3 of 37

Our trifold mission to care, teach, �and discover is achieved daily through an integrated academic culture devoted to excellence in patient care, education, and research.

TECHNOLOGY IS A KEY ENABLER

NYU Langone Health’s Mission

CARE

TEACH

DISCOVER

NYU Langone Health

Values and Culture

Performance

Respect

Integrity

Diversity

Excellence

We are a culture of exceptionalism.

(PDF-087) Has your organization developed a written IT strategic plan aligned to your organization’s business plan to include the following?

(PDF-088) Does your organization have an IT Steering Committee for the following?

(PDF-089) Are all IT investment proposals for the following reviewed and approved by an IT Steering Committee?

(PDF-090) Do all IT investment proposals for the following include a documented business case?

(PDF-091) Do IT investment proposals for the following involve an executive sponsor from the impacted business area who is responsible for overseeing the business case for IT investments in their area?

(PDF-092) Does your organization have an IT Steering Committee overseeing and tracking the progress of each approved IT implementation project, for the following?

4 of 37

NEW YORK

THE BRONX

STATEN ISLAND

QUEENS

NASSAU

WESTCHESTER

SUFFOLK

CONNECTICUT

NEW JERSEY

NYU Langone Hospital–Brooklyn

BROOKLYN

MANHATTAN

FLORIDA

PALM BEACH BROWARD

MIAMI-DADE

(3)

NYU Langone Hospital–Long Island

(1)

EAST 34TH ST

EAST 36TH ST

EAST 37TH ST

EAST 38TH ST

EAST 39TH ST

EAST 40TH ST

EAST 35TH ST

EAST 30TH ST

EAST 29TH ST

EAST 28TH ST

EAST 26TH ST

EAST 25TH ST

EAST 24TH ST

EAST 23RD ST

EAST 22ND ST

EAST 21ST ST

EAST 20TH ST

EAST 19TH ST

EAST 18TH ST

EAST 17TH ST

FDR DR

1ST AVE

2ND AVE

3RD AVE

LEXINGTON AVE

1ST AVE

2ND AVE

3RD AVE

LEXINGTON AVE

PARK AVE

Kimmel Pavilion & Hassenfeld Children’s Hospital—34th Street

Tisch Hospital

MANHATTAN

NYU Langone Orthopedic Hospital

NYULH— Suffolk (affiliate)

(1)

FAIRFIELD

BERGEN

Locations

320+

NYU Langone Inpatient Locations Tisch Hospital

Kimmel Pavilion

Hassenfeld Children's Hospital NYU Langone Orthopedic Hospital NYU Langone Hospital—Brooklyn

NYU Langone Hospital—Long Island

NYU Langone Health— Suffolk (affiliate)

5 of 37

Our Rankings & Honors

Five-Star Quality Rating

by the Centers for Medicare & Medicaid Services (CMS)

Magnet Designation

The Only Health System*

in New York to Earn

Magnet^® Recognition

Across All Its Hospitals

*Refers to multi-specialty and �multi-site hospital organizations

Grade A Leapfrog Hospital Safety Grade

Straight A's for Safety

at Our Hospitals in Manhattan, Brooklyn, �and Long Island

Vizient’s Bernard A. Birnbaum, MD, Quality Leadership Award

Ranked #1 Out of 118 Participating Comprehensive Academic Medical Centers for Quality

The Gold Seal of Approval^®

by The Joint Commission

Reflects a Commitment to

Continuous Improvement �and Delivering Safe, High-Quality Care

Vizient’s Ambulatory �Care Quality and Accountability Award

Ranked 1 Out of 69 Networks for Demonstrating Excellence in Delivering High-Quality Outpatient Care

Most #1 Ranked Specialties in the Nation and 13 Specialties in the Top 20 by U.S. News & �World Report

Quality & Safety are at the heart of everything we do.

6 of 37

Vocabulary

Generative AI: Set of algorithms, capable of generating seemingly new, realistic content — such as text, images, video, or music — from examples
Frontier Model: Reference to Gemini, GPT, or Claude.
Anthropic: Company behind the Claude line of Frontier Models
Gemini: Frontier model built by Google.
OpenAI: Company run by Sam Altman who built GPT4. The company has its own products.
ChatGPT: The actual chatbot run by OpenAI.
GPT: Generative Pretrained Transformer
Large Language Model (LLM): Technical name for Frontier Models (will use interchangably during talk)
Parameters: Size of a model. Ranges from 1 billion to 1 trillion. Compute cost scales in these ranges. Larger parameters generally means a better model.

8 of 37

Artificial Intelligence. 1. What is Artificial Intelligence… | by DSLAB GLOBAL | Medium

Generative AI

9 of 37

The brief history of artificial intelligence: The world has changed fast – what might be next? - Our World in Data

10 of 37

Generative AI – Images and Text (a.k.a. ChatGPT)

Test Yourself: Which Faces Were Made by A.I.? - The New York Times (nytimes.com)

Joke from https://arxiv.org/pdf/2204.02311v2

11 of 37

AI at NYU Langone

12 of 37

MCIT Department of Health Informatics: �Division of Applied AI Technology

Develop, translate, evaluate, apply, train the workforce in artificial intelligence, machine learning, and predictive analytic solutions in support our corporate, clinical, research, and education missions.

