How Overdiagnosis Is Measured, and Why the Estimates Vary

Overdiagnosis means finding real disease that would never have become clinically apparent or shortened life. It is not a false-positive test and not simply an incorrect pathology label. The core difficulty is counterfactual: once a screen-detected lesion is treated, no one can observe what would have happened without detection. Researchers therefore estimate overdiagnosis from differences across populations, and each method makes assumptions that shape the result.

Keep four concepts separate#

A screening test can lead to several very different outcomes. A false positive is an abnormal screen followed by evaluation showing no target disease. Misdiagnosis means the disease label itself is wrong; overdiagnosis means the disease is present under accepted criteria, but it would never have produced symptoms or death during that person's lifetime. Overtreatment is the unnecessary treatment that can follow overdiagnosis.

These distinctions matter because the harms and measurement methods differ. False-positive rates can be measured by following abnormal tests through diagnostic workup. Overdiagnosis cannot be verified that way. A biopsy may prove that a lesion meets the definition of cancer while leaving its untreated future unknown.

The National Cancer Institute includes overdiagnosis among the recognized harms of screening because treatment can cause physical, psychological, and financial harm without improving health; that does not make screening inherently harmful. It means mortality benefit, earlier-stage diagnosis, false positives, procedure harms, and overdiagnosis all belong in the same benefit-harm assessment.

The hidden counterfactual#

Imagine two identical futures for one screen-detected cancer. In one, it is treated. In the other, it is never found and never causes symptoms before the person dies of another cause. Only the first future is observable. The second is the counterfactual needed to classify overdiagnosis.

Some lesions have features associated with slower growth, and active-surveillance studies can clarify natural history. But a probability of indolence is not certainty about one individual's course. Competing mortality also matters. The same slow-growing lesion may represent overdiagnosis in a person with limited life expectancy but become clinically important in someone who lives several decades, which is why careful sources give you an overdiagnosis fraction for a screened population, rather than declare, after routine treatment, that a particular person was definitely overdiagnosed.

Method 1: long follow-up of a randomized screening trial#

Randomization creates groups that should initially have similar underlying disease risk. Screening shifts some diagnoses earlier. During the active screening period, the screened group therefore accumulates more cases, and if those additional cases are only earlier versions of disease that would later appear clinically, the control group should catch up after screening stops and enough time passes.

A persistent excess of diagnoses in the screened arm after adequate catch-up time can estimate overdiagnosis. This design reduces confounding from screening choice and provides a clear comparison population. It is often treated as the strongest direct design when conduct, adherence, and follow-up are sound.

The method still has limits. Screening in the control arm dilutes the contrast. Technology and diagnostic thresholds may change before long-term results arrive. Follow-up must be long enough to cover the relevant lead time, which can differ by tumor and age, and deaths from other causes can prevent latent disease from becoming apparent and need appropriate treatment in the analysis.

Method 2: cohort and ecological incidence comparisons#

Researchers can compare cancer incidence in regions, periods, or groups with different screening intensity. These studies are practical for monitoring real programs and modern technology. A rise in early-stage diagnoses without the expected later fall in advanced disease can support concern about overdiagnosis.

The raw excess is not automatically the overdiagnosis fraction. Screening advances the date of diagnosis, creating a temporary rise even if every case would eventually have appeared. Background incidence can also change because risk factors, diagnostic practice, registries, or disease definitions change. People who choose screening may differ from those who do not.

Credible analyses address lead time, underlying incidence trends, age, competing mortality, screening uptake, and contamination. Their assumptions should be visible to you. An estimate that merely subtracts one contemporary incidence curve from another can mistake timing and population differences for overdiagnosis.

Method 3: natural-history modeling#

A model represents unobserved stages of disease, such as onset, preclinical detectable disease, clinical presentation, and death from other causes. Researchers calibrate transition rates using trial, registry, and screening data, then simulate how many detected lesions would never become clinically apparent.

