Using casual inference and machine learning with exposure determinant modeling to identify important workplace...

Area of Science:

• Occupational Health and Safety
• Environmental Science
• Data Science and Machine Learning

Background:

• Exposure determinant modeling is crucial for industrial hygienists to control occupational exposures.
• Traditional methods are limited by selection bias and the "small n, large p" problem.
• Informal electronic waste (e-waste) recycling presents unique occupational exposure challenges.

Purpose of the Study:

• To explore the application of causal inference and machine learning in exposure determinant modeling.
• To identify key determinants of heavy metal concentrations among informal e-waste recycling workers.
• To overcome limitations of traditional modeling approaches in occupational health studies.

Main Methods:

• A case study involving 41 e-waste workers was analyzed using inverse probability weighting to address selection bias.
• Forty-four potential exposure determinants were quantified through video monitoring.
• Machine learning algorithms (LASSO, boosted regression trees, random forests) and traditional models were compared using leave-one-out cross-validation.

Main Results:

• The random forest model demonstrated the best performance (lowest LOOCV-RMSE).
• Preventing workers from bending their backs during e-waste dismantling was identified as the most significant determinant of heavy metal concentrations.
• This intervention was estimated to reduce blood lead (Pb) by 0.81 µg/dL, a finding missed by traditional regression models.

Conclusions:

• The causal inference framework combined with machine learning effectively models exposure determinants and overcomes common statistical limitations.
• This approach yields interpretable estimates for reducing biomarker concentrations via hypothetical workplace controls.
• Findings aid industrial hygienists in selecting and contextualizing the most effective hazard controls in specific work environments.