What is overdiagnosis?
Overdiagnosis is finding a real abnormality that meets the definition of a disease but that would never have caused symptoms, harm or death if it had gone undetected. The diagnosis is accurate, not a mistake; the problem is that the “disease” was harmless for that person, so detecting and treating it can only subtract from their health rather than add to it. In short, overdiagnosis turns a healthy person into a patient for no benefit.
The word describes a population-level pattern more than any single case. For any one person, it is usually impossible to know whether their diagnosis was an overdiagnosis, because the abnormality was treated and can no longer prove what it would have done if left alone. Overdiagnosis is visible only in the aggregate, when large groups of people are tracked and it turns out that far more disease was detected than ever went on to cause trouble.
Why overdiagnosis is not the same as being wrong
Overdiagnosis is often confused with error, but it is a distinct idea. A misdiagnosis is a wrong label: the condition named is not the one present. A false positive is a test result that points to disease that further testing then rules out. Overdiagnosis is neither, because the finding is genuinely there and would survive any amount of confirmation. A biopsy really does show abnormal cells; a scan really does show a nodule; a blood test really is out of range. What is “wrong” is not the finding but the assumption that finding it must help.
This is why overdiagnosis is so counterintuitive. Medicine is built on the reasonable belief that catching disease earlier is better. That belief holds for conditions that progress. It breaks down for conditions that do not, and the same sensitive tests detect both kinds without announcing which is which.
The reservoir of harmless disease
A useful way to picture overdiagnosis is to imagine a large hidden reservoir of abnormalities sitting silently in the population. Sensitive screening lowers a net into that reservoir and pulls up whatever it can reach. Some of what it catches is genuinely dangerous and worth removing. But the reservoir also holds a great deal that would have stayed harmless forever, and the net cannot tell the two apart.
H. Gilbert Welch, whose book Overdiagnosed did much to popularize the concept, uses an animal comparison to explain why detected disease behaves so differently from case to case. Some abnormalities are like birds: fast and already escaping, so that even early detection rarely catches them in time. Others are like rabbits: capable of progressing, and worth catching early, because acting promptly makes a real difference. But many are like turtles: so slow that they will never move far enough to matter within a person’s lifetime. Screening is good at penning the turtles, which never needed penning, while the birds have often already flown. The mismatch between how fast a disease moves and how well screening catches it is the engine of overdiagnosis.
Overdiagnosis versus lead-time and length-time bias
Two statistical effects make overdiagnosis look, from the outside, like a success story, and it helps to separate all three.
- Lead-time bias is the illusion of longer survival created simply by diagnosing earlier. If a disease is found five years sooner but the person dies at exactly the same moment they always would have, they appear to “survive” five years longer, even though nothing changed except the start of the clock.
- Length-time bias is the tendency of screening to preferentially catch slow-growing disease. Because slow cases spend longer in a detectable but silent state, a single scan is more likely to land on them than on fast, aggressive cases that come and go between screens. The screened group therefore looks like it does unusually well, partly because it is stocked with mild cases.
- Overdiagnosis is the extreme end of length-time bias: the detected disease is so slow, or so static, that it would never have surfaced at all. It is not a bias in a survival statistic so much as the detection of something that was never going to be a problem.
The three overlap, and all three can make screening appear more effective than it is. Distinguishing them is central to reading claims about cancer screening honestly.
Why it is so hard to see
Overdiagnosis is nearly invisible to the person who experiences it, and that is what makes it persist. Someone whose slow-growing cancer is found and removed feels, understandably, that screening saved their life. They cannot know that the same cancer would never have harmed them, because it was taken out before it had the chance to prove otherwise. The individual becomes a grateful advocate for the very process that, in their particular case, may have done net harm. Multiply that across many people and the result is a strong cultural belief that more detection is always better, even where the evidence at the population level is mixed.
The concept applies well beyond cancer. Lowering the numerical threshold for conditions such as high blood pressure, prediabetes or reduced bone density can, overnight, reclassify millions of healthy adults as having a disease, some of whom will be treated without ever having been at meaningful risk. The same logic drives debate over mental-health labels, including whether there is overdiagnosis of ADHD in children and adults. In each case the underlying question is the one this definition points to: not “is the finding real?” but “would this person have been better off never knowing?”
A tool for asking better questions
Understanding overdiagnosis is not a reason to distrust medicine or to refuse tests. It is a way to ask sharper questions before agreeing to screening or treatment: how likely is this to find something that actually threatens me, how likely is it to find something harmless, and what happens next if it does. Framed that way, overdiagnosis becomes a literacy skill rather than a source of fear. It restores the missing half of every diagnostic decision, the half concerned not with what a test can detect, but with whether detecting it will help.