TL;DR: Among people with abnormal findings on Taiwan’s subsidized lung cancer screening, nearly nine in ten were not diagnosed with lung cancer during the reporting period. Some screening programs now propose a “third path” that uses induced sputum plus high-sensitivity genetic testing to help this group decide whether to worry. The problem is that non-cancerous lung tissue frequently carries oncogenic mutations. The more sensitive the test, the more likely it is, in theory, to return a positive result in someone who is healthy. Until prospective follow-up data in a ground-glass nodule population are available, this approach may simply produce one more positive report that is harder to interpret.

According to the Ministry of Health and Welfare’s 2024 mortality statistics, lung cancer has been Taiwan’s leading cause of cancer death for 21 consecutive years (CNA report). My mother also died from lung adenocarcinoma.

In July 2022, the Health Promotion Administration (HPA) launched a subsidized low-dose CT (LDCT) lung cancer screening program. Screening can detect early-stage lung cancer, but it also brings a new problem: among people whose reports note a nodule, nearly nine in ten were not diagnosed with lung cancer during the reporting period.

A screening program targeting this population has recently proposed a “third path.” Rather than waiting three to five years to see whether a nodule grows, or operating the moment one appears, the idea is to first collect a lung specimen through induced sputum, then run a PCR that amplifies only mutant signals. The goal is to assess how far that person is from lung cancer.

Compared against published research, the premise is sound: most people with abnormal findings are indeed not diagnosed with lung cancer, and they genuinely need better triage tools. But the current evidence does not yet support the task being assigned to this approach.

How Many People with Abnormal LDCT Findings Are Diagnosed with Lung Cancer?

HPA statistics show that in the first year of the subsidized LDCT program (July 2022 to June 2023), 49,508 people were screened. About 9.2% had suspected abnormal pulmonary nodules requiring periodic follow-up or further workup; 531 had been diagnosed with lung cancer at the time of reporting. That works out to roughly 12% of the abnormal-finding group, with nearly nine in ten not diagnosed. The HPA’s own press release reminds readers that most pulmonary nodules are benign.

Taiwan’s TALENT study lands in the same range. The paper opens by noting that Taiwanese lung cancer patients are predominantly non-smokers. The study screened people aged 55–75 who had never smoked or had smoked very little and quit more than 15 years ago, and who carried risk factors such as family history. Of 12,011 enrolled participants, 2,094 had positive results. The reported positive predictive value was 14.0%: about one in seven positive cases was diagnosed with lung cancer during the study’s follow-up period, leaving roughly six out of seven without a diagnosis.

These nine-tenths of people leave with something new on their minds. Those who proposed the third path observed two common responses: some avoid coming back rather than face the uncertainty; others prefer to have the nodule removed rather than let it linger. Between those extremes, the default is returning every few months for imaging and letting the nodule’s behavior speak for itself.

The anxiety of waiting and the time spent on follow-up appointments are real costs, even if they rarely appear in screening outcome statistics. Still, one in eight people with abnormal findings is diagnosed with lung cancer, so follow-up cannot be skipped. Anyone who develops symptoms should not wait for a scheduled appointment and should see a doctor.

Does a Higher Rate of Early Detection Mean Screening Is Working?

When HPA marked three years of the subsidized program, it reported that through June 2025, 217,744 people had been screened and 2,506 diagnosed, of whom 2,069 (about 83%) were identified at an early stage. By comparison, Taiwan’s 2021 cancer registry showed that only 35.3% of all lung cancers across the country were found at an early stage.

These two groups differ in composition: one is a high-risk screened population, the other is every diagnosed case nationwide. The contrast shows that screening finds early-stage lung cancer; it does not directly prove that screening reduces mortality.

A separate study examined the other side of the same trend. A 2022 paper in JAMA Internal Medicine used Taiwan’s cancer registry to analyze female lung cancer from 2004 to 2018. Taiwan’s female smoking rate has been below 5% since 1980. Over that period, early-stage (stage 0–I) incidence rose from 2.3 to 14.4 per 100,000, an increase of more than sixfold. Late-stage (stage II–IV) incidence moved from 18.7 to 19.3, essentially unchanged. Mortality was flat as well.

