Abbey AI

What is the name of your organization?

Genetic Alliance

What is the name of your solution?

Abbey AI

Provide a one-line summary of your solution.

We offer an AI-powered rare-disease patient-centric chatbot and an automated web browser to lessen the cognitive load for patients and caregivers.

In what city, town, or region is your solution team headquartered?

What specific problem are you solving?

Our solution addresses the significant challenges faced by the rare disease community, particularly in accessing accurate information, coordinating care, and connecting with support resources. The immense scale of this problem can be categorized into, but not limited to the following:

1. Rare disease is globally prevalent. Rare diseases collectively affect over 300 million people worldwide, or 3.5-5.9% of the global population. This equates to 1:17 to 1:29 people living with a rare disease. Language barriers often prevent patients from gathering helpful knowledge and outreach.

2. Patients are faced with prolonged diagnostic delays. The average rare disease patient consults with 7.3 physicians over 4.8 years before receiving an accurate diagnosis. Due to the rarity of their condition and limited knowledge among healthcare professionals, patients are faced with biases. This prolonged "diagnostic odyssey" exacerbates medical and emotional burdens on patients and families. 

3. Fragmented care leads to inconsistencies in treatment. Once diagnosed with a rare disease, the “odyssey” does not end. 78% of rare disease patients see 3 or more different specialists for their condition, with 38% seeing 5 or more. This lack of coordination increases the burden on families to manage complex care.

4. There is a lack of centralized and accurate information. With over 7,000 identified rare diseases, 72% of which are genetic and 70% starting in childhood, there is a severe lack of centralized, accurate information resources. 72% of patients and caregivers struggle to find reliable and up-to-date information about their condition. This issue presents itself in research as well, where hurdles such as limited funding, small patient populations, and a lack of natural history data hinders progress in rare disease research. 

5. People are not connected to appropriate programs. Only 15% of rare disease patients are connected to their disease-specific support groups or patient advocacy organizations that could provide valuable assistance. The same problem is seen in clinical trials, where there are logistical challenges with patient recruitment.

Our AI-powered platform aims to tackle these interconnected issues by providing personalized information access in 50+ languages, automating care coordination tasks, facilitating connections to support resources, and enhancing overall accessibility for the rare disease community. By addressing these core challenges, we hope to significantly improve quality of life and health outcomes for millions of individuals and families affected by rare diseases globally. 

What is your solution?

Our solution is an AI-powered patient-centered chatbot and automated AI web agent designed to automate information retrieval, knowledge extraction, and web browsing for rare diseases, addressing the significant challenges faced by patients and caregivers. It will leverage Genetic Alliance's existing tools and the open-source solutions of its network partners. We have named it Abbey AI after Abbey Meyers, the founder of NORD and Genetic Alliance who led the initiative for rare diseases in the US.

Key Features:

1. Information retrieval 

   1. The AI chatbot accesses scientific papers and trusted web sources to provide accurate, up-to-date information on rare diseases. It uses novel low-data citation methods, ensuring reliability even in understudied areas.

   2. Users can click on any sentence in the AI-generated response to view a ranked list of citations according to a relevance score between 0 and 1 and expert source quality rating graded on a letter scale. Furthermore, users can access the original source by clicking on the given URL for further interpretation.

   3. By using Disease InfoSearch, a crowd-sourced information repository of Genetic Alliance, the AI agent informs patients of their disease and connects patients to their respective patient organizations.

2. Knowledge extraction

   1. The AI agent extracts relevant personal and medical information from patient health records to tailor its responses. For example, it can identify the user's geographic location to find local specialists or use medical data to narrow down potential rare diseases based on symptoms.

   2. The AI agent is also effective at parsing documents and websites, curating this information, and generating summaries related to user queries.

   3. Like a case manager, the AI agent will be eventually able to organize and schedule appointments with healthcare providers, helping reduce the chance patients miss the care they need due to a logistical issue.

3. Automated Web Browsing

   1. The AI agent handles web tasks such as filling out forms and organizing outreach, significantly reducing the logistical burden on patients and caregivers. Importantly, the agent does not need to be supervised while accomplishing these tasks.

   2. This feature is particularly beneficial for those with cognitive or physical impairments, who often face barriers in accessing support programs.

   3. When deployed on a task, the AI agent can be enabled to speak to the user using Deepgram’s text-to-speech API. Similarly, leveraging speech-to-text, the user can speak to instruct and correct the agent between steps.This is both a novel user experience requiring 0 graphical interface, but also an important accessibility feature for those with motor issues.

   4. Multiple AI agents can also be deployed at once, allowing multiple tasks to be accomplished in parallel, multiplying the productivity of patients and caregivers for bureaucratic tasks.

Our AI-powered web agent aims to significantly improve the quality of life and health outcomes for millions of individuals and families affected by rare diseases globally, providing a comprehensive, accessible, and continuously improving resource.

