Why Screening Selects Slower Cancers and Inflates Survival Statistics

Better survival among screen-detected cancers does not show that screening caused the improvement. Screening preferentially samples the cancers that spend most time in a detectable, presymptomatic state, and those tumors often have slower biology, so the comparison is already favorable before anyone is treated.

Key points#

• The chance of detection rises with the length of the presymptomatic detectable phase.
• Fast-growing cancers can arise and cause symptoms between scheduled screens, becoming interval cancers.
• Lead-time bias adds time to measured survival by moving diagnosis earlier; length-time bias changes which cancers enter the screen-detected group.
• Overdiagnosis is detection of disease that would never have caused symptoms or death during the person's lifetime.
• Randomized comparisons of disease-specific mortality and serious harms are more informative than survival among diagnosed cases.

A sampling problem hidden inside a screening program#

Imagine two tumors that become visible on imaging at the same size. One remains asymptomatic for six years. The other causes symptoms after six months. With screening every two years, the slow tumor has several opportunities to be found. The fast tumor can pass from undetectable to symptomatic between visits.

The screen-detected group will therefore contain a disproportionate number of slow tumors. The symptom-detected and interval-cancer groups will contain more aggressive disease. Comparing their survival is like comparing two groups selected by tumor behavior and then attributing the difference to the act of detection. The “length” in length-time bias is the length of that detectable presymptomatic period, and a longer period creates a larger sampling window.

Why five-year survival can improve without fewer deaths#

Five-year survival uses diagnosis as time zero. Screening can improve that statistic through at least three mechanisms that do not necessarily prevent death.

First, lead time moves diagnosis earlier. If a cancer would have been diagnosed from symptoms in 2029 and caused death in 2032, detecting it in 2026 changes measured survival from three years to six years even if death still occurs in 2032.

Second, length-time bias fills the diagnosed group with slower tumors whose natural prognosis was already better.

Third, overdiagnosis adds cancers that would never have become clinically important. People with those lesions have excellent cancer-specific survival because the disease was not destined to harm them, and adding them to the denominator can make the program appear successful while increasing biopsies and treatment. All three mechanisms can operate together, so a rise in incidence, a shift toward early-stage disease, and improved survival are not sufficient evidence that screening reduces mortality.

Interval cancers reveal the other side of selection#

An interval cancer is diagnosed after a negative screen and before the next scheduled screen. Some were missed by the earlier test. Others were not yet detectable and grew rapidly. As a group, interval cancers may have less favorable biology than screen-detected cancers.

That difference is sometimes described as proof that screening found the “good” cancers early and failed only on unavoidable aggressive disease. The description misses the sampling mechanism. The screening schedule itself is more likely to capture long-duration lesions, and a program must be judged by outcomes across everyone invited or assigned to screening, not by comparing selected cancer types after diagnosis. Improving test sensitivity can detect more fast-growing disease, but it can also find more indolent lesions, and the balance depends on the cancer, the test, the interval, who is eligible, and what happens downstream.

Overdiagnosis is related but not identical#

Length-time bias covers enrichment with slower disease. Some slow cancers still would eventually cause symptoms and may benefit from earlier treatment. Overdiagnosis is the subset that would never have become clinically apparent during the person's lifetime, either because it does not progress meaningfully or because another event occurs first.

At diagnosis, a clinician generally cannot tell you with certainty which lesion is overdiagnosed. The person still carries the cancer label and may undergo surveillance, surgery, radiation, or medication. That is why overdiagnosis is a population-level harm even when an individual treatment is technically successful.

Estimating overdiagnosis is difficult. Long follow-up is needed because screening advances the timing of diagnoses. Contamination of control groups, changing diagnostic thresholds, and background incidence trends can alter estimates. A simple count of “extra cancers” immediately after screening begins can overstate the lasting excess, while short follow-up can miss it entirely.

The trial metric that resists the illusion#

Randomized screening trials compare outcomes based on assignment or invitation, preserving comparable groups at baseline, and disease-specific mortality asks whether fewer people die from the target cancer, regardless of when diagnosis occurred. It is not affected by starting the survival clock earlier and is less vulnerable to selecting favorable tumor biology.

All-cause mortality avoids uncertainty in assigning cause of death but usually requires much larger trials because deaths from the target cancer are a small part of all deaths. Disease-specific mortality can be affected by misclassification, especially if knowledge of screening influences attribution. Blinded cause-of-death review and complete follow-up help.

The National Lung Screening Trial illustrates the design principle. It randomized eligible participants to low-dose CT or chest radiography and found fewer lung-cancer deaths in the CT group; comparing survival only among cancers found by CT would not have established that benefit. Even mortality benefit does not erase harm. A complete evaluation also includes false positives, invasive procedures, overdiagnosis, treatment complications, incidental findings, resource use, and the size of the absolute mortality difference you are being offered.

How to read a screening claim#

Be cautious when the evidence is framed as:

• screen-detected cancers have higher five-year survival,
• screened communities have more early-stage diagnoses,
• people who attend screening fare better than nonattenders,
• survival improved after a screening program began,
• or a new test finds more cancers than the old test.

Each statement can be true without showing net benefit. Attendance comparisons add healthy-volunteer and access differences, before-after comparisons add changes in treatment, classification, and background risk, and detection yield rewards finding disease whether or not that disease matters.

A stronger appraisal asks:

1. Was the study randomized by invitation or screening strategy?
2. Did disease-specific mortality fall, and by how much absolutely?
3. Was follow-up long enough for benefit and overdiagnosis to emerge?
4. Were deaths classified without knowledge of assignment?
5. How many false positives and invasive procedures occurred?
6. Was excess incidence assessed after screening ended?
7. Does the studied population match current eligibility?