Buoy Health’s mission to debunk Dr. Internet
Buoy Health is improving the way you Google your symptoms
Why is machine learning important in healthcare?
In the last couple decades, the Internet has presented us with several pseudo-doctors guised as search engines that we have used to feverishly (pun-intended) self-diagnosis, self-triage, and research symptoms whenever we have felt a sickness coming on. But the proliferation of medical information to the general public comes at a price and is often misinterpreted when not combined with the advice of a trained physician, resulting in misdiagnosis and unnecessary anxiety.
In the last four years, machine learning and AI have taken the health industry by storm. In 2014, the AI healthcare market was valued at $600 million and in 2018, an Accenture report[ii] was released at HIMMS, estimating that the market may increase eleven-fold by 2021 to $6.6 billion.
In 1976, Maxmen predicted that artificial intelligence would bring about the end of physicians[iii]. While the possibility of living in a physician-less world seems unlikely, this megatrend is fundamentally changing how consumers find and receive care and how health providers administer care.
What is Buoy Health doing?
Buoy Health is a Boston-based, Series A startup that is using machine learning and AI to develop an online symptom and cure tracker that uses intelligent algorithms to help patients diagnosis and triage themselves. For the past five years, Buoy Health has been developing algorithms that analyze thousands of real world data points drawn from 18,000 clinical papers covering 5 million patients to resemble the dynamic and nuanced experience of chatting with a doctor[iv],[v]. Their hope is to use AI to augment decision making for doctors and help improve the gap between a patient googling for systems online and having to go in to see a physician.
As shown in Exhibit 1, the Buoy algorithm interacts with a patient like a provider would, asking individualized follow up questions and eliminating potential disease choices based on their answers. As a patient inputs symptoms, Buoy is dynamically picking 1 of 30,000 possible questions to ask the patients based on which one, from a statistical perspective, is going to reduce the uncertainty of what they have the most. At the end, Buoy will spit out potential diagnosis (3 max) and different triage options. Sharing personal information like sex and age eliminates potential options that a patient never could simply by Googling their symptoms. Not only will this result in better diagnosis of patients, but patients will be able to triage themselves better and come better prepared when they see their physician. It can even affect reduce unnecessary costs like ER visits or urgent care.
Exhibit 1: Sample screenshot of Buoy Health chatbot[vii]
What can Buoy do in the future?
Now that Buoy has developed an initial working application, they now need to work on 1) refining their algorithms based on their continuous influx of data and 2) gaining credibility in the industry. Because of the complexity of healthcare, the users are usually not the ones that pay for services. Therefore, because their business model will most likely depend on developing partnerships with payers and employers, they will need to follow these steps to strengthen their ability to forge partnerships with payers, employers, and patients, ultimately fueling their business model.
With several million patients using Buoy every month, they now have a growing data set that can be used to continually train and test their algorithms, adding to its accuracy and building out its capability of the types of conditions that it can diagnosis. Buoy Health recently partnered with Boston Children’s Hospital to improve the way parents diagnose their children[vi]. Not only will this be new data to help improve their algorithms for diagnosing children, but it will also give them access to a whole network of new providers to gain input.
Questions moving forward
Machine learning and AI have the potential to improve care delivery, streamline processes, decrease costs, and improve decision making. The possibilities are endless and the only questions are how the changes will manifest. How will the role of providers transition from being the main decision maker for data and instead become a voice in a cognitive-computing fueled cycle that can make more personalized, accurate decisions for large populations? How can professions such as radiologists, anesthesiologists and pathologists remain relevant as these technologies permeate the industry?
(800)
[i] The Medical Futurist. “Could AI Solve the Human Resources Crisis in Healthcare?” https://medicalfuturist.com/could-a-i-solve-the-human-resources-crisis-in-healthcare. August 2, 2018. Accessed November 2018.
[ii] Dale Van DeMark, HIT Consultant. “AI and Machine Learning is Shaping the Future of Healthcare Delivery”. https://hitconsultant.net/2018/06/27/ai-machine-learning/. June 27, 2018. Accessed November 2018.
