Multimodal data integration with machine learning for predicting PARP inhibitor efficacy and prognosis in ovarian cancer.

BACKGROUND

Poly(ADP)-ribose polymerase inhibitors (PARPi) have brought a significant breakthrough in the maintenance treatment of ovarian cancer. However, beyond BRCA mutation/HRD, the direct impact of other prognostic factors on PARPi response and prognosis remains inadequately characterized.

METHODS

We assessed PARPi prognostic factors from clinical characteristics, pathological findings, and biochemical indicators from 251 ovarian cancer patients. Cox univariate and multivariate analyses were employed to identify the factors which influencing PARPi efficacy and patients prognosis. Feature screening was conducted using correlation analysis, significance analysis, Variance Inflation Factor (VIF), and Elastic Net stability analysis. Patient-specific efficacy and prognosis prediction models were then constructed using various machine learning algorithms.

RESULTS

Total bile acids (TBAs) and CA-199 present as an independent risk factor in Cox multivariate analysis for primary and recurrent ovarian cancer patients respectively (P < 0.05). TBAs emerged as a risk factor, with each unit increase associated with a 10% rise in recurrence risk. The best-performing model has an AUC of 0.79 ± 0.09 and an AUC of 0.72 ± 0.03 for primary and recurrent ovarian cancer patients respectively. External validation(n=36) in multicenter cohorts maintained robust performance with AUC of 0.74 and an AUC of 0.70 for primary and recurrent ovarian cancer patients respectively.

CONCLUSIONS

We identified TBAs and CA-199 as a significant prognostic factor in primary and recurrent ovarian cancer patients respectively. The integration of multimodal data with machine learning holds significant potential for enhancing prognosis prediction in PARPi treatment for ovarian cancer.