Well: What We Need to Talk About When We Talk About Health
Oxford University Press, $24.95 (cloth)
Oxford University Press, $42.95 (cloth)
Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again
Basic Books, $32 (cloth)
Criticisms of medicine are as old as medicine itself. Those at any particular time and place reveal something about the prevalent health problems, medical practices, and social, economic, and political currents. During the 1960s and 1970s in the U.S. and Western Europe, for example, criticisms of the privileged status of the medical profession and medicine’s harms and limited efficacy eclipsed older, Progressive-era concerns over the failure to distribute the benefits of modern medicine more equitably. A touchstone for that era’s criticism of medicine is the work of British physician and epidemiologist Thomas McKeown, whose influential book The Role of Medicine (1976) situated what ailed health and health care within the larger spectrum of ideas developed during the previous decade. Comparing that era’s criticisms and solutions to more recent analyses illuminates some noteworthy changes—as well as frustrating continuities.
A myopic commitment to individual health services has endured—despite several decades of critique.
Examining the declining mortality in England and Wales since the early nineteenth century, McKeown concluded that the trend was due not to specific clinical or public health measures but rather to general socioeconomic improvements, especially better nutrition. Critics soon pointed out that McKeown had missed the role of declining fertility, that his nutritional hypothesis seemed more placeholder than proven cause, and that at smaller scales some medical and public health interventions did matter. His work nevertheless became an Ur-text for the nascent field of population health, which focused on understanding and influencing the non-medical, “social” determinants of health—including poverty, neighborhood resources, education, environmental hazards, and crime.
McKeown’s subtitle—Dream, Mirage, or Nemesis?—spoke back to ideas from the previous decade. “Dream” evoked physician Lewis Thomas’s hope, articulated in a much-cited essay reprinted in The Lives of a Cell (1974), that technological improvements, driven by better understandings of basic science and disease mechanisms, would supplant the expensive and ineffective “halfway technology” that characterized much of contemporary medicine. “Mirage” evoked microbiologist René Dubos’s pessimistic ecological view of human limits, articulated in Mirage of Health (1959), according to which health gains against specific diseases run up against limits of biology. For Dubos, curing or preventing problem x inevitably reveals vulnerability to problem y. And “nemesis” evoked the work of philosopher and ex-priest Ivan Illich, who followed up his scathing takedowns of schooling and other large societal institutions in Western capitalist societies by publishing an attack on medical harm, Medical Nemesis (1975).
McKeown rejected all three of these models: Thomas’s techno-utopian solutions, Dubos’s pessimism about biological limits, and Illich’s flattening nihilism. While he did not disavow a role for health care—there was a reason his book was not titled No Role for Medicine—he also argued that the institution of medicine should focus more on care, especially on rehabilitation, relief of pain, better quality of life, and disease prevention. (As part of this vision, he emphasized medical abortion for fetal abnormalities in ways that are uncomfortable to read now.)
It may be striking to readers today that access to medical care did not figure prominently in McKeown’s critique, or those he drew on. But McKeown was writing against the background of decades of de jure universal access in Britain via the National Health Service. Unlike the “old left” focus on guaranteeing equal access to the fruits of medicine and techno-science, countercultural voices largely questioned these institutions—their inner workings, claims of efficacy, and societal dominance. Across the pond in the United States, the enactment of Medicare and Medicaid may have seemed a down payment on soon-to-come universal access, but here too skepticism of the cultural authority of medicine and techno-science loomed large.
A lot has changed in the interim—or has it? Western societies have embraced “evidence-based medicine” to separate the harmful and ineffective from the smaller number of interventions that actually work. Technological innovation has continued unabated. We have sequenced the human genome. We have some targeted therapies against cancer. We have pills that cure Hepatitis C, an infection stealthily spread by iatrogenic (physician-caused) and other means in these interim decades. Mortality continued its modest decline—whether driven by medical advances or by social determinants—until suddenly increasing this decade, driven in part by an iatrogenic opioid crisis. While the social determinants of health are now more widely acknowledged and given lip service in public health, policy, and medical education, the commitment to acute health care services that McKeown questioned has endured; health care now constitutes nearly a fifth of the U.S. economy. The health care access problem in the U.S. did not fade away, as it has in other advanced economies, while trouble within the beast of what has come to be called biomedicine remains. Three recent books underscore this continuity in societal angst about health and health care.
