Letter to the Editor: Risk of cancers in women with polycystic ovary syndrome: Cohort study based on health insurance database in South Korea

Yuk et al. have provided valuable nationwide data exploring the association between polycystic ovary syndrome (PCOS) and subsequent cancer risk using Korea's comprehensive insurance database. The large sample size, inclusion of multiple cancer types, and use of propensity-score matching strengthen the statistical precision of their findings.1 Their work contributes meaningfully to the growing literature on long-term outcomes in women with PCOS. Nevertheless, key methodological considerations merit attention and clarification, as they may substantially affect interpretation. In the study, PCOS was defined solely by the ICD-10 code E28.2 recorded on three or more visits, with cohort entry at the first such visit, without clinical, biochemical, or ultrasonographic confirmation against the Rotterdam 2003 consensus criteria. Within the NHIS and HIRA databases, diagnostic codes are entered primarily for reimbursement purposes. Data quality management (DQM) is not required, and codes are not independently validated for research or clinical accuracy.2 Although requiring at least three separate codes for PCOS likely reduces single-entry or provisional miscoding and improves specificity, it does not ensure diagnostic accuracy. Similarly, cancers were identified only by repeated diagnostic codes rather than by registry linkage or pathology confirmation. These choices risk both exposure and outcome misclassification, potentially biasing the estimates toward or away from the null. Moreover, the requirement of at least three PCOS-coded visits likely introduces selection bias, as it preferentially captures women who are more symptomatic or healthcare-engaged. This approach limits generalizability to milder or undiagnosed PCOS and may inflate effect sizes for hormone-sensitive cancers. Additionally, several important confounders are not captured in administrative claims and were therefore not accounted for in the analysis. Although the authors adjusted for demographic and clinical variables, certain critical cancer-related factors were not captured. Considering the high variety of cancers included, it may not be feasible to adjust for every site-specific risk factor; however, well-established determinants, including smoking, alcohol consumption, chronic infections such as human papillomavirus (HPV), and family history of malignancy, play a major role in cancer development. These factors are known to be associated with increased cancer risk and may confound or mediate the reported associations with PCOS.3, 4 Surveillance bias also deserves consideration. Women with PCOS tend to have more frequent medical evaluations for menstrual, metabolic, and fertility issues, which increases their opportunity for incidental or earlier cancer detection compared with women without PCOS.5 Consequently, part of the observed excess risk may reflect greater diagnostic intensity rather than biological susceptibility. Compared with smaller, single-center, or self-reported studies, the nationwide cohort by Yuk et al. offers substantial statistical power and broader representativeness, enabling more precise, site-specific estimates across diverse subgroups. At the same time, the retrospective design and claims-based ascertainment limit causal inference, invite exposure and outcome misclassification, and risk selection toward more symptomatic or healthcare-engaged patients. These limitations merit attention and clarification, as addressing them through prospective confirmation of PCOS and stronger cancer case verification would yield more accurate and generalizable risk estimates, provide a stronger basis for clinical implementation of risk-based care, and better inform prevention strategies. The author used ChatGPT, an AI language model, for editorial assistance to improve clarity and style. All content was independently reviewed, verified for factual accuracy and originality, and the author takes full responsibility for the final manuscript. The author declares no conflicts of interest. No new data were generated or analyzed in this letter.