Response: Risk of cancers in women with polycystic ovary syndrome: Cohort study based on health insurance database in South Korea

We sincerely thank Dr. Mohammad Azizi for his thoughtful and constructive comments on our recent article.1 His letter highlights important methodological considerations that are inherent to large-scale administrative database research, and we appreciate the opportunity to clarify these points and further discuss the strengths and limitations of our study. Dr. Azizi correctly notes that polycystic ovary syndrome (PCOS) was identified using the ICD-10 code E28.2 recorded on three or more separate visits, without independent clinical, biochemical, or ultrasonographic confirmation of the Rotterdam 2003 criteria. Throughout the entire study period (2011–2020), the clinical guidelines of the Korean Society of Gynecologic Endocrinology explicitly recommended only the Rotterdam 2003 criteria for PCOS diagnosis. We deliberately required at least three separate coded visits to substantially increase specificity and reduce provisional or single-entry miscoding. While we fully acknowledge that this approach cannot guarantee perfect diagnostic accuracy and may exclude milder or undiagnosed cases, it represents the most rigorous method currently feasible in nationwide claims data to minimize false positives. We transparently addressed this limitation in our discussion section and agree that it may preferentially capture more symptomatic patients, which could potentially influence effect estimates for hormone-sensitive cancers. Regarding cancer ascertainment, we employed the same validated operational definition (≥3 healthcare visits with a cancer diagnostic code) that has been widely adopted and validated in Korean NHIS/HIRA-based oncology studies. Although linkage to the national cancer registry or pathology confirmation would be ideal, such linkage was not available for this customized dataset. However, the repeated-code requirement has demonstrated high positive predictive value in previous Korean validation studies for most cancer sites, providing reasonable confidence in our outcome classification. We share Dr. Azizi's concern about residual confounding. We adjusted for a comprehensive set of covariates using both propensity-score matching and stratified Cox regression: age, socioeconomic status, region, Charlson comorbidity index, parity, hypertension, diabetes, dyslipidemia, menopause, prior adnexal surgery, uterine fibroids, and endometriosis. However, important cancer risk factors including smoking, alcohol consumption, HPV infection status, and family history of malignancy are not recorded in the HIRA claims database and could not be included. We fully acknowledge these unmeasured confounders as important limitations that may influence the observed associations. Notably, our sensitivity analysis restricted to medium-to-high socioeconomic status groups yielded consistent results, suggesting some robustness of our findings across different population strata. Regarding surveillance bias, we believe this concern is considerably mitigated in our study design compared to many previous PCOS cohorts. Critically, the non-PCOS control group was not a general population sample but comprised women who had at least one health checkup or medical visit during the same period, ensuring comparable healthcare access and engagement in both groups. The propensity-score matching further balanced overall comorbidity burden and healthcare utilization patterns between groups, making substantial differential detection intensity unlikely. While we cannot entirely exclude surveillance bias, the magnitude of risk elevation observed for endometrial cancer (hazard ratio [HR]: 4.96) substantially exceeds what could plausibly be explained by increased detection alone, particularly given the symptomatic nature of this malignancy. We completely concur with Dr. Azizi that prospective studies with clinically confirmed PCOS diagnoses, extended follow-up duration, registry-confirmed cancer outcomes, and more complete risk-factor adjustment are needed to definitively delineate cancer risk patterns in women with PCOS. Nevertheless, nationwide administrative data, despite their inherent limitations, currently provide the largest and most representative cohorts available in Korea. Our findings of markedly increased risks of endometrial and ovarian cancer in PCOS are biologically plausible, consistent with established mechanistic pathways involving chronic anovulation and hormonal dysregulation, and have been replicated across multiple international cohorts. These convergent lines of evidence support the clinical relevance of our findings for risk counseling and surveillance strategies in women with PCOS. We are grateful to Dr. Azizi for his careful reading and valuable critique, which strengthens the interpretation of our results and highlights important directions for future research. Jin-Sung Yuk: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Roles/Writing – original draft, and Writing – review & editing. Ji Hyun Noh: Project administration, Resources, Software, Supervision, Validation, Visualization, Roles/Writing – original draft, and Writing – review & editing. Gwan Hee Han: Project administration, Resources, Software, Supervision, Validation, Visualization, Roles/Writing – original draft, and Writing – review & editing. Sang Hee Yoon: Project administration, Resources, Software, Supervision, Validation, Visualization, Roles/Writing – original draft, and Writing – review & editing. Myounghwan Kim: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Roles/Writing – original draft, and Writing – review & editing. We utilized ChatGPT for the purposes of English writing and proofreading. The authors declare no conflicts of interest.