A methodological framework for missing data in occupational NIHL surveillance and risk modelling
Abstract
Background: Noise-induced hearing loss remains a major occupational health issue. Large audiometric surveillance datasets offer opportunities to study risk patterns, but challenges such as missing data, outcome definitions and modeling approaches complicate analysis. This study develops a reproducible analytical framework using a national hearing surveillance database in the U.S.
Methods: We conducted a retrospective observational study analyzing 1,114,966 audiometric records from workers across multiple industrial sectors. Data preprocessing included recoding sentinel values, harmonizing demographic and occupational variables and defining exposure categories. The primary outcome was better-ear PTA4 hearing loss (>25 dB HL). Additional phenotypes included detection of a notch at 4 kHz, asymmetry between ears, and WHO-based severity categories. Missing data were addressed using multiple imputation, complete-case analysis and inverse probability weighting.
Results: PTA4-related frequencies showed near-complete data (99.97%), while thresholds at 8 kHz showed 28.4% structured missingness, likely due to testing protocols. Overall hearing loss prevalence was 5.5%, increasing from 0.6% among workers aged 18–25 to 23.3% among those aged 56–65. Age was the strongest predictor, while female sex was protective. Exposure categories had limited effect after age-sex adjustment, although sector differences were more pronounced for high-frequency outcomes. Different missing-data strategies produced consistent results for the primary outcome.
Conclusions: Methodological choices strongly influence findings in occupational hearing data. For PTA4-based outcomes, missing-data strategy has minimal impact. For outcomes relying on high frequencies, structured missingness requires transparent handling, favoring complete-case analysis. Surveillance should adopt richer phenotypes and sector-specific modeling to improve prevention strategies and inform occupational health policies globally.
Key messages: • This work shows that PTA4-based NIHL surveillance is highly robust to missing data, while 8 kHz-dependent outcomes require careful, transparent handling in large occupational datasets in real-world.
• It demonstrates that richer phenotyping and sectorsensitive models improve detection of occupational hearing damage, support better prevention targeting and more precise workplace health surveillance.
Related
- Relational Multiplicity in Group Synchrony: Properties, Cognitive Demands, and Methodological Framework
- Beyond abundance: an evidence-aware framework for environmental conditioning, surveillance and infection potential in Vibrio vulnificus
- A coupled geometric surface and surface reaction-diffusion framework for modelling nuclear mechanics during confined cell migration
- A spatiotemporal deep learning framework for detecting unusual human activity in surveillance videos
- A general mathematical framework for modelling subnetworks of the nuclear auxin pathway
BibTeX
@article{vasileiadis2026methodological,
title = {{A methodological framework for missing data in occupational NIHL surveillance and risk modelling}},
author = {G P Vasileiadis and E Jelastopulu},
journal = {The European Journal of Public Health},
year = {2026},
month = oct,
doi = {10.1093/eurpub/ckag169.915},
url = {https://doi.org/10.1093/eurpub/ckag169.915},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - A methodological framework for missing data in occupational NIHL surveillance and risk modelling
AU - G P Vasileiadis
AU - E Jelastopulu
PY - 2026
DA - 2026/10/01
JO - The European Journal of Public Health
DO - 10.1093/eurpub/ckag169.915
UR - https://doi.org/10.1093/eurpub/ckag169.915
AB - Background: Noise-induced hearing loss remains a major occupational health issue. Large audiometric surveillance datasets offer opportunities to study risk patterns, but challenges such as missing data, outcome definitions and modeling approaches complicate analysis. This study develops a reproducible analytical framework using a national hearing surveillance database in the U.S. Methods: We conducted a retrospective observational study analyzing 1,114,966 audiometric records from workers across multiple industrial sectors. Data preprocessing included recoding sentinel values, harmonizing demographic and occupational variables and defining exposure categories. The primary outcome was better-ear PTA4 hearing loss (>25 dB HL). Additional phenotypes included detection of a notch at 4 kHz, asymmetry between ears, and WHO-based severity categories. Missing data were addressed using multiple imputation, complete-case analysis and inverse probability weighting. Results: PTA4-related frequencies showed near-complete data (99.97%), while thresholds at 8 kHz showed 28.4% structured missingness, likely due to testing protocols. Overall hearing loss prevalence was 5.5%, increasing from 0.6% among workers aged 18–25 to 23.3% among those aged 56–65. Age was the strongest predictor, while female sex was protective. Exposure categories had limited effect after age-sex adjustment, although sector differences were more pronounced for high-frequency outcomes. Different missing-data strategies produced consistent results for the primary outcome. Conclusions: Methodological choices strongly influence findings in occupational hearing data. For PTA4-based outcomes, missing-data strategy has minimal impact. For outcomes relying on high frequencies, structured missingness requires transparent handling, favoring complete-case analysis. Surveillance should adopt richer phenotypes and sector-specific modeling to improve prevention strategies and inform occupational health policies globally. Key messages: • This work shows that PTA4-based NIHL surveillance is highly robust to missing data, while 8 kHz-dependent outcomes require careful, transparent handling in large occupational datasets in real-world. • It demonstrates that richer phenotyping and sectorsensitive models improve detection of occupational hearing damage, support better prevention targeting and more precise workplace health surveillance.
SN - 1101-1262
SN - 1464-360X
ER -