Lung cancer screening with low-dose CT, by the numbers

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

In the National Lung Screening Trial (NLST), 96.4% of positive low-dose CT results were false positives. The same trial found a 20% lower lung-cancer death rate with low-dose CT than with chest X-rays. Both findings are real, and they do not conflict. The first describes what one positive scan means for the person who gets it; the second, what a screening program did for tens of thousands of people over several years. This page keeps each on its own denominator: per positive scan, per person screened, per death prevented.

Who the screening is for

The U.S. Preventive Services Task Force (USPSTF) recommends annual screening with low-dose CT (LDCT) for adults aged 50 to 80 who have a 20 pack-year smoking history and currently smoke or quit within the past 15 years. (A pack-year is an average of one pack, 20 cigarettes, a day for a year.) Screening should stop once a person has not smoked for 15 years, or develops a health problem that substantially limits life expectancy or the ability or willingness to have curative lung surgery. The grade is B: moderate certainty of a moderate net benefit.

The Task Force names smoking and older age as the two most important risk factors for lung cancer, and found the evidence insufficient to judge whether more elaborate risk-prediction models would improve outcomes. In this site's terms, eligibility sets the pre-test probability, and nearly every number below depends on it.

NLST itself enrolled a narrower group, ages 55 to 74 with at least 30 pack-years, and the evidence review behind the 2021 recommendation noted that trial participants were more likely to benefit than the U.S. screening-eligible population. Trial figures are reference points, not forecasts.

What a positive scan usually means

In NLST, any non-calcified lung nodule at least 4 mm across counted as a positive CT screen: a finding that needs a closer look, not a diagnosis.

In the first round, 7,191 CT participants (27.3%) screened positive. Among people with a positive screen, 90.4% had at least one follow-up diagnostic procedure, mostly more imaging (81.1%); 4.2% had surgery. Lung cancer was diagnosed in 292 people in the CT group after that round, 1.1% of the group. More than a quarter screened positive; about 1 in 90 was diagnosed with lung cancer. This page explains that gap.

One screening round, 1,000 people

The arithmetic uses an NLST analysis by Pinsky and colleagues: with the trial's definition of a positive screen, LDCT had 93.1% sensitivity and 76.5% specificity (the first-round report gives 93.8% and 73.4%). That specificity flags 23.5% of people without lung cancer each round, close to the 23.3% average per-round false-positive rate the National Cancer Institute (NCI) reports for NLST. The positive likelihood ratio is 0.931 ÷ 0.235 ≈ 3.96: a positive scan multiplies the odds of cancer by about four.

For prevalence, this example uses 1.1%, 11 of 1,000 people: the share of NLST's CT group diagnosed with lung cancer in the first screening round. That count includes cancers diagnosed after a negative scan as well as those the scan found, so, unlike a detection rate, it does not leave out the missed cases. Pinsky's accuracy figures pool all three rounds, though, so the table below is a model built from trial figures, not a replay of that round.

Of the 11 people with cancer, 11 × 0.931 = 10.241 are flagged and 0.759 are missed. Of the 989 without, 989 × 0.235 = 232.415 are flagged anyway and 756.585 are cleared. The share of positives that are real is the positive predictive value (PPV):

PPV = 10.241 ÷ (10.241 + 232.415) = 4.22%

About 242 of the 1,000 test positive and roughly 10 of them have cancer: nearly 23 false positives for every true one. The model makes 95.8% of positives false; NLST observed 96.4% over three rounds, which is close but not a like-for-like check.

A negative is more informative. Of the 758 people cleared, about 1 has cancer, a remaining risk of 0.759 ÷ (0.759 + 756.585) = 0.100%, down from 1.1% (LR− = 0.069 ÷ 0.765 ≈ 0.09). For that round, the scan rules cancer out far better than it rules it in.

Three rounds, not one

NLST screened each person three times, a year apart. Positive rates were 27.3%, 27.9% and 16.8% by round, and 39.1% of CT participants had at least one positive result. In the final round, radiologists could call a screen negative if a nodule had been stable across all three exams.

The calculator's repeat-testing view gives a benchmark under stated assumptions: if each round were an independent 23.5% chance of a false alarm, a cancer-free person's chance of at least one in three rounds would be 1 − 0.765³ ≈ 55.2%. Treating NLST's own round-by-round positive rates as independent gives a similar 56%. The observed 39.1%, true positives included, is well below both. Because this comparison stays within one trial and one group of people, the gap is informative: positives tended to recur in the same participants. The repeat-testing guide explains how dependence between rounds moves the total.

Moving the line between positive and negative

The 4 mm rule is one choice among many. Pinsky and colleagues ranked NLST scans by radiologists' follow-up recommendations; using those as the cutoff, specificity could reach 92.4% while sensitivity fell only to 86.9%, the sensitivity–specificity trade-off in concrete form.

Per 10,000 people at 1.1% prevalence (110 with cancer), NLST's rule flags about 102 cancers and 2,324 people without cancer; the stricter cutoff flags about 96 and 752. That is roughly 7 more cancers missed that round and 1,572 fewer false positives, with PPV rising from 4.22% to 11.28%. Whether the exchange is worthwhile depends partly on what the missed cancers do before the next scan, which this arithmetic cannot show.

