Korea's cancer registry told a journal that a study of rising thyroid cancer deaths had not measured mortality

In September 2024 the International Journal of Surgery published a study with an arresting conclusion: after years of decline, thyroid cancer deaths in Korea had started rising again among patients diagnosed since 2015 — at the same time as diagnoses were falling. It read as a warning that the campaign against overdiagnosis had gone too far.

Six months later the Korea Central Cancer Registry replied in the same journal. Its objection was not to the arithmetic. It was to the name.

What the study reported#

Researchers used the National Health Insurance Service database to follow 434,228 people diagnosed with thyroid cancer, with a median follow-up of 7.4 years, and calculated what they called the thyroid cancer-specific mortality rate by year of diagnosis.

Over the same span the standardised incidence rate went from 91.9 per 100,000 in 2012 down to 50.6 in 2015. Diagnoses nearly halved while this figure more than tripled, and the authors read that as evidence that appropriate diagnosis of high-risk disease matters alongside the effort to reduce overdiagnosis.

The registry's first objection: this is not mortality#

Statisticians at the Korea Central Cancer Registry, part of the National Cancer Center, wrote that the analysis counts deaths among people already diagnosed with thyroid cancer, which is a case fatality rate. A mortality rate counts deaths in the general population over a period.

The registry added that case fatality is a reasonable measure for an acute illness, where progression is fast and death follows soon, and a misleading one for a cancer where the risk of dying of something else grows with every year of follow-up.

The second objection: the rate's shape#

Expressing the figure per person-year assumes the risk is constant over time. The registry points out that it is not: the risk of death after a cancer diagnosis is highest in the first year and falls thereafter, becoming minimal after about five. Patients diagnosed in 2018 had been followed for a fraction of the time of those diagnosed in 2013, so a larger share of their observed follow-up sits in the high-risk window — which can inflate the later figures without any change in the disease.

The third objection: who is in the cohort changed#

As screening of low-risk people fell away, the registry argues, the remaining diagnosed population contained a higher proportion of advanced disease. A case fatality rate calculated over a sicker cohort rises even if every individual's prognosis is unchanged. Its published table shows the shift.

The localised share falls from 50.4 percent in 2005 to 42.0 percent in 2016 as the registry describes — and then moves back to 47.2 percent in 2018, which their own table records and their argument does not address. Read strictly, the case-mix explanation fits the years to 2016 better than it fits the last year in the series.

The column that settles what the population experienced#

The most direct answer in that table is not a rate at all. It is the raw count of deaths.

Diagnoses tripled and then halved. Deaths stayed inside a band of sixty-four. The age-standardised mortality rate fell from 1.28 per 100,000 in 2005 to 0.80 in 2018 — the opposite direction to the figure that made the original study's headline, because it is a different figure.

What this exchange does not resolve#

1. The study's authors published a response in the same journal in November 2025, titled "Understanding thyroid cancer-specific mortality metrics: a response to recent commentary". It is behind a subscription and we have not read it, so nothing here should be taken as representing their answer.
2. The registry's objection is to naming and interpretation, not to the underlying counts. Whether case fatality among diagnosed Korean patients genuinely rose after 2013 is a separate question, and the confounders the registry names — follow-up length and case mix — would need to be handled before it could be answered.
3. Falling population mortality is not by itself proof that reduced screening caused no harm. A small number of avoidable deaths could sit inside a total that stays between 330 and 394 without moving it detectably.
4. The registry's case-mix argument fits the stage distribution through 2016 and not the 2018 figure in its own table.