Who does your solution serve, and in what ways will the solution impact their lives?

In 2020, a father of a boy who had around 100 seizures a day reached out to Genetic Alliance. Our goal was to reduce the number of seizures per day. Through Community-Driven Innovation™, an online patient platform, the families found immediate interventions that could be widely adopted to improve the children's quality of life. For instance, excessive drooling was impeding their development, but Genetic Alliance found that salivary gland Botox injections reduced drooling, enabling the children to sit and crawl without choking. Additionally, special teas were found to reduce constipation and lower the number of seizures. While these discoveries were significant, it took months to discover, disseminate, and put them into practice. Abbey AI, our AI platform consisting of a patient-oriented chatbot and automated web browsing agent will significantly reduce the time, resources, and complexity required for these patients and caregivers to access the information they need hidden in our existing systems.

Our patient-centered chatbot has several innovative features designed to help people at different stages of the diagnostic process. Currently, individuals have to search through various sources, which can sometimes provide conflicting information, and then try to make sense of it all on their own. With our advanced causal citation and patient feedback alignment algorithms, newly diagnosed patients can access a large amount of information instantly, process it, and receive verified, accurate, and useful information. Disease InfoSearch, a crowdsourced repository from Genetic Alliance, will be a primary source, ensuring that the information is verified by experts: disease advocates who have the most current and accurate information about a specific condition. This is a significant improvement over the outdated and inaccurate information available on the internet and from general-purpose chatbots like ChatGPT. With data and patient testimonials from our Community Driven Innovation™ platform, those further along can easily access practical care management tips. Making this data more accessible to patients will also facilitate the exchange of information throughout the biomedical ecosystem, including clinicians and researchers. For undiagnosed patients, querying our diverse bank of patient testimonials across 2000 of diseases provides grounded experiences that they can reference to generate hypotheses along their diagnostic odyssey. Our patient-oriented chatbot has key innovations over existing resources only possible due to our previous initiatives spanning 38 years.

Our solution, Abbey AI, also includes automated web browsing capabilities designed to assist undiagnosed patients in accessing critical resources. For example, if a patient needs free genetic testing resources, Abbey AI's web agent can search the web, fill out necessary forms, and summarize both the findings and any encountered issues, all autonomously. This significantly reduces the bureaucratic burden on patients and caregivers. Users can initialize the agent by specifying their objectives or using pre-existing workflows. Additionally, multiple agents can operate simultaneously, enabling parallel exploration. We hope this parallelization will speed up the diagnostic odyssey. In theory, Abbey AI can autonomously complete web tasks concerning any condition in over 50 languages, potentially benefiting hundreds of millions of people worldwide.

Which dimension of the Challenge does your solution most closely address?

What is your solution’s stage of development?

Please share details about why you selected the stage above.

Genetic Alliance has developed more than 20 resources and tools since 1986. These include DiseaseInfoSearch.org, a comprehensive directory of rare diseases, as well as registries, biobanks, and tools like the advocacy bootcamp. We offer services such as matchmaking, referrals to advocacy and support organizations, and train-the-trainer programs. Additionally, we operate iHope Genetic Health, a program dedicated to diagnosing undiagnosed children in underserved communities worldwide. These tools were created before and during the early days of the internet. Integrating the very specific and highly trained AI tool into these resources will significantly enhance their availability and effectiveness.

In terms of Abbey AI, we have built a fully working patient-centered chatbot and begun testing our AI web agent on real-world tasks. The data pipeline for the chatbot (ingestion of PDFs & patient testimonials, parsing, chunking, embedding, retrieval algorithm, evaluation) has been built out in its entirety for one rare disease. Multiple citation mechanisms have been implemented (Cohere Rerank, Context-Cite). Feedback mechanisms such as thumbs-up, thumbs-down and context feedback mechanisms have been implemented in our frontend and logged using LangSmith in the backend. The reinforcement learning from human feedback algorithms on the backend have not yet been written given the early stage of the project and current lack of user data. Evaluations on LLM hallucination rate, factual recall ability, document relevance, and answer helpfulness have begun using LangSmith for logging and LLM-as-a-judge techniques for scoring.

Our agent currently has been built and works on toy tasks. We have completed adding speech-to-text and text-to-speech capabilities and frontend integration for creating new agents. We are currently testing the boundaries of this new and exciting but still immature technology for this problem.

What makes your solution innovative?

Innovations and Impact

• Accessibility to knowledge is greatly enhanced.

  • Our AI model, GPT-4o, supports real-time voice translation for over 50 languages. This makes our model accessible to a global audience. 

  • We include text-to-speech for the visually impaired and speech-to-text for those with motor disabilities, ensuring inclusivity in the services the agent can provide.

• The model reaches a wider audience in the diagnostic odyssey, including undiagnosed patients.

