1 of 28

BRIDGE(ing) the Gap: Leveraging technology in genetic services

Rachelle Chambers, MS, CGC

12/3/2025

Perlmutter Cancer Center

2 of 28

Disclosure: This work was supported by National Cancer Institute, National Institutes of Health, U01CA23282603-S1 and U24CA204800�

3 of 28

Objectives

Review the BRIDGE study and the development of chatbots for genetics education

Describe the process, challenges and opportunities of automated algorithms for identifying patients who are eligible for hereditary cancer testing

Describe the role of engagement with the health care system in patients’ response and receptiveness to outreach about genetic service

Perlmutter Cancer Center

4 of 28

Current State of Genetic Testing, Counseling & Services (in the US)

Perlmutter Cancer Center

4

Advances in technology

Breakthroughs in Genomic Precision Medicine

Cost

public awareness & interest

Increased demand

Exponential growth

Institutional awareness

Genetic counseling/ testing

- Time intensive

- Complex

- Manual

- Precise

Bottlenecks, wait times, patient impact, non-genetic provider shift, etc.

5 of 28

Utilization of technology in Genetic Services

Goal: Leverage technology to develop sustainable and scalable workflows to enhance accessibility, efficiency, and impact of genetic counseling and testing services.

Maintain high quality of patient care & empower patients and providers

Technology currently utilized in healthcare systems to streamline care

Identification of at-risk patients
Pre- and post test genetic education and return of results (ROR)
Integration of genomic testing in the EHR
High risk patient navigation & management

Perlmutter Cancer Center

6 of 28

Automated tools/strategies in genetics services

As of 2023, 87 digital tools for providing some aspect of genetic services

International Cancer Syndrome Consortium (ICSC)

Convened to describe, demonstrate and compare digital tools for delivering genetic services

Domain 1:

Tool Development

Domain 2: Process and context

Domain 3: Evaluation

7 of 28

BRIDGE Study

Division/Department Name

7

8 of 28

Division/Department Name

8

9 of 28

Enhanced standard of care control arm

Outreach through patient portal
Standard of care genetic counseling

Eligibility screening

Randomization

Chatbot intervention arm

Outreach through patient portal
Pre-test education and return of negative results by chatbot
Positive and VUS results returned by genetic counselor

4-week questionnaire

Satisfaction with counseling and responses to results

12-month questionnaire

Adherence to clinical recommendations

Aim 3

Aim 2

Aim 1

Study design

N=445,910

24 to 60 years old
UHealth or NYU

n=22,208 (5.0%)

10 of 28

Division/Department Name

10

Rules-based algorithm (clinical decision support)

Population Health Management dashboard

Allows for managing patient contacts

Chatbot

Scripted chatbot

Completed chats returned to EHR as a clinical encounter

Del Fiol G et al. JCO Clin Cancer Inform 2020, Bradshaw RL J Am Med Inform Assoc 2022

GARDE-Genetic Cancer Risk Detector

11 of 28

Script based on genetic counseling standard of care

All patients see core content

Can select options for additional content

Natural language processing allows patients to ask questions

Scripted bank of responses

Chatbot approach

12 of 28

Chatbot approach

Patients indicate interest in testing at end of pre-test chat

Follow up with genetic counseling assistant

13 of 28

Enhanced standard of care control arm

Outreach through patient portal
Standard of care genetic counseling

Eligibility screening

Randomization

Chatbot intervention arm

Outreach through patient portal
Pre-test education and return of negative results by chatbot
Positive and VUS results returned by genetic counselor

Aim 2

Aim 1

Study design

Primary outcomes

Completing pre-test genetic services
Completing genetic testing
Hypotheses of equivalence

Secondary outcomes

Starting pre-test genetic services
Ordering genetic testing
Hypotheses of equivalence

14 of 28

14

BRIDGE Outcomes

Chatbot intervention N=1554
1017 (69%)
481 (31%)
400 (26%)
233 (15%)
Total: 13% tested

Enhanced Standard of Care N=1519
1023 (67%)
444 (29%)
361 (24%)
275 (18%)
Total: 14% tested

