This post also appears on TechCrunch.
The “practice of medicine” developed through tradition and the experiential evolution of best practices with small-scale medical research studies can be substantially improved through the “science of medicine” with statistically better-validated data and conclusions. Much of the current practice is driven by conclusions derived from partial information about a patient’s history and symptoms, incomplete medical understanding based as much on opinions as validated science, interacting subjectively with known and unknown biases of the physician, hospital and healthcare system.
Technology will reinvent healthcare. Healthcare will become more scientific, holistic and consistent; delivering better-quality care with inexpensive data-gathering techniques and devices; continual monitoring and ubiquitous information leading to personalized, precise and consistent insights. New medical discoveries will be commonplace, and the practices we follow will be validated by more rigorous scientific methods. Although medical textbooks won’t be “wrong,” the current knowledge in them will be replaced by more precise and advanced methods, techniques and understandings.
Hundreds of thousands or even millions of data points will go into diagnosing a condition and, equally important, the continual monitoring of a therapy or prescription. Companies like Quanttus are proposing 10,000 vital sign readings per hour, not to mention Applied Proteomics, which already gets 300,000 biomarkers from a blood sample, in addition to thousands of genes, their epigenome, microbiome and more!
During the next decade, we will see systems providing “bionic assistance” to physicians and other healthcare professionals, allowing them to perform at substantially improved levels of expertise like the very best specialists in multiple domains. Inevitably, over twenty years, the majority of physicians’ diagnostic, prescription and monitoring functions will be replaced by smart hardware, software and testing. With bionic assistance, in addition to handling more patients, a physician or nurse practitioner will operate at the level of six specialists managing six areas of care for one patient with multiple comorbidities in a more coordinated, holistic and comprehensive manner. Nurses, enabled by technology, will replace many of the functions doctors perform today. Patients will act as the CEO of their own health: they will be better informed and able to make more educated choices to select a course of therapy. Care will become cheaper, faster, more optimal, accessible and consistent across practitioners.
Early versions of these systems will appear underwhelming and clumsy. By comparison, my first cell phone was the size of a sewing machine and floor mounted to a car with a cumbersome handset cord but has since evolved into today’s iPhone. Similarly, in fifteen or twenty years, changes to healthcare will seem obvious, inevitable and well beyond what we envision compared to the somewhat crude digital health technologies we see today.
The role healthcare professionals will play is hard to define, but I suspect it will change dramatically and involve the empathetic and ethical elements of medicine. A more cooperative system leveraging humans and technological systems may evolve, but the core functions necessary for complex diagnoses, treatments and monitoring will be driven by machine judgment instead of human judgment.
As Atul Gawande notes, studies show “our attempt to acknowledge and deal with human complexity [in human ways] causes more mistakes than it prevents.” In a review of more than a hundred studies, researchers found that “in virtually all tests, statistical thinking equaled or surpassed human judgment.” The inconsistency and biases of humans, coupled with the inability to accurately consider many factors simultaneously, leads to statistical methods outperforming humans.
According to Gawande, “the machine, oddly enough, may be holistic medicine’s best friend….As expert systems begin to take on more of the technical and cognitive work of medicine, generalist physicians will be in a position to embrace the humanistic dimension of care.”
Already, many early versions of these systems have been developed. Alivecors’ ECG iPhone case enables cardiac patients to take hundreds of ECG’s at virtually no cost and is FDA approved to diagnose atrial fibrillation, which it can detect as efficiently as a cardiologist looking at the same tests. Continuous and convenient at-home monitoring of atrial fibrillation becomes economically feasible with this low-cost device, possibly avoiding many strokes. Similarly, Cellscope’s iPhone case imager will soon be able to diagnose a child’s ear infection and suggest a prescription. Lumiata may be able to help decide when to see a doctor and ensure they do not miss a symptom that may affect a diagnosis. Ginger.io monitors mental health patients and can reduce suicides, depression and bipolar episodes by working with a psychiatrists’ nurse. The app can characterize a patients’ behavior more accurately and effectively when the patient is outside of the psychiatrists’ office than a human could.