Medical Nihilism, by the philosopher of medicine Jacob Stegenga, echoes Illich’s Medical Nemesis in its sweeping scope and deflation of medical pretentions, but it also takes a more analytical approach. Where Illich emphasized the ways that medicine hollowed out the spiritual dimension of suffering, Stegenga takes aim at what he considers its shaky epistemology. What exactly do doctors know about the treatment of disease? How reliable, really, are their interventions?
What exactly do doctors know about the treatment of disease? How reliable, really, are their interventions?
Stegenga’s answer is mostly negative; he seeks to expose the methodological and epistemic flaws in medical evidence making and regulation. These are large issues, but he focuses almost exclusively on drugs—though similar problems could be said to beset medical devices, surgery, and diagnostic technology and practices. He also shies away from the contextual factors underlying pharmaceutical efficacy, even though pills can work only with the right infrastructure, maintenance, and professional competence. Instead he relies heavily on a preexisting critical literature within modern medicine—especially work by the physicians Marcia Angell, Richard Horton, and John Ioannidis. He explains how and why most medical research is “malleable,” subject to many nefarious influences that direct the way studies are done and interpreted. Given the biased and weak evidentiary foundation for a large fraction of pharmaceuticals, Stegenga argues we should have “little confidence in the efficacy of medical interventions.”
Central to Stegenga’s thesis is the idea that many drugs today do not treat genuine disease (as opposed to non-genuine targers of therapy, such as risk factors for heart disease). This is the least persuasive part of his argument. The idea that drugs and other therapeutics should affect the causal chain of a disease to be truly efficacious is a narrow and idealized view of efficacy. It ignores a wide literature on so-called “social efficacy,” and it flies in the face of our ignorance of the causal chain behind much of the human suffering that motivates patients to seek clinical help. People take pills to not lose a day of work, not just to cure a cold. There is also the non-trivial problem that so much suffering arises not from diseases per se, but rather from medical interventions to treat them. The CDC estimates, for example, that over a million ER visits annually are for adverse drug reactions. Unsurprisingly, Stegenga finds that most drugs fall far short of his ideal of a magic bullet targeting a highly specific molecular mechanism of disease and little else.
I am sympathetic to Stegenga’s meta-Bayesian argument (drawing on the mathematical theory that relates our knowledge of prior probabilities to estimates of present ones in light of new evidence). If we accept—for all the reasons he describes—that the received effectiveness of medical practices is in general exaggerated and harms downplayed, then we should approach any new evidence of efficacy as if it were an evaluation decision made under conditions of uncertainty. We should assume a very low “pre-evaluation” probability of the “true” effectiveness and safety such that any particular new claim has a low predictive value of being truly effective. In other words, add a large, extra dose of skepticism to your evaluation of any new evidence. This skepticism has a very long history, captured in a nineteenth-century French medical aphorism to use drugs when they are new, while their effects last.
Stegenga does not dream of a better, more effective medicine driven by new insights into basic science; he emphasizes the brake over the accelerator. He ends his book with a heartfelt call for “gentle medicine”—some mix of doing less, being more skeptical, and redressing health problems by non-medical means (including addressing social determinants).
Cardiologist Eric Topol, in Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, shares some of Stegenga’s criticisms of modern medicine’s wastefulness and harm, but he places more emphasis on the dehumanized way patients are treated and the stress imposed on physicians by the bureaucratic aspects of care. His cure is artificial intelligence (AI).
Book after book has declared a new revolution in therapeutics just around the corner.
Topol believes that the efficiencies of AI will lead to more time for humane medicine. Maybe this prophecy will be accurate, but he presents little evidence to support it. In fact, so many of his accurate observations about our troubled present suggest the opposite conclusion. Topol knows that the positives of electronic medical records—for example, “one click” prescribing and instant access to X-rays and laboratory tests—are counterbalanced by all kinds of inefficiencies for physicians: the labor involved entering information, say, or the dehumanizing impacts of typing rather than listening, and the massive signal-to-noise problem introduced by pre-constructed templates to document levels of service for best billing. Such ironies of technology—nicely captured in the title of the historian Ruth Cowan’s 1983 history of household technologies, More Work for Mother—are a consistent theme in historical studies, and Topol does little to overcome my suspicion that virtual medical assistants and other AI-based technologies coming down the pike will not simply free up time for making “healthcare human again,” but rather create new kinds of labor and distractions.