Open portal message

Started/Scheduled

Completed

Ordered test

Test complete

Kaphingst KA et al. JAMA Netw Open 2024

CITY OF HOPE

15 of 28

Identification of high-risk patients

Division/Department Name

15

16 of 28

ALGORITHM CRITERIA – BREAST CANCER (GARDE)

FHx of BRCA1/2, CHEK2, ATM, PALB2, TP53, PTEN, CDH1, Cowden syndrome, or Li-Fraumeni syndrome 1^st or 2^nd degree relative with breast cancer AND age of onset < 45 1^st or 2^nd degree relative with ovarian OR pancreatic cancer Three or more 1^st or 2^nd relatives with breast or prostate cancer on same side of family Breast cancer in a male 1^st or 2^nd degree relative Ashkenazi Jewish ancestry and any family member with breast, ovarian, prostate, or pancreatic cancer

*Adapted from NCCN guidelines for breast cancer risk reduction

17 of 28

Automated patient identification through the EHR�

Scale
Timeframe
Required no change to primary care workflows
Process fully integrated into the EHR

Division/Department Name

17

18 of 28

Patients meeting criteria based on EHR data�

Division/Department Name

18

Demographics	NYU Population	NYU Criteria	Utah Population	Utah Criteria
Met >1 NCCN family hx criteria	-----	6%	----	4%
Female	57%	74%	64%	84%
Male	43%	26%	36%	16%
White	55%	65%	71%	80%
Black	13%	9%	3%	1%
Asian	6%	4%	5%	2%
Hispanic	1%	1.5%	14%	11%
English preferred language	89%	96%	92%	97%
Spanish preferred language	6%	2%	5%	2%
Age (Mean)	43	44	40	41

Chavez-Yenter D et al. JAMA Netw Open 2022

Disparities in family history collection

19 of 28

Disparities in family history collection�

Division/Department Name

19

Availability of any family history information

Availability of cancer family history information

Availability of age of onset

Availability of type of cancer

Availability of type of relative

20 of 28

EHR family history vs. Population-based data�

133,764 Utah primary care patients also in the UPDB

20

Patient with more family information available, more likely to meet criteria

Del Fiol G et al. JCO Clinical Cancer Informatics 2025

In order to better understand how well the EHR family history reflects “real” family history, we compared family history in the EHR to these same patient’s family history in the Utah population database. The Utah Population Database is a statewide resources that links together state vital records like birth and death certificates to the state cancer registry. I would like to some air quotes around “real” family history because this is still limited the relatives and cancers that were diagnosed in Utah.

We added the additional family information available from the UPDB, the % of primary care patients meeting criteria when from 4% based on EHR data only, up to 9%. The factor that was most significant for determining if a patient would meet criteria was having more family information. About 20% of those patients who had more than 10 relatives in the UPDB met criteria.

21 of 28

Criteria met by algorithm�

Most people meet criteria due to a relative with ovarian or pancreatic cancer, regardless of family size

Complex criteria, particularly Lynch syndrome/polyposis, are difficult to identify

21

22 of 28

Machine learning

Data driven

Can handle complex relationships that are hard to express in a “rule”

Have been used for screening for familial hypercholesterolemia in the EHR, less so for hereditary cancer

Rules-based algorithms

“if-then” rules

Natural language processing minimal

Difficult to account for context, limited information, and uncertainty

Rules-based algorithms vs. Machine learning/AI

Harris WR J Am Med Inform Assoc 2025, Bradshaw RL, et al. J Biomed Inform 2024

And this is a draw back to rules based algorithms. These are if-then rules. We did incorporate NLP to evaluate comments such as noting that mom was diagnosed in with breast cancer in her late 40s or early 50s. However, many comments could not be matched to a rule. There was recent review of automated tools for identification of patients meeting criteria for Tier One conditions. There were two published algorithms for identifying patients, ours and one from the Healthy Nevada/Renown health study. Both were rules-based. There was several examples of machine learning models that had been used to identify patients with familial hypercholesterolemia. Machine learning/AI can look at more complex relationships. Such as a patients lab values or resonse to medications. To make a classification.