Medical literature is rife with studies about how the practice of medicine does not meet expectations for acceptable care. Physicians often will make different diagnoses and recommend different therapies for the same patient (according to a study of the Cleveland Clinic’s second opinion line, more than 50-percent of the time!). Purported experts in their respective fields frequently disagree on the effects of basic procedures. For example, a study involving colon cancer experts showed that there was no consensus on the value of colon cancer screening. In another study, cardiologist were given the same patient information and half recommended cardiac surgery whereas the other half did not. Two years later with the same data, 40-percent of cardiologists reversed their recommendation
Medical fact often ends up being wrong yet continues to persist. For instance, prescriptions for antipyretics such as aspirin are typically given to individuals with a fever as is the “practice of medicine”. Yet recent studies show prescribing antipyretics for fever reduction is significantly riskier than allowing the fever to run its course. Stanford researcher, Dr. John Ioannidis, has studied the phenomenon of medical research studies extensively and found that, shockingly, it’s “more likely for a research claim to be false than true”.
Today, misdiagnosis, conflicting diagnoses and general diagnostic error are common: ICU misdiagnoses cause as many deaths as breast cancer and adverse drug interactions cause as many deaths as automobile accidents. Preventable medical errors, often with clinical findings already in the medical record, are common. Biases frequently affect patient care: “premature closure” or coming to a diagnosis too quickly and “recency bias,” where a doctor is influenced by a recent case or article, just to name a few, are common. Conflicting recommendations are even less surprising if you consider that the average U.S. Medicare patient sees seven specialists, and the prescriptions of each specialist are seldom coordinated with the others. Today’s diagnostic error rate is the equivalent of Google’s driverless car having one accident per week; while this would be unacceptable for self-driving cars; this failure rate is permissible in healthcare.
The use of data science will add meaning, and over time, there will be many distinct improvements: (i) validation of what we accept in medical practice about therapies, prescriptions and procedures; (ii) more data feeding into comprehensive and holistic diagnosis and prescriptions; (iii) patient centric decision making with a better understanding and matching of choices to individual preferences; and (iv) invention of new prescriptions, therapies and procedures based on more holistic data about a patient.
The transition will start incrementally and develop slowly in sophistication, much like an MD who starts with seven years of medical school and then spends a decade training with practitioners by watching, learning and experiencing. Expect many laughing-stock attempts by “toddler computer systems” early in their evolution. A three-year-old child makes laughable errors, but this does not imply they will make the same errors as a 21 or 40 year old. We won’t let initial toddler systems make decisions while they are developing. Their mode will be assist, learn and amplify with new generations of systems every two or three years with radical improvements over five to seven generations. To equate early “toddler digital health systems” to what may be possible is naïve.
Innovation will occur swiftly, more like a tsunami than a linear change, but each version — from v0 to v7 and maybe more – will feel incremental and logical, much like how the first mobile phone that was floor mounted to a car looks nothing like iPhone today.
The accumulation of data will accelerate these improvements. The transition to the automated science of medicine will likely occur as an organic process of trial and error, starting with initial technologies and ideas that go through multiple iterations. In 15 to 20 years, data will transform diagnostics to the point where automated systems may displace up to 80-percent of physicians’ medical work. Computers are much better than people at organizing, recalling and synthesizing complex information and keeping up with the latest research. Tasks previously thought to be very difficult such as facial recognition on Facebook can be done by systems just as well as humans. Other examples include self-driving cars, which require more complex and rapid intelligence than medical decision-making does.
Within a decade, we will have doctors making data based decisions using the personalized medical equivalent of a Bloomberg financial terminal for each patient. While many will opine on these possibilities, today’s doctors are not qualified to judge what surprising software technologies may emerge in the future even though they understand the “problem of patient care” better than most of us. Just because a technology does not exist today does not mean it won’t exist in the future. My optimistic view is that we are on a path to create digital health innovations that will exceed human judgment in medical competency.
