AI-driven app shows ‘benefits across the board’ for managing cancer symptoms through diet
Click Here to Manage Email Alerts
Key takeaways:
- The majority of patients with cancer who used the AI nutrition program self-reported benefits.
- Next steps include integrating other languages, comparing program with human dieticians.
CHICAGO — Most individuals with cancer who used an AI-based nutritional platform reported the application helped them regulate their diets and improved their quality of life, according to study results presented at ASCO Annual Meeting.
Findings simultaneously published in JCO Clinical Cancer Informatics detailed how the text messaging service, named Ina, provided instantaneous advice to users on foods to eat and avoid for managing symptoms, meal plans and recipes, expert tips, and more.
“We were impressed by the magnitude with which patients felt that the platform was helping them,” Keshav Goel, MSc, medical student at Perelman School of Medicine at University of Pennsylvania, told Healio.
“The vast majority of patients who filled out the survey felt the platform provided significant self-symptom management help, and the majority stated that they were utilizing the platform’s nutrition advice, recipes and more,” Goel added. “Most users reported benefits across the board from all the measurements that we were looking at.”
Hallucination-free AI
In the U.S., there is only one registered dietician for every 5,800 people with cancer, but individuals with cancer can have numerous nutritional concerns, including appetite issues, weight loss and cachexia, among others, Goel said during his presentation.
“These can result in poor outcomes, specifically prolonged and repeated hospitalizations, increased treatment-related toxicity, reduced response to cancer treatment, impaired quality of life and overall worse cancer prognosis,” he said.
Digital solutions could provide more access to services and may help financially because patients are not always reimbursed for nutritional care.
The Ina platform currently contains over 114,000 guidelines-based interventions that were developed by a team of oncology-credentialed registered dietitians, registered nurses, oncologists, and social workers based on clinically validated guidelines, experience and training. The platform's matching algorithms identify the most appropriate intervention for each individual based on their symptoms, comorbidities, medications and other clinical variables.
A user can text Ina and get a response instantly.
“The technology continuously learns from interactions with patients,” Goel said during his presentation.
“This is a machine-learning system that uses natural language processing and, importantly, it is not large language model-based,” he added. “There is no issue with hallucinations as might be seen with other AI systems. Instead, Ina is a self-contained, closed system with experts retaining total control over the system for both quality and for safety.”
The issue of hallucinations Goel referred to is a phenomenon where the output of a large language model produces information that is nonsensical or incorrect.
If Ina did not have enough data to provide an automated answer to a user query, the query was triaged to a human registered dietician. Human-generated registered dietician responses took an average of 2.5 minutes.
Goel provided multiple examples of Ina’s abilities. He discussed potential messages regarding user nausea and mouth sores, and Ina responded with foods to eat and avoid, as well as recipes. Ina answered another message about antioxidant supplements with up-to-date information.
User characteristics
Researchers implemented Ina across 25 cancer advocacy organizations and had 3,310 users (median age, 57 years; 73% women). Half of the users had stage III or IV cancer and at least two additional long-term health conditions. The most common cancer types among users included genitourinary, breast, gynecologic, gastrointestinal and lung.
Large percentages of users had nutritional issues, including overweight or obese (58%), food allergies or intolerances (42%) and special diets (52%).
Additionally, 83% of patients had at least one barrier to healthy eating, including cost (34.2%), time (27.3%) and nutrition knowledge (26.8%).
Users had the opportunity to answer six different types of surveys — baseline, symptom, treatment, dietary preference, quality of life and satisfaction.
Patients rate experience
Users sent a total of 19,024 messages,2,255 sent at least one text message, and 910 opted out of the system.
Patients used Ina for an average of 8.8 months, with researchers reporting a 94% satisfaction rate.
Additionally, 98% found the tips to be helpful, 88% felt the AI assisted them in managing symptoms and 84% used advice to guide their diets. A full 82% of users felt their quality of life improved because of Ina and 47% used the recipes it provided.
Goel noted several study limitations, including the self-selected population, self-reported outcomes and users may have had potential bias because they engaged the platform.
“There are opportunities to further tailor recipes and guidance based on context, social determinants of health, cultural and ethnic preferences, and language,” Goel said. “We believe that context is the difference between success and failure in any behavior modification program, so we are working with experts and local communities to understand what they need and how they want it delivered.”
Next-gen Ina
Researchers are continually enhancing the platform to incorporate new features. Ina has been translated into Spanish and will be extensively tested within the diverse Hispanic community prior to being made broadly available beyond research purposes. It is currently at a fifth-grade reading level.
“We would love to examine how well it can meet the needs of Hispanic communities that predominantly speak Spanish and may have barriers to access health care because of their primary language,” Goel told Healio.
“Where we think that the platform may really prevail in not only being able to meet this unmanned need, but also, the Hispanic population is incredibly diverse, whether they hail from Mexico, South America, Spain and Portugal, or different areas,” he added. “We believe the platform might be able to provide context-dependent, linguistically sensitive advice that might be able to reach these different populations.”
Ina could provide “cultural literacy” and “cultural competency” in nutrition that some dieticians may not possess.
“Although American medicine is getting a lot better about this and is taking huge strides and making sure that American physicians and nurses come from more diverse backgrounds ... there’s still a lot of ground that needs to be covered,” Goel told Healio. “If a patient has someone as their health care provider who’s great, but may not come from their particular background, they may not be able to provide very culturally sensitive advice when it comes to recipes. That is somewhere we believe Ina could make headway.”
Over the next several months, researchers plan to conduct a validation study with Ina and registered nutritionists and determine if the AI-powered service could produce advice equivalent to that of human practitioners.
They also want to see if Ina performed better in certain areas, such as with patients dealing with fatigue vs. mouth sores.
“Our goal is to have the validated study completed by the end of 2024 to prove that Ina is as robust in delivering nutrition advice to patients with cancer as oncology-credentialed registered dieticians and health care providers,” Goel told Healio. “We would also hope that Ina has been proven through research to provide culturally competent advice, reach patients regardless of geography or background, and be shown to provide meaningful health benefits.”
References:
- Buchan ML, et al. Abstract 1500. Presented at: ASCO Annual Meeting; May 31-June 4, 2024; Chicago.
- Buchan ML, et al. JCO Clin Cancer Inform. 2024;doi:10.1200/CCI.24.00085L.
For more information:
Keshav Goel, MSc, can be reached at keshav.goel@pennmedicine.upenn.edu.
Collapse
Buchan ML, et al. Abstract 1500. Presented at: ASCO Annual Meeting; May 31-June 4, 2024; Chicago.
Click Here to Manage Email Alerts