And Then You're Dead

The Autopsy / Case File · Medical Error & Preventable Care

The Diagnosis That Came Too Late

An estimated 371,000 Americans die every year not because a disease was untreatable, but because it was misdiagnosed, missed, or caught too late to matter.

A 2023 Johns Hopkins analysis, built from a decade of malpractice and patient-safety data, put a number on something the medical system had never fully measured: how many Americans are seriously harmed each year by a wrong or delayed diagnosis. The number is 795,000. Most of that harm concentrates in three disease categories, and the gap between how often it happens and how rarely it gets counted is itself part of the story.

And Then You're Dead · September 2026 · 8 min read

Join the discussion →
A physician holds up an X-ray and reviews it with a patient and family member in an exam room
A physician reviews imaging with a patient. Most diagnostic error isn't a single dramatic mistake — it's a process failure spread across visits, specialists, and handoffs.

Diagnostic error doesn't look like the medical mistakes that make the news. There's rarely a single villain, a single chart, a single moment someone can point to. Instead there's a patient with chest pain sent home from an emergency department twice before the heart attack on the third visit; a cough treated as bronchitis for months before the CT scan finds a tumor that was operable a year earlier; a fever and confusion labeled a urinary tract infection while the sepsis underneath it spreads unchecked. Individually, each case reads as an unfortunate judgment call. In aggregate, across the entire American healthcare system, they add up to one of the largest unmeasured causes of death in the country.

Putting a Number on the Invisible

For decades, diagnostic error was the patient-safety problem researchers agreed was serious but couldn't quantify with any precision — unlike medication errors or surgical complications, a misdiagnosis rarely generates its own discrete, codeable event in a hospital's records. In July 2023, a team led by Dr. David Newman-Toker at the Johns Hopkins Center for Diagnostic Excellence published the most rigorous national estimate to date in BMJ Quality & Safety, building a bottom-up model from disease-specific error rates rather than relying on self-reported incident data.

0
Americans estimated to die or suffer permanent disability from diagnostic error annually, across all clinical settings — Johns Hopkins, BMJ Quality & Safety, 2023
0
of those harms specifically classified as deaths, with the remaining 424,000 classified as permanent disability

The study's plausible range ran from roughly 598,000 to just over a million serious harms a year, reflecting genuine uncertainty in disease-specific error rates rather than sloppy math — the researchers were explicit that this was a first rigorous approximation, not a precise census. Even at the low end of that range, the total dwarfs the death tolls of most conditions that receive far more public health attention and far more dedicated federal surveillance infrastructure.

The "Big Three"

The paper's most useful finding for understanding where to intervene isn't the topline number — it's the concentration. Newman-Toker's team, building on an earlier methodology paper that coined the term "Big Three," found that three broad disease categories account for three-quarters of all serious diagnostic-error harm: vascular events (strokes, heart attacks, blood clots), infections (sepsis chief among them), and cancers.

Vascular events, infections, and cancers — the "Big Three" — account for 75% of all serious harms from diagnostic error, not because doctors miss them more often than other diseases, but because they occur so much more frequently in the first place.

The mechanism behind that concentration is arithmetic, not incompetence. The study estimated roughly 6 million vascular events, 6.2 million infections, and 1.5 million cancers occur in the U.S. annually. Diagnostic error rates across all three categories were comparable — a weighted mean of about 11.1% of cases involved some diagnostic error, with roughly 4.4% of cases resulting in serious harm. But because vascular events and infections are so much more common than cancer, they generate far more total missed or delayed diagnoses in absolute terms, even though cancer diagnoses are individually more likely to prove fatal when missed.

0
serious harms, including deaths, attributed to just five specific conditions: stroke, sepsis, pneumonia, venous thromboembolism, and lung cancer — 38.7% of the study's total estimate
0%
of death-or-disability harms specifically attributable to misdiagnosed cancers, versus 22.8% for vascular events and 13.5% for infections

That last split matters for where the clinical stakes are highest. Cancer makes up the smallest share of total case volume among the Big Three, but produces the largest share of the worst outcomes — consistent with the pattern that a delayed cancer diagnosis often converts a treatable early-stage disease into a terminal one, where a delayed stroke or sepsis diagnosis, while still often fatal, has a narrower window in which the delay is the decisive factor.

