Establishing gender- and age-specific reference ranges for eight tumor markers in adults based on real-world data from Southern China

Objectives To establish and validate age- and sex-specific reference intervals (RIs) for eight common tumor markers using large-scale real-world data (RWD) and to compare the performance of five indirect statistical algorithms for RI derivation. Methods This retrospective study included 404,347 individuals undergoing health examinations at The University of Hong Kong-Shenzhen Hospital between October 2022 and December 2024. Distribution characteristics and age-related trends were evaluated using kernel density estimation and generalized additive models. After age and sex stratification, RIs for AFP, CA125, CA15-3, CA19-9, CEA, fPSA, tPSA, and the fPSA/tPSA ratio were established using the Tukey, Z-score, Gaussian mixture model (GMM), isolation forest (IF), and local outlier factor (LOF) methods. Performance was assessed by bias ratios and validated using an independent cohort. Results Most tumor markers exhibited markedly right-skewed distributions, whereas the fPSA/tPSA ratio showed only moderate skewness. AFP and CEA showed negligible sex differences. Age-specific stratification was identified for male AFP, female CA125 and CA15-3, CA19-9, CEA, and prostate-related markers. Tukey and Z-score methods showed the greatest agreement and achieved validation pass rates of ≥90 % across all demographic and analyte subgroups. In contrast, GMM and IF failed validation in 11 subgroups, and LOF failed in 6 subgroups, mainly because of the over-exclusion of high-end values. Conclusions Large-scale RWD enabled robust age- and sex-specific RIs for eight tumor markers. For highly right-skewed tumor marker data, the Tukey and Z-score methods outperformed GMM, IF, and LOF, providing reliable and practical approaches for clinical RI establishment.