Failure to Rescue
Why Medicine Is Not the Third Leading Cause of Death
The claim that medical error is the third leading cause of death in the United States has become one of the most persistent and corrosive slogans in modern medicine. It is repeated in academic settings, policy discussions, media coverage, and patient safety campaigns as though it were a settled fact. The claim originates with a 2016 paper by Martin Makary and Michael Daniel published in The BMJ1 and has been reinforced more recently by a 2023 Johns Hopkins–affiliated modeling study on diagnostic error.2 Together, these papers have shaped a public narrative in which physicians are portrayed as killing hundreds of thousands of patients each year.
That narrative is wrong. Not because medicine is free of error, and not because care is always good. The problem also is not merely methodological, although the methodology is weak. The deeper issue is conceptual. These studies quietly redefine what it means to cause death, collapsing the distinction between actively harming a patient and failing to rescue one from a lethal disease process. That distinction is foundational to medicine, law, and moral reasoning. Erasing it produces impressive numbers at the cost of intellectual honesty. In doing so, they mislead the public and distort how medicine is judged.
The 2016 BMJ paper is the origin of the headline. Makary and Daniel did not examine individual deaths or review patient charts. Instead, they reviewed prior literature estimating preventable adverse events among hospitalized patients, extrapolated those estimates to national admission data, and concluded that medical error would rank third among causes of death if it were included on death certificates. Because medical error is not currently coded as a cause of death, they proposed that it should be.
That proposal quietly reframed the question. Rather than asking how many patients die from disease despite medical care, the paper asked how many deaths occur in the presence of care that was imperfect. Those are not equivalent questions. The leap from the latter to the former is what allows medical error to be rhetorically elevated to the status of a leading cause of death, even though no new deaths are identified and no causal mechanisms are demonstrated.
The 2023 Johns Hopkins diagnostic error study follows the same pattern, albeit with more refined modeling and narrower scope. The study does not examine actual patients. It does not review charts, reconstruct clinical timelines, or adjudicate physician decision-making. Instead, it begins with national incidence estimates for a selected group of serious diseases and applies assumed diagnostic error rates derived from heterogeneous prior studies. It then assumes that a portion of resulting deaths or permanent disabilities would have been prevented with earlier diagnosis and attributes those outcomes to diagnostic error. Through a series of statistical transformations, they arrive at an estimate of how many patients experience “serious harm,” including death, attributable to diagnostic error.
At no point do the authors demonstrate that any specific patient who died experienced a missed diagnosis. At no point do they show that earlier diagnosis would have changed the outcome for any individual. The patients counted are statistical abstractions. The attribution of error is assumed rather than observed, and the preventability of death is presumed rather than proven. The deaths are hypothetical in the only sense that matters clinically: no one can say whether any given person who died was misdiagnosed, diagnosed late, or diagnosed as early as was reasonably possible.
The second problem follows directly. Even if some proportion of patients experienced delayed or missed diagnoses, the study assumes that earlier diagnosis would have prevented death or permanent disability. That assumption is doing nearly all of the moral work in the analysis, and it is unsupported.
Many of the conditions included in the model carry substantial mortality even under ideal circumstances. Sepsis kills despite rapid recognition and guideline-concordant care. Stroke outcomes vary widely even with immediate intervention. Aggressive cancers often progress lethally regardless of detection timing. Earlier diagnosis does not guarantee rescue. To treat all delayed diagnoses as preventable deaths is to confuse possibility with probability and hope with causation.
But even these flaws are secondary to the most important error, which is categorical rather than statistical: the conflation of failing to rescue with actively causing death.
Even if every assumption in the study were granted, even if every death counted represented a real patient who received suboptimal care, it would still be incorrect to say that doctors killed those patients.
A physician who fails to rescue a patient from a lethal disease has not caused that patient’s death. The disease is what has killed the patient. This distinction is foundational to medicine, law, and moral reasoning. It is the difference between action and inaction, between harm inflicted and harm not averted.
If a physician administers a lethal medication, performs a wrong-site surgery, or introduces a harmful intervention that directly kills a patient, then the physician has caused the death. That is iatrogenic harm. If a physician fails to diagnose a disease in time to prevent its progression, the physician may have provided poor care. The physician may even have committed malpractice. But the cause of death remains the disease.
Failure to stop a process is not the same as initiating it. This distinction is not semantic. Medicine is not an obligation to guarantee survival. It is an attempt, under uncertainty, to alter probabilities. When that attempt fails, the failure does not retroactively convert disease into homicide.
None of this is a denial that bad care exists. Diagnostic error, missed diagnosis, and delayed diagnosis are real phenomena. They can reflect incompetence. They can constitute malpractice. They can prolong suffering, and in some cases directly worsen outcomes. A substantial portion of modern medicine is practiced under political, financial, corporate, and bureaucratic constraints that actively obstruct good care even when competent physicians are prepared to provide it. Much of my own work has focused on precisely these failures and on the policies and incentives that make them more common than they need to be.
