Is Finding It Early Always Better? — An Introduction to Overdiagnosis

Tenfold More Cancers, Fewer Than 1 Death per 100,000 People

In 1999, thyroid cancer in South Korea was diagnosed in 6.3 people per 100,000, age-standardized. By 2012, that figure was 63.4 (PMID 33979040). It had multiplied tenfold in 13 years. Using South Korean thyroid cancer as our stage, this article follows whether finding disease early is always beneficial, and expands outward into numbers from randomized trials of prostate, breast, and lung cancer screening.

Mortality moved on an entirely different scale from diagnosis. In a study looking at National Health Insurance data and cause-of-death data from 2006 to 2015, South Korea’s thyroid cancer incidence more than doubled before dropping by nearly half after 2012, while the thyroid cancer mortality rate did not fluctuate over the same period (PMID 33176753). Only one thing increased alongside screening. When comparing 16 cities and provinces from 2008 to 2010, regions where more people received thyroid cancer screening had higher incidence rates (correlation coefficient 0.77), and screening was associated only with the rise of papillary thyroid cancer, showing no relation to thyroid cancer mortality (PMID 27627550). Because this was an ecological study grouping data by region, it cannot speak to individual-level causality. The authors concluded that screening strongly suggested it was the most important driver of this increase, and wrote that we must closely examine whether we are turning thousands of symptomless people into cancer patients without a known benefit.

Mortality trends vary depending on the data and the analysis period. While the aforementioned 2006–2015 analysis reported that mortality did not fluctuate, a study applying segmented regression analysis to national cancer registry and cause-of-death data from 1999 to 2016 reported that age-standardized mortality began to decline in the early 2000s (PMID 33979040). One review pointed to a mortality rate that rose from 0.17 per 100,000 in 1985 to 0.85 in 2004, then dropped to 0.42 in 2015, interpreting this decline as suggesting that early detection reduced deaths (PMID 32207260). The same review added that, given how common thyroid cancer is and how mild its course is, early detection may be less cost-effective. Although the direction of the trend conflicts across datasets, looking at the review that reported concrete mortality figures, the mortality rate hovered below 1 person per 100,000 while the incidence rate moved in the tens of people.

If a screening test is effective, it should not merely increase the diagnosis of early-stage disease; it should reduce the occurrence of advanced and metastatic cancers and cut deaths from those cancers (PMID 36568283). If it fails to do this, screening may simply be scooping up lesions that would never have caused symptoms or suffering. A picture where diagnoses surge enormously while deaths do not decrease correspondingly is the first signal raising suspicion of overdiagnosis.

A Real Disease, But One That Would Have Remained Quiet for a Lifetime

If a fine-needle aspiration biopsy revealed papillary thyroid cancer cells and the tissue removed during surgery confirmed the same cancer, the diagnosis is correct. The authors of a paper addressing the definition and ethics of overdiagnosis also defined an overdiagnosed condition as a diagnosis that a relevant group of experts would accept as correct (PMID 27402883). Overdiagnosis is the labeling of a condition that would never have caused symptoms or harm during a person's lifetime. The disease is real; what is wrong is the guess about what that disease would do to the person's life.

In this regard, overdiagnosis differs from a false positive. A false positive is a result where the test says a disease is present when it is not, and "What a Test Actually Tells You — An Introduction to Sensitivity, Specificity, and Pretest Probability" covered that calculation. A misdiagnosis is mistaking another condition for that disease. In overdiagnosis, the test, the pathology, and the name of the disease are all correct. The authors who provided the definition earlier also separated overdiagnosis from false positives in a response paper (PMID 27573152).

For overdiagnosis to occur, two things must be present. There must be a reservoir of disease sitting quietly inside the body without symptoms, and there must be an activity that scoops from that reservoir, particularly screening (PMID 20413742). There are two reasons why the water in the reservoir stays quiet for a lifetime (PMID 25663695). One relates to the tumor. If left alone, it is a cancer that spontaneously disappears, does not grow, or grows slowly enough that it will not threaten a life even over the course of the longest possible human lifespan. The other relates to the person. It is a cancer that grows fast enough to threaten a life within a normal lifespan, but the person dies of another cause before the day when the cancer would have been diagnosed through symptoms had there not been screening. The same authors wrote that in most organs, screening produces both types of overdiagnosis, and the ratio between the two likely varies greatly depending on the organ, the screening method, and patient characteristics.

A cancer removed by surgery does not show what would have happened if it had been left alone. Therefore, overdiagnosis cannot be measured directly and is estimated through indirect methods (PMID 38865364). The authors who established the definition repeatedly stated that overdiagnosis will continue to need estimation at the population level for a considerable period (PMID 27573152). Overdiagnosis cannot be confirmed for an individual; it is estimated only at the population level. Methods include counting the excess diagnoses in a screened group, as well as using models to estimate how much screening advances diagnosis or to estimate the natural history of the disease (PMID 29117353). There are also Korean figures estimated in this way. In a study calculating figures for 63 countries using previously developed and validated methods, more than 85% of thyroid cancer cases in South Korean women between 2013 and 2017 were estimated to be overdiagnosed (PMID 39389067). The combined figure calculated by the same study across 63 countries was 75.6% out of 2,297,057 cases. All of these are population estimates produced by models and do not tell the share of any single patient.

What Increased Survival Rates Do Not Tell You — Lead Time and Length Bias

Let us follow an imaginary person. This person's cancer is destined to cause symptoms and be diagnosed at age seventy, and to cause death from this cancer at age seventy-five. Suppose a screening test finds this same cancer at age sixty-seven, but treatment changes nothing. This person still dies at age seventy-five, but the survival time after diagnosis increases from five years to eight years. Lead time is the period from the day diagnosed by screening to the day it would have been diagnosed by symptoms if there had been no screening. This person's lead time is three years, and the three added years are simply time known earlier, not a single extra day lived.

