Identifying Pediatric Patients at Risk of Severe Hearing Impairment After Treatment for Malignancies of the H&N/CNS with Proton Therapy

Objectives

Recent work by the Pediatric Normal Tissue Effects in the Clinic (PENTEC) hearing loss task force provides guidance on benchmark dose constraints to reduce hearing loss but does not generate individualised risk estimations for patients. Uniquely, a risk calculation model was presented in 2021 by Keilty et al. for determining the likelihood of severe hearing impairment (HI) for paediatric patients treated with photon radiation therapy. This study aimed to validate their risk-prediction model for our paediatric patient cohort treated with proton therapy (PT) for malignancies of the head and neck (H&N) or central nervous system (CNS).

Methods

This was a single-institution study which extracted data on all patients aged ≤ 18 years treated with PT between Feb 2010 – Feb 2022 for malignancies of the H&N/CNS, either via the UK Proton Overseas Programme or at The Christie PT centre. The factors required for input into the Keilty model were extracted: age at PT, time since end of PT, mean cochlea dose, and platinum chemotherapy doses. Validation was performed using the statistical software R v 4.3.1, which analysed event discrimination and model calibration.

Results

587 patients met inclusion criteria. Validation of the model demonstrated excellent discriminative ability, with an “optimal” cut-off value of 16% at a specificity and sensitivity of 82%. However, model calibration was less satisfactory, indicating an overestimation of risk of severe HI by the model as compared to clinically observed events in our cohort, possibly linked to differences in event scoring between the model developers and this study, and short follow-up time in this study.

Conclusion

The published (photon-based) model of Keilty et al. was validated in a PT context, demonstrating a high discriminative ability to determine patients at high risk versus low risk for severe HI. However the overall observed risk was lower than model predictions.