FASTER HORSES

In the height of his innovative stardom, having revolutionized the automobile industry by taking a luxury item reserved for the elite and making it available to the middle class, Henry Ford is alleged to have said, “If I had asked people what they wanted, they would have said ‘faster horses’.” Despite a seeming inability to prove Ford ever made that statement, it lives on in the annals of business philosophy and echoes in the hearts of innovators who long to dream the unimaginable, achieve the impossible and to change the world along the way. Most true innovators are in it to make the world, and human lives, a little longer, a little better, and a lot more meaningful. And for centuries, tech innovators did exactly that. Their vision and creation put automobiles where horses once roamed, telephones where there were once messengers, the Internet in what was once a vast emptiness. Communication now happens in split seconds, not weeks. Information is now available at the touch of a screen or the sound of your voice rather than requiring travel to a physical building where that knowledge was once housed. Each new generation of innovators has built on the progress of the past, creating something newer, faster, better and more unbelievable than its previous iteration.

And if ever there was an industry in need of the power and promise of innovation, it is the American healthcare system. Facing a need for 1.2 million new registered nurses by 2030, an opioid crisis costing $35 billion and nearly 70,000 lives per year, chronic diseases that affect 150 million Americans, a budgetary strain turned crisis under the demands of the Covid pandemic, and a ranking of last among comparable countries on the Healthcare Access and Quality Index (measuring amenable mortality, or “premature death that is preventable and treatable by effective and timely care”), to name just a few, we have reached a critical healthcare tipping point. It is a system begging for technological transformation and innovation, with nearly 330 million American lives on the line. 

But despite notable technological achievements in our collective history, despite an abundance of innovators still willing to dream and create, and despite a healthcare industry in desperate need of that innovation, public doubt threatens to ground the process. The 2022 Edelman Trust Barometer demonstrates one bold and concerning key point: we have entered a new “cycle of distrust” where “nearly 6 in 10 now say their default tendency is to distrust something until they see evidence it is trustworthy.” While both technology and healthcare industries remain high on the Edelman report, at 74% and 69% respectively, a societal undercurrent of distrust can undermine the innovation and collaboration necessary to meet the critical healthcare challenges of today.

That undercurrent is likely worse than the numbers suggest, in part because of the tendency of the human mind to conflate much of the tech sector with those tech giants who dominate the news. The constant negative press surrounding Meta, with their subsidiaries Facebook and Instagram, Twitter and its buyout debacle with Tesla CEO Elon Musk, Google and their data-grab on an almost inconceivable level, and government crackdowns spreading far and wide across social media. When we think of technology companies, we often envision social media, Edelman’s least trusted sector at just 44% and on a continual decline.

This mistrust of social media is bleeding into an industry that bears no real association with it. Even with an overall decline in trust, the healthcare and technology sectors saw gains in public trust, with healthcare gaining three points and technology gaining four in 2022. Those two sectors continue to rank at the top of the trust survey, though perhaps largely because their controversies have received less press than their social media counterparts. In the spirit of “no news is good news,” we continue to place our faith in the teams of medical researchers, scientists, engineers and doctors, believing in their desire to do the right thing and to build a better version of wellness for those in need of it. To find solutions to the problems that have so plagued human bodies. To cure cancer. To meaningfully improve mental health. To give us longer lives that are worth living. 

The answer to this societal state of declining trust is not to abandon the potential of healthcare technologies altogether but rather to require that companies provide the data—the evidence—that their products and technologies work as intended, are equitable to all, and provide the promised benefit to patients. And, perhaps even more importantly, for patients to demand a seat at the table and a part of the conversation concerning the technologies we want and those we need for the betterment of our health. To demand the right to share in the vision of the roles technology will assume in our healthcare system and as an integral part of our personal wellness. 

Sometimes, the answer is the Model T, an innovation the public didn’t know we needed but that transforms the world as we know it. But sometimes, the answer is faster horses—those solutions requested by patients and providers that will make better use of the expertise, technologies and resources already available to us.