An optical illusion also arises from the nature of the cancers that screening selects. Length bias is a phenomenon where, because slower-growing diseases stay without symptoms for a longer period and are thus caught more easily by screening, diseases found through screening appear milder than they actually are. If screening happens once a year, a cancer that causes symptoms in a few months easily reveals itself as symptoms between screenings, while a cancer that stays quiet for years is likely to be caught by screening. Therefore, slow cancers are mixed in more heavily among cancers found by screening. This does not mean every cancer found by screening is a slow cancer. In a model estimating the natural course using American breast cancer screening data, the proportion of pre-symptomatic breast cancers that do not grow was estimated at 4.5% (uncertainty interval 0.1 to 14.8%) (PMID 35226520). If the period staying without symptoms is longer than the remaining life expectancy, that cancer is one that would have remained quiet for a lifetime like water sitting in a reservoir.

Both lead time and length bias drive up survival rates after diagnosis without reducing a single death. Therefore, the success or failure of screening is not judged by the survival rate of patients found through screening. A group invited to screening and a group not invited are randomly divided, and the proportions of people who died from that cancer are compared. The strengths and limitations of the randomized trial itself were covered in "Is the Randomized Controlled Trial Really the 'Gold Standard'?". Whether a screening that increased survival rates reduced deaths can only be known through the mortality rates of the two groups.

How to Count Overflowing Diagnoses — Overdiagnosis

The European Randomized study of Screening for Prostate Cancer (ERSPC), started in 1993, divided 162,236 men aged 55 to 69 across several European countries into two groups, recommending repeated PSA testing to only one side (PMID 41160819). When followed for a median of 9 years, the cumulative incidence of prostate cancer was 8.2% in the screening group and 4.8% in the control group (PMID 19297566). Overdiagnosis is the number of diseases diagnosed more in the group that received screening than in the group that did not.

Overdiagnosis is not equal to the size of overdiagnosis from the beginning. The excess in the screening group contains cancers that were diagnosed earlier by the lead time but would someday have been diagnosed as symptoms in the control group as well. As time passes and those cancers are diagnosed in the control group, the difference shrinks, and the portion that remains unfilled to the end is overdiagnosis. The 16-year follow-up report in the ERSPC also noted that overdiagnosis decreased (PMID 30824296), and at a median of 23 years, the cumulative incidence in the screening group was 1.30 times that of the control group (PMID 41160819).

That is why how long the follow-up lasted greatly changes the numbers. In a meta-analysis pooling 26 low-dose CT lung cancer screening studies, the proportion of overdiagnosis among cancers found by screening went down from 37% (95% confidence interval 14% to 60%) in studies with 5 years or less of follow-up to 7% (-7% to 22%) in studies followed for more than 5 years (PMID 40679325). In a model mimicking breast screening in Flanders, Belgium, overdiagnosis per 100,000 women starting screening at age 50 was calculated to be 98.5 cases 2 years after stopping screening, and 13.4 cases 15 years after (PMID 35839596). The authors of this model concluded that to obtain unbiased estimates, follow-up must continue for 10 years or more after stopping screening. A pooled analysis by the International Cancer Screening Network noted that to measure breast cancer overdiagnosis without bias, cohorts that received screening and cohorts that did not must be compared for at least 30 years, and such randomized data will not emerge (PMID 33872390).

There are also multiple methods. A systematic review summarizing 40 studies on breast screening overdiagnosis divided the methods adjusting for lead time into those directly estimating lead time and those counting excess incidence, and further split the former into average lead time, lead time distribution, and natural history models. Methods for obtaining an unexamined comparison group were also split into randomized trial control groups, non-attenders, other regions, extensions of pre-screening trends, and natural history models (PMID 29117353). The authors concluded that each method has strengths and limitations. Excess incidence approaches the true magnitude of overdiagnosis only after being followed long enough for the lead time to fully elapse.

From 0% to 50% — Why the Number Is Never Just One

Estimates of overdiagnosis in breast cancer screening range from 0% to well over 50% in published literature (PMID 34158298). Even across the same trials, the numbers change depending on what is chosen as the denominator. An independent panel in the UK calculated excess incidence by pooling three trials that did not recommend screening to the control group after the screening phase ended: when using all cancers diagnosed over a long term in the invited group as the denominator, the figure was 11% (95% confidence interval 9–12%); when using cancers diagnosed during the screening period as the denominator, it was 19% (15–23%) (PMID 23117178). The exact same excess turns into two different numbers depending on the denominator.

The type of data also shapes the numbers. Among studies using individual-level data, no estimate exceeded 17%, whereas studies using aggregate data frequently produced estimates above 40% (PMID 34158298). The author of this review considered this difference too systematic to be a coincidence, and concluded that many estimates were heavily inflated. What counts as overdiagnosis also alters the number. In a simulation model of ductal carcinoma in situ in the breast, 20% of ductal carcinoma in situ cases diagnosed under the base definition were overdiagnosed, but changing the definition shifted the estimate from 18% to 94% (PMID 41082841).

Both low and high answers have been put forward for the same breast cancer question. If the aforementioned aggregate data studies represent the high side, a model estimating the natural history using data from the Breast Cancer Surveillance Consortium calculated that for biennial screening between ages 50 and 74, 15.4% of screen-detected cancers (uncertainty interval 9.4–26.5%)—about 1 in 7 cases—were overdiagnoses (PMID 35226520). Of those, 6.1% were due to finding cancers that would not grow, and 9.3% were due to finding cancers that would have progressed but were found in women who died first from other causes. Between the two branches seen earlier, the former side was larger in point estimate, but the uncertainty intervals of the two values (0.2–20.1% and 5.5–13.5%) overlap heavily. An analysis by the International Cancer Screening Network noted that overdiagnosis of invasive breast cancer in populations aged 50–69 is under 10%, and that higher estimates are likely the result of inaccuracies in study design (PMID 33872390). Moving beyond breast cancer to a broader range of malignancies, a reanalysis of 19 randomized trials of breast, lung, liver, prostate, and ovarian screening calculated—in post-hoc analyses excluding trials with high risk of bias—that overdiagnosis ranged from 17% to 38% depending on the screening program, and concluded that the true magnitude is likely even higher (PMID 36963292). The authors noted that trials are generally not designed to measure overdiagnosis, and that bias may have pulled estimates toward 0.

