Proposal

AI is rapidly transforming many professional fields. Occupational hygiene and occupational health will not escape this as businesses drive for efficiency. The aim is to investigate how this quickly evolving technology can enhance hazard detection, exposure assessment, and risk management. There is currently limited critical analysis of the practical, ethical, and policy implications of these technologies for occupational hygiene. This research will focus on developments since 2010; however, it is expected that the relevant research in this topic will be much newer, from 2020 onwards. Research into AI uses for occupational hygiene is far more limited than general AI topics and thus should be suitable for investigation in a suitable time period and within the word count.
Despite the promise of AI, uncertainty remains regarding the impact, reliability, and ethical challenges of the technology in occupational hygiene. There is also a need to understand how different types of AI might extend the reach of the discipline, especially in resource-limited settings.

Methodology

Approach

This research essay will use a systematic review of secondary sources, focusing on peer-reviewed scientific literature, authoritative reports, and policy documents published since 2010. The cited research will be limited to secondary sources due to time and resource constraints. The review will be structured around the main types of AI, examining their definitions, capabilities, and applications in occupational hygiene. Critical appraisal of current and potential uses for each AI type in practice, research, and policy will be included. Particular attention will be paid to evidence of real-world impact and ethical considerations.

Search Strategy
• Databases: Google Scholar, Medline, PubMed and the Annals of Work Exposures and Health
• Search terms: “artificial intelligence”, “machine learning”, “deep learning”, “natural language processing”, “computer vision”, “expert systems”, “occupational hygiene”, “exposure assessment”, “risk management”, “policy”, “ethics”, “UK”, “algorithm”, “wearable technology”, “data privacy”.
• Inclusion criteria: Publications from 2010 onwards, in English, relevant to occupational hygiene and AI.
• Exclusion criteria: Non-scientific sources, opinion pieces without evidence, publications before 2010 unless foundational.

All sources will be referenced using the Harvard system.#

Declaration on the use of AI

The guidance provided for this research essay submission states that ‘AI must not be used, and if you have used AI for research or drafting, then it will not be your own work’. However, it then recommends Google Scholar as a good tool for compiling resources. While Google Scholar does not present itself as an “AI tool” in the way that newer platforms do, it quietly leverages AI and machine learning to improve search relevance, making academic search more intuitive, as do most search engines.

This research will utilise specific search engines such as Google Scholar to find articles and research based on the subject. This may utilise AI but it will not be used to draft, write or summarise. AI is also now sometimes used for the writing of research and presented in journals and is normally indicated by a disclaimer in the article. Papers and research that uses AI will not be discounted from referencing in this essay and will be critiqued in the same way as other research.

Introduction

Artificial intelligence (AI) is a buzzword throughout business, industry and increasingly, our day to day lives. AI refers to computer systems capable of performing tasks that typically require human intelligence. These tasks include learning from experience, understanding natural language, recognizing patterns, solving problems, and making decisions. By mimicking cognitive functions, AI enables machines to adapt to new inputs and carry out complex operations with minimal human intervention (Turing, 2009). Advocates of AI praise it as a transformative technology across the healthcare, finance, transportation, and education sectors. Whilst challenges include data security, questions over accuracy and the psychosocial and psychological stresses.

Although the concept and application of AI are decades old, recent years have seen impressive advances, particularly in the areas of image and speech recognition, natural language processing, translation, reading comprehension, computer programming, and predictive analytics.

AI’s growing importance is reflected in major scientific awards: Two Nobel Prizes recognized work that led to deep learning (physics) and to its application to protein folding (chemistry), while the Turing Award honoured groundbreaking contributions to reinforcement learning (Institute for Human-Centered AI, Stanford University, 2025).

There is currently limited critical analysis of the practical, ethical, and policy implications of these technologies for occupational hygiene, a multidisciplinary scientific field focused on identifying, evaluating, and controlling workplace hazards that may affect the health and well-being of workers (Shah & Mishra, 2024). These hazards can include chemical, physical, biological, and ergonomic risks. By systematically assessing and mitigating these dangers, occupational hygiene plays a crucial role in preventing work-related illnesses and injuries, promoting a safer and healthier working environment, and ensuring compliance with health and safety regulations.

AI will play an ever-increasing role in many of the systems we use daily. This essay will review the research and investigate how it will impact the future of occupational hygiene. Already, AI enables real-time monitoring of workplace hazards, allows proactive hazard identification, and, in the public health sector, provides predictive analytics to forecast health trends (Niehaus, et al., 2022). There is great potential on offer to revolutionize the field by enhancing hazard detection and exposure assessment, streamlining research through advanced data analytics, and informing evidence-based policy development (Pishgar, et al., 2021). By leveraging AI technologies, occupational hygienists can improve workplace safety outcomes, respond more effectively to emerging risks, and support proactive, data-driven decision-making at both organizational and regulatory levels. Recