“I felt like nobody cared or listened. Thank you for treating me today with respect and acting in a manner that says every life has value.”
Immersed in a healthcare system with profound inequities, every good doctor has heard similar statements from their patients. But this particular message wasn't directed to a doctor. This is a message from a woman who, after multiple rushed, dismissive experiences at various hospitals, decided to use MDandMe—the world's first conversational AI symptom checker—to help her understand the decision-making behind a recent hospital discharge that she felt was premature.
And she is not alone. Of the more than 2,500 people we helped in the first month of our launch, the vast majority would recommend MDandMe to friends and family (NPS 60+). More than 150 users came to us with potentially life-threatening conditions and were accurately directed to seek care as soon as possible. And every user interviewed reported that MDandMe was the most powerful and seamless symptom checker they had ever used.
So what is MDandMe?
MDandMe is the world's first conversational symptom checker powered by proprietary generative AI technology. In other words, it is the closest web-based experience to chatting with a doctor, besides chatting with a real doctor. Usually this is over text, though voice is also available with a subscription.
What we provide:
A thorough text-based medical interview at any time and at your own pace
An immediate assessment of potential causes of your symptoms
Information on where to get care and how urgently you need it
Personalized information on what you can do to alleviate your symptoms prior to your appointment
A medical note to bring to your doctor
What we do not provide:
Medical advice
Medical orders such as prescriptions
Structure of MDandMe interviews with example dialogue and assessmentMyth-busting
When we first started MDandMe—with a mission to build technology to empower a healthcare system free from judgment, long waits, and time constraints—healthcare professionals were, for good reason, skeptical. This was a pitch they had heard before, from questionnaire-based legacy AI symptom checkers such as Ada, Buoy, and in-house offerings from hospitals including Mayo Clinic and Cedars-Sinai. However, symptom checkers have historically been unable to effectively triage patient concerns, often perpetuating long healthcare wait times by sending too many patients to the emergency room. They were also too limited to provide the quality and depth of a medical interview that would be useful for clinical decision-making.
Enter generative AI. This new approach resolves many of the issues that plagued older generation medical chatbots, as many physicians have personally experienced through using applications like ChatGPT. However, some physicians still had reservations. Let's take a look at a few common ones.
Myth 1: “Patients barely fill out a questionnaire. They will probably drop off after max 5 responses to an AI.”
As it turns out, only 1.2% of our users drop off within 5 responses.
Length and number of responses per chatThe average MDandMe user talks through just over 4 symptoms (range 1-15 symptoms) across 29 turns. In fact, we had to adjust our length limit and conversation structure because 22% of users were maxing out on our previous maximum conversation length.
We were pleasantly surprised by the eagerness users showed toward engaging with our AI, especially given that many physicians warned us of the opposite. Our hypothesis was that by creating a free-flowing and human-like conversational experience, we could retain patient engagement and gather orders of magnitude more information than any questionnaire or form. Indeed, MDandMe's sister product AuxHealth, used for patient intake for doctor's practices, has a >95% completion rate and similarly positive reviews.
Myth 2: “Patients won't always know the proper name of a symptom. Plus, it's not possible to build a flow to capture every possible history and combination of symptoms.”
We agree! Generative AI allows us to handle any set of symptoms and any patient history rather than older technologies that relied on pre-defined flows and multiple choice questions.
From “brain zaps” to “not sure how to get home” to “fishy odor and cottage cheese”, we understand it all. In fact, 96.8% of users reported that our A.I. understood what they were saying “almost always” and 0% of patients reported not being able to accurately describe their symptoms. That being said, we are continuously improving our system to ask precise, medically relevant questions in language that is easily understood by the general public. On the horizon are simplified English and multilingual capabilities.
How our symptom checker worksMyth 3: “Patients just want to talk to a real doctor”
Possibly! MDandMe allows those with limited healthcare access to make more informed decisions about their health. We explicitly state that we are not a substitute for real medical care at the beginning of each interview.
Over 90% of patients search online prior to seeking medical care, but existing resources are not tailored to a patient's specific situation and thus can often be misleading. MDandMe was made to help patients better understand their symptoms and make more informed choices about where and when to seek care. Additionally, we can help patients better understand their diagnoses after a visit with a doctor, monitor symptoms over time, and also provide behavior change coaching.
1 in 4 of our users present with symptoms that are unlikely to warrant immediate workup or prescription; for example, uncomplicated upper respiratory infection, primary headache, muscle ache, chronic back pain, gastroenteritis. These represent opportunities to follow up with patients and refer them to a healthcare provider should symptoms worsen or persist. We never recommend against seeking medical care, but can suggest self-management techniques and over-the-counter medications for low-risk situations.
On the other hand, 10% of our users provide a medical history consistent with serious conditions that may require immediate hospital-based care, such as a post-injury loss of consciousness, symptoms consistent with acute blood cancer, heart inflammation, bowel obstruction, kidney infection, and severe anemia. Importantly, approximately 1 in 3 of these patients report that before using our service they were not intending to seek care at all or were only considering booking a long-wait primary care appointment.
Clinical subject and urgency of user conversationsWhat's next?
Medical care is not readily accessible for many patients across the United States and the world. High costs, long travel distances, lack of availability, and bad prior experiences are just a few of many barriers to care access that push patients to research their symptoms before and after seeking medical care. MDandMe enhances this search, tailors it to each patient's specific current and past medical history, and acts as a bridge to the appropriate level of medical care.
As our AI improves by the day, we seamlessly incorporate the newest expert guidance, clinical guidelines, and research findings, and distribute personalized medical information universally and uniformly.
“This is exactly what's needed to help the medical system finally reach people in low-access regions,” one primary care physician from Boston said. “You are creating the type of technology that could change the entire practice of medicine for the better,” added a former medical chief at a large academic hospital.
Less than 30,000 will graduate from medical schools across the country this year. But empowered with the scalability of AI, we at MDandMe will tackle healthcare inequity with the vigor of a million doctors still scarred by the patients they saw turned away due to lack of time, limited resources, or just inadequate insurance. We will keep building until everyone, regardless of income or geographical location, can access world-class healthcare whenever they need it.
You can try MDandMe for yourself at https://mdme.ai/chat.
Have a comment? Email our co-founder, William, at william@mdme.ai.
