AI-Enabled Chatbot Replicates Efficacy of SOC Approaches in Aiding Receipt of Cancer Genetic Testing

A chatbot-based approach was found to be equivalent to standard-of-care (SOC) approaches in terms of the completion of pretest cancer genetic services and genetic testing among individuals who were eligible for cancer genetic testing, according to data from the interventional BRIDGE trial (NCT03985852).1

Findings published in JAMA Network Open demonstrated that individuals in the chatbot group (n = 1554) and the SOC group (n = 1519) completed genetic testing (12.3% vs 13.6%) and completed pretest genetic services (25.7% vs 23.8%), representing respective estimated differences of –1.3% (95% CI, –3.7% to 1.1%; P = .002) and 2.0% (95% CI, –1.1% to 5.0%; P < .001). Patients in both arms also ordered genetic testing (15.0% vs 18.1%) and began pretest services (31.0% vs 29.2%) at similar rates, conferring estimated differences of –3.1% (95% CI, –5.7% to –0.5%; P = .16) and 1.7% (95% CI, –1.5% to 5.0%; P < .001), respectively.

“There’s greatly increasing demand for pretest cancer genetic services for unaffected patients, [including individuals] who might not be seen in a cancer center as a part of a diagnosis but have a family history of cancer suggesting that they may be eligible for cancer genetic testing,” Kimberly Kaphingst, ScD, the lead author of BRIDGE, said in an interview with OncologyLive. “The rationale [for BRIDGE] was to test whether we could use a chatbot as part of a genetic counseling workflow to supplement the role of certified genetic counselors, saving their time for the return of the complicated results, the pathogenic findings, and so forth with the chatbot by doing some of the automated education.”

Kaphingst is the director of cancer communication research as well as a professor and an associate chair in the Department of Communication at the Huntsman Cancer Institute at the University of Utah in Salt Lake City.

BRIDGE was an equivalence trial conducted from August 2020 to August 2023. The study enrolled patients aged 25 to 60 years who underwent a primary care visit in the University of Utah Health or NYU Langone Health systems over the prior 3 years. Patients needed to be eligible for cancer genetic testing according to the modified National Comprehensive Cancer Network guidelines based on their family history in the electronic health record (EHR). To be included in the study, patients also had to speak English or Spanish, had not received prior cancer diagnoses other than nonmelanoma skin cancer, and had not undergone previous genetic counseling or testing related to hereditary cancer.

A pool of eligible patients was identified using GARDE, an open-source platform that extracts and examines information related to family history of cancer from the EHR. Eligible patients were randomly assigned 1:1 to the chatbot intervention group or the enhanced SOC control group.

In the chatbot intervention group, patients received a patient portal message recommending genetic services that included a hyperlink to pretest genetics education administered via chatbot. The chatbot used text, images, and video to convey key information delivered during SOC pretest genetic counseling appointments. After a video message from the lead genetic counselor at the patient’s center, the patient progressed through a core set of scripted information and was able to ask for additional information on predetermined topics or ask open-ended questions; open-ended questions that did not have a scripted response were emailed to the genetic counseling team.

Patients received a reminder message in 1 week in the event of an incomplete chat in addition to up to 2 follow-up calls from a genetic counseling assistant. At the end of the script, patients were offered the choice to continue with genetic testing and were then contacted by a genetic counseling assistant to confirm their decision and collect additional information.

“The chatbot that we tested in this study was an entirely scripted, rules-based chatbot, which is designed to have interactive conversations with patients,” Kaphingst explained. “It is artificial intelligence [AI] enabled, so it was able to answer questions using a natural language processing part of the platform. It did not [use] any generative AI.”

In the enhanced SOC group, patients received a patient portal message that recommended they receive genetic services, encouraging them to contact the genetics clinic at their site to schedule a pretest genetic counseling appointment. Those who did not respond received a reminder message a week later and 2 follow-up calls from a genetic counseling assistant. Patients who were interested in and eligible for an appointment were scheduled for one with a certified genetic counselor; most appointments were performed via telephone.

Samples from patients in both arms who opted for genetic testing were analyzed via pan-cancer, multigene panels for cancer susceptibility of approximately 34 to 36 genes. The primary outcomes of BRIDGE were the completion of pretest cancer genetic services, such as a pretest genetics education discussion or the scheduling of a pretest genetic counseling appointment, and the completion of genetic testing. Secondary outcomes included initiating pretest cancer genetic services and ordering genetic testing.

The baseline patient characteristics were well balanced between the chatbot and SOC arms; the mean patient ages were 43.5 years (range, 26-63) and 44.1 years (range, 26-63), respectively. Most patients in both groups were female (74.1% vs 71.7%) and White (74.9% vs 75.1%), spoke English (98.6% vs 98.8%), were seen at NYU Langone Health (53.0% vs 53.1%), and resided in an urban area (96.3% vs 96.5%).

Additional findings from BRIDGE revealed that patients in the chatbot (n = 731) and SOC (n = 713) groups from the University of Utah Health system completed genetic testing (14.2% vs 14.2%), completed pretest genetic services (28.0% vs 26.5%), ordered genetic testing (17.2% vs 20.9%), and began pretest services (33.2% vs 32.5%) at similar rates. In the NYU Langone Health system, the rates of completing genetic testing (10.6% vs 13.0%), completing pretest genetic services (23.7% vs 21.3%), ordering genetic testing (13.0% vs 15.6%), and beginning pretest services (28.9% vs 26.3%) were also similar in the chatbot (n = 823) and SOC (n = 806) groups.

“Having these types of tools to help supplement the genetic counseling process is going to be extremely helpful,” Whitney F. Espinel, MS, CGC, a coauthor of BRIDGE from the Breast Cancer Risk Clinic at Huntsman Cancer Institute at the University of Utah, said in the interview. “We can help the traditional process [with] a pretest and a posttest counseling session. We can utilize chatbots [during the] pretest and cut out some of that need for the genetic counselor up front and still utilize our skills to interpret test results and connect patients to resources. There are lot of unique ways that we can implement [chatbots] to see more patients and expand access to populations we haven’t been able to access before.”

Expanding Chatbot Access to Additional Patient Populations

Kaphingst and her coinvestigators are planning several additional studies to further evaluate the utility of the chatbot investigated in BRIDGE for other patient populations. “We’re thinking about how to improve access to these sorts of automated tools for Spanish-speaking patients, patients who speak other languages, and to reach out to rural and frontier populations who may have more limited broadband access,” she said.

In January 2024, the Huntsman Cancer Institute announced that Kaphingst and her coinvestigators had received additional funding from the National Cancer Institute to study the chatbot in Spanish-speaking women.2

“[This is] an area I’m enthusiastic about, because I think you can really incorporate chatbots into the places that individuals who may not speak English as a primary language much easier,” Espinel added. “We’re utilizing a cancer screening mobile clinic for that and there’s a lot of ways we can bring it to the community instead of them having to come to us, which is often a barrier [to care].”

References

1. Kaphingst KA, Kohlmann WK, Lorenz Chambers R, et al. Uptake of cancer genetic services for chatbot vs standardof- care delivery models: the BRIDGE randomized clinical trial. JAMA Netw Open. 2024;7(9):e2432143. doi:10.1001/ jamanetworkopen.2024.32143
2. Using chatbots to advance Spanish-speaking patient outreach. News release. University of Utah Huntsman Cancer Institute. January 9, 2024. Accessed November 4, 2024. https://healthcare.utah.edu/huntsmancancerinstitute/ press-releases/2024/01/using-chatbots-advance-spanishspeaking- patient-outreach