Age also drastically changes the numbers. In a cohort study using US Medicare data comparing women aged 70 and older who continued screening versus those who stopped, the share of breast cancers that might be overdiagnosed among screened women was estimated at 31% for ages 70–74, 47% for ages 75–84, and up to 54% for ages 85 and older (PMID 37549389). This study showed no statistically significant reduction in breast cancer mortality associated with screening. The authors noted that the study was not designed to estimate overdiagnosis, limiting any definitive conclusion on the overall benefits and harms of screening, that differences in risk between women who continued screening and those who stopped may have confounded the results, and that the findings were sensitive to model settings. Whether the harms of overdiagnosis are offset by benefits, and for whom, remains an important question, conclude the authors. Calculations pointing in the opposite direction also exist. A model analysis for the US Preventive Services Task Force estimated that the ratio of benefits to harms when continuing screening up to age 79 is similar to or better than stopping at age 74 (PMID 38687505). Overdiagnosis on the former branch has more room to occur the shorter the remaining life expectancy is.

Although there is a meta-analysis pooling 30 studies to arrive at an overdiagnosis rate of 12.6% for women aged 40 and older, the $I^2$ representing variation between studies was 99.99% (PMID 36983705). How to read this statistic is covered in "I² Is Not the Size of the Scatter — How to Read Heterogeneity Statistics". When you hear an estimate of overdiagnosis, you must first ask what the denominator was, what kind of data was used, how long people were followed, and what was counted as overdiagnosis. If those four things differ, the numbers for the exact same screening program can diverge by several-fold.

The Other Side of the Same Scale — Preventing Deaths and Overdiagnosis

Following up the ERSPC for a median of 23 years, prostate cancer deaths were 13% lower in the screening group (rate ratio 0.87, 95% confidence interval 0.80 to 0.95). The absolute risk reduction was 0.22% (PMID 41160819). For every 456 men (95% confidence interval 306 to 943 men) invited to screening, and for every 12 men diagnosed with prostate cancer, 1 prostate cancer death was prevented. This is an improvement over the figures from the shorter 16-year follow-up, which were 628 men and 18 men. Although the relative reduction decreased from 20% in the 16-year report (rate ratio 0.80) to 13% (PMID 30824296), the number of people needed to prevent 1 death fell. The authors concluded that longer follow-up sustained the reduction in mortality and improved the ratio of benefits to harms. In the same concluding paragraph, the authors wrote that future screening should reduce overdiagnosis by selecting target populations based on risk. This does not mean that out of the 12 diagnosed men, the 11 who did not avoid death were all overdiagnosed. Relative risk, absolute risk, and how to read such numbers were covered in What Does "50% Lower Risk" Mean in People? — Reading Relative Risk, Absolute Risk, and NNT.

Data from the Rotterdam region of the ERSPC with 21 years of follow-up showed that patients in the screening group were less likely to be diagnosed with advanced stage disease than control patients (12% vs. 23%), and less likely to receive palliative treatment (21% vs. 55%) (PMID 37690917). In exchange, radical prostatectomies were more frequent at 32% vs. 14%, which the authors noted was the price paid for more overdiagnosis and likely more overtreatment. Over time, localized treatments decreased and active surveillance increased.

The weight of evidence also shifted over time. A 2013 Cochrane systematic review combined five trials and found no significant reduction in prostate cancer deaths, noting that only the ERSPC reported a 21% reduction in its pre-specified cohort aged 55 to 69 (PMID 23440794). The 2026 update, based on sensitivity analyses assessing risk of bias using the ERSPC, judged that screening likely reduces prostate cancer mortality (moderate certainty, 2 fewer deaths per 1,000 people) (PMID 42134821). It noted that overall mortality might be reduced, but the certainty is low, the confidence interval includes no effect, and the interpretation of the results is highly sensitive to where one sets the clinically meaningful difference. It also noted a lack of evidence regarding harms such as biopsy and treatment complications. The US PLCO trial, unlike the ERSPC, found no reduction in mortality; however, when re-analyzed by matching how much screening and diagnostic workup advanced diagnosis in each group using average lead time, no evidence emerged that the two trials had different effects (PMID 28869989). The average lead time was longer in the PLCO control group than in the ERSPC control group, and for every 1-year increase in average lead time, the risk of prostate cancer death was estimated to be 7% to 9% lower. Noting the limitation that average lead time is a simple metric, the authors concluded that, accounting for differences in implementation and settings, the two trials provide mutually consistent evidence for the mortality reduction of screening.

Breast cancer screening also calculates benefits and overdiagnosis together. An independent UK panel reported in a meta-analysis of 11 randomized trials that breast cancer mortality among invited women was 20% lower, calculating that inviting 10,000 women aged 50 to screening for 20 years prevents 43 breast cancer deaths and leads to 129 overdiagnosed cases (PMID 23117178). This is about 3 overdiagnoses for every 1 death prevented. The panel noted that these numbers come from studies with many limitations, carry large uncertainty, and should be viewed only as a rough guide. A Cochrane review found no significant reduction in breast cancer mortality in three trials where randomization was adequate (PMID 23737396). When all seven trials were pooled together, the relative reduction was significant with a relative risk of 0.81, but the authors considered this an implausible result where breast cancer deaths were skewed toward the screening arm due to differences in cause-of-death classification. Assuming a 15% reduction in mortality and 30% overdiagnosis, this review calculated that inviting 2,000 people for 10 years prevents 1 breast cancer death while 10 receive unnecessary treatment (PMID 23737396). A previous modeling analysis for the US Preventive Services Task Force, assuming 100% screening uptake, estimated that screening with digital breast tomosynthesis every 2 years from age 40 to 74 prevents 8.2 breast cancer deaths per 1,000 women, while producing 1,376 false-positive recalls and 14 overdiagnoses (PMID 38687505).

