The test reported cancer. Every scan said otherwise. He was never told the number that mattered.
A 58-year-old man with no symptoms takes a blood test that claims to detect 50 cancers. It comes back positive. Over the next six weeks he undergoes a CT, an MRI, an endoscopic ultrasound, and a PET scan. The pancreas is normal. A biopsy needle nicks a vessel. He spends a night in hospital. Three months and several thousand dollars later, the working conclusion is: false positive. He was terrified, harmed, and bankrupted by a number nobody computed before sending him down that path.
Because most screened people are healthy, the test must be judged across the whole population — not by one person's result.
A screening test is applied to people without symptoms, drawn from a defined risk group. This is categorically different from a diagnostic test. Because most screened people do not have the disease, the test must be judged by what happens across the whole population: deaths prevented, false alarms generated, healthy people harmed by follow-up. When a test meets a person, four outcomes are possible. True positive, false negative, true negative, false positive. The false positive is the person sent for six weeks of invasive workup who had nothing wrong.
The worried patient wants PPV. PPV is not on the test's spec sheet.
Two properties describe the test itself. Sensitivity is the fraction of people with the disease who are correctly flagged. Specificity is the fraction without the disease who are correctly cleared. These are measured once and applied everywhere. But the number a worried patient actually wants is the positive predictive value, or PPV: given a positive result, the probability that disease is truly present. PPV is not a property of the test. It depends on prevalence, the fraction of the screened population that actually has the disease. That dependence is the fact nobody told the man at the opening of this chapter.
Length-time bias: screening catches cancers during their detectable preclinical phase. Slow-growing tumors spend a long time in that phase, so they are over-represented among screen-detected cancers. Fast, lethal tumors produce symptoms between rounds and are under-represented. Screen-detected cancers would have done better on average even without treatment. And then overdiagnosis: the hardest bias to see. Screening can detect cancers that would never have caused symptoms. The patient is diagnosed, treated, potentially harmed, but would have lived just as long without any of it. Only population-level studies reveal overdiagnosis as excess diagnoses above expected incidence.
In 1968, Wilson and Jungner laid out criteria still in use today. The disease must have a detectable preclinical phase. Effective treatment must exist for early disease: finding something earlier only matters if earlier treatment works better. The test must perform well enough that benefits outweigh harms. And harms include false positives, unnecessary procedures, overdiagnosis, and anxiety. The accepted screening programs are those where randomized trial evidence shows benefit clearly dominates harm in a defined population: low-dose CT for lung cancer in heavy smokers, cervical cytology, mammography with appropriate qualification, colonoscopy for colorectal cancer.
A pervasive error in the public understanding of screening is treating improved survival statistics as proof of benefit. They are not. Lead-time bias guarantees improved survival even with zero therapeutic benefit. Length-time bias adds to it. Overdiagnosis inflates the denominator with patients who were never going to die of the disease. The correct question is: did the screened group die of this cancer less often than the unscreened group? The National Lung Screening Trial showed that low-dose CT in heavy smokers reduced lung cancer mortality by approximately 20 percent. That is the kind of endpoint that licenses a recommendation.
The curable early-stage cancers shed the least circulating tumor DNA — precisely the cancers screening most needs to find.
The finding that would revise this chapter: a large randomized trial of a multi-cancer early detection blood test that demonstrated a clear reduction in cancer-specific mortality, not just more cancers found, not just longer survival from diagnosis, but fewer deaths, with an acceptable false-positive and overdiagnosis burden. The NHS-Galleri trial and NCI prospective studies are designed to answer exactly this question. There is a hard biological constraint: the curable early-stage cancers shed the least circulating tumor DNA, precisely the cancers screening most needs to find. Whether sensitivity at stage I can ever be high enough to matter is set partly by tumor biology, not only by assay technology.
Still open: what informed consent means when overdiagnosis is invisible at the individual level, and how to balance depth versus breadth in risk-stratified screening.
So here is the chapter's claim. Screening must be judged by population-level mortality reduction, because survival statistics, detection counts, and even positive predictive value can each mislead in isolation. The only endpoint immune to lead-time bias, length-time bias, and overdiagnosis simultaneously is: did fewer people die of this cancer in the screened group than in the unscreened group? What remains open: what informed consent means when overdiagnosis is invisible at the individual level, and how to balance depth of coverage against population breadth in risk-stratified screening programs.
Cancer Research · Chapter 1 · Cancer Screening — Finding It Early Enough to Cure
That is the frame for everything ahead. Earlier detection is only beneficial when it leads to treatment that changes the outcome, in a population where the prevalence is high enough to keep false positives from overwhelming true positives, evaluated by an endpoint that cannot be gamed by moving the clock. Prevalence governs PPV. Mortality, not survival, is the measure. Benefits must outweigh harms in the specific population being screened. Everything else in this series is simply what happens when those conditions are, or are not, met.