Mission

Statement

13 of 37

NYULH Vision – AI and Generative Models

Using AI and Generative AI ethically and equitably at scale to improve: efficiency, quality and safety.

We believe that:

Every piece of data captured and/or text written for or by the workforce will go through an AI based model one day.

Efficiency

Quality

Safety

14 of 37

Generative AI Innovation and Progress

One of the first healthcare institutions to collaborate with Microsoft for a private, secure ChatGPT.
Allows safe and responsible AI access across NYUL.
Trains workforce in Generative AI with promptathons across operational, research, and clinical units [%].
Builds custom Generative AI models from scratch published in Nature, NYUTron [$].

AI + GenAI Models Currently Live (Homegrown + Vendor)	AI + GenAI Models in Development (Homegrown + Vendor)
128 Models	153 Projects

$ - Jiang, L.Y., Liu, X.C., Nejatian, N.P. et al. Health system-scale language models are all-purpose prediction engines. Nature 619, 357–362 (2023).

% - Small WR, Malhotra K, Major VJ, et al. The First Generative AI Prompt-A-Thon in Healthcare: A Novel Approach to Workforce Engagement with a Private Instance of ChatGPT. PLOS Digit Health. 2024 Jul 23;3(7):e0000394.

15 of 37

UltraVioletAI

Secure, HIPAA compliant Available to All 50,000 employees.

Launched October 1.

Deployed: 45k Windows devices, 3.7k Mac devices, and 12.5k smartphones

Unique Users: 19,601

New Chats Per Day: ~4,000

Mobile

Epic View

Desktop

Confidential – Do Not Distribute

17 of 37

Early Predictions of Skilled Nursing Facility Discharges To Accelerate Discharge Planning

Objective: Integrate early predictions of inpatient discharges to skilled nursing facilities (SNFs) into Epic to facilitate discharge planning and PT/OT resource triaging.

*Model: Language model (nyutron) predicts SNF discharge risk from AI summaries of H&P notes.

EHR Intervention: Patient list column describing SNF risk as "High", "Intermediate" or "Low". �Hover bubble displays AI summaries of H&P note that provide end-users with relevant context.

End-Users: case managers, physical/occupational therapists, providers, medical directors
When: Live for general internal medicine inpatients; planning expansion to other services

Enhancing the prediction of hospital discharge disposition with extraction-based language model classification. �Small, W.R., Crowley, R.J., Pariente, C., Eaton K, Jiang L, Oermann E, Aphinyanaphongs Y. npj Health Syst. 3, 4 (2026). https://www.nature.com/articles/s44401-025-00059-8#citeas

18 of 37

Epic Integration: AI Predictions of Skilled Nursing Facility Discharges from Clinical Documentation

Running in batch mode every morning on 18,000 patients.

20 of 37

What defines a quality note? The 5Cs��

“C”	Description
COMPLETE	Does this note contain an appropriate history, physical examination, and plan?
CLINICAL ASSESSMENT & REASONING	Does this note contain a differential diagnosis OR a commitment to a clear diagnosis with a trajectory on how the patient is doing?
CONTINGENCY PLANNING	Does this note contain an articulation of what milestone must be fulfilled to change management or Fulfill a discharge?
CONCISE	Does this note contain only pertinent data/information?
CORRECT	Is this note internally consistent AND free of critical deficiencies?

21 of 37

Powerful Dashboard Promotes Transparency��

Runs daily in batch mode on about 5,400 notes.
Unit directors estimated to save on avg 6h per month using the dashboard
To populate dashboard via direct physician review would take ~10,000 clinician hours per year

27 of 37

More Use Cases!

28 of 37

GPT to classify in Immunosuppressant Reconciliation (Collaboration with Dr. Paawan Punjabi)

Clinical Scenario	Verbiage	Intent to Order today?	Alert Provider if Order NOT placed.
55 y/o M w cirrhosis who underwent liver transplant	"Cellcept 1000 mg BID IV starting today"	Yes	Yes
48 y/o M w Acute Lymphatic Leukemia presents for Bone Marrow Transplant	“Tacrolimus Start Day +5”	No	No (intent is future state)
14 y/o with new-onset AKI	"Should she have proliferative lupus nephritis I would add...CellCept"	No	No (intent is conditional)

Order today: NO

Do not alert clinician.

Classify the intent of verbiage in a clinical note as to whether the medication will be ordered today.

17 fires between 1/3 to 2/3 for any immunosuppressant across the system.

29 of 37

Physician Ambient Documentation: Abridge

AI platform for providers to capture patient conversations
Available to ambulatory providers since September 15!
Automates transcription and summarization of patient-provider interactions

�Why are we investing in Abridge?