A review spanning literature on breast, prostate, and thyroid screening noted that benefits and harms cannot be separated. Many screening programs simultaneously produce the benefit of reducing mortality from that cancer at the population level, and the harms of overdiagnosis and resulting unnecessary treatment (PMID 28104314). Overdiagnosis does not erase the benefits of screening. It is a harm that sits on the opposite side of the same scale as the benefit. Exactly what goes onto the two pans of the scale and in what amounts varies by screening test, age, and duration of follow-up.

Low-Dose CT and Low-Risk Populations

Low-dose CT lung cancer screening was covered in detail in "Lung Cancer and Low-Dose CT Screening — Reading a Small Spot in the Picture Over Time", so we will leave only a few numbers. In a meta-analysis of 9 randomized trials, low-dose CT screening reduced lung cancer deaths by 16% (relative risk 0.84), and the overdiagnosis estimate considered most valid by the authors was 8.9% (PMID 32583338). In the meta-analysis seen earlier, the share of overdiagnosis was 66% in cancers without a solid component and 7% in solid cancers, while it was 38% in Asian countries and 22% in Western countries (PMID 40679325).

When the screening target is expanded toward lower-risk groups, a picture similar to thyroid cancer in Korea appears. In Shanghai, China, where low-dose CT became widely used in workplace health checkups after 2011, the incidence of early-stage lung cancer in women increased by 16.1 per 100,000 people between 2002 and 2017, but there was no significant decrease in the incidence of advanced lung cancer, while the lung cancer mortality rate steadily declined over the same period (PMID 35998705). For men, early-stage lung cancer increased by 6.9 per 100,000 people while advanced lung cancer decreased by 5.5. The authors interpreted this as population-level evidence that increased screening in low-risk groups led to lung cancer overdiagnosis among Chinese women. Data from a hospital in Taiwan also showed that as the number of low-dose CT scans increased, stage 0 lung cancer increased while stage 4 lung cancer remained similar, with the effect being greater in women (PMID 38867342). Even with the same test, the tilt of the scale changes depending on whom you screen.

When You Lower the Line, or Scan for Another Reason

"A Line Painted on a Slope — An Introduction to Reference Ranges and Diagnostic Thresholds" followed an example where a single kidney function line drawn regardless of age increased the diagnosis of chronic kidney disease in older adults. The diagnoses of older adults who fell below that line were not wrong by pathology or measurement. The group-level definition we saw earlier encompasses situations like this. They defined overdiagnosis at the group level as a case where a condition is diagnosed in a way that an expert group accepts as correct, but the disease label or the subsequent intervention does more harm than good (PMID 27402883). A response paper stated that this definition captures a broader type than definitions that only count the diagnosis of harmless diseases, noted that a diagnosis that was once an overdiagnosis can become a beneficial one if medical knowledge and practice change, and wrote that the effectiveness of treatment partly determines whether something is an overdiagnosis (PMID 27573152). Viewed through this definition, where the line is drawn also changes the size of overdiagnosis.

The disease label itself is also a subject of debate. An international symposium addressed whether the name "cancer" should be attached to Grade Group 1 prostate cancer (PMID 39350309). According to this report, Grade Group 1 is very common in autopsies of older men, and pure Grade Group 1 lacks the ability to metastasize. While the attendees put forward a proposal to give it a different name, they noted that Grade Group 1 should not be considered normal and that active surveillance is strictly necessary. A majority of the attendees believed that public health would improve significantly if lesions incapable of causing symptoms or threatening life were given a name other than "cancer," but they also left open the question of whether patients would continue surveillance if the label "cancer" were not attached. The weight of the word was also revealed in a survey of the general public. Members of the general public associated "cancer" with death and fear much more than the synonymous term "carcinoma" (PMID 41190609).

In the activity of pumping the reservoir, alongside screening, there are images taken for other reasons. An incidentaloma is a lump discovered incidentally on an image taken for another reason without suspecting that disease. Most lumps captured incidentally on abdominal CT, MRI, and ultrasound do not harm the patient, but when unnecessary follow-up tests tail one another, anxiety, discomfort, cost, and complications arise (PMID 34767479). The author of this article argued that there is also an argument to be made for not reporting small, simple cysts with an extremely low likelihood of clinical significance. The same article noted that for lumps that appear benign but are not definitive, follow-up imaging, more precise imaging, or sometimes a biopsy may be necessary, and that for more concerning lumps, the need for additional tests is clear. A review summarizing incidentaloma guidelines in the gastroenterology field identified cascading care and overdetection as major ethical issues (PMID 42893844).

In South Korea's thyroid screening, there was also a pathway of receiving it while coming in for other screenings. In a 2014 national cancer screening survey, among 2,270 women over the age of 30, 25.1% had received a thyroid cancer screening, and women who received only a breast ultrasound were 2.71 times (odds ratio) more likely to have received a thyroid screening than women who did not receive a breast screening (PMID 36782225). The authors concluded that thyroid screening in South Korea is likely happening opportunistically. Overdiagnosis arises not only from screening programs, but also from moving the line and looking inside for other reasons.

Labels and Over-treatment

In a cohort following 6,265 low-risk prostate cancer patients in Germany, Austria, and Switzerland, patients who underwent nerve-sparing radical prostatectomy saw their urinary control score drop by 18 points and their sexual function score drop by 35 points 12 months later (PMID 41698790). Patients who were monitored without immediate treatment maintained their function across all domains. This is an observational study that did not randomly assign people to choose treatment versus surveillance. Over-treatment is the treatment of an overdiagnosed disease; because the disease would not have caused harm, treatment leaves only harm with nothing to be gained. For an overdiagnosed person, this loss of function is a cost for which nothing was prevented. The problem is that we cannot know in advance who that person is.