Topol shares some of the techno-optimism from a half century earlier, but it springs from advances in information technology rather than better molecular understandings of health and disease. We currently have helpful (deep) and harmful (shallow) diagnostics and therapeutics, he argues, but we are often at pains to distinguish one from the other, and to determine for which individuals a drug or practice might or might not work. Topol impressively corrals disparate problems into this information frame: overdiagnosis and overtreatment, the challenges of clinical diagnosis, and the poor evidentiary basis of therapeutics choice. In doing so, he is onto something about our zeitgeist. Unlike a half century ago, we now suffer from information overload—an exacerbated crisis in knowing and judging as a result of an explosion of interventions, many launched into the marketplace and onto our bodies on the shaky evidentiary basis Stegenga presents.
Like older and present futurists, Topol believes that we can innovate our way out of the hole we have dug for ourselves—all while saving money, reducing suffering, and liberating time. As for the latter, not only are there better and less technology-driven ways to save time than AI, but inhumane care is much more than a problem of time. Topol also sidesteps the way the technologies he describes inevitably privilege one group over another. The surgeon, writer, and now health care executive Atul Gawande, for example, convincingly demonstrates that the electronic medical record (EMR) privileges the needs of health provider systems (by enabling better surveillance of practitioners, more portable operations, and improved billing yield) and partly patients (who have better access to their records) over the needs of clinicians. Many U.S. clinicians cannot buy or innovate their way out of the EMR hassles with scribes in India and the like. Not only do clinicians stare at screens rather than patients; they also lose their ability to quickly scan records for key information.
Topol believes that we can innovate our way out of the hole we have dug for ourselves.
Today we are obsessed less with “halfway” technologies and incomplete fixes than with overtreatment and signal-to-noise problems in existing evidence and data. While book after book has declared a new revolution in therapeutics—thanks to new genetic and immunological knowledge—just around the corner, Topol’s vision represents a longing more specific to our time—the desire for a better way of knowing and judging. Underwriting this desire is a tremendous faith that probabilistic knowledge—linking aspects of individual biological characteristics to outcomes—can help to direct preventive and therapeutic actions, even as there is good evidence that many of the conditions we most want to prevent and treat, including many prevalent cancers, are the result of largely unpredictable stochastic events, evolving within the individual and shaped by poorly understood selective forces. A soft, unstated determinism lurks in Topol’s hope that AI technologies—sift through as much complex patient data as they may—will lead to significant, actionable medical responses, hyped here and elsewhere as a new age of personalized and precision prevention and treatment.
It is striking that Topol has little to say about a very different solution, increasingly advocated by clinicians and researchers: large, simple clinical trials that ask important, basic questions (“are anti-depressants doing net good or harm?”), dictated by clinical need rather than the financial interests of pharmaceutical companies, device makers, and other self-interested parties. (On this front, Topol’s objectivity is threatened by the many consultant and other roles he plays with the industries he studies; his interactions with these firms constitute a lot of the book’s content.) I remain skeptical of Topol’s and others’ faith that the solution to the limitations of existing clinical experimental methods such as the randomized controlled trial—how to translate their findings to real-world conditions and individual variation, or how to evaluate evidence from limited endpoints to ones that matter—will be overcome with analyses of data collected under non-experimental conditions. Confounding, selection problems, and other biases run amok, and the very recording of data is subject to impossible-to-disentangle nefarious influences. “Those same biases,” Topol acknowledges, “as part of human culture, can become embedded into AI tools.” There is also the problem of how much confidence clinicians, patients, and consumers will have in conclusions produced by black-box technologies such as machine learning.
A soft, unstated determinism lurks in Topol’s hope that AI technologies will lead to significant medical responses.
For all this hope, Topol is highly critical of the hype of existing AI-influenced technologies. His chapter “The skinny on deep learning” is one of his best. Topol cringes when others use the success of neural networks in gaming to argue this technology will be a “‘game changer’ when it’s applied to purported medical progress.” This skepticism falters, however, when it comes to the promise of AI in the near and not-so-near future. The book begins with an anecdote about how Topol’s orthopedist missed the diagnosis of a rare problem Topol suffered that should have led to a different type of postoperative care. Topol argues that a machine learning analysis of the existing literature and his own personal history could have led to the diagnosis that his orthopedist missed.
