificial Intelligence for Learning Health Systems

The SAIL 2025 CALL FOR ABSTRACTS is open! Join us on May 6-9 in Río Grande, Puerto Rico! In-person attendance is limited to those with accepted work. NEJM_AI will award cash scholarships to the top abstracts. Travel support is available. Submit by January 17: sail.health/event/sail-2.... #SAIL25

25.11.2024 21:48 👍 10 🔁 2 💬 1 📌 0
Post image

Sky is blue(kinda)
@bsky.app

21.11.2024 14:04 👍 3 🔁 0 💬 0 📌 0
Preview
a man says you look pretty stressed in a showtime ad Alt: a man says you look pretty stressed in a showtime ad

We’ve all been under a lot of stress lately.

But did you know that stress affects blood coagulation?

If you didn’t before prepare to now with another #hemesky #skeetorial

1/

20.11.2024 20:48 👍 263 🔁 51 💬 19 📌 7

There’s a block list for which you have to subscribe

20.11.2024 15:19 👍 0 🔁 0 💬 1 📌 0

It had to be done

20.11.2024 02:03 👍 1 🔁 0 💬 0 📌 0
Post image

Sky is blue @bsky.app

19.11.2024 14:57 👍 8 🔁 0 💬 0 📌 0
Post image

Good morning

19.11.2024 12:28 👍 6 🔁 0 💬 0 📌 0

I’m so happy @profgalloway.bsky.social is on @bsky.app

19.11.2024 03:11 👍 2 🔁 0 💬 0 📌 0

#AI #health #digital #systems

18.11.2024 19:52 👍 1 🔁 0 💬 0 📌 0
Preview
Study Identifies Strategy for AI Cost-Efficiency in Health Care Settings

Ben Glicksberg Eyal Klang

www.nature.com/articles/s41...
www.mountsinai.org/about/newsro...

18.11.2024 19:46 👍 3 🔁 0 💬 1 📌 0

🧑‍⚕️ Real-world impact? Think:
Efficient resource allocation reports.
Bulk patient analysis for clinical trials.

Next steps?
Apply these findings in prospectively
Test new models.
Address regulatory hurdles. Safe & efficient AI deployment is key. 🏗️🔐

18.11.2024 19:46 👍 1 🔁 0 💬 1 📌 0

Economics matter: The study shows how token-based pricing (OpenAI's GPT-4: $0.24/task) makes concatenation strategies a game-changer. Every $ counts at system scale! 🩺💸

18.11.2024 19:46 👍 1 🔁 0 💬 1 📌 0

Failures were mostly formatting (JSON errors) or skipped questions (omissions). The good news? These issues are fixable with re-queries, meaning fewer false answers in critical settings. 🔄✔️

18.11.2024 19:46 👍 0 🔁 0 💬 1 📌 0

But performance depends on the model:
GPT-4-turbo-128k and Llama-3-70B? 👌 High accuracy, low errors.
Others? Struggled as task size increased. 😓

18.11.2024 19:46 👍 0 🔁 0 💬 2 📌 0

Key insight: By grouping (concatenating) up to 50 tasks in a single prompt, we can reduce costs 17x—while staying accurate. A win for health systems scaling AI use! 🏥💻

18.11.2024 19:46 👍 1 🔁 0 💬 1 📌 0

We tested 10 LLMs (proprietary/open source of different sizes) through 300k+ experiments to test their ability to process real clinical data under heavy workloads. The challenge? Accuracy + cost-efficiency. Here's what they found. 🧪

18.11.2024 19:46 👍 0 🔁 0 💬 1 📌 0

LLMs promise to transform healthcare: summarizing patient records, predicting outcomes, and streamlining operations. But... they're EXPENSIVE (due to $/API call) and not built for system-scale use. 💰⚡

18.11.2024 19:46 👍 0 🔁 0 💬 1 📌 0