The harm is not distributed fairly. In South Korea between 2006 and 2015, higher-income individuals had more thyroid screenings, thyroid cancer diagnoses, and post-surgical complications, while lower-income individuals had higher thyroid cancer mortality rates (PMID 33176753). In the same data, lung cancer and stroke, examined as comparisons, showed lower incidence and mortality together among higher-income individuals, differing from the pattern of thyroid cancer. The authors concluded that this trend in disparities supports the overdiagnosis of thyroid cancer in South Korea.

There are also costs to communicating overdiagnosis. The authors who formulated the definition of overdiagnosis noted that messages about overdiagnosis could harm people who had been unaware they might be overdiagnosed, and they also pointed out a distributive justice issue in which the burden of overdiagnosis and its prevention may fall on the least well-off (PMID 27402883). In a qualitative study interviewing 29 South Korean women, participants were largely unaware of thyroid cancer overdiagnosis, and hearing about it did not significantly change their intentions to get screened in the future (PMID 26546276). Some found the explanation of overdiagnosis difficult to understand.

Watching, Finding Less, Stopping the Search

In 1993, the Kuma Hospital in Japan first introduced the approach of watching low-risk thyroid microcarcinomas under 1 cm instead of operating on them right away (PMID 33224846). It was one of the methods meant to prevent overdiagnosis and overtreatment of low-risk microcarcinomas (PMID 35256266). Active surveillance is an approach where, instead of treating a low-risk cancer immediately, doctors check it at set intervals and watch it, moving on to treatment only if the cancer progresses or the patient wants it. Japanese clinical guidelines approved active surveillance as a primary management option for asymptomatic microcarcinomas in 2010, and the 2015 American Thyroid Association guidelines accepted it as an alternative to immediate surgery for very low-risk papillary cancers confirmed by cytology (PMID 33224846). A 2026 review noted that long-term data have accumulated showing that in appropriately selected low-risk patients, the progression rate is very low and cancer outcomes are comparable to immediate surgery (PMID 42556947).

The Korean Thyroid Association's 2025 guidelines recommend active surveillance for adult microcarcinomas confirmed as malignant or suspicious for malignancy (Bethesda V-VI) by cytology, provided there are no lymph node or distant metastases, no clear extrathyroidal extension, no invasion of the trachea or recurrent laryngeal nerve, and no aggressive histologic type (PMID 40598902). The same guidelines state that both immediate surgery and active surveillance can be considered, that initial evaluation should be done with high-resolution neck ultrasound by an experienced examiner, that age, comorbid diseases, and the ability to be followed for a long time should be weighed, and that the pros and cons of surgery and surveillance should be balanced through shared decision-making. Follow-up is done every 6 months for 2 years, and annually thereafter, using neck ultrasound and thyroid function tests, and surgery is performed if the tumor clearly grows or new metastases appear. The guidelines note on their own that uncertainties remain. Even for small thyroid cancers, if there is clinically evident lymph node or distant metastasis, or invasion of surrounding organs, surgery is indicated (PMID 35256266). A consensus statement from the Japanese Association of Endocrine Surgery also recommended immediate surgery for tumors attached to the trachea or lying along the path of the recurrent laryngeal nerve (PMID 33023426).

There is also a path to finding less. In a screening trial in Gothenburg, Sweden, instead of the previous approach where all men with a PSA of 3 or higher underwent an MRI followed by systematic biopsies together, performing targeted biopsies of only the suspicious areas on MRI cut the diagnosis of clinically insignificant cancers in half, from 1.2% to 0.6% (PMID 36477032). As a trade-off, the detection of intermediate-risk tumors was delayed in some patients. In a Stockholm trial as well, the MRI pathway lowered the proportion of insignificant cancers from 12% to 4% while detection of clinically significant cancers did not fall behind conventional biopsies (PMID 34237810).

There is also an age to stop the search. A 2003 Cochrane review concluded that because it can take 10 years for a reduction in prostate cancer deaths to appear, men with a life expectancy of less than 10 to 15 years should be informed that screening is unlikely to benefit them (PMID 23440794). In the ERSPC Rotterdam study, the cumulative probability of dying from prostate cancer up to age 85 was 0.54% for 7,052 men who were screened up to ages 70–74 and not diagnosed, but it was 6.8% in men with a PSA of 6.5 or higher and no history of a benign biopsy (PMID 37919190). The authors suggested that screening can be stopped if the PSA is under 3 or there is a history of a benign biopsy, while noting that men with a higher PSA and no biopsy history can consider continuing screening if their life expectancy exceeds 10 years.

The incidence of thyroid cancer in South Korea began to decline in 2012, ahead of 2014 when the overdiagnosis debate began (PMID 33979040). The authors wrote that changes in thyroid nodule examination guidelines may have influenced this inflection, and the 2014 debate may have accelerated the decrease.

In a cohort tracking 434,228 South Korean thyroid cancer patients, the thyroid cancer mortality rate per 1,000 person-years dropped from 1.94 in 2005 to 0.76 in 2013, and then rose to 2.70 in 2018 (PMID 38874484). This rise appeared in patients who underwent total thyroidectomy or no surgery, whereas the mortality rate for patients who underwent hemithyroidectomy or subtotal thyroidectomy remained flat throughout 2005–2018. The authors concluded that properly diagnosing and treating patients at high risk of progression while reducing overdiagnosis and overtreatment in low-risk patients are both important. Active surveillance is one of the options chosen together by the medical team and the patient within the conditions set by the guidelines.

One Person in the Clinic

Let us follow a hypothetical forty-six-year-old woman. During a health checkup, she had a breast ultrasound and also received a thyroid ultrasound, which revealed a 7mm nodule. The first fork in the road is here. According to a review, most guidelines recommend that nodules under 1cm should not undergo fine-needle aspiration regardless of ultrasound findings, and the review notes that patients watched in this way are effectively receiving active surveillance (PMID 33224846). While watched without being poked, this nodule is not given the name of cancer. This spot, where it splits whether a disease name is attached or not, is where overdiagnosis can occur. This person underwent a fine-needle aspiration biopsy, and the result was papillary thyroid cancer. This diagnosis is correct. Whether this cancer would have remained quiet throughout this person's lifetime is something no one knows right now. Many prognostic markers can only be evaluated after surgery, and finding preoperative markers for small cancers is cited as the most important unmet need in thyroid cancer (PMID 32207260). The estimate that over 85% of thyroid cancers in South Korean women between 2013 and 2017 were overdiagnosed is a number from a modeled population, and it does not contain this person's nodule size, age, or ultrasound findings.