Why a Missed Stroke Looks Different From a Missed Diagnosis Everywhere Else

Stroke deserves particular attention because it's one of the conditions Newman-Toker's own prior research has focused on most closely. Strokes that present with clear, classic symptoms — sudden one-sided weakness, slurred speech, facial drooping — are diagnosed accurately at high rates in emergency settings. The diagnostic failures concentrate overwhelmingly in atypical presentations: dizziness, vertigo, and imbalance, symptoms that overlap heavily with much more common and far less dangerous conditions like inner-ear disorders. A patient who arrives at an ED with vertigo and is sent home with a diagnosis of benign positional vertigo, only to have suffered a posterior-circulation stroke, is a documented and recurring failure pattern in the diagnostic-error literature — one specific enough that emergency medicine researchers have built bedside exam protocols aimed directly at distinguishing the two.

Sepsis carries a related but distinct problem: its early symptoms — fever, elevated heart rate, confusion, general malaise — are nonspecific enough to mimic dozens of far more benign conditions, and the physiological window in which early antibiotic treatment meaningfully changes the outcome is measured in hours, not days. A diagnosis that's only slightly late can still be a diagnosis that arrives after the window that mattered has closed.

Cognitive Error, Not Negligence

The dominant driver identified across the diagnostic-error literature isn't a rogue clinician or a broken piece of equipment — it's a category researchers call cognitive error: a physician's reasoning process anchoring on an initial, plausible-sounding explanation and failing to revisit it as new information arrives that should have prompted reconsideration. Premature closure, confirmation bias, and failure to consider a broad enough differential diagnosis show up again and again in root-cause analyses of serious diagnostic-error cases. These aren't failures unique to careless doctors; they're documented features of human clinical reasoning under time pressure, high patient volume, and incomplete information — which is precisely why the fix, in the patient-safety research literature, tends to point toward systems (structured checklists, mandatory second-look triggers for high-risk presentations, better diagnostic decision support) rather than individual blame.

Why the Number Stayed Hidden

Part of why 795,000 is a genuinely new figure, rather than a widely cited one, is structural: death certificates record a cause of death, not whether that cause was identified correctly or on time. A patient who dies of sepsis that was originally misdiagnosed as a urinary tract infection has "sepsis" recorded on the certificate — the diagnostic failure that preceded it leaves no trace in vital statistics. The same is true of most disability outcomes; a stroke survivor left with permanent impairment because treatment was delayed by a missed diagnosis is coded, in most systems, identically to a stroke survivor who received textbook-fast care and suffered the same outcome anyway. The underlying event that made the difference — the diagnostic error itself — is invisible to the data systems built to track cause of death and disability.

0%
weighted-mean diagnostic error rate across the Big Three disease categories, per disease case
0%
weighted-mean rate at which a diagnostic error in these categories results in serious harm — death or permanent disability

That invisibility has downstream consequences beyond just public awareness. Research funding, hospital quality metrics, and malpractice-insurance risk models have historically been built around the causes of death and injury that are directly codeable — medication errors, surgical complications, hospital-acquired infections — precisely because those events are countable in ways diagnostic error, until recently, was not. A problem this size, without a matching data infrastructure, has effectively been competing for institutional attention while invisible on the scoreboard that determines where attention goes.

And Then You're Dead

Nobody's death certificate says "diagnostic error." It says stroke, or sepsis, or metastatic cancer that was operable eighteen months earlier — the disease that eventually did the killing, with no official record of the missed appointment, the symptom read as something else, or the scan that got ordered too late to change anything. The scale only becomes visible when someone builds the model backward from disease-specific error rates, and even then, the number is an estimate bounded by real uncertainty, not a body count anyone is actually keeping. For three-quarters of the people this happens to, the disease itself was never the unsolvable part.

That's it. That's the whole thing.

Comments

Loading…

Comments are anonymous and moderated.

Sources

Photo: "Doctor Explains X-ray to Patient," by Rhoda Baer, courtesy of the National Cancer Institute, via Wikimedia Commons (public domain, U.S. government work).