It will be helpful to see how AI can be used in patient identification. Rules-based algorithms do not account for context, limited information and uncertainty.

23 of 28

Patient Engagement

23

24 of 28

24

BRIDGE Outcomes

Chatbot intervention N=1554
1017 (69%)
481 (31%)
400 (26%)
233 (15%)
Total: 13% tested

Enhanced Standard of Care N=1519
1023 (67%)
444 (29%)
361 (24%)
275 (18%)
Total: 14% tested

Open portal message

Started/Scheduled

Completed

Ordered test

Test complete

Kaphingst KA et al. JAMA Netw Open 2024

CITY OF HOPE

25 of 28

Patient Engagement

Zhong L, et al Health Serv Res 2025, Bather JR Health Lit Res Pract 2025

Predictors of opening a portal message and starting genetic services
Number of previous portal logins
Having a recorded primary care provider
More primary care visits in the previous 3 years
Low socioeconomic vulnerability*

Factors that did not impact opening a portal message or starting genetics services
Most sociodemographic characteristics Age Race Ethnicity Urbanity
Clinical characteristics
Intervention arm

*2018–2022 American Community Survey 5-year estimates

Factors that were associated with looking at the portal message and starting genetics services, either starting the chat or scheduling an appointment were how often they open portal messages in general, having a primary care provider, having more visits, and having low socioeconomic vulnerability. Overall, these can be summarized as those patients who open portal messages and go to the doctor are more likely to open this message. We saw that those patients with less social vulnerability were also more likely to have a test ordered and complete testing compared to patients with high social vulnerability.

Going into the project we concerns about whether there would be populations that were not comfortable using the portal or patients tools, but we did not note differences in most demographic factors, clinical characteristic where defined as meeting only one testing criteria versus multiple, or whether they were being offered an appointment versus a chatbot.

26 of 28

Summary and Future Direction

Chatbot tools can be a core strategy to improve access to genetic counselling services and testing
Clinical decision support tools and other technology are available to streamline the identification process, genetic education/testing, support providers, and navigate high-risk patients in healthcare systems
However, there are complexities to these tools and more research is needed to understand the impact of technology on access and health care disparities

27 of 28

Saundra Buys, MD

Huntsman Cancer Institute

Kimberly Kaphingst, ScD

Huntsman Cancer Institute

Meenakshi Sigireddi, MD

NYU Langone Health

Primary Care

Mike Flynn, MD

University of Utah

Rachel Hess, MD

University of Utah

Biomedical Informatics

Guilherme Del Fiol, MD, PhD

University of Utah

Ken Kawamoto, MD, PhD, MHS

University of Utah

Rick Bradshaw, PhD, MS

University of Utah

Devin Mann, MD, MS

New York University

Javier Gonzalez

New York University

Broadening the Reach, Impact, and Delivery of Genetic Services

Advisory Board

Kola Okuyemi, MD

University of Utah

Will Dere, MD

University of Utah

Jeff Botkin, MD

University of Utah

John Barrett, MD

University of Utah

Neli Ulrich, PhD

Huntsman Cancer Institute

Jessica Everett, MS, CGC

New York University

Genetics

Research Team

Rachel Monahan

Perlmutter Cancer Center

Molly Volkmar

Huntsman Cancer Institute

Lauren Kaiser-Jackson

Huntsman Cancer Institute

Melody Goodman, PhD

New York University

Jemar Bather, PhD

New York University

Adrian Harris, MS

New York University

Daniel Chavez-Yenter, MPH

University of Utah

NCI U01CA232826

Wendy Kohlmann, MS, CGC

Huntsman Cancer Institute

Rachelle Chambers, MS, CGC

NYU Langone Health

Sang Lee

NYU Langone Health

Amanda Gammon, MS, CGC

Huntsman Cancer Institute

Sarah Colonna, MD

Huntsman Cancer Institute

Josh Schiffman, MD

Huntsman Cancer Institute

28 of 28

Contact:Rachelle.Chambers@nyulangone.org

Perlmutter Cancer Center

Thank you