Applying Machine Learning Models to Predict Occupational Noise-Induced Hearing Loss: A Systematic Review and...

Area of Science:

• Occupational health
• Biomedical informatics
• Machine learning applications

Background:

• Occupational noise-induced hearing loss (ONIHL) is a significant public health concern.
• Machine learning (ML) offers potential for developing predictive models for ONIHL.
• Critical appraisal of existing ML models for ONIHL prediction is needed.

Purpose of the Study:

• To systematically map and critically appraise ML-based prediction models for ONIHL.
• To evaluate the reporting quality and risk of bias in these models.
• To assess the performance of ML models in predicting ONIHL.

Main Methods:

• Systematic literature search across seven databases.
• Reporting quality assessed using TRIPOD-AI checklist.
• Risk of bias and applicability evaluated using PROBAST tool.
• Meta-analysis of ML model performance (AUC) with subgroup analyses.

Main Results:

• Sixteen studies included, with Support Vector Machine (SVM), Random Forest, and Artificial Neural Network (ANN) as common algorithms.
• Input variables often included age, noise level, and exposure duration, with limited consideration of noise type.
• Meta-analysis showed a pooled AUC of 0.788 for ML models; XGBoost and ANN achieved the highest AUC (0.846).
• Significant risk of bias (56%) and applicability concerns (31%) were identified.

Conclusions:

• ML models show potential for ONIHL prediction.
• High risk of bias, heterogeneity, and lack of external validation necessitate cautious interpretation of current findings.
• Further well-designed studies are required to improve the reliability and applicability of ML models for ONIHL prediction.