Published in the Blog of the APA
Health care in the United States is extraordinarily expensive. To be precise, in 2024, total health spending in the US was about $5.1 trillion or 18.2% of our GDP. Projections to 2031 put that figure at $7.2 trillion, or about 20% of projected GDP. The Medicare program surpassed $1 trillion in 2024, and it is projected to cost more than $15 trillion over the next decade. Approximately 15% of the US population is currently over the age of 65, yet they are responsible for 37% of all health spending. The sickest 5% of our population in any given year will use almost 50% of health care spending in any given year. In 1960, we were spending only 5.2% of our GDP on health care.
What explains this decades-long escalation in healthcare spending? The short answer is that half of those annual increases are due to very costly, emerging, life-prolonging technologies. A prime example of what I have in mind are the targeted cancer therapies (more than 150) and immunotherapies for metastatic cancer patients. Collectively, they comprise precision medicine because they are designed to target the specific genetic mutations that drive various cancers. Unfortunately, most of these cancers become resistant to these drugs within a year, which means the cancer grows unless another of these targeted therapies is available. None of these drugs are curative, and most will have costs exceeding $200,000 per year. For most patients, the gains in life expectancy will not be much more than a year. More than 610,000 Americans died of cancer last year. Notably, 87% of all cancers occur in individuals over the age of 50.
We might be inclined to heave a sigh when we read this and comment on the misfortune of these individuals. However, there might be something ethically problematic about that sentiment. There are today 32 million Americans without any health insurance at all, and another 45 million who have very inadequate, employer-provided health insurance. What “inadequate” means is that none of those 77 million individuals and their families could afford any of these targeted cancer therapies. It also means that they are not among the 5% of the population with extraordinary health needs requiring costly care. Many in the 5% group will have their lives saved by some costly medical technology, perhaps for more than $1 million. Others, however, will perish despite extraordinary medical efforts and expenditures. This is where we will encounter some significant challenges associated with distributive justice and the rapid expansion of AI in medicine as a predictive and prognostic tool.
Let us start with a quick definition of a wicked problem. It is “wicked” because there is a conflict of reasonable, legitimate ethical or social values regarding some policy issue. No matter what we do to address the issue, the result will be as ethically problematic or worse than the situation we started with. In 1972, Congress passed the End-Stage Renal Disease amendments to the Medicare program. Those amendments would have the federal government pay for a kidney transplant or dialysis for anyone with end-stage kidney disease. Age, employment, insurance status, and so on were all irrelevant.
The problem at the time was that few people could afford the annual cost of dialysis ($90,000 per year today) or a transplant ($200,000). That program today costs $54 billion and sustains the lives of 570,000 patients. This is a wonderful outcome, except for those other patients with end-stage liver disease, or heart disease, or lung disease who wonder why the federal government does not have a program to pay for their life-prolonging transplants (costing as much as $500,000). That seems unjust. However, if the government paid for all transplants, metastatic cancer patients would expect the federal government to pay for these targeted cancer therapies. The same would be true for patients with comparable, very expensive, life-threatening medical problems. This is a paradigmatic wicked problem.
We return to the 77 million Americans and their families who cannot afford the very effective but costly life-saving or life-prolonging interventions. These are individuals who would gain a decade or two of extra life if they could afford the costs, while enormous sums are lavished on individuals who can only achieve an additional year or two of life. This appears to be an unfair or unwise use of limited healthcare resources. It also appears to be indecent, wholly lacking in compassion, when we have a very effective life-prolonging technology from which these individuals could benefit significantly. Of course, many will say that among those 5% needing very costly medical care, we simply do not have the ability to identify whose life will be saved and whose life will be lost before the fact. However, that may be about to change because of the emerging use of AI as a predictive and prognostic tool. Should it change?
Consider this example. We have something called CAR T-cell therapy to treat advanced blood cancer patients (leukemia, lymphoma). The technology involves extracting the T-cells from a patient and having them re-engineered over four weeks to attack the distinctive features of their cancer. The front-end cost of that technology is $475,000, and 40% of those patients will experience a very costly reaction when those T-cells are re-infused, either cytokine release syndrome or some form of neurotoxicity, which will add $200,000 to the front-end costs. What we know statistically is that 30% of those patients will fail to survive another year. The remaining 70% of those patients will gain one to four extra years of life. There are 70,000 patients each year in the US who will die of a blood cancer.
What we are on the edge of accomplishing is using AI and many predictive biomarkers to identify with 90% confidence before the fact hat 30% that will not survive a year with CAR T-cell therapy. Obviously, 90% is not 100%. AI could get it wrong 10% of the time, and we would not know who suffered because of some inherent unpredictability in much of medicine, even precision medicine. This is part of the wickedness. Those patients would be denied CAR T-cell therapy because the AI protocol told us that this did not represent a cost-effective use of limited health care resources. It becomes a justice issue if those saved resources could be redirected to providing more effective and cost-effective health care to those 77 million individuals at risk of a premature death or greatly diminished quality of life. Should we do that?
Obviously, we have no reason to believe that this use of AI would be limited to these cancer patients. We should imagine that it would be applied to the 5% of patients responsible for 50% of all healthcare costs. That would likely generate enormous savings that could be redirected to those 77 million individuals and their families with no insurance or only very marginal insurance. Should we do that?
We face a wicked problem. I say that because we would have to tell individuals who could experience some gain in life expectancy, because the technology was there that could accomplish that, that we were going to deny them access to that technology for the sake of nameless, faceless, statistical others. We were confident in the predictive capacities of the AI technology that generated that result. In other words, we would be using this technology to deliberately bring about the “premature” death of these patients (who many would regard as being among the “medically least well off”).
Perhaps we, as philosophers, would soothe our consciences with the thought that this was an omission, allowing nature to take its course, rather than an active killing of these patients. Physicians, patients, and their families are unlikely to embrace with equanimity this philosophic nicety. After all, the resource was not scarce in some absolute sense. It would be because the cost of saving those life-years was too high relative to the millions of other potential life-years that could be saved at a much lower cost among the 77 million other individuals. We might imagine in the US that AI would be the key to making a comprehensive package of healthcare available to everyone, given the potentially achievable savings. I ask my readers if you would endorse this solution for your future possible self (not knowing your future health needs or insurance status)? Is this bit of wickedness ethically tolerable for you?
