Skip to main content
AAO-HNSF 2025 Annual Meeting & OTO EXPO Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Scientific Oral Presentations

Otology/Neurotology

Session: Scientific Oral Presentations 5 - Otology/Neurotology

Analyzing Cardiovascular Risk Factors Using Machine Learning and Their Associations with Hearing Loss

Monday, October 13, 2025
8:36 AM - 8:42 AM EDT
Location: 244
Introduction: This study explores machine learning's ability to predict hearing loss using cardiovascular risk factors. Analysis of NHANES 2012–2018 data reveals links between cardiovascular conditions and hearing, offering insights for personalized prevention and early intervention.

Methods: This retrospective study analyzed data from the NHANES (2012–2018) to explore the link between cardiovascular risk factors and hearing outcomes. The study included 7,996 participants aged ≥20 years, excluding those with self-reported hearing loss or incomplete data. Hearing loss was assessed via pure-tone audiometry and categorized into clinical thresholds. Independent variables included hypertension, diabetes, smoking, obesity, demographics, and behaviors. Using LightGBM and neural networks, models predicted hearing thresholds and pure-tone averages. Data preprocessing involved imputation, feature selection, and stratified sampling. Performance metrics such as accuracy, F1-score, and mean absolute error showed robust predictive capabilities.

Results: The study highlights the effectiveness of machine learning in predicting hearing outcomes using cardiovascular risk factors. The LightGBM model achieved 80.1% accuracy (AUC-ROC: 0.820, F1-score: 0.800) for mild hearing impairment (>25 dB HL) and strong performance for >16 dB HL (78.8% accuracy) and >40 dB HL (86.1% accuracy). A multi-layer neural network predicted pure-tone averages with a 3.05 dB error (95% CI: 3.04–3.06). Age, gender, blood pressure, and waist circumference were key predictors, enabling clinical integration for early intervention.

Conclusions: The study concludes that LightGBM and neural networks accurately predict hearing outcomes using cardiovascular risk factors, supporting their integration into clinical workflows for early detection and personalized hearing loss management. However, limitations like the cross-sectional design and need for external validation emphasize the importance of future studies. These models complement, not replace, traditional evaluations.

    Presenting Author(s)

  • AN

    Ali Nabavi, MD

    Shiraz University of Medical Sciences, Iran

    Disclosure information not submitted.

  • FS

    Farimah Safari, MD

    Shiraz University of Medical Sciences
    Shiraz, Iran

    Disclosure information not submitted.

  • MK

    Mohammad Kashkooli, MD

    Shiraz University of Medical Sciences, Iran

    Disclosure information not submitted.

  • LA

    Leo Anthony Celi, MD

    Massachusetts Institute of Technology, United States

    Disclosure information not submitted.

    • Senior Author(s)

  • AN

    Ali Nabavi, MD

    Shiraz University of Medical Sciences, Iran

    Disclosure information not submitted.

    • Co-Author(s)

  • AF

    Ali Faramarzi, MD

    Shiraz University of Medical Sciences, Iran

    Disclosure information not submitted.

  • MA

    Meskerem Aleka Kebede, MD

    Global Surgery Policy Unit, LSE Health, Department of Health Policy, United Kingdom

    Disclosure information not submitted.

  • TA

    Tesfamariam Aklilu, MD

    Addis Ababa University, Ethiopia

    Disclosure information not submitted.

  • LA

    Leo Anthony Celi, MD

    Massachusetts Institute of Technology, United States

    Disclosure information not submitted.