Repeated screening and the arithmetic of false alarms

Worth the Test · Written by an emergency physician · Last reviewed September 2026

Every accuracy number on this site describes a single round of testing. Almost no screening program consists of one. Mammography is invited every one to two years for decades; FIT is annual; PSA, low-dose CT, cervical screening — all recur. And repetition quietly transforms the arithmetic: a specificity that sounds reassuring per round compounds into a false-alarm rate per screening career that surprises nearly everyone, including the clinicians doing the inviting. This guide works that compounding in both directions — because the same mechanism that piles up false alarms is also, run the other way, how confirmation works.

The compounding, idealized

Take a test with 90% specificity: each round, a disease-free person has a 90% chance of being correctly cleared. Ask instead: what is the chance of getting through n rounds without a single false alarm? If each round erred independently, it would be 0.9 × 0.9 × … — that is 0.9ⁿ — and the chance of at least one false positive would be

P(≥1 false alarm in n rounds) = 1 − specificityⁿ.

The numbers escalate faster than intuition expects. At 90% specificity: 10% after one round, 27% after three, 41% after five, 65% after ten. Even an excellent 99%-specific test comes close to 10% after ten rounds. The per-round error never changed; exposure did. The same person keeps re-rolling the same dice, and screening programs are the re-rolling of dice, decade after decade, on millions of healthy people. The calculator's serial-testing panel draws this curve live for whatever specificity you set — drag the slider and watch the decade-long picture swing.

What real programs actually observe

The formula rests on three assumptions: errors that are independent from round to round, the same specificity at every round, and a person who stays disease-free. Independence is an assumption, not a bound — real cumulative rates can come out below or above the curve. If each round has false-positive probability q, all that can be guaranteed for n ≥ 1 rounds is q ≤ P(at least one false positive) ≤ min(1, nq). At 90% specificity and ten rounds, that range runs from 10% to 100%, and independence's 65.1% is just one point inside it. If the same 10% of people are flagged every time, cumulative risk stays at 10%; if a different, non-overlapping 10% is flagged each round, it reaches 100%.

• Mammography: In the New England Journal of Medicine, Elmore and colleagues reviewed ten years of screening records for 2,400 women. They estimated the cumulative risk of at least one false-positive result after ten screening mammograms at 49.1%, and estimated that 18.6% of women without breast cancer would undergo a biopsy after ten mammograms. For comparison, the independence formula at the 88.9% per-round specificity measured in modern U.S. digital screening by the Breast Cancer Surveillance Consortium gives about 69%. The populations, eras and reading practices differ, so the gap between 69% and 49% does not measure how correlated the rounds are.
• Multi-organ screening: Croswell and colleagues analyzed the screening arm of the PLCO trial, where each participant could receive up to 14 tests over three years — for lung and colorectal cancer, plus prostate cancer in men or ovarian cancer in women. After 14 tests, the cumulative risk of at least one false positive was 60.4% for men and 48.8% for women. The cumulative risk of undergoing an invasive diagnostic procedure prompted by a false positive was 28.5% for men and 22.1% for women — measured across all participants, not only those who had a false alarm.

Observed cumulative rates can sit below or above the independence estimate; only the range from q to min(1, nq) is guaranteed, so neither study confirms or refutes the curve. What both show is scale: for a disease-free person, a screening career is closer to a coin flip than a formality. In each study, the estimated chance of at least one false alarm over a full course of screening was about half or more — and each false alarm means further evaluation, from repeat imaging or testing to, sometimes, a biopsy or other invasive procedure, before the all-clear. The anxiety, cost, and occasional complication of those work-ups belong on the harms side of any screening ledger; the screening-harms guide weighs them alongside overdiagnosis.

None of this is an argument that repetition is a mistake. Repeating a screen is how programs patch imperfect sensitivity — the cancer a mammogram misses this round gets another chance to be caught next round, and repetition is what lets a 79%-sensitive FIT anchor a respectable program despite its per-round misses. Repetition buys real detection; it pays in accumulated false alarms. Both sides of that trade compound with n, which is why neither can be read off a single-round accuracy table.

The flip side: repetition as confirmation

Now run the same compounding on the positive side, and the villain becomes the hero. If a positive result is followed by a second, independent test — and the second is also positive — the errors multiply against each other. A false alarm now requires being unlucky twice: with 90% specificity per round, one person in ten is falsely flagged once, but only about one in a hundred twice. Meanwhile most true cases keep testing positive. Sequential positives concentrate truth.

The odds arithmetic from pre-test to post-test probability makes it exact. A 90%/90% test has a positive likelihood ratio of 9. At 1% prevalence, one positive lifts the probability of disease from 1% to just 8.3% — the base-rate fallacy in action. But if the second test's mistakes are unrelated to the first's — among people with the disease and among people without it — a second positive multiplies the odds by 9 again: from 8.3%, the odds 0.083 ÷ 0.917 ≈ 0.09 become 0.82, and the probability lands near 45%. Two cheap, mediocre tests in agreement approach what one excellent test achieves. This is why confirmatory testing is the universal design pattern for screening programs: the two-assay HIV algorithm, the colonoscopy after a positive FIT, the diagnostic work-up after a screening mammogram. No screen is asked to be right; each is asked to be right enough that the next, better test is worth running.

The same independence caveat applies in this direction too — and here it cuts against you. Whatever caused a false positive once (a cross-reacting antibody, a benign calcification) may cause it again, so repeating the identical test confirms less than the formula promises. Good confirmation changes the mechanism: a different assay, a different modality, a tissue diagnosis. The math wants a second opinion, not an echo.

The asymmetry worth remembering

One mechanism, two morals. Under the independence model, repeated screening of the disease-free compounds false alarms toward near-certainty of a scare, while repeated confirmation of a positive compounds specificity toward near-certainty about the truth. Which one you get is a matter of design — who gets retested, with what, and why. So when a program is described to you as "a simple yearly test," the single-round accuracy is the least of what you need: ask what ten years of it does to a healthy attendee, and what happens after the first positive. Those two answers — not the per-round percentages — are the program.

Correction, September 2026. Elmore's 18.6% biopsy estimate applies to all women without breast cancer, and Croswell's invasive-procedure risks apply to every participant; this guide had described both as shares of people with a false alarm. It had also attributed the 69% comparison to Elmore's own data and presented both studies' lower rates as confirmation that rounds are correlated. The corrections log lists every change.