Preliminary development and internal evaluation of a plasma SOX1 methylation-based model bladder cancer detection

Abstract

The evaluation of individuals with suspected bladder cancer (BCa) continues to rely on cystoscopy, while urine cytology demonstrates limited sensitivity. This single-center retrospective two-gate case-control study included 90 individuals (30 with BCa and 60 non-cancer controls). SOX1 methylation signals in plasma-derived DNA were quantified using quantitative methylation-specific PCR. Logistic regression was used to generate a SOX1 -only model and a combined model incorporating SOX1 methylation, age, and sex. Model performance was described using apparent ROC-AUC, sensitivity, specificity and other operating characteristics at the Youden-derived cut-off, together with internal bootstrap, five-fold cross-validation, calibration, and exploratory decision-curve analysis (DCA). All resampling procedures used the same 90 participants; no independent validation cohort was available. PAX1 methylation was assessed as a prespecified exploratory single-marker comparator. In this selected sample, the combined model had a higher apparent AUC than the SOX1 -only model (0.882 vs 0.828). At the Youden-derived threshold, sensitivity was 83.3 % and specificity was 88.3 %. Exploratory DCA favored the combined model within the study sample, but this estimate is dependent on the fixed case-control sampling fraction. The PAX1- only AUC methylation was 0.845, numerically higher than the SOX1 -only AUC, although PAX1 was not included in multivariable modeling. Adding age and sex to the plasma SOX1 methylation signal improved apparent discrimination within this small selected cohort. This preliminary, hypothesis-generating model is not validated for clinical use and should not be assumed to generalize to consecutive hematuria populations. Independent prospective evaluation in larger, clinically representative cohorts is required.