Technology is radically, and rapidly, transforming the healthcare industry. While this partnership had its beginnings in the early 1970s, as experts in both industries saw the potential in merging technology—particularly artificial intelligence—with the scientific advancements in the medical field, it wasn’t until the “data tsunami just beginning to crest” around 2007 that we began to see an explosion of healthcare tech innovation, according to Andrew Beam, assistant professor of Epidemiology at Harvard with an appointment in the Department of Biomedical Informatics at Harvard Medical School. 

A major milestone in this emerging partnership seemed to come in 2013 with the integration of IBM’s Watson, a question answering supercomputer that had bested human counterparts Brad Rutter and Ken Jennings on Jeopardy! just two years prior. Following that $1 million victory, IBM sought other uses for Watson’s advanced data mining capabilities, and they found a goldmine in medical research data. With more than 29 million references to biomedical journal articles on Medline and more than 34 million on PubMed, “no person or team of persons can absorb all the material gushing from this firehose…. Enter Watson. You instruct Watson, the data-miner, to dig into the literature and bring you papers relevant to your case…. Then Watson, its dirty work done in a fraction of a second, climbs out of the data mine and brings you all the publications relevant to your work,” says Robert Marks, author of Non-Computable You: What You Do That Artificial Intelligence Never Will. At least that’s how the partnership was supposed to play out.

In theory, this partnership seems like the perfect merger of the technological speed and data digestion of Watson with the intelligence and experience of medical doctors, bringing a narrowed list of best diagnostic and treatment options to a doctor trained to discern between them. And that potential was enough to keep the partners involved in this project committed and at work for four years and to the tune of $62 million. The media caught Watson fever, too, claiming that Watson was helping to “revolutionize cancer care at MD Anderson Cancer Center.” IBM itself called Watson “the future of knowing.”

The lead scientist responsible for the creation of Watson had a different message for IBM executives as they moved forward with plans to capitalize on his technology: “Beware what you promise.” According to an article by New York Times contributor Steve Lohr, that scientist—David Ferrucci—“explained that Watson was engineered to identify word patterns and predict correct answers for the trivia game. It was not an all-purpose answer box ready to take on the commercial world, he said. It might well fail a second-grade reading comprehension test.”

Despite this warning, IBM moved forward, and Watson predictably failed, not because the Watson technology was unimpressive but because of the company’s missteps in emphasizing “big and difficult initiatives intended to generate both acclaim and sizable revenue for the company.” Innovation does not always go as intended. Manoj Saxena, once a manager for IBM’s Watson business, notes that their original objective was “to do pioneering work that was good for society” but that “the challenges turned out to be far more difficult and time-consuming than expected.” Those challenges ultimately led to a system that was incompatible with MD Anderson’s electronic medical records system and that provided “multiple examples of unsafe and incorrect treatment recommendations” to physicians.

In addition to Watson not being developed and created to work in the healthcare industry, and so not having programming effective for that purpose, Watson’s team of programmers lacked the medical knowledge necessary to build such a program. While the potential for technology and healthcare collaborations is profound, experts in each industry have highly specialized knowledge specific to their industry with few having gained needed knowledge in their partner discipline. If these healthcare and technology collaborations intend to earn the public’s trust, after notable failings like Watson and a history of excluding the broader public from knowledge of these activities, they will need to deepen the collaborations and include the public in those conversations.

The process of deepening the collaborations by getting medical experts and engineers to work together to build comprehensive systems is well underway. An example of these collaborations is beamlab, “an interdisciplinary group of computer scientists, epidemiologists, statisticians, and physicians” based out of the Harvard T.H. Chan School of Public Health. Beamlab is “working to change healthcare” and is headed by husband-and-wife team Andrew Beam, a machine learning specialist, and Kristyn Beam, a clinical fellow in the Harvard Neonatal-Perinatal Medicine Training Program. They are “principally concerned with improving, stream-lining, and automating decision-making tools that doctors can use to better care for their patients.”

As part of their work, they are developing a natural language processing and understanding system, the core technologies used in the earlier Watson project, to sort through the “world’s medical knowledge in unstructured sources such as textbooks, websites, and biomedical journals.” Their system, not yet publicly named, is being trained on a “unique collection of biomedical texts containing 4.3 million articles, 50,000 pages of reference material, 15,000 flash cards, dozens of medical textbooks, and 20,000 multiple choice medical questions.”