The next step for this person is to check the conditions of the previous section one by one. If an experienced examiner's ultrasound shows lymph node metastasis, clear extension outside the thyroid, problems toward the trachea or recurrent laryngeal nerve, or if an aggressive tissue type is suspected on cytology, this cancer falls out of the target for active surveillance and becomes a target for surgery. When such findings are absent, guidelines immediately place both immediate surgery and active surveillance as options, and have them weigh age, comorbidities, ability to be followed long-term, and anxiety and preferences together. If active surveillance is chosen, ultrasound is performed every 6 months for 2 years, and if the tumor clearly grows, it moves on to surgery.

The questions this person will ask in the clinic are summarized as follows. Does my cancer have findings that require surgery? If I choose surveillance, what and how often is checked, and what changes lead to moving on to surgery? If I choose surgery, what do I gain and what do I endure? Whether this person is an overdiagnosis is never confirmed, and overdiagnosis remains only as a number in the population to which this person belongs. That number does not tell this person's destiny; it tells why such questions must be asked.

Where to Look Further

At the moment of diagnosis, separating cancers that will stay quiet for a lifetime from those that will progress remains an open question. In thyroid cancer, preoperative markers are an unmet need (PMID 32207260), and in ductal carcinoma in situ, current markers also fail to accurately predict the risk of progressing to invasive cancer, while active surveillance trials are ongoing (PMID 34931856). With multi-cancer early detection tests that screen for multiple cancers with a single blood draw, technologies change so rapidly that a new problem has arisen: overdiagnosis must be estimated from short-term data (PMID 39592444). There is no verified answer as to whether changing a disease name reduces harm (though many symposium attendees thought it would improve things), and we still do not know whether people continue surveillance even without the name "cancer" (PMID 39350309).

Earlier installments and related articles in this series are as follows. False positives and test performance are in "What a Test Actually Tells You — An Introduction to Sensitivity, Specificity, and Pretest Probability", pre-test probability is in "Where Does Pretest Probability Come From? — History, Physical Examination, and Clinical Prediction Rules", and places where numbers and outcomes diverge, such as survival rates versus mortality rates, are in "Better Numbers, Better Patients? — An Introduction to Surrogate Endpoints", while the business of drawing lines is in "A Line Painted on a Slope — An Introduction to Reference Ranges and Diagnostic Thresholds". How to account for people who die of other causes first is covered in "When People Don't Stay in Line — The Ones Who Switched, and the Ones Who Left for Another Reason", and why we remove lumps that mostly never become cancer is covered in "Why Remove Colon Polyps, and What Do Colorectal Cancer Tests Look For? — A Floor Relaid Every Day, Bumps That Mostly Never Become Cancer".

The next installment, "Why Medicines Pile Up — An Introduction to Polypharmacy and Time to Benefit", lifts the same scales on the treatment side, comparing the time it takes for the benefits of each drug to appear when multiple medicines accumulate against the remaining time a person has left.

