Recurrence timing and exploratory prediction after surgery for mucinous-dominant borderline ovarian tumors: implications for postoperative surveillance in a single-center retrospective cohort

Abstract Background Postoperative surveillance for borderline ovarian tumors with mucinous and related institutional histology labels remains heterogeneous, particularly after organ-sparing surgery. Whether routinely collected clinical variables can identify patients who require intensified surveillance is uncertain. Methods We retrospectively analyzed 178 patients treated at a single center from 2010 to 2022. Index surgery was the first tumor-removing operation. First recurrence required a new ovarian lesion after postoperative imaging-defined no evidence of disease and subsequent pathological confirmation. One patient who never achieved a postoperative disease-free state was excluded. Twenty-seven patients experienced a first recurrence and seven later experienced a second recurrence. Age, FIGO stage IC or higher, and broad organ-sparing surgery were evaluated using ridge-penalized Cox regression and a random survival forest under repeated nested cross-validation. Results Twenty-seven of 178 patients (15.2%) experienced a pathology-confirmed first recurrence; seven experienced a second recurrence, yielding 34 recurrence episodes. Median time to first recurrence was 41.0 months (IQR 20.5–66.0). Ten first recurrences occurred more than 5 years after index surgery, including three at 10 years or later. The C-index was 0.719 (95% CI 0.612–0.807) for penalized Cox and 0.742 (95% CI 0.655–0.816) for the random survival forest; the paired difference was 0.023 (95% CI -0.102 to 0.136). At 36 months, AUC was 0.666 for Cox and 0.580 for the random survival forest, while average precision among 73 patients with known 36-month status was 0.373 and 0.207, respectively. Stable clinical net benefit was not demonstrated. Conclusions Late recurrence was observed, but routine baseline variables did not provide clinically actionable risk stratification and the random survival forest did not show a reliable advantage over penalized Cox. These findings identify uncertainty concerning the duration and intensity of surveillance after organ-sparing surgery but cannot establish an optimal follow-up schedule. The models remain exploratory and must not alter clinical care.