  • The solution serves as a go-to resource for patients who have yet to receive a clear diagnosis, potentially reducing the average 6 physician visits before an accurate diagnosis. As a starting guide, patients are able to be connected to potential care-providers and advocacy groups, avoiding the hassle of doing the research themselves.

• Automated AI web browsing significantly reduces bureaucratic barriers for patients

  • AI web agents are currently cutting-edge innovative technology. In a world where AI web agents significantly cut down on logistical tasks, rare disease patients and caregivers would get care faster, more consistently, and at better times.

  • Organizations working with other groups of less-abled and disadvantaged people will learn from our example, seeing the value that these agents bring to those who have difficulty advocating and managing their own lives while navigating systems. 

• Human biases are reduced.

  • The AI provides an objective perspective, reducing human bias in the diagnostic process. This immensely mitigates the average 4-year delay in rare disease diagnosis. This can improve diversity by increasing accessibility to communities that are wary of the medical system.

• Model continuously learns to improve tailored responses from simple feedback

  • Through user interactions and feedback input, the agent improves over time by providing better solutions for the patient communities it works for. This ensures that the solutions provided are tailored to meet the specific needs of each rare disease group, while also learning personalized responses to fit a user’s needs.

  • Importantly, we expect our feedback mechanism to scale better than previous technologies. As a wiki, Disease InfoSearch is updated when users write and edit the page. But this need for active maintenance often leads to disrepair. In contrast, generative AI models are maintained simply by usage. These models easily generate content. Instead, this content needs to be aligned with facts and values. To do so, we use Reinforcement Learning from Human Feedback (RLHF) algorithms that employ a binary thumbs-up and thumbs-down, textual annotated feedback, and user engagement metrics to align the models. This means all users provide feedback for the product through usage and simple feedback forms. We see this innovation in feedback as being key to our success due to the difficulty of performing manual maintenance at scale as a nonprofit group.

Why are you applying to the Prize?

1. We  have mature and relevant tools deployed in the rare disease community,. These will greatly benefit from a high tech solution stitching them together, reducing the burden to use them, and iteratively improving them without the enormous amount of real time person-power needed at this time.

2. Financial Support:  Our team would have the resources to advance the development and deployment of our AI-powered solution, ultimately enhancing its impact on the rare disease community. The prize money would also help accelerate the delivery of our innovative solution to those who need it the most.

3. Validation and Credibility: The Amgen Prize lends significant credibility to our solution, as this recognition can open doors to additional funding opportunities, partnerships, and collaborations, which are critical for solving this complex challenge.

4. Overcoming Market and Adoption Barriers: This field suffers from significant market and adoption barriers due to the diverse patient populations, and limited knowledge and awareness. The recognition and support from a prominent organization like Amgen can help overcome these barriers by facilitating partnerships, excess to networks and potentially opening doors to new markets or distribution channels.

5. Collaboration and Access to Expertise: MIT Solve network provides us with the opportunity to build integration with like-minded innovators and to have the opportunity to engage with subject matter experts, researchers, and industry professionals.

Who is the Team Lead for your solution?

Sharon Terry

How are you and your team well-positioned to deliver this solution?

Genetic Alliance, a community-based organization, provides training and capacity building for thousands of communities who serve those affected by this “diagnostic odyssey”. Since 1986, Genetic Alliance’s members and leadership have worked tirelessly to alleviate this suffering, drawing on their personal experiences of their own children who have struggled to be diagnosed. They help accelerate the diagnosis process and access to services. Genetic Alliance also creates scalable tools and trains advocacy organizations. They work directly with inner-city, rural, and indigenous communities and partner with community leaders to better understand their community's needs. They enable communities to collect data, influence policy-makers, pass legislation, and equip local clinicians and community health organizations with specific technologies such as genome sequencing. Genetic Alliance matches individuals and families with available services and builds community-based services where they do not exist or require improvement. As a result, communities determine health priorities, drive the agenda, “build the table” (rather than secure a seat at the table) in health services and research, and advance policies to improve their communities' health.

Partnership is key to Genetic Alliance’s work. We work with over 2,000 advocacy organizations and thousands of other grassroots communities. We provide training, tools, and resources to build their capacity. In addition, we partner with major professional societies in healthcare and research. We serve on various committees and councils for the National Institutes of Health, the Health Resources and Services Administration, Centers for Disease Control and Prevention, and Food and Drug Administration. We provide a conduit for other organizations to reach thousands of communities about various policy issues in our network. 

Sharon Terry, CEO. In 1994 her children were diagnosed with a genetic condition after a diagnostic odyssey. Within days of their diagnosis, she realized competition and greed were impeding health. Without any science training, she gathered the community of people affected by PXE, solicited their priorities, built a robust scientific system for the condition. Parents of children with other conditions asked that she help them. She joined Genetic Alliance and scaled these solutions to enable thousands of other communities to advance research. Together they advanced community advocacy wherein the people control of the process of determining priorities for their communities.

What type of organization is your solution team?

Nonprofit