An effective screening program should increase early-stage diagnoses while also reducing late-stage ones. Because late-stage incidence did not fall, the authors argued that almost all of the additional early-stage cases represented overdiagnosis: lesions that would never have caused symptoms or death during the patient’s lifetime were being found and treated. Mortality over the same period did not decline. Five-year survival did rise, from 18% in 2004 to 40% in 2013, but the authors note that this figure is distorted by the large volume of slow-growing early cancers being detected. Their recommendation was that LDCT should be restricted to heavy smokers unless randomized trials demonstrate benefit in lower-risk populations.

This is an ecological study examining aggregate trends; it cannot prove that any specific individual was overdiagnosed. The TALENT authors also acknowledge in their paper that overdiagnosis is possible, particularly among those diagnosed with adenocarcinoma in situ.

In March 2026, the question became a public argument in Taiwan. A public health researcher published a commentary describing Taiwan as a “kingdom of lobectomies,” citing an increase in lobectomies from fewer than 2,000 per year before 2006 to 26,444 in 2024. This researcher was also a co-author of the JAMA paper. The National Health Insurance Administration subsequently published its own figures: 4,193 lobectomies in 2024 and 21,378 total lung resections, with wedge resections (13,342 cases) accounting for the majority; total lung resections across 2023 to 2025 grew by roughly 2% (Business Today summary).

The two sets of numbers diverge considerably. Assessing whether overtreatment is occurring requires knowing what proportion of resected nodules proved to be malignant, not just total case counts. The same report cited NHIA data showing that at several major medical centers, on average more than 70% of resected lung nodules were malignant on post-operative pathology; it also noted that at some hospitals the figure was below 50%. The Minister of Health and Welfare stated that the ministry would continue monitoring variation across hospitals and physicians, and had already fed the data back to individual institutions (CNA).

LDCT itself has an evidence base. The HPA cites international research showing that LDCT reduces lung cancer mortality in heavy smokers by approximately 20% (HPA). The debate centers on what happens after the program expands to non-smokers: specifically, how many of the nodules found in that population actually require intervention. TALENT’s published one-year follow-up results demonstrate that early-stage lung cancer can be found in non-smokers; mortality data are not yet available.

Does a Mutation in Sputum Mean Lung Cancer Is Coming?

The logic behind the third path runs as follows: lung cancer is driven by genetic mutations, so if those mutations can be detected before a nodule grows, it should be possible to identify earlier who actually needs to worry. Specimens are collected through induced sputum, avoiding invasive procedures like bronchoscopy. The assay uses mutation-enrichment PCR, which amplifies only mutant signals. This class of technology has been in the research literature since 2008; its sensitivity for detecting mutations can reach up to a hundredfold that of conventional methods (study). The technology has a legitimate foundation.

The trouble starts once the test can detect these mutations. A 2023 paper in Nature used ultra-deep sequencing to examine histologically normal lung tissue. In 295 samples, 18% carried EGFR driver mutations; in a separate set of 81 samples, 53% carried KRAS driver mutations. Most samples came from non-cancerous lung tissue taken from patients who had lung cancer or other cancers elsewhere; among 19 individuals who died from non-pulmonary cancers with no lung cancer at all, 3 still had EGFR driver mutations in their lung tissue. The study’s main focus was PM2.5: air pollution acts on cells that already carry these mutations, promoting their proliferation through inflammation and setting cancer in motion. The popular claim that “PM2.5 causes Taiwan’s rise in lung adenocarcinoma” gets only partial support: the study supports PM2.5 as a promoting factor and does not say it is the main cause.

What follows is my own inference, not a test the authors performed: if non-cancerous lung tissue so often carries these mutations, a high-sensitivity test could return positive results in people who do not have lung cancer. Whether it does depends on the assay’s detection limit and on how many mutant cells are present in the specimen. The same study estimated that roughly one in 550,000 lung cells carries an oncogenic EGFR mutation. How often sputum-based tests of this type would return positive results in a lung-cancer-free population is a number I have not been able to find.

