Innovative Machine Learning Model Predicts Risk of Hearing Loss in Children Treated with Cisplatin

A groundbreaking machine learning-based tool, developed by an international research team led by Children's Hospital Los Angeles, now enables clinicians to predict a child's individual risk of cisplatin-induced hearing loss with remarkable accuracy. Cisplatin, a potent chemotherapy agent used since the 1970s to treat various solid tumors, remains a cornerstone in pediatric oncology for conditions such as brain and spinal cord tumors, neuroblastoma, and rhabdomyosarcoma. However, its significant side effect—permanent hearing loss—affects up to 80% of treated children, impairing their social interactions, educational performance, and long-term career prospects.

The new model, named PedsHEAR, leverages routine clinical data and advanced machine learning techniques, including ensemble predictors, to assess the individual risk of hearing loss prior to treatment initiation. Published results in the Journal of Clinical Oncology demonstrate that PedsHEAR can provide risk estimates with 95% confidence, offering crucial insights for personalized treatment planning.

This development builds on over two decades of research aiming to mitigate cisplatin’s ototoxic effects. The effort included pivotal clinical trials and the FDA approval of sodium thiosulfate in 2022, a chemoprotective agent. Despite such advances, understanding each patient’s specific risk remains critical, especially since not all children need adjunct therapies like sodium thiosulfate.

The study utilized data from more than 1,400 patients across the US and Canada, analyzing various risk factors to train the machine learning model. Validation involved additional real-world data sources, confirming the model’s robustness across diverse populations. The approach incorporates multiple strategies, including logistic regression, elastic net, and random forests, combined into an ensemble that enhances predictive accuracy.

Dr. Etan Orgel emphasizes that this tool is designed for routine clinical use, relying solely on data commonly collected at diagnosis, making it accessible for immediate application. The goal is to empower clinicians and families to make informed decisions, optimize protective strategies, and implement targeted hearing monitoring during treatment.

Future plans include expanding the model’s applicability to young adults and adults up to 65 years, incorporating genomic data, and extending predictions to other chemotherapy-related side effects. Ultimately, this innovation represents a significant shift towards personalized, data-driven pediatric cancer care, promoting better quality of life outcomes for young patients.

Source: https://medicalxpress.com/news/2025-05-machine-tool-child-personal-cisplatin.html