AI Machine Learning Models Accurately Predict Ovarian Cancer Recurrence and Platinum Sensitivity
Artificial intelligence (AI) machine learning models have shown promising results in predicting the outcomes of epithelial ovarian cancer according to a recent study titled “AI machine learning survival models for the prediction of epithelial ovarian cancer outcomes.” The study was presented at the 2024 Society of Gynecologic Oncology (SGO) Annual Meeting on Women’s Cancer.
The study aimed to develop and assess machine learning models for predicting time to recurrence (TTR) probabilities and platinum sensitivity using patient attributes and treatment outcomes available at the conclusion of front-line chemotherapy.
A cohort of 269 patients diagnosed with epithelial ovarian cancer between January 2017 and December 2021 was analyzed. Various machine learning survival models, including penalized Cox elastic-net (PenCox), random survival forest (RSF), gradient boosting survival analysis (GBSA), FastCPH, and DeepSurv, were trained on patient data. FastCPH and DeepSurv, both deep learning neural network models, emerged as top performers in most metrics.
The models were trained to predict TTR survival curves and recurrence hazard scores using input features such as cancer stage, histology, BRCA status, debulking outcome, chemotherapy type, CA-125 measurements, and maintenance therapy.
Evaluation metrics were performed using the concordance index (C-index), Uno index, Cumulative-Dynamic AUROC (CD-AUC), and Brier Score. The study found that FastCPH exhibited the highest sensitivity (90%) in predicting recurrence within six months post-chemotherapy, with a specificity of 0.89 and percent positive value of 0.62.
The study noted that all models performed best at predicting recurrence risk between 5 and 21 months post-chemotherapy, after which performance declined due to limited recurrence events beyond the 21-month mark.
The ability of FastCPH to accurately predict early recurrence at the conclusion of front-line treatment holds significant prognostic value, enabling clinicians to tailor treatment plans accordingly.
In conclusion, the findings suggest that AI machine learning models, particularly deep learning approaches like FastCPH, hold promise in accurately predicting recurrence risk and platinum sensitivity in epithelial ovarian cancer patients across all stages. Such advancements provide prognostic insights and the potential for optimizing treatment strategies to improve patient outcomes.
Source:
Nakayama J, McGaughey M, Pindzola GM, Summerscales T. AI machine learning survival models for the prediction of epithelial ovarian cancer outcomes. Presented at the 2024 Society of Gynecologic Oncology (SGO) Annual Meeting on Women’s Cancer; March 16-March 18, 2024. San Diego, California.