External validation of prediction models for early relapse in advanced epithelial ovarian cancer using Australian and Dutch population-based data

Objective

To externally validate the published postoperative and BRCA models predictive of early relapse in patients with advanced-stage epithelial ovarian cancer (EOC) using independent Australian and Dutch cohorts.

Methods

Advanced-stage EOC patients diagnosed between January 1, 2002, and June 1, 2006, in Australia, and between January 1, 2016, and December 31, 2017, in the Netherlands were included. Data from patients who underwent cytoreductive surgery and platinum-based chemotherapy were used to validate both models. Missing data were addressed through multiple imputation. Model updates included recalibration-in-the-large, recalibration, and model revision, with a closed testing procedure to identify the most suitable approach. Model performance was assessed for calibration, discrimination, and the Brier score.

Results

The Australian cohort (N = 1334) included 457 early relapsers and 859 late or non-relapsers, showing baseline differences compared to the development cohort. Discrimination was adequate for both the postoperative and BRCA models (c-statistics: 0.69 and 0.70, respectively). The postoperative model required full revision, while recalibration-in-the-large was sufficient for the BRCA model in the Australian cohort. The Dutch cohort (N = 1212) included 283 early relapsers and 929 late or non-relapsers, with baseline characteristics similar to those of the development cohort. Both models demonstrated adequate discrimination (c-statistics: 0.71 and 0.70, respectively). Recalibration-in-the-large corrected miscalibration in the Dutch cohort.

Conclusion

The postoperative and BRCA model were successfully validated for predicting early relapse in advanced-stage EOC patients, confirming their robustness. However, local data updates are advised to enhance accuracy across clinical settings. Online calculators were built for clinical use (Link 1; Link 2).