Evaluating targeted versus universal BRCA testing in Asian women with ovarian cancer

Background: Germline genetic testing for BRCA1 and BRCA2 pathogenic variants (PVs) is recommended for all ovarian cancer patients, as identifying PVs informs treatment planning and facilitates family cascade testing. However, in resource-limited settings, high testing costs often limit feasibility and uptake. An alternative approach is to use predictive models to prioritize patients with a higher likelihood of carrying PVs, optimizing resource allocation. Existing models are largely Western-derived and underperform in Asians; while we have a validated model for breast cancer, no equivalent model currently exists for ovarian cancer. Methods: Using data from a multi-center study of 1,126 Asian ovarian cancer patients (including 147 BRCA PV carriers), we developed predictive models incorporating routinely collected information such as cancer history and clinicopathological features to estimate likelihood of carrying BRCA PVs. We evaluated model performance in terms of discrimination, calibration, overall accuracy, sensitivity, and specificity, and compared the associated genetic testing volumes and costs to those of universal testing. Results: Our final model demonstrated good calibration and strong discriminatory power, with an area under the curve of 0.80 (95% confidence interval: 0.74-0.87). Factors included in the model were age at diagnosis, ethnicity, personal and family cancer history, and clinicopathological features. At the optimal threshold, the model achieved 77% accuracy (73% sensitivity and 73% specificity), compared with 13% accuracy for universal testing (100% sensitivity but 0% specificity). In practice, this translates to identifying one carrier for every three patients tested, versus one in eight under universal testing, reducing the genetic testing cost per carrier identified from USD 4,000 to USD 2,000. From a budget perspective, even when we need to detect every carrier, as in universal testing, the model reduces testing volume by 15%, yielding potential savings of USD 70,000 for every 800 patients screened annually. Conclusions: Targeted testing using a mutation prediction model offers a more efficient alternative when universal testing is not feasible, with adjustable risk thresholds that can be tailored to local resources to optimize impact in resource-limited settings.