Bridging the Gap between External and Internal Exposure: Challenges and a Machine Learning Approach Validated Using Benzene Ring-Containing Pollutants

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: External validation Qext² = 0.784, indicating that approximately 78.4% of variance in internal exposure (blood concentrations) is explained by the model in the external validation dataset

• Primary Outcome: Predictive performance of the ML-EIExpLink framework (gradient boosting decision trees) for estimating blood concentrations of benzene ring-containing pollutants from external exposure variables

• Nnt Or Sensitivity: Not applicable (methodological framework paper); key predictive variables identified include external air concentration, molecular planarity, and temperature, all exhibiting nonlinear and threshold effects on predicted blood concentrations

• Confidence Interval: Not reported — a significant methodological omission

Clinical Application

The ML-EIExpLink framework is currently at proof-of-concept stage and is not ready for direct clinical or regulatory application. Implementation would require: (1) prospective validation in independent, diverse population cohorts; (2) integration with existing biomonitoring infrastructure; (3) uncertainty quantification for individual-level predictions; and (4) regulatory acceptance of ML-based exposure modelling. The computational requirements are manageable with standard research infrastructure. In Australia, benzene exposure is regulated by Safe Work Australia (workplace exposure standard: 1 ppm TWA) and monitored through the Australian Institute of Health and Welfare environmental health frameworks. The RACGP does not currently have specific guidelines for ML-based environmental exposure assessment. The TGA and NICNAS (now AICIS — Australian Industrial Chemicals Introduction Scheme) regulate chemical risk assessment but have not endorsed ML-based external-internal exposure linkage tools. The National Environmental Health Strategy and state-based environmental health units could potentially benefit from validated versions of this framework for population-level benzene biomonitoring, particularly in communities near petroleum refineries, coking plants, or high-traffic urban corridors. PBS implications are indirect — improved exposure assessment could inform occupational health screening and cancer surveillance programs. Before adoption in Australian practice, validation using Australian population biomonitoring data (e.g., from the Australian Health Survey or state environmental health studies) would be essential. Populations with documented or suspected exposure to benzene ring-containing environmental pollutants, including occupationally exposed workers (petroleum, chemical, rubber industries), residents near industrial facilities, and urban populations with high traffic-related air pollution exposure

Abstract

Establishing quantitative linkages between external and internal exposures is essential for accurate exposure and health risk assessment, yet this goal remains hindered by multiple challenges. This Perspective systematically reviews major challenges limiting the quantitative linkage across three dimensions: (i) External exposure: complex multimedia and multipathway exposure, environmental factor variability, and substantial data gaps; (ii) Internal exposure: difficulties in biological sample collection, large-scale biomonitoring, and appropriate biomarker selection; and (iii) External-internal exposure linkage: interindividual variability, scarce toxicokinetic parameters, and methodological limitations. In this context, the Perspective proposes ML-EIExpLink (machine learning-based external-internal exposure linkage), a framework integrating multisource, multidimensional variables, including pollutant physicochemical properties, ADME parameters, exposure factors, environmental conditions, and socioeconomic indicators, to model complex external-internal exposure relationships. An exploratory case study on benzene ring-containing pollutants demonstrates the framework's feasibility and reliability. Using gradient boosting decision trees as the core predictive engine, the model achieved robust predictive performance, with an external validation Qext2 of 0.784. Most key variables, including external air concentration, molecular planarity, and temperature, exhibited nonlinear and threshold effects on predicted blood concentrations. This Perspective introduces a novel ML-based framework for establishing robust, interpretable external-internal exposure linkages, thereby providing a useful basis for improved exposure assessment and health risk evaluation.

References

1. 1.Zhang, X., Wang, X., Ma, J., Wang, S., Wang, X., Cao, Q., Wu, F., & Zhao, X. (2026). Bridging the gap between external and internal exposure: Challenges and a machine learning approach validated using benzene ring-containing pollutants. Environmental Science & Technology. https://doi.org/10.1021/acs.est.6c02845