Lung-RADS, a stricter reporting system from the American College of Radiology that is now in wide U.S. use, was later applied retrospectively to NLST scans. It cut the first-round false-positive rate from 26.6% to 12.8%, while first-round sensitivity fell from 93.5% to 84.9%. NLST's accuracy figures are therefore best read as historical.

Why the eligibility rules matter

Back at NLST's 93.1% / 76.5% operating point, lower prevalence to an illustrative 0.3%, a stand-in for a lower-risk group rather than an estimate for any real population.

True positives fall from 10.241 to 2.793 per 1,000, but false positives barely move, from 232.415 to 234.295, because nearly everyone is cancer-free in both groups. PPV = 2.793 ÷ (2.793 + 234.295) = 1.18%, about 84 false positives for each true one. The scan did not change; the population did. That is the base-rate fallacy at work.

Risk shapes the benefit too. Kovalchik and colleagues split NLST participants into fifths by five-year risk of dying from lung cancer, from 0.15–0.55% in the lowest fifth to more than 2% in the highest. Lung-cancer deaths prevented per 10,000 person-years rose from 0.2 in the lowest-risk fifth to 12.0 in the highest, and people with false-positive results per prevented death fell from 1,648 to 65. The lowest-risk fifth accounted for only 1% of prevented deaths. Since every participant already met the trial's age and smoking criteria, eligibility rules are a coarse filter: risk still varied widely inside them.

The benefit, on its own denominator

NLST enrolled 53,454 people at high risk at 33 U.S. centers from 2002 to 2004 and randomly assigned them to three annual screens with LDCT (26,722) or single-view chest radiography (26,732). Lung-cancer deaths ran at 247 per 100,000 person-years with CT and 309 with radiography, a 20.0% relative reduction (95% CI 6.8% to 26.7%). Deaths from any cause were 6.7% lower (95% CI 1.2% to 13.6%).

The comparator was chest X-ray, not no screening. NELSON, a Dutch–Belgian trial that compared CT screening with no screening, reported a lung-cancer mortality rate ratio of 0.76 in men after at least 10 years (95% CI 0.61 to 0.94).

In absolute terms, NCI gives a number needed to screen (NNS) of 320: one lung-cancer death prevented per 320 people assigned to CT rather than X-ray screening over a median of 6.5 years, roughly 3 per 1,000. Extended follow-up (a median of 12.3 years) found 3.3 fewer lung-cancer deaths per 1,000, an NNS of 303.

An NNS describes a whole program; its denominator is everyone assigned to screening, including people who never tested positive. The calculator's treatment NNT box asks how many of the true positives already counted must be treated for one to benefit, so 320 does not belong there: it would apply prevalence and sensitivity twice, shrinking roughly 3 prevented deaths per 1,000 to about 0.03. The NNT, NNH and NNS guide explains why.

The harms, on theirs

Procedures. The Task Force's evidence review calculated that for every 1,000 people screened in NLST, false-positive results led to 17 invasive procedures and fewer than one major complication. NCI cautions that complication rates in community practice may be higher: one study found them more than twice NLST's.

Radiation. One LDCT in NLST averaged 1.4 mSv; counting follow-up imaging, participants received an estimated average of 8 mSv over three years. Modeling cited by NCI suggests there could be one radiation-related cancer death per 2,500 screens, against about one lung-cancer death avoided per 960 screens. NCI judges the benefit substantially larger in an NLST-like program, but says that for younger people and those without significant lung-cancer risk the balance may reverse.

Overdiagnosis. Some screen-detected cancers would never have caused symptoms or death. Welch and Black estimated from randomized trials that about 50% of lung cancers found by chest X-ray and/or sputum screening were overdiagnosed, and reviewed observational data suggesting overdiagnosis with CT too. For NLST, Patz and colleagues estimated an 18.5% probability (95% CI 5.4% to 30.6%) that a CT-detected lung cancer was overdiagnosed after 6.4 years, or 1.38 overdiagnosed cancers per 320 people screened. By a median 11.3 years, the CT arm's excess of cancers had shrunk to a non-significant 20, which NCI puts at about 3%. Across trials, the Task Force review found estimates from 0% to 67%. NLST's estimates are also relative to X-ray screening, not to no screening.

Incidental findings and distress. The Task Force review found incidental findings, abnormalities unrelated to lung cancer, in 4.4% to 40.7% of people screened, and listed increased distress among the harms.

What generalizes

Low-dose CT shows that a test can be weak at ruling cancer in and still belong to a program that reduced deaths in a large trial, because the two facts sit on different denominators; the screening-harms guide shows how to keep them apart. Three habits carry to any screen: ask which denominator a figure uses, ask who was screened, and keep benefits and harms side by side rather than netted into one score. How to weigh one prevented death against the follow-up, procedures and overdiagnoses that come with it is a judgment the numbers can inform but not make.

Correction, September 2026. An earlier version of this page entered the trial's number needed to screen (320) into the calculator's treatment NNT box, which applies prevalence and sensitivity twice; that example has been removed. The corrections log lists every change.