Reduces administrative burden for healthcare prov
Supports better work-life balance for healthcare professionals
Allows more focus on the patient

Goal: 1,000 providers using Abridge by December 2025

Active Physicians by Month

July 2025

August 2025

September 2025

October 2025

Confidential – Do Not Distribute

30 of 37

Patient Digital Experience – Patient Friendly Discharge Summary Pilot (Colloboration with Dr. Feldman, Dr. Zaretsky)

https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2815868

31 of 37

Case timeline

High risk

"likely"

High risk

"definitely"

AI monitoring and summarizing from real-time ED data

Safety & Quality - ED High Risk Copilot: The alert starts a conversation (Colloboration with Dr. Genes, Dr. Simon, Dr. Femia)

Candidate Conditions:

Necrotizing Fasciitis
Cardiogenic Shock & Tamponade
Septic Shock
Aortic Dissection

32 of 37

Summarization – ED visit summaries for post-visit calls (Collaboration with Dr. Major and Dr. Silberlust)

ED patients receive a phone call after their visit. It takes the nurses a few minutes to review each chart to prepare.

33 of 37

Quality – Communicating ED Incidental Findings (Collaboration with Dr. Kar-mun Woo, Dr. Greg Simon)

34 of 37

Existing: Prioritizing high-risk messages to Physicians for fastest review (Collaboration with Dr. Major, Dr. Austrian).

Hi Dr xxx., xxx has been fussy for the past two days. She’s been crying almost every hour. She has no fever though. We tried giving her gripe water, which usually works but this time her crying is non-stop. We also noticed her cry has changed with episodes like she’s gasping for air. This is the first time we’ve noticed this. Should we take her to the ER at this point?

Model score: 78.98

RN called – Advised to go to ER

Subject: Fussiness

Patient age: 3 months

Prioritizing clinically concerning, high-acuity Epic InBasket messages

35 of 37

Efficiency – Speeding Connection with Physicians (Dr. Szerencsy)

Confidential – Do Not Distribute

36 of 37

Patient #3 scheduled for surgery on 7/17, required cardiac clearance, saw the cardiologist, BUT GPT detects no documentation of clearance, escalate. (Dr. McKeever)

Surgery Details

All appointments in last 90 days

'.Chief Complaint and History of Present Illness: Ms. X is a 42 y.o. female who comes for a follow up visit to discuss test results.Patient is a 42 year old woman non-smoker with h/o Printzmetal angina. CCTA showed no CADStarted using Norvasc 2.5mg po daily. No episodes of CP since. However, does note occasional episodes of low BP.S/p echo and ETT

Assessment / Plan: Diagnosis1. Other chest pain: ETT is WNL. Started amlodipine 2.5 mg daily. No more CP since. However, does note occasional episodes of low BP. Can take 1/2 tab if needed if SBP<110-refer to GI to r/o esophageal spasm as the cause for CP2. SOB (shortness of breath) on exertion : 2D echo shows normal LVEF, no significant valvular disease. She has gained some weight recently. Advised to exercise daily3. Mixed hyperlipidemia :Low fat diet and regular walk daily.4. Overweight 5. Primary insomnia 6. Hypothyroidism due to medication: She is on replacement therapy.

Tests Ordered This Visit:No orders of the defined types were placed in this encounter. Medications Ordered This Visit: Patient was made aware of her medical status; all questions, concerns and alternatives were outlined and discussed in details; patient was informed about risks and outcomes of untreated conditions or poor compliance.Special instructions, diet, exercise, weight control and plan of care were discussed with patient who verbalized his/her understanding and agreed with above. CQM CMS 69 / NQF 0421 - Adult BMI Screening and Follow Up: The patient\'s BMI is above normal. Counseled patient regarding BMI, healthy eating, portion control and exercise importance of diet to maintain a healthy weight.Compliance adherence was discussed [ x ] Risks of refusing the recommended test/treatment that include but not limited to death, life threatening 5 arrhythmia, stroke, heart attack, limb gangrene and amputation were outlined to the patient. Patient understands. Return in about 6 months (around 12/24/2024). Patient seen and evaluated with PA-Milana Bubis Tariq Jamil, MDAssistant professor of medicine/cardiology at NYUBoard certified in cardiology, nuclear cardiology, echocardiography and RPVI.

1 of 37

2 of 37

3 of 37

4 of 37

5 of 37

6 of 37

7 of 37

8 of 37

9 of 37

10 of 37

11 of 37

12 of 37

13 of 37

14 of 37

15 of 37

16 of 37

17 of 37

18 of 37

19 of 37

20 of 37

21 of 37

22 of 37

23 of 37

24 of 37

25 of 37

26 of 37

27 of 37

28 of 37

29 of 37

30 of 37

31 of 37

32 of 37

33 of 37

34 of 37

35 of 37

36 of 37

37 of 37