Yet here and in other places, Topol fails to give enough evidence for why this will be the case. Indeed, untrue and unhelpful diagnoses might be just as likely a result of big data crunching. Topol calls his orthopedist’s approach “robotic,” without any irony, as he extols the future of virtual medical assistants and hypothesis-free data massaging. In his discussion of biomarkers (measurable indicators of disease), Topol accurately points out all the limitations and the hype from present-day promoters, but reflexively states “there’s hope we’ll get there someday, and, for those eager to see substantive progress, I think there’s no reason to be depressed.”
There is little argument that AI-based technologies can improve upon human pattern recognition and lead to efficiencies and improvements in diagnostic accuracy. I dictated part of this essay on Dragon Naturally Speaking, a piece of software I have used for more than twenty years, during which time its accuracy has markedly improved. AI-based technologies are similarly capable of examining radiological images and pathological specimens and, along with telemedicine, leading to improvements in accuracy and efficiency. None of this, however, adds up to an advance in thinking or judgment, as Topol foretells. Clinical judgement involves a wide mix of skills not accounted for here: eliciting histories, assessing who and what is reliable, creating a sequence of diagnostic and therapeutic maneuvers that might help the patient before you (including ones that might bypass diagnosis altogether), discounting or emphasizing different sources of extant knowledge on the basis of likely biases and historical knowledge, discerning what problems can be ignored and which need urgent attention, and so on.
It is important to evaluate hopes and dreams soberly and seriously. Medical history teaches us that imagined futures cast their shadows in the present. What we believe most about the future dictates present investments and priorities.
Tracking closely the policy implications of McKeown’s work from fifty years ago, Sandro Galea’s Well: What We Need to Talk About When We Talk About Health asks why we still avoid talking about health in population terms, instead focusing myopically on the provision of clinical health services for individuals. His cure is to change the conversation by persuasion and research, and he offers suggestions for employing political and social power in different ways.
Galea challenges us to face the reality that your zip code is much more predictive of your health than your genetic code.
The book consists of short, accessible chapters that read like a secular catechism of the old and new faith in social medicine, social epidemiology, and population health. Galea skillfully distills complex research findings and connecting our present overinvestment in health care to our values, psychological flaws, and politics. He hones in on the consequences of our faith in unfettered individualism and the importance of framing health as a collective value. He marshals evidence and common sense to show that we are all swimming in the same water, that our health is shaped by our past, and that our most fateful individual health choices are not socially or politically unfettered. He shows, in particular, how public preferences for common-sense policies (say, gun control) can be blocked by the power of special interests, and offers a theory of political change within which he situates public health research and activism. This is a welcome corrective to arguments about inequality and poor health that end with a naïve call for redressing economic and racial inequality, without much thought as to how this might get accomplished.
Occasionally Galea, like other population health advocates, comes down too hard on our commitment to investments in health care as the lone problem we face. My objection is largely political. We need to meet the public where they actually are, and many Americans are understandably focused on access to individual health services, preventative as well as curative. Linking our failure to think and act more communally to our blind faith in health care has not been a winning argument. Investments in health care and the social determinants of health do not necessarily form a zero-sum game. Galea acknowledges that health care spending can lead to health gains, especially for those over the age of seventy-five. It is perhaps a shrewder political move to link guarantees to better health care access to investments in public health and social and economic policies that lead to improved population health.
Investments in health care and the social determinants of health do not necessarily form a zero-sum game.
Galea adds a persuasive epicurean touch to the argument in contending that the politics of health and health care necessarily involve some agreement on what we mean by a good life. “The ideal life,” he argues, “would combine the avoidance of pain with the moderate, enduring joys that come with friendship, community, and other goods that can be sustained over a lifetime.” “Public health informs a different, more expansive notion of pleasure,” he continues, “one rooted in the epicurean ideal of moderation, community, and long-term sustainability, to the end of having better health and fuller, richer lives.”
The book thus mounts an implicit challenge to observers of the health scene like Topol, who stay largely within the arena of health care and individual behavior. Galea challenges promoters of personalized and precision medicine to explain and respond to the reality that your zip code is much more predictive of your health than your genetic code.
Galea concludes that we have a profound need for more evidence of how social determinants work, in order to better understand just how inequality gets under the skin. “Our knowledge of these conditions,” he argues,
is caught in the middle of a “spiral” trajectory—we know enough to be certain about the link between health and social conditions (housing, money, power, politics, transportation, and racial injustice—the topics in this book), but we do not yet have all the facts we need in order to answer the next tier of questions about the nature of this influence.