[iii] C. David Naylor, American Medical Association. “One the Prospects for a Deep Learning Health Care System.” http://www.fsk.it/attach/Content/News/6636/o/jama_naylor_2018.pdf. August 30, 2018. Accessed November 2018.
[iv] Buoy Health website. https://www.buoyhealth.com/. Accessed November 2018.
[v] Andrew Le, Product Hunt. “Buoy Health”. https://www.producthunt.com/posts/buoy-health. 2016. Access November 2018.
[vi] PRNewswire. “Buoy Health Partners With Boston Children’s Hospital To Improve The Way Parents Currently Assess Their Children’s Symptoms Online”. https://www.prnewswire.com/news-releases/buoy-health-partners-with-boston-childrens-hospital-to-improve-the-way-parents-currently-assess-their-childrens-symptoms-online-300693055.html. August 8, 2018. Access November 2018.
[vii] Avery Hartmans, Business Insider. “This easy-to-use app eliminates the scary feeling of looking up your health symptoms online”. https://www.businessinsider.com/buoy-app-health-symptom-checker-photos-2017-5#i-added-one-more-symptom-feeling-foggy-headed-then-buoy-asked-me-about-my-lifestyle-and-any-other-unusual-feelings-or-symptoms-just-like-a-regular-doctor-would-5. May 2, 2017. Accessed November 2018.
This is an excellent overview of Buoy’s technology. Having personal experience with the company at its founding, its core technology is based on a static algorithm that was based on inputting medical literature as a provider would. The machine learning component comes only as Buoy starts to get more unbiased data. Obtaining such data is difficult as many users of buoy are people on the internet who do not ultimately report what their final diagnosis is or report how “reasonable” their possible diagnoses are. The data that they do get are people who go to seek medical care and therefore it is likely biased toward more serious conditions.
Although the technology is excellent and becoming more “conversational,” it misses one crucial element in real-life interactions with a physician. The ability for a patient to “free-text” rather than answer multiple choice questions. Valuable information is often gained from understanding both the more qualitative nature of a patient’s symptoms as well as the story/time-course of symptoms which is slightly more difficult to capture in nuanced detail over a chat.
Finally, Buoy was created to serve as a gateway to providers to filter out those that don’t need medical assistance and to encourage those that are seriously ill to seek care immediately. For the latter camp, Buoy’s algorithm, graphics, explanations, and recommendations are missing one thing: comfort. There is nothing quite like looking a parent in the eye and telling them that in a day or two their child will be back to normal. There are some patients who will always crave that sort of interaction, even if the ultimate recommendation is rest and hydration.
Despite the reservations that I’ve expressed above, I think that the algorithm Buoy is currently refining has huge potential in the healthcare space. The problem to solve now is where best in the health ecosystem it adds the most value.
This was an extremely interesting read! Individuals are continuously looking to the internet for guidance on various medical ailments and Buoy’s product offers a better alternative than WebMD and other less interactive models to diagnose and triage. With regard to application, this product could help alleviate supply and demand issues in markets where patient to physician ratios are high, specifically in rural areas and developing countries. Patients who would have otherwise not been able to seek treatment are provided some baseline diagnosis. Additionally, patients who lack access to healthcare (i.e. no insurance coverage) have avenues for treatments.
However, there are some major concerns that Buoy should continue to think through. Patient/doctor relationships require a high degree of trust. How will Buoy’s app develop trust amongst users? What is the marketing strategy around ramping patient use at the onset of launch of this project. Second, will insurance companies be incentivized to aggressively push this product on their patient populations to save costs? Will they start mandating the use of the app prior to any doctor visit, which could adversely affect individuals who are severely ill, especially if the algorithm incorrectly diagnoses them.
