The VA HEARS Model for Predicting Audiogram Changes in Adults Undergoing Cisplatin Chemoradiotherapy

Develop and validate a model for predicting complete audiograms of individual patients that also supports cisplatin-ototoxicity forecasting and early identification using any existing diagnostic and classification criteria.

Prospective observational study.

Veterans Administration hospital cancer ambulatory care.

One hundred thirty-one Veterans undergoing cisplatin chemotherapy or chemoradiation.

Predicted 0.5 to 8 kHz audiogram thresholds in both ears. Hearing changes were derived from the predicted audiograms using American Speech-Language Hearing Association (ASHA) and Common Terminology Criteria for Adverse Events (CTCAE) criteria.

The VA Hearing Loss Evaluation and Risk Screener (VA HEARS) is a Bayesian multivariate linear mixed model trained on cisplatin patient audiograms collected from 2010-2022 and validated with data from 2019-2024. Audiogram thresholds are forecast conditional on the pretreatment audiogram, patient age, and prescribed cisplatin chemoradiation.Each patient had 1 to 8 audiograms measured before, during, and after cancer treatment. Forty-seven percent of patients received concurrent cochlear radiation.Forecasted thresholds achieved average prediction errors of 5.2 to 10.6 dB across frequencies. Ears were accurately diagnosed as having changed hearing with an area under the receiver operating characteristic curves of 0.79 and 0.78 for ASHA and CTCAE grade 1 predictions in validation data. After 280 mg/m² cumulative cisplatin, 8 kHz thresholds increase by an average of 12 dB with concurrent radiation versus 8 dB without.

The VA HEARS is the first validated model for forecasting cisplatin-related audiogram changes in adult cancer patients and to output multiple ototoxicity classification criteria simultaneously. This model is intended to personalize pretreatment counseling and guide timely interventions for patients receiving ototoxic medications.