He contends that “to reproduce the tobacco effect on other fronts in the pursuit of health, we need the best possible data so that you will have a knowledge base on which to proceed.” There is some irony in this analogy, since one thing that the tobacco story tells us is that you can have political success without knowing the precise pathophysiological mechanisms by which tobacco leads to lung cancer (and many other terrible health outcomes, for that matter). As the historian Allan Brandt has shown, we just needed an iota of evidence that individual smoking could damage innocent others, especially children, to undermine the libertarian arguments mobilized by tobacco interests against regulation.
So what is the actual advantage of creating more precise knowledge about the mechanisms by which population-level associations work? Do we really need a weathervane to tell us the way the wind blows? I am skeptical that more, inevitably contested evidence of how these associations work is going to change hearts and minds, let alone policies. As Galea acknowledges elsewhere, the major causal role for social determinants in individual and population health is now settled for the audience that thinks this insight is actionable. If we need more evidence, maybe it is to determine which population-level interventions work—a clinical trial of social and economic policies on a broad array of health measures.
The underlying arguments for an upstream approach to the health of populations—redressing social determinants, the historical, political, and environmental factors that put people at risk of risks—have hardly changed in the last few decades. These ideas, if not the policies and reforms they imply, have become more mainstream.
The challenge remains to answer questions that matter to suffering patients rather than special interests.
The fall of the Berlin Wall catalyzed this diffusion. While the sudden collapse of communist regimes largely deflated hard-left arguments focused on the injustice of inequality per se (though strains of this rhetoric can once again be found at the more left-leaning pole of contemporary Democratic party politics), it may have drawn attention to the impact on inequality on health. In many Western liberal nations, there was for a time still traction to be gained by pointing out that the different types of social and economic inequalities could have substantial health consequences, even when equality per se as a goalpost was drifting away. There remained some legitimacy in lamenting that one’s health and lifespan were determined by one’s social position, race, and neighborhood. The simple message that “inequality was killing us” became, for a period, the last legitimate inequality argument with wide appeal (and continually reinforced by new evidence, such as the recent downward slide of American mortality).
There may be nothing much new under the sun in the angst about medical harm, but intervening decades of evidence-based medicine—the overpromising and the underdelivering—have led to widespread skepticism that better evidence making and evaluation alone will navigate us out of the problem of knowing what medical interventions are safe and efficacious. This is the fertile soil upon which utopian ideas about the potential of AI and big data to come to the rescue.
Unlike fifty years ago, though, today we arguably face a complexity crisis in clinical and consumer judgement. The massive rise in medical interventions has made bodies different from one another, making the translation from the aggregate to the individual—a constant challenge in medicine—all the more difficult. Add to this the sheer numbers of diagnostic and therapeutic products, devices, and practices to evaluate, the often limited endpoints used to initially gain regulatory approval (which might or might not correlate with outcomes that do matter to us), and the fast pace of innovation and knowledge production. Health care costs continue to skyrocket, too. Our concern with overdiagnosis, overtreatment, and similar problems suggests we know that something bad is happening at the aggregate level, but have few ways to respond as individuals—or to influence the upstream flow of medicines, devices, practices, screening programs.
There are no easy solutions to this crisis. I doubt robots or other AI-decision aids will help very much. A shift in the priorities of medicine along the lines of what McKeown advocated—toward care, relief of pain and suffering, rehabilitation and better functioning—would help to slow the upstream creation and mass diffusion of interventions we are often ill prepared to evaluate. Americans’ health care and health would also benefit from different elements of Stegenga’s “gentle medicine” and the social and economic policies that follow from core social medicine truths rearticulated by Galea.
Yet a political strategy for change remains elusive. We have cooked the values of the market into our system of research and development and provision of health care, but we also want medicine to deliver only real benefit and minimize harm. If our faith in the power of medicine to deliver something that matters were a complete fantasy, it would be easier to embrace medical nihilism, whether old or new. Some promising changes have recently emerged from clinicians themselves; one is the Choosing Wisely campaign, which enlisted specialty societies to find ways of curbing the use of low marginal benefit procedures. However, such responses are necessarily limited. They amount to tinkering with better ways to close the barn door after the horse has left. Fifty years on from the “question authority” cultural moment, the challenge remains to find ways to ask and answer questions that matter to caring clinicians, suffering patients, and bewildered consumers, rather than those of special interests.