Acknowledging those realities, however, does not require accepting the claim that medicine itself is a leading cause of death. Bad care does not automatically imply causative killing. Poor performance, systemic obstruction, and even negligence do not transform disease mortality into physician homicide. One can condemn malpractice without rewriting the cause of death.
The distinction matters because medicine operates under uncertainty. Earlier diagnosis improves probabilities. It does not guarantee survival. Retrospective certainty does not imply prospective negligence. To treat every bad outcome as proof of culpability is to deny the probabilistic nature of clinical care.
There is a simple counterfactual test that exposes the flaw in the “medical error kills” narrative. Would these patients have been better off had they never sought medical care at all?
For the deaths counted in these studies, the answer is plainly no. These studies are not about healthy patients who are killed by a physician administering a poison or surgically removing a vital organ by mistake. These are patients who were sick and dying from a disease, and the supposed error is that their trajectory was not identified and reversed quickly enough. Clearly, had these patients stayed home, avoided hospitals, and never encountered physicians, they still would have died of the same diseases. The outcome would have been unchanged. The only difference is that their deaths would have been recorded honestly, as deaths due to sepsis or cancer or stroke, rather than reassigned after the fact to the category of medical error.
A cause of death cannot coherently be labeled a medical error if the outcome is identical in the absence of medical care. To suggest otherwise implies that abolishing the healthcare system would reduce mortality by eliminating diagnostic delay. That conclusion is absurd, yet it is the logical endpoint of the third-leading-cause framing.
The popularity of this narrative cannot be separated from a broader cultural shift. Patients are increasingly told that they are entitled to medical care, that they should not have to pay for it directly, and that it should be comprehensive, immediate, and effective regardless of circumstance. Lifestyle choices are treated as morally neutral. Outcomes, however, are treated as obligations owed.
Within that framework, death becomes unacceptable. If care is a right and care is assumed to be capable of saving you, then dying must represent a failure. And if it is a failure, someone must be responsible. Reframing deaths from disease as deaths due to medical error implicitly asserts that patients have a right not to die from these conditions and that if they do, it must be someone else’s fault.
But no such right exists. There is no entitlement to rescue from biology. Medicine can alter probabilities, not abolish mortality. Treating death as evidence of wrongdoing is a denial of biological inevitability.
This entitlement logic also explains why compensation is so often reframed as greed. If medical care is a right, then those who provide it are not offering a service but discharging an obligation. To expect payment for fulfilling an obligation appears immoral. To fail in that obligation appears culpable. Physicians are thus cast as both indispensable and suspect, powerful enough to save lives yet morally obligated to do so on demand.
The downstream consequences are already visible. When every bad outcome is framed as error and every error as culpability, clinicians respond rationally. They practice defensively. They order excessive tests, pursue low-yield imaging, and follow rigid protocols not because they believe these actions improve care, but because documentation and conformity provide legal shelter. Over time, judgment gives way to checklists, and experience yields to algorithms.
Algorithmic medicine is often presented as progress. In reality, it is a shield. It offers protection against liability by allowing clinicians to say they followed the pathway, even when the pathway is poorly suited to the individual patient. In a culture that treats death as evidence of wrongdoing, discretion becomes dangerous. Deviation becomes reckless. Human judgment becomes a liability.
The claim that medical error is the third leading cause of death persists not because it is true, but because it satisfies a cultural demand. It reassures the public that death is optional, that survival is owed, and that when biology prevails it must be because someone else failed.
The 2016 BMJ paper and the 2023 Johns Hopkins modeling study do not show that physicians are killing patients. They show that disease remains lethal despite medical care and that medicine operates under uncertainty with imperfect tools. Reframing those limits as culpability does not make care safer but rather more dishonest.
Medicine cannot promise rescue. It can offer effort, skill, and probability, but not guarantees. A society that treats every death as a failure and every failure as a moral offense will not eliminate mortality. It will only ensure that fewer people are willing to accept the responsibility of trying to fight it.
Makary MA, Daniel M. Medical error-the third leading cause of death in the US. BMJ. 2016 May 3;353:i2139. doi: 10.1136/bmj.i2139. PMID: 27143499. https://pubmed.ncbi.nlm.nih.gov/27143499/
Newman-Toker DE, Nassery N, Schaffer AC, Yu-Moe CW, Clemens GD, Wang Z, Zhu Y, Saber Tehrani AS, Fanai M, Hassoon A, Siegal D. Burden of serious harms from diagnostic error in the USA. BMJ Qual Saf. 2024 Jan 19;33(2):109-120. doi: 10.1136/bmjqs-2021-014130. PMID: 37460118; PMCID: PMC10792094. https://pubmed.ncbi.nlm.nih.gov/37460118/