The other concern is false negatives. A 2013 study recruited 10 patients already diagnosed with EGFR-mutant lung cancer and tested their sputum; depending on the method, only 30–50% tested positive. Twenty controls (10 EGFR-negative lung cancer patients and 10 with chronic obstructive pulmonary disease) all tested negative. These were patients with confirmed diagnoses, the sample was small, and the methods may not match those used in programs promoting this test today. Ground-glass nodules typically sit at the lung periphery, far from the major airways, so the number of lesion cells that sputum can carry out is presumed to be even smaller. Direct data are absent.

Both ends of the problem lack data: people without cancer may test positive, and people with cancer may test negative. Until those two numbers are clarified, the results of this test should not drive a decision about surgery.

Four Questions to Ask Before Agreeing to a Lung Genetic Test

The semiconductor industry faces an analogous situation. In a wafer fab, the more sensitive the inspection tool, the more defects it finds, most of them “nuisance defects” that have no effect on yield. The engineer’s skill lies in classification: which are the killer defects that will fail the chip, and which can be left alone. Yield depends on classification ability, not detection sensitivity alone.

If someone recommends this kind of test, four questions are worth raising first:

  1. What happens after a positive result? Another LDCT, a biopsy, or immediate surgery? If the protocol is not spelled out clearly, a positive result adds anxiety without adding direction.
  2. Are there prospective follow-up data in a ground-glass nodule population? What share of test-positive people saw their nodules worsen within a few years, and what share of test-negative people did? Only if those two numbers differ enough does the test have discriminating power.
  3. Does a negative result allow one fewer LDCT? If routine follow-up continues regardless of a negative result, the test has not changed any decision.
  4. Is there evidence that it reduces unnecessary surgery or lowers mortality? This is the question any screening tool ultimately has to answer.

At the time of writing (October 2026), I could find no prospective follow-up data for this type of test in a ground-glass nodule population in PubMed or any other public source. The data may exist but remain unpublished. The post “Once AI Enters the Hospital, Who Shows That It Works?” covers the same principle: performance numbers need to connect to specific clinical contexts and be followed by post-deployment tracking. New diagnostic tools are held to the same standard. For how to check claims yourself, “Before You Click on a Stem Cell Ad, Spend Three Minutes Doing This” walks through the method: check the claimed indication against the regulator’s approval list and registration databases.

What Does It Take for a New Test to Prove Itself?

The problem those proposing the third path identified is real. Nearly nine in ten people with abnormal findings are not diagnosed with lung cancer, and they deserve something better than “come back and wait.”

That is precisely why a new tool should be held to the same standard: register a study, enroll participants, follow them, and publish the outcomes for both the positive and negative groups. Once that step is taken, the third path has a real chance of standing on its own.

Eligibility for subsidized LDCT covers men aged 45–74 and women aged 40–74 with a first-degree family history of lung cancer (parent, child, or sibling), or heavy smokers aged 50–74 with at least 20 pack-years of smoking who are still smoking or quit fewer than 15 years ago, with screening available once every two years. For people outside these criteria, no evidence currently supports using a genetic test in place of LDCT or as a basis for treatment decisions. If someone in that situation still chooses to be tested, what happens after a positive result is a question that should be settled before the test is run. This article discusses testing added on top of LDCT; it does not suggest that eligible individuals skip the subsidized program. Whether to screen and how to follow up are decisions to make with a physician.

If you receive an abnormal lung imaging report, the most reliable next step is working with a pulmonologist to agree on a follow-up plan. Clinical judgment about whether to observe or intervene is based on nodule size, density (solid vs. ground-glass), and rate of growth. Small nodules with a benign appearance are typically managed with periodic imaging (Science Media Center Taiwan expert commentary). If someone hands you a more sensitive test, ask first what happens if the result comes back positive.

For more on health and the body: Reflections & Memory → Health and the Body.