Abstract
BACKGROUND: Poly (ADP-ribose) polymerase inhibitors (PARPis) are the standard of care for primary ovarian cancer (OC), yet adverse events (AEs) frequently necessitate dose modifications. The prognostic impact of these modifications remains unclear. This study aimed to evaluate the impact of AEs-associated dose modifications on progression-free survival (PFS) and identify predictive biomarkers. METHODS: This retrospective study included 117 patients with primary OC receiving PARPi maintenance. To mitigate immortal time bias, the association between dose modifications and PFS was evaluated using a landmark analysis at 24 and 48 weeks, complemented by a time-dependent Cox regression model that treated dose modification as a time-varying covariate. To find predictors for dose modification, prognostic nutritional index (PNI), hemoglobin (HGB), and platelet-to-lymphocyte ratio (PLR) were evaluated. A Cox model was developed and externally validated in an independent cohort (n = 45). RESULTS: Landmark analysis at 24 weeks demonstrated that patients in the dose-maintenance group had significantly superior PFS compared to those in the dose-modification group (P = 0.002). Crucially, in the time-dependent Cox analysis adjusted for confounders, the occurrence of a dose modification event remained a significant independent risk factor for disease progression (HR 2.10, 95% CI 1.13–3.91; P = 0.019). Specifically, permanent discontinuation was significantly associated with worse PFS (HR 2.80, P = 0.008), whereas dose reduction/interruption was not (HR 1.69, P = 0.169). Regarding the prediction of dose modifications, multivariate analysis identified low PNI, low PLR, low HGB as independent predictors. The prediction model demonstrated good discrimination in the training cohort (C-index 0.79) and moderate discrimination in the external validation cohort (C-index 0.65). CONCLUSIONS: This study confirms that AEs-associated dose modifications, specifically permanent discontinuation, are associated with compromised PARPi efficacy independent of baseline prognostic factors. PNI, PLR and HGB serve as robust predictors, supporting early risk stratification to optimize patient management. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s13048-026-02064-3.