Source Papers

• "Decreasing trends in thyroid cancer incidence in South Korea: What happened in South Korea?" (2021) — PMID 33979040 — Age-standardized incidence of thyroid cancer in South Korea went from 6.3 (1999) to 63.4 (2012) per 100,000, declining since 2012 (prior to the 2014 debate), with mortality declining since the early 2000s, likely influenced by changes in screening guidelines for nodules.
• "Income differences in screening, incidence, postoperative complications, and mortality of thyroid cancer in South Korea" (2020) — PMID 33176753 — Incidence from 2006 to 2015 went from 46.6 to 115.0 to 63.5 per 100,000 with no change in mortality, higher income associated with screening, incidence, and postoperative complications while lower income associated with higher mortality, contrasted with negative controls (lung cancer and stroke), pointing to overdiagnosis.
• "Thyroid Cancer Screening in South Korea Increases Detection of Papillary Cancers with No Impact on Other Subtypes or Thyroid Cancer Mortality." (2016) — PMID 27627550 — Ecological analysis of 16 cities and provinces, screening rate and incidence correlation r=0.77, associated only with papillary cancer, unrelated to mortality.
• "Unmet Clinical Needs in the Treatment of Patients with Thyroid Cancer." (2020) — PMID 32207260 — South Korean mortality went from 0.17 (1985) to 0.85 (2004) to 0.42 (2015) per 100,000, interpreted as mortality reduction from early detection, but cost-effectiveness may be low, with pre-surgical prognostic markers being the greatest unmet challenge.
• "Cancer overdiagnosis: a challenge in the era of screening." (2022) — PMID 36568283 — Effective screening must reduce advanced and metastatic cancer and mortality, otherwise it may be finding a reservoir of non-progressing lesions.
• "A definition and ethical evaluation of overdiagnosis." (2016) — PMID 27402883 — Population-level definition (diagnosis accepted as correct by a panel of experts + benefit-harm balance of the disease label and intervention), distributive equity, and harms of overdiagnosis communication.
• "A definition and ethical evaluation of overdiagnosis: response to commentaries." (2016) — PMID 27573152 — Distinguished from false positives, a broader definition than diagnosing harmless disease, and overdiagnosis must be estimated at the population level.
• "Overdiagnosis in cancer." (2010) — PMID 20413742 — Two preconditions of overdiagnosis: a reservoir of quiet disease and activities that find it (especially screening).
• "Conceptualizing overdiagnosis in cancer screening." (2015) — PMID 25663695 — Two branches: tumor-related (regression, stagnation, slow progression) and person-related (death from other causes comes first), with proportions varying by organ, method, and patient.
• "USPSTF Recommendations and Overdiagnosis." (2024) — PMID 38865364 — Overdiagnosis is directly unmeasurable and indirectly estimated, requiring consideration of lead time and incidence trends (authors estimate breast overdiagnosis at 10% or less).
• "Evolving epidemiological patterns of thyroid cancer and estimates of overdiagnosis in 2013-17 in 63 countries worldwide" (2024) — PMID 39389067 — Over 85% of thyroid cancer in women estimated to be overdiagnosed in South Korea and others, totaling 75.6% of 2,297,057 cases overall.
• "European Study of Prostate Cancer Screening - 23-Year Follow-up." (2025) — PMID 41160819 — 162,236 men, mortality rate ratio 0.87, absolute reduction 0.22%, inviting 456 or diagnosing 12 to prevent 1 death (16-year figures 628 and 18), cumulative incidence 1.30 times, suggesting risk-based strategies to reduce overdiagnosis.
• "Screening and prostate-cancer mortality in a randomized European study." (2009) — PMID 19297566 — Median 9-year cumulative incidence 8.2% versus 4.8%, 20% mortality reduction, high risk of overdiagnosis.
• "A 16-yr Follow-up of the European Randomized study of Screening for Prostate Cancer." (2019) — PMID 30824296 — 16-year mortality rate ratio 0.80, greater absolute benefit and decreased excess incidence with long follow-up.
• "Overdiagnosis in low-dose CT lung cancer screening: A systematic review and meta-analysis..." (2025) — PMID 40679325 — 26 studies, overdiagnosis of screen-detected cancer dropped from 37% to 7% with follow-up exceeding 5 years, non-solid 66% versus solid 7%, Asia 38% versus Western 22%.
• "Overdiagnosis of invasive breast cancer in population-based breast cancer screening: A short- and long-term perspective." (2022) — PMID 35839596 — Flanders model, women starting at age 50, 2-year 98.5 versus 15-year 13.4 per 100,000, requiring follow-up of 10 years or more, greater in older ages.
• "Breast cancer screening and overdiagnosis." (2021) — PMID 33872390 — Unbiased measurement requires comparison over 30 years or more, invasive cancer overdiagnosis under 10% for ages 50–69, higher figures likely due to design inaccuracies.
• "Quantifying Overdiagnosis in Cancer Screening: A Systematic Review to Evaluate the Methodology." (2017) — PMID 29117353 — 40 studies, lead-time approach and excess-incidence approach, methods for obtaining non-screened comparison groups, each with strengths and limitations.
• "Estimations of overdiagnosis in breast cancer screening vary between 0% and over 50%: why?" (2021) — PMID 34158298 — Individual data studies show 17% or less, aggregate data studies frequently exceed 40%, many estimates are overestimates.
• "The benefits and harms of breast cancer screening: an independent review." (2012) — PMID 23117178 — 20% mortality reduction, excess incidence 11% (long-term total denominator) or 19% (screening-period denominator), inviting 10,000 women for 20 years prevents 43 deaths while causing 129 overdiagnosed cases, with substantial uncertainty.
• "Overdiagnosis of ductal carcinoma in situ by grade and definition in population-based screening: A modeling study." (2025) — PMID 41082841 — Ductal carcinoma in situ baseline 20%, ranging from 18% to 94% depending on definition.
• "Estimation of Breast Cancer Overdiagnosis in a U.S. Breast Screening Cohort." (2022) — PMID 35226520 — 15.4% (UI 9.4–26.5) overdiagnosis of screen-detected cancer, 6.1% non-progressing and 9.3% dying from other causes.
• "Quantification of overdiagnosis in randomised trials of cancer screening" (2023) — PMID 36963292 — 19 trials, post hoc analyses 17%–38%, trials not designed to measure overdiagnosis, likely an underestimate.
• "Estimating Breast Cancer Overdiagnosis After Screening Mammography Among Older Women in the United States." (2023) — PMID 37549389 — Ages 70–74 at 31%, 75–84 at 47%, 85 and older up to 54%, no significant reduction in breast cancer mortality, limited by confounding and model sensitivity.
• "Overdiagnosis Due to Screening Mammography for Breast Cancer among Women Aged 40 Years and Over" (2023) — PMID 36983705 — Combined 30 studies at 12.6%, I² 99.99%.
• "From Screening to Mortality Reduction... ERSPC Rotterdam After 21 Years of Follow-up..." (2024) — PMID 37690917 — Advanced stage 12% versus 23%, palliative care 21% versus 55%, curative resection 32% versus 14%, price of overdiagnosis and overtreatment, increasing active surveillance over time.