While the intention of this project bears marked similarities to what executives at IBM likely hoped for with their Watson project, the beamlab system was built from the start as a support for overburdened medical professionals who want to offer more and better to their patients within a system that currently offers them little in the way of time, funding, or resources to do so. The beamlab system is built to empower doctors with a broader range of knowledge and available options. And it is designed and built based on high-level interdisciplinary knowledge, rather than the exclusive technical knowledge of engineers, and through a funding partnership with the Robert Wood Johnson Foundation(RWJF), an organization whose mission is to “support efforts to build a national Culture of Health rooted in equity that provides every individual with a fair and just opportunity for health and well-being, no matter who they are, where they live, or how much money they have.” 

That is a step forward. At its heart, though, it is machine learning making the decisions about the most relevant research, most likely diagnoses and most effective treatment paths without human input. While there is human oversight in the doctor’s review of those recommendations, the path to those recommendations is solely the work of machine learning. The tech industry promotes this path as a groundbreaking solution for the overburdened and often fallible healthcare industry, but it is worth remembering that the Model T was not the only available solution to the transportation problem facing America at that time. It was simply the solution that won, the idea that succeeded. In the case of machine learning in medical diagnostics, we need to carefully consider each of our options and articulate what we, as the human beings the system serves, want in this process.

Beamlab rightly points out that one of the key challenges to the integration of machine learning into medical diagnostics is that “nearly all of modern machine learning techniques are designed to give predictions, but what doctors often want are decisions.” Missing from this statement is the desired source of those decisions, for medical practitioners and patients alike. 

Neither of those groups is suggesting they want to hand the diagnostic reigns over directly to machine learning. In a 2019 study exploring patient openness to medical AI, researchers Chiara Longini, Carey K. Morewedge and Andrea Bonezzi found substantial resistance, even in cases where patients were provided evidence that AI was statistically more reliable in the diagnostics or more successful in surgical procedures. They attribute this reluctance to uniqueness neglect, “a concern that AI providers are less able than human providers to account for consumers’ unique characteristics and circumstances.” 

Perhaps it is less neglect than connection that is the core of patients’ concern. A patient wants to be much more than the electronic medical record denoting their health history or a collection of complex symptoms. They want to be known. Understood. Heard. It is a core human tenant to want to be seen as an individual, to be recognized and treated as such. To achieve that, patients seek out a fellow human being who understands what it is like to live with the complexities of a human body, someone who has born witness to or experienced the pain of a fractured bone or the physiological chaos of cancer, the slow loss of language and self in Alzheimer’s or the loss of our sensory touchstones with blindness or deafness. They want to look in the eyes of a fellow human being and know the understanding and empathy that comes from the shared challenge of navigating life in a human frame that can sometimes fail. In short, they want faster horses. They want human doctors who have at their fingertips a broader base of knowledge to track down the specifics of their challenging diagnostic case and to provide the most efficient and effective treatment for it. 

The majority of doctors seem to share this vision. They are not seeking technologies that will replace their role in the diagnostic work. They are seeking technologies that enable them to do more for their patients: to enhance their own understanding, to expand their network of collaborators, to enrich their efforts. As Tom Lawry, National Director of AI for Health and Life Sciences at Microsoft, recently said, “Automation means highly repetitive work done by humans today is going to be done by a smart machine today or in the future. But the biggest part of healthcare today is augmentation…the idea of augmentation is ‘how do we bring AI in behind the humans to make them better at something they care about?’”

What doctors care about is their patients, those unique individuals who have come to them for hope and a path forward to healing. Yet too many are currently denied access to the doctors and the medical knowledge necessary to restore health and save lives. As founder and chair of the Human Diagnosis Project (Human Dx), Jayanth Komarneni said, “Millions in this country and more than a billion people worldwide lack access to the health care they need, so they choose between paying for it themselves and being forced into poverty, or not getting it and becoming sicker or dying as a result. Thousands of doctors from over 70 countries are tired of this and have come together to build a solution.” That solution is Human Dx, a worldwide collaborative effort that currently involves more than 100,000 medical professionals from more than 500 medical institutions in more than 100 countries. A massive, living, human network of medical experience and information from which doctors can draw to enhance their diagnostic efforts.

