South Korean Thyroid Cancer Trends: Good News and Bad

The Good News: Better Identification of Cancers Initially Decreased Rates of Persistent and Recurrent Disease

Between 1996 and 2005, cancer care in the institutions included in this study was likely getting better as a result of the government's initiative. The proportion of thyroid cancer cases with cervical lymph node metastases at the time of diagnosis decreased, and the proportion of patients with high volume cervical metastases also decreased. Furthermore, rates of persistent and recurrent disease fell. All of this suggests that, overall, clinically significant disease was being caught at an earlier stage. Table 3 shows that for patients in period 2 compared to period 1, the difference in hazard rates for structural recurrent or persistent disease was not significantly different when adjusted for stage at diagnosis and treatment. This suggests that earlier disease identification and effective care were the reasons for the difference, further supporting the hypothesis that cancer care was likely getting better.

The Bad News: As Non-Sanctioned Thyroid Screening Took Hold, Population Benefit Declined

The proportion of newly identified cases that were of papillary histology and of small size increased, suggesting there was likely some identification of clinically indolent disease. As rates of thyroid cancer detection continued to climb through 2005 to quintuple what they had been in 1996, the falling rates of recurrent/persistent disease and what the authors labeled as disease-specific mortality were unlikely to be due to improvements in treatment. In their adjusted analysis, they controlled for all the factors that would be expected to favorably impact these rates. The differences were still significant for Period 3 compared to Period 1, and for deaths. This means catching aggressive disease earlier or treating it better was not why these rates fell. Some unmeasured factor(s) is/are the likely reason for the improvement. Common epidemiologic reasons for this are record keeping changes (how deaths are reported or attributed), or increasing numbers of deaths from other competing causes before people can die of their thyroid cancer (increasing cardiovascular deaths, for example). To be sure, a type II error could also be the reason; the follow-up length and case numbers were just too small to show the value of the increased detection. Increasing follow-up length is unlikely to solve this, however. The additional cases detected were mainly small cancers, the vast majority of which would likely remain indolent, keeping mortality low.

Cautionary Lessons

First, epidemiologic measures of mortality are tricky. Jeon et al. use the term “disease-specific mortality” to label the deaths, but mortality is usually defined as a rate per population at risk (e.g., per 100,000 in the population at large). The authors gave a percentage, and while they indicate in the discussion that they used National Statistics data, the methods are not clear to us. We wonder if the authors are reporting a death-to-case ratio—the number of cancer deaths due to disease divided by the number of people who had the disease. In this case, the improvements they observed would be expected based on the increasing number of patients with low-risk disease being included in the denominator in later years. If they were using the population of the institutions' catchment areas as the denominator, it is unclear how this was determined, and how age and population adjustment over time were managed.

Second, survival rates are not a reliable way to measure success in the war on cancer when the main approach to the battle is screening. The improving disease-specific survival rates observed by Jeon et al. were primarily due to lead time bias. Since survival is measured from the time of diagnosis, patients who are given a diagnosis earlier will have an improved survival (5). As the probability that a disease will be detected by screening is proportional to the length of its detectable preclinical phase (5), thyroid cancers detected with screening are more likely to be small, indolent tumors instead of rapidly growing, aggressive tumors. Thus, length-time bias explains both the greater proportion of low-risk disease and the improved disease-free survival seen in the cohort. This phenomenon has been observed in United States thyroid cancer cohorts as well (6). Challenging the concept that lead and length-time biases explain the entire phenomenon demonstrated by Jeon et al. is the fact that improved outcomes were also seen in patients with stage III and IV disease. It is possible, as proposed by the authors, that earlier treatment led to improved outcomes in these higher risk patients. However, since patients with stage III and IV disease are a heterogeneous cohort, it is also possible that detection of less aggressive disease still explains the improved outcomes in this cohort (7). In addition, although this study had a respectable length of follow-up, the follow-up was shorter in the later years, thus potentially contributing to the improved outcomes.

Last, beware of the unintended effects of major health policy initiatives. As thyroid cancer care improved in South Korea between 1996 and the early 2000s as a result of major health policy advancements, the population's health likely improved. But as elective, nonendorsed thyroid cancer screening took hold in the last years of study of this cohort, the improvements in care were accompanied by huge amounts of overdiagnosis, leading to spurious improvements in survival and no net benefit for patients. The data in the study by Jeon et al. are an excellent demonstration of one of the enduring challenges of modern cancer care at the population level—how to balance appropriate rates of cancer detection against the risk of finding disease not destined to cause harm, a history lesson to which we should all be paying attention.