Modeling can compare screening schedules, ages, or technologies long before decades of follow-up are available. It can also make lead time and competing mortality explicit, and its weakness is dependence on the assumed natural history, which is exactly the process that cannot be observed directly for treated cases.

Two models can fit the same observed data yet assign different rates of progression or regression. Their overdiagnosis estimates can therefore differ. Model validation, sensitivity analyses, alternative assumptions, and transparent parameter sources matter more than the sophistication of the software.

Method 4: pathology, imaging, and reservoir studies#

Pathology and imaging studies examine features associated with indolent behavior, such as size, grade, growth, or molecular pattern. Autopsy studies identify occult cancers in people who died from unrelated causes, showing that a reservoir of clinically silent disease can exist. Incidental findings after highly sensitive imaging can offer another view of that reservoir.

These approaches establish biological plausibility and can help stratify risk. They do not directly show the fraction of screen-detected cases that would remain harmless under current practice; an occult lesion found at autopsy is not identical to a lesion detected by a particular screening program, and a favorable marker does not guarantee nonprogression. Their greatest value is often explanatory: they show you why a more sensitive test can increase diagnosis without producing a proportionate reduction in late disease or mortality.

Lead time is the central correction#

Lead time is the amount by which screening advances diagnosis. Suppose a cancer would have caused symptoms in 2030 but screening detects it in 2026. Counting cases only through 2028 makes the screened population appear to have an excess even though the control population will begin catching up by 2030.

Adequate follow-up or modeling must allow for that shift. The required period is not universal because growth rates vary. If catch-up is incomplete, the estimated overdiagnosis fraction tends to be too high. If control participants receive substantial screening, the groups converge and the estimate may be too low.

Lead-time bias in survival statistics is related but distinct. Earlier diagnosis can lengthen measured survival from diagnosis without changing the date of death, and overdiagnosis can inflate survival further because cases that were never destined to cause death enter the denominator. Disease-specific mortality and advanced-disease incidence are therefore more informative screening outcomes than survival after diagnosis alone.

The denominator can transform one numerator#

An estimated number of excess diagnoses can be divided by several quantities:

• all cancers diagnosed in the screened population;
• cancers detected during screening;
• the number of people invited to screening;
• the number who actually attended; or
• the number of deaths prevented.

Each produces a different percentage and answers a different question. “Twenty percent overdiagnosed” is incomplete unless you are told twenty percent of what, over which period, and under which method. Comparisons across studies should place the numerator, denominator, age range, screening schedule, and follow-up side by side.

Definitions also differ. Some analyses include in situ lesions and others count only invasive cancer. Some use cumulative incidence from the start of screening, while others begin after a prevalence round. These choices can materially change the result without any arithmetic error.

How overdiagnosis can change over time#

Overdiagnosis is not a fixed property of a test. It depends on who is screened, how often, the detection threshold, disease prevalence, competing mortality, and what happens after an abnormal result. More sensitive imaging may find smaller lesions. Extending screening to older ages may increase the chance that another cause of death occurs first. Risk-based screening can alter the balance again.

Management also changes the harm. If clinicians can identify a low-risk lesion and monitor it safely, the detection may still meet a population definition of overdiagnosis, but the downstream burden may be less than when every lesion receives immediate invasive treatment. Measuring diagnosis and measuring treatment harm are related but separate tasks.

A six-part appraisal#

Identify the disease and what counted as a diagnosis. Determine whether the design was a randomized trial, cohort comparison, model, or pathology-based inference. Check the follow-up and lead-time adjustment. Find the numerator and exact denominator. Look for contamination, background incidence, competing mortality, and sensitivity analyses. Finally, ask whether the screening technology and population still resemble current practice.

The result should be presented as a range or method-bound estimate, not a property of every detected lesion. Uncertainty is not a reason to ignore overdiagnosis. It is the reason to explain the method before using the number in a screening decision.