Meet Dave, the first AI oncology mentor. Is it the future of medicine?

10 October 2023
How does Dave work? Dave marks a significant milestone in our journey to revolutionize the healthcare industry. Dave provides smart, personalized, and accessible information instantaneously, which can greatly improve the quality of care and life for millions of patients worldwide. Dave was trained on proprietary datasets containing unique de-identified patient-to-physician and patient-to-patient interactions, de-identified medical documents, and PROs, which were aggregated from Belong's Beating Cancer Together app, the world's largest social and professional network for cancer patients. Over seven years, these distinct datasets, unique to Belong, have generated billions of data points that were utilized as a training set to elevate Dave to an oncology mentor, where accuracy is combined with responsibility. In order to establish Dave as a healthcare-grade solution, we have developed a distinctive approach known as ARCH, which is an acronym representing the four fundamental challenges a generative AI platform must address to attain readiness for healthcare applications: Accuracy, Responsibility, Compliance and Hallucinations. Right, hallucinations represent the biggest danger of using LLM in medicine. To mitigate hallucinations and effectively address accuracy concerns, Dave is powered by several LLM engines that collaborate through a proprietary LLM-multiplexer. Furthermore, an additional "Responsible AI" layer has been implemented atop Dave's architecture, employing a range of in-house algorithms. This layer ensures compliance with the stringent regulations and safety requirements essential in the healthcare industry, thereby elevating Dave to the status of a healthcare-grade generative AI platform. Dave is also available as a SaaS solution for hospitals, providers and patient support programs, who can customize the mentor characteristics and fields of expertise to include in-house clinical guidelines for their optimal and exclusive use. Dave has a long-them memory of previous correspondence, distinguishing it from classical generative AI tools. How personalized is the performance over time? Indeed, this posed a significant challenge: how to sustain a conversation spanning the entire patient journey, composed of numerous data points, and ensure that Dave consistently provides precise responses, taking into account the complete patient history and journey. In contrast to most conversations with common, publicly available, and generic generative AI solutions like ChatGPT and Bard, interactions with Dave often entail hundreds of follow-ups over extended periods, sometimes spanning months. This unique characteristic fosters an "intimate" relationship between users and Dave, as they perceive Dave as already acquainted with them. Consequently, users need not repetitively reintroduce themselves and their concerns each time they initiate a conversation. To address this challenge, we've developed a highly scalable mechanism referred to as internal "Context." This mechanism breaks down lengthy conversations into smaller, digestible segments, allowing our LLM to process substantial amounts of data when responding to inquiries. Furthermore, we've made substantial investments in "Context tagging." This feature enables Dave to reference only specific and pertinent portions of the patient history when formulating responses, thereby reducing the volume of data it needs to analyze, ensuring more efficient and contextually relevant interactions. Do you consider integrating it one day with EHR so the bot could, for example, answer questions regarding lab results? Integrating Dave with EHR is something we're working on. We are currently in discussions with health systems on the subject—and are happy to discuss this further with interested parties. Furthermore, Dave is already providing explanations of lab results today! Users have the capability to upload various types of medical documents, such as lab result documents, imaging reports (including CT, PET, and X-ray), genomic sequencing reports, and summaries of doctor visits, and Dave will handle them for you. Initially, for verification purposes, Dave will identify the document type. It will then provide a simplified explanation of the document's content in plain language. Afterwards, Dave will address any specific questions you may have regarding the document.
If I hadn't known that Dave is an AI mentor, I would have been convinced he is a physician
What's more, Dave can also manage images of documents and PDFs. If a user captures a photo of a medical record, Dave will enhance the image and apply OCR (Optical Character Recognition) to extract the text content. This functionality has proven to be a standout feature, delivering significant value to patients who receive reports they find challenging to comprehend, causing stress. Within seconds, Dave can simplify the information and provide answers to related questions, alleviating the uncertainty and anxiety experienced by patients. What is the feedback so far? Are cancer patients ready for such a completely new model of support? What are the things to be improved on? Dave is already accessible to a user base of 500,000 individuals, and our efforts to expand our reach are ongoing. The feedback we've received has been overwhelmingly positive! Engagement with Dave has shown remarkable growth as more and more people discover the value in his accurate responses, immediate availability, compassionate interactions, and informative answers. One noteworthy endorsement comes from a prominent doctor at Mount Sinai, who remarked, "If I hadn't known that Dave is an AI mentor, I would have been convinced that he is a physician with expertise in various cancer subtypes." Additionally, patients who have interacted with Dave have expressed their gratitude by saying "Thank you, Dave" more than 40,000 times to date. Mentoring cancer patients requires a lot of empathy. How did you address this challenge? This is a fundamental aspect of Dave's character. We've equipped Dave with the ability to express empathy as an integral part of his communication skills. Cancer patients are navigating an extremely delicate phase in their lives and, therefore, require a unique approach. It's not sufficient to merely present them with information; Dave excels in bridging this gap by conveying the facts while also emphasizing the positive aspects whenever feasible. It also requires keeping the balance between being informational, emphatic, honest, meantime careful not to give ungrounded hope. Indeed, Dave's commitment lies in adhering to fact-based information and refraining from offering false hope. However, facts can be conveyed through diverse means. For instance, a frequently posed question is, "Dave, what are my odds?" One approach involves presenting patients with raw statistical figures and hoping they can absorb the information. Alternatively, Dave opts to share the statistics while elucidating the factors underlying these figures. He stresses that statistics are constructed from a broad spectrum of patient outcomes and cautions against assuming that these statistics invariably apply to one's individual circumstances. What are the biggest gaps in cancer care, and will breakthrough technologies be able to bridge them? The primary challenge we face is the constraint of time. Each year, there is an increasing number of cancer diagnoses, and the available medical workforce is insufficient to manage this influx. Generative AI solutions, such as Dave, have the potential to address this issue by automating numerous routine tasks and processes, thereby enabling healthcare professionals to concentrate on the most crucial aspects of patient care. These AI solutions can handle various tasks, including patient education, where patients often have numerous questions as they seek to comprehend the decision-making process, treatment protocols, and more. They can simplify the communication of medical information to patients, streamline appointment scheduling, generate alerts, and perform many other essential functions in the medical field. How do you see the role of tools like Dave in the future? Our vision at Belong is that in another year or two, every patient will embark on their health journey with the help of a professional AI mentor like Dave. Hospitals and health insurance companies will also support it because solutions like Dave's significantly improve understanding and communication, keeping patients calm and confident, improving quality of life and facilitating better treatment outcomes. From the author: Dave was recently joined by Tara – an AI-powered clinical trials matching agent designed to aid patients in discovering suitable clinical trial opportunities.