How does the company think about the possibility of false negatives occurring, especially if patients are severely ill and may rely on this product for primary treatment. How will Buoy incentivize patients to report mis-diagnoses and continue using the app if misdiagnoses occurs. Lastly, the gut/intuition that physicians develop over many years of training seems hard to replicate across AI/machine learning. Given these concerns, I believe Buoy is a great aid to physicians but not a substitute.
I believe there is immense value in having more targeted, relevant information available to patients. There is a great deal of difficulty and danger in attempting to diagnose yourself by typing your symptoms into a search engine instead of seeking professional assistance. However, as we all know, seeking medical advice in the US is extremely complicated given the financial burden, long waits, full schedules, transportation challenges, inefficient appointment scheduling processes, poor understanding of insurance benefits and cost sharing, and general complexity surrounding which type of doctor to see. So in the context of our healthcare structure, Googling symptoms is often the first logical step that many people take, and Buoy Health therefore has an intriguing value proposition.
Incorporating personal characteristics, such as sex and age, in addition to symptoms is crucial. For example, in the case of upper abdominal pain, a heart attack would be higher up on the list of possible diagnoses for a woman than for a man, since more women with heart attacks present with abdominal pain than men with the same diagnosis. In terms of age, it would be highly unlikely for a child or young adult with upper abdominal pain to be experiencing a heart attack!
I see Buoy as a means to encourage individuals to seek care who otherwise wouldn’t, since many individuals wait too long to seek care. On the flip side, I would be very hesitant to tell someone who thinks they need medical assistance to stay home based on reassuring Buoy results. In other words, I think this should be marketed towards being a safety net and improving outcomes for individuals who consider themselves fairly healthy and low risk, as opposed to a tool for more high-risk individuals with complicated medical problems to search for a reason to not seek professional care. Because of this belief, I’m not sure I agree with marketing this to employers and insurers, since if people use it as I described, it would actually increase costs. Certainly, having greater access to the healthcare system can lower costs in the long-run by keeping people’s medical problems more controlled, but insurers and employers generally don’t keep the same individuals for the whole course of their life and therefore may not see a financial benefit from this. (It’s similar to the debate about technology like Fitbit, where many insurers feel they won’t actually realize financial gains from the incremental health benefits of Fitbit because people may switch insurers multiple times in their lifetime as their employment changes. In other words, is the insurer paying more upfront to make someone healthier for their next insurer?)
An advantage I see for Buoy is that patients may be more likely to enter more private, sensitive information on their platform than voice that information to a doctor. For example, a person may be more likely to be honest on Buoy about how many cigarettes per day they smoke than when they have to voice that number out loud to a physician. However, there are many subtleties about the patient-doctor interaction that will be hard to replace with technology and AI and there is much that a physician can glean from talking to or even observing an individual.
Thanks for writing about this interesting technology. Buoy is an interesting idea and certainly has potential to be helpful in the overburdened current healthcare system. There are many algorithms that physicians follow and automating many of these algorithms into a tool such as Buoy seems like it could be promising.
However, I believe there are some concerns that need to be addressed with a technology likes Buoy. To begin, many patients I’ve treated often come stating that they believe they have “x infection” or “Y illness” but with the caveat that they googled their symptoms, and they know google is not a doctor.
My concern would be that with Buoy, patients will try to self-diagnose, and assured by the fact that Buoy is a sophisticated, AI based device, they would go on to self-treat. This could lead to two concerning outcomes. One could be poor outcomes for the patient. The other could be increased liability for the physician. If a physician recommends Buoy or a similar device to their patients, they may have to shoulder some, or all the liability shall something go wrong. Physicians have become overburdened by an extensive amount of documentation, requests for electronic refills, e-consultations and the like, and I am not sure many physicians would want to use Buoy as a substitute for their judgement, especially if they are held liable for the negative outcomes that result. Buoy, I believe, could function better as an aide to physicians or mid-level practitioners to speed up data gathering during patient visits. The physician or medical provider would be better equipped to look at Buoy’s recommendations but take these with a grain of salt. After all, they have the requisite training to question Buoy.