Leading Cancer Center — Boston

Dana-Farber Cancer Institute

A multi-year collaboration combining predictive risk stratification and LLM-powered clinical summarization to improve goal-concordant care, support serious illness conversations, and generate real-world evidence for oncology patients.

BRIDGE-SIC
Pilot Program
Launched October 2025
90-day
Mortality Prediction
For patient selection
Seconds
SIC Summary Generation
vs. minutes of manual review
20,000
Medicare Patients
In JAMA-published lung cancer study

The Challenge

Patients with cancer often receive care near the end of life that may not align with their preferences. Serious illness conversations between patients and clinicians improve well-being and reduce aggressive care — but documentation of these conversations is buried in voluminous free-text medical records.

When patients are admitted to the hospital, inpatient teams rarely have visibility into prior goals-of-care discussions that happened in outpatient settings. The result: care decisions made without awareness of what patients actually want.

The BRIDGE-SIC Program

Better Real-time Information on Documentation of Goals of care for Engagement in Serious Illness Communication — a pilot study combining HDAI’s predictive platform with LLM-powered summarization to surface patient preferences at the point of care.

Predictive Patient Selection

HealthVision's validated risk models stratify patients at highest risk of 90-day mortality, identifying those most likely to benefit from goals-of-care awareness during hospitalization.

LLM-Powered SIC Extraction

HDAI's secure LLM infrastructure identifies and extracts prior serious illness conversations from across the medical record, generating summaries in seconds rather than minutes of manual chart review.

Real-Time Clinical Delivery

AI-generated summaries are shared with both inpatient and outpatient clinical teams when eligible patients are admitted, prompting awareness of preferences and enabling goal-concordant care.

From the Research Team

“A key challenge in ensuring high-quality, goal-concordant care in acute hospitalization settings is effectively identifying and communicating patient preferences to clinical teams providing inpatient care. LLMs are helping us to drastically reduce the time and effort needed to search for documentation of these conversations.”

Christopher Manz, MD

Assistant Professor in Medical Oncology, Dana-Farber
Sponsor-Investigator & Co-Principal Investigator, BRIDGE-SIC

“We believe deeply in the goal of every patient having their wishes front and center. We are on track to show that LLMs for real-time identification of Serious Illness Conversations work and represent a major breakthrough in our field.”

Charlotta Lindvall, MD, PhD

Palliative Care Physician & Researcher, Dana-Farber & Harvard Medical School
Co-Principal Investigator, BRIDGE-SIC

Published Research

The HDAI–Dana-Farber collaboration has generated multiple peer-reviewed publications in leading oncology and medical informatics journals.

JAMA Network Open2021

Association Between First-Line Immune Checkpoint Inhibition and Survival for Medicare Patients With Advanced Non-Small Cell Lung Cancer

Real-world evidence study of nearly 20,000 Medicare patients examining immuno-oncology effectiveness in older populations — finding shorter survival times than clinical trials suggested.

JCO Oncology Practice2025

Feasibility Study for Using Large Language Models to Identify Goals-of-Care Documentation at Scale in Patients With Advanced Cancer

Demonstrating that LLMs can reliably identify serious illness conversation documentation across large patient populations.

Journal of Pain and Symptom Management2025

Large Language Models to Identify Advance Care Planning in Patients With Advanced Cancer

Validating LLM-based extraction of advance care planning documentation from unstructured clinical notes.

The Technology

Dana-Farber has implemented HDAI’s HealthVision platform viewable within the EHR, applying validated prediction models for patient stratification alongside a secure, performant LLM infrastructure that creates clinical summaries in seconds.

  • Validated 90-day mortality prediction models for trial patient selection
  • Secure LLM infrastructure with governance wrapper for explainable, traceable output
  • EHR-embedded delivery of AI summaries to inpatient and outpatient teams
  • Hybrid AI approach — combining targeted NLP with LLM capabilities for accuracy and efficiency

Bring Goal-Concordant Care to Your Organization

Learn how HealthVision’s predictive platform and LLM capabilities can support serious illness care, palliative identification, and clinical decision-making.

Request a Demo