Using artificial intelligencepowered evidence-based molecular decision-making for improved outcomes in ovarian cancer

This white paper details a study by Predictive Oncology and UPMC Magee-Womens Hospital. By combining the assets and domain expertise of both organizations, this collaboration aimed to improve ovarian cancer prognosis and treatment through accurate patient stratification and survival outcome predictions.

Background and objective
High grade serous carcinoma (HGSC) of the ovary is a complex and deadly disease. Although frontline treatment of chemotherapy and surgery is often effective, with nearly all patients experiencing remission, over 80% of patients will experience disease recurrence in the first two years and ultimately die of the disease.

Only 20% of patients with ovarian HGSC will be long-term survivors, but no current clinical data can accurately predict which patient will be a short- or long-term survivor. There is a pressing need for alternative precision medicine solutions to better characterize patients for individualized treatment decisions. If clinicians could better define the prognosis for these patients, their clinical management, monitoring, and systemic treatment decisions could be optimized to improve survival outcomes.

To address this unmet clinical need, Predictive Oncology (POAI) completed a retrospective study in collaboration with UPMC Magee-Womens Hospital to build a data-driven artificial intelligence solution leveraging machine learning (ML) to predict survival outcomes for ovarian HGSC patients. Patient tumor samples were characterized by a diverse set High grade serous carcinoma (HGSC) of the ovary is a complex and deadly disease. Although frontline treatment of chemotherapy and surgery is often effective, with nearly all patients experiencing remission, over 80% of patients will experience disease recurrence in the first two years and ultimately die of the disease. of multiple “omics” data sets with the motivation to identify the best combination of features for predicting survival outcome. Artificial intelligence solutions using predictive modeling are a compelling avenue for addressing this problem. Predictive models use algorithms to search for patterns in data and identify relationships between tumor sample features and survival outcome. These models are particularly useful for their ability to extrapolate information from large and complex datasets and show much promise for their potential use in personalized healthcare solutions.