If I am right, consumers will become the CEO of their own health because of such systems. Over time, mobile sensors, devices and apps will increase data collection and data science sophistication to offer insights to outperform the average physician, although not in every case. While many will choose this path, consumer choice should always be a priority for those who want to use traditional doctors and the current healthcare system.
Not all doctors will change how they practice medicine, but the thought leaders will. The direction of medicine will be self-evident and advantages to patient outcomes will be well documented. Already, we see this shift with people on the fringes of medicine and a growing number are taking steps to facilitate the future. Doctors and nurses will be devoted to the human element of care, while “Dr. Algorithm” systems will assist with diagnostic and prescription work. Systems will improve by learning from the best doctors and becoming increasingly more competent. Doctors at the forefront of medicine and technology will provide better care and show markedly better treatment results; over time, the rest of the world will accept the “science of medicine”.
Sadly, the major problems in healthcare are systemic and do not involve doctors, many of whom are accomplished, caring, honest and compassionate. In the U.S., there is a misalignment of incentives, where organizations maximize revenue at the expense of care (extra surgeries anyone? a hospital inpatient who develops an ulcer increases revenue by $40,000!).
Further, the incredible increase in the amount and complexity of newly enabled data, vast amounts of research, longitudinal health records and medical histories are well beyond what humans can reasonably understand. Add to that the incoming deluge of 10,000 vital sign reads per hour, thousands of genomic data points and expression data, microbiome, proteomic and other data. New sensors and testing will allow for more integrative analysis. Utilizing this data will enable much better and more holistic care that will get progressively better with time.
Because this will go against established thought and practice, I believe deep innovation will most likely come from outside the system. This often occurs with innovators acting naïvely, failing and then realizing they need more knowledge and collaboration within the system. Entrepreneurial teams often add domain expertise to their naïve “fresh piece of paper” re-invention ideas.
Further, the media and society broadly try to assign power to larger entities, like governmental institutions and Fortune 500 behemoths; however, radical innovation seldom comes from them. Did Walmart reinvent retail or Amazon? Did General Motors reinvent electric cars or Tesla? Did SpaceX reinvent space launches or NASA and Lockheed Martin? Most importantly, did big pharmaceutical companies reinvent biotechnology pharmaceuticals or Genentech?
The biggest risks to slowing medical innovation are policies from government agencies, too large a belief in traditional institutions, in addition to resistance from those who may be hurt by these positive social disruptions. In the worst cases, it can substantially slow technology-driven medical innovation, which will force innovation to less developed geographies where traditional medical resources are limited.
There will be many improbable attempts and possibilities for how data and consumer-driven systems will transform healthcare. Although many of these attempts will fail, the few that succeed will determine the future of healthcare as it is driven and transformed by digital health technologies. Improbable does not mean unimportant: we just don’t know which improbable outcome is important. Some improbable scenario today will become tomorrow’s reality.
Biological sciences will continue to be significant, as fundamental scientific research will improve our understanding of biological systems and support complex data science systems. That said I expect innovation cycles to be much longer than those for the digital sciences, which will improve every two to three years. I do believe that digital technologies may do more for medicine in the next decade or two than all the biological sciences combined.
Over time, we will see a 5×5 improvement across healthcare: 5x reduction in doctors’ work (shifted to data-driven systems), 5x increase in research (due to the transformation to the “science of medicine”), 5x lower error rate (particularly in diagnostics), 5x faster diagnosis (through software apps) and 5x cost reduction.
I want to emphasize that my hypotheses and forecasts are only meant as directional guesses rather than precise predictions. These are not absolutes, but rather “truer than not” speculations. Although many disciplines will contribute to innovations in medicine like biological research or new device development, I am primarily focused on the contributions of digital health technologies to medical innovation. They are the most variable and impactful and therefore the hardest to predict in direction, timelines and scope.
We have to imagine what might be possible: Then, we must have the courage to try and make the possibilities a reality.