Integrated Clinical and Genomic Models to Predict Optimal Cytoreduction in High-Grade Serous Ovarian Cancer

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Advanced high-grade serous ovarian cancer (HGSC) treatment involves surgery and chemotherapy, with surgical decisions based on clinical assessment.
• Optimal cytoreductive surgery is crucial for improved patient survival, but current clinical prediction accuracy is limited to approximately 70% for optimal outcomes.

Purpose of the Study:

• To develop and validate predictive models using comprehensive genomic data and machine learning to improve the prediction of optimal and complete cytoreductive surgery in advanced HGSC.
• To assess if integrating biological tumor data can enhance the accuracy of predicting surgical outcomes compared to clinical assessment alone.

Main Methods:

• Genomic data (gene expression, SNVs, CNVs, methylation, etc.) were collected from 83 HGSC patients.
• Statistical learning (lasso regression) and machine learning (TensorFlow) were used to build predictive models integrating genomic and clinical data.
• Models were validated using The Cancer Genome Atlas (TCGA) HGSC database.

Main Results:

• Of 124 models for optimal cytoreduction, 21 met or exceeded the historical clinical prediction rate.
• Of 89 models for complete cytoreduction, 37 demonstrated potential to surpass current clinical decision-making accuracy.
• The study identified specific genomic features and integrated models that show promise in predicting surgical outcomes.

Conclusions:

• Integrating multi-omic tumor data with clinical information can significantly improve the prediction of optimal and complete cytoreductive surgery in advanced HGSC.
• These validated predictive models offer a potential tool for objective decision-making, aiming to enhance patient outcomes in ovarian cancer treatment.
• Prospective validation is recommended to confirm the clinical utility of these novel predictive models.