Development and Validation of a Novel Prediction Model for Hearing Loss From Cisplatin Chemotherapy

Abstract

Purpose: Cisplatin treats many common tumors but causes permanent and debilitating hearing loss (HL). The objective of this study was to develop and externally validate a predictive model of HL in cisplatin-treated children and adolescent cancer survivors.

Methods: The Pediatric Holistic Evaluation of Auditory Risk (PedsHEAR) model architecture used several machine learning approaches followed by an ensemble predictor. The primary end point was post-treatment communication-affecting HL (International Society of Pediatric Oncology Ototoxicity Scale [SIOP] Grade ≥2). PedsHEAR was developed from a multicenter data set of cisplatin-exposed patients up to 21 years old (1984-2017) and externally validated using data from the Children's Oncology Group ACCL05C1 study (2007-2012) and two combined institutional cohorts (1988-2022). The model predicts post-treatment HL in each patient (probability [%], 95% CI) and classifies patients as low, intermediate, or high risk for HL (probability HL <0.33, 0.33-0.60, >0.60, respectively).

Results: In the training data set (n = 1,115, median age 6.3 years, SIOP Grade ≥2 HL 44%), PedsHEAR demonstrated excellent discrimination (AUC, 0.93 [95% CI, 0.92 to 0.95]) and then successfully validated within the internal (testing; AUC, 0.79 [95% CI, 0.74 to 0.85]) and two external validation cohorts (AUC, 0.74 and AUC, 0.67). In an aggregate validation cohort (n = 631), the model predicted the probability of HL (AUC, 0.76 [95% CI, 0.72 to 0.79]) and classified 22% (141/631), 71% (447/631), and 7% (43/631) of patients as low, intermediate, or high risk for HL.

Conclusion: PedsHEAR predicted SIOP Grade ≥2 HL in pediatric cisplatin-treated patients. This is the first validated model to successfully predict cisplatin-induced HL in a broadly representative population treated with diverse regimens across a range of treatment settings.