Dubbed a “collective superintelligence” by its creator, Human Dx allows primary care physicians who are uncertain about their diagnostic conclusion or are baffled by a particularly challenging case to input details about the case—including general patient characteristics, symptomology, and patient-approved test results and images—into the system in order to request help from the collective wisdom of worldwide specialists. Within 24 hours, the program returns an aggregated report with the most likely diagnosis and, at times, additional testing and treatment options. 

Underpinning this collective superintelligence is machine learning, including natural-language processing algorithms that search the request for keywords needed to distribute it to appropriate specialists. As responses from those specialists are returned, the machine learning algorithms compare them against an impressive body of previous case studies in order to “validate each specialist’s finding, weight each one according to confidence level, and combine it with others into a single suggested diagnosis,” says Shantanu Nundy, director of Human Dx. “Every part of our process is a combination of human intelligence and machine intelligence to improve results.”

And the results have, to date, been impressive. In order to validate the success of the system, Human Dx partnered with a team of researchers from Harvard Medical School. After testing their system on a wide sampling of challenging teaching cases, they found that “the collective of the Human Dx community paired with machine intelligence arrived at the correct assessment 85% of the time while individual doctors were correct only 60% of the time."  

This is a project meant to be generational, to go the distance, in order to help close the gaps in care faced by the roughly 30 million uninsured or underinsured Americans relying on safety net hospitals that provide low- or no-cost services but lack access to needed specialists and those patients in rural and remote areas around the globe for whom a lack of proximity prevents adequate care. Using innovative technologies to gather, assess and aggregate the collective wisdom of human specialists allows the Human Dx project to promote the type of care patients want and providers want to give and to do so in a way that trims the time to diagnosis and reduces the potential of a misdiagnosis that may lead to substantial harms. 

What the Model T approach to medical diagnostics misses is the importance of collaboration, of partnership, to practitioners and patients alike. As Richard J. Baron, MD and President and CEO of the American Board of Internal Medicine, has said, “Collaborating to find solutions to deliver the right care at the right time is a deeply held value in the medical community.” The success of the Human Dx project is demonstrating that deeply held value in action, revealing the powerful impact of machine learning innovations on the healthcare system when they are used to augment, rather than automate, the diagnostic work of doctors. 

But the absence of meaningful spaces for patients to participate and collaborate in these processes remains. Human Dx is restricted to those with medical expertise, and rightly so to ensure greater accuracy in the diagnostics. There is no patient-centered equivalent, though. No way for patients to collaborate and innovate, no entry point for them to enable better outcomes for their own health and for others around the globe. The absence of patient spaces and technologies may be the result of an industry perspective that the public lacks the knowledge needed to meaningfully participate in these conversations or that they blindly turn all trust over to their healthcare provider, leaving the full responsibility of research, diagnostics and evaluation in their doctors’ hands. Increasingly, however, that is not the case. The tech industry is far from the only benefactor of the data explosion of the past few decades. With access to a global internet and to the technologies that make locating and assessing information faster, easier and more intuitive, answers are available at our fingertips any time of day. 

That access has created an informed and empowered public now aware of the need for their participation in their own wellness journey and now asking for a seat at the medical diagnostics table. Whether or not the industry will be willing to grant that collaborative access to patients is yet to be seen. But, as noted by Scott Melville, CEO of the Consumer Healthcare Products Association, in an interview with McKinsey & Company, “we’ve seen a huge demand in consumers wanting to take charge of their health. This has been driven by lots of different factors. One is technology: consumers have been empowered by technology, they’re accessing more information than ever before, and they’re able to make choices about their own healthcare, whereas in the past perhaps they would have relied upon their doctors to make those choices for them.”

The right people need to be in the room. All the right people, including patients who deserve a voice in these discussions that so shape their lives, who deserve to know how their diagnosis has been reached and, importantly, by whom. Patients who deserve to participate in meaningful ways in that diagnostic effort.  

Written by Marcia Young.