Predicting benefit from PARP inhibitors using deep learning on H&E-stained ovarian cancer slides

• Published in: European journal of cancer (1990) 2024-12, Vol.216, p.115199, Article 115199
• Main Authors: Marmé, Frederik, Krieghoff-Henning, Eva I., Kiehl, Lennard, Wies, Christoph, Hauke, Jan, Hahnen, Eric, Harter, Philipp, Schouten, Philip C., Brodkorb, Tobias, Kayali, Mohamad, Heitz, Florian, Zamagni, Claudio, González-Martin, Antonio, Treilleux, Isabelle, Kommoss, Stefan, Prieske, Katharina, Gaiser, Timo, Fröhling, Stefan, Ray-Coquard, Isabelle, Pujade-Lauraine, Eric, Brinker, Titus J.
• Format: Article
• Language: English
• Subjects: digital pathology; homologous recombination deficiency (HRD); ovarian cancer; PARP inhibitor
• ISSN: 0959-8049; 1879-0852; 1879-0852
• DOI: 10.1016/j.ejca.2024.115199

Ovarian cancer patients with a Homologous Recombination Deficiency (HRD) often benefit from polyadenosine diphosphate–ribose polymerase (PARP) inhibitor maintenance therapy after response to platinum-based chemotherapy. HR status is currently analyzed via complex molecular tests. Predicting benefit from PARP inhibitors directly on histological whole slide images (WSIs) could be a fast and cheap alternative. We trained a Deep Learning (DL) model on H&E stained WSIs with “shrunken centroid” (SC) based HRD ground truth using the AGO-TR1 cohort (n=208: 108 training, 100 test) and tested its ability to predict HRD as evaluated by the Myriad classifier and the benefit from olaparib in the PAOLA-1 cohort (n=447) in a blinded manner. In contrast to the HRD prediction AUROC of 72% on hold-out, our model only yielded an AUROC of 57% external. Kaplan-Meier analysis showed that progression free survival (PFS) in the PARP inhibitor treated PAOLA-1 patients was significantly improved in the HRD positive group as defined by our model, but not in the HRD negative group. PFS improvement in PARP inhibitor-treated patients was substantially longer in our HRD positive group, hinting at a biologically meaningful prediction of benefit from PARP inhibitors. Together, our results indicate that it might be possible to generate a predictor of benefit from PARP inhibitors based on the DL-mediated analysis of WSIs. However, further studies with larger cohorts and further methodological improvements will be necessary to generate a predictor with clinically useful accuracy across independent patient cohorts. •Developed a deep learning model to predict HRD status from H&E-stained ovarian cancer slides.•Achieved 72% AUROC for HRD prediction on internal validation and 57% on external datasets.•Kaplan-Meier analysis indicated PARP inhibitor benefit linked to HRD positive model predictions.•Demonstrated potential for AI-based predictors to support personalized ovarian cancer treatment.•Highlighted the need for larger datasets and methodological improvements for clinical utility.