• "Screening for prostate cancer." (2013) — PMID 23440794 — Cochrane review, pooled mortality reduction not significant, ERSPC alone 21% reduction, overdiagnosis and overtreatment common, unlikely to benefit those with life expectancy under 10–15 years.
• "Prostate-specific antigen (PSA) test for prostate cancer screening." (2026) — PMID 42134821 — Updated Cochrane review, likely mortality reduction (moderate certainty, 2 per 1,000 men), effect on all-cause mortality includes no effect, insufficient evidence on harms.
• "Reconciling the Effects of Screening on Prostate Cancer Mortality in the ERSPC and PLCO Trials." (2017) — PMID 28869989 — No evidence of a difference in trial effects when adjusted for mean lead time, 7%–9% reduction in mortality risk per year of lead time, findings that reconcile with each other.
• "Screening for breast cancer with mammography." (2013) — PMID 23737396 — Three adequately randomized trials showed no significant reduction, seven-trial pool showed 0.81 but skewed toward screening due to cause misclassification, assuming 15% or 30%, inviting 2,000 women for 10 years avoids 1 death while causing 10 unnecessary treatments.
• "Collaborative Modeling to Compare Different Breast Cancer Screening Strategies" (2024) — PMID 38687505 — Assuming 100% participation, continuing to age 79 yields similar or better benefit-harm ratios, digital breast tomosynthesis every 2 years for ages 40–74 prevents 8.2 deaths per 1,000 women while causing 1,376 false-positive recalls and 14 overdiagnosed cases.
• "Breast, prostate, and thyroid cancer screening tests and overdiagnosis." (2017) — PMID 28104314 — Many screening tests produce both benefits (mortality reduction) and harms (overdiagnosis and overtreatment) at the population level.
• "Lung Cancer Screening with Low-Dose CT: a Meta-Analysis." (2020) — PMID 32583338 — 9 studies, lung cancer mortality RR 0.84, most valid overdiagnosis estimate 8.9%.
• "The Introduction of Low-Dose CT Imaging and Lung Cancer Overdiagnosis in Chinese Women." (2023) — PMID 35998705 — Shanghai, women early-stage cancer +16.1 per 100,000, no significant reduction in advanced cancer, continued drop in mortality, men early-stage +6.9 and advanced -5.5.
• "Impact of annual trend volume of low-dose computed tomography for lung cancer screening on overdiagnosis..." (2024) — PMID 38867342 — Taiwan hospital data, as LDCT volume increased stage 0 increased while stage 4 remained constant, greater in women.
• "When is prostate cancer really cancer?" (2025) — PMID 39350309 — International symposium, grade group 1 is common in autopsy and lacks metastatic potential, discussion of renaming, but it is not normal and active surveillance is mandatory, unknown whether renaming will affect surveillance adherence.
• "Linguistic Framing in "Cancer" and "Cancer"-Adjacent Terms." (2026) — PMID 41190609 — General public associates cancer more with death and fear than carcinoma.
• "Incidentalomas at abdominal imaging." (2023) — PMID 34767479 — Most abdominal incidentalomas are harmless, cascades of unnecessary extra testing cause harm, rationale for not reporting very small simple cysts.
• "Incidentalomas in Gastroenterology and Hepatology: A Narrative Review..." (2026) — PMID 42893844 — Ethical issues such as diagnostic cascades and over-detection, 6-step management framework.
• "Role of breast cancer screening in the overdiagnosis of thyroid cancer: results from a cross-sectional nationwide survey." (2023) — PMID 36782225 — 25.1% of 2,270 people in 2014 had thyroid screening, breast ultrasound alone OR 2.71, opportunistic screening.
• "Patient-reported outcomes: active surveillance vs radical therapies in low-risk prostate cancer." (2026) — PMID 41698790 — Prospective cohort of 6,265 men, nerve-sparing radical surgery urinary function -18 and sexual function -35, maintained with active surveillance.
• "A qualitative study of women's views on overdiagnosis and screening for thyroid cancer in Korea." (2015) — PMID 26546276 — Interviews with 29 women, almost no awareness of overdiagnosis, information had little effect on screening intention, some found it difficult to understand.
• "Active surveillance for low-risk small papillary thyroid cancer in North American countries..." (2020) — PMID 33224846 — Introduced at Kuma Hospital in 1993, Japanese guidelines primary management in 2010, ATA alternative in 2015.
• "Active Surveillance Versus Surgery for Low-risk Well-differentiated Thyroid Cancer." (2026) — PMID 42556947 — Progression rate very low in appropriately selected patients and outcomes comparable to immediate surgery.
• "2025 Korean Thyroid Association Clinical Management Guideline on Active Surveillance for Low-Risk Papillary Thyroid Carcinoma." (2025) — PMID 40598902 — Eligibility criteria, considering both immediate surgery and surveillance, shared decision-making, follow-up every 6 months for 2 years and annually thereafter, surgery upon progression, uncertainties remain.
• "Active surveillance of low-risk papillary thyroid microcarcinoma." (2023) — PMID 35256266 — Surgery for microcarcinoma with lymph node or distant metastasis, or invasion of adjacent organs.
• "Indications and Strategy for Active Surveillance of Adult Low-Risk Papillary Thyroid Microcarcinoma: Consensus Statements from the Japan Association of Endocrine Surgery..." (2021) — PMID 33023426 — Immediate surgery recommended even for tumors attached to the trachea or in the path of the recurrent laryngeal nerve.
• "Prostate Cancer Screening with PSA and MRI Followed by Targeted Biopsy Only." (2022) — PMID 36477032 — Göteborg-2, insignificant cancer 0.6% versus 1.2% (RR 0.46), some delay in finding intermediate-risk tumors.
• "MRI-Targeted or Standard Biopsy in Prostate Cancer Screening." (2021) — PMID 34237810 — STHLM3-MRI, non-inferior for significant cancer, insignificant cancer 4% versus 12%.
• "Prostate Cancer Mortality Among Elderly Men After Discontinuing Organised Screening..." (2024) — PMID 37919190 — 7,052 unscreened men ages 70–74, prostate cancer mortality 0.54% by age 85 (6.8% for PSA ≥ 6.5 with no biopsy history), can stop if PSA under 3 and benign biopsy history, consider continuing if PSA high and no biopsy history and life expectancy exceeds 10 years.
• "Thyroid cancer-specific mortality during 2005-2018 in Korea, aftermath of the overdiagnosis issue" (2024) — PMID 38874484 — 434,228 patients, deaths per 1,000 person-years 1.94 (2005) to 0.76 (2013) to 2.70 (2018), increase driven by total thyroidectomy and no-surgery groups while hemithyroidectomy and subtotal thyroidectomy groups remained flat, highlighting the importance of both proper diagnosis and treatment for high risk and reducing overdiagnosis for low risk.
• "Ductal Carcinoma in Situ: State-of-the-Art Review." (2022) — PMID 34931856 — Current markers inadequately predict progression risk, active surveillance trials underway.
• "Quantifying Overdiagnosis for Multicancer Detection Tests: A Novel Method." (2024) — PMID 39592444 — Multi-cancer detection tests are particularly difficult to estimate because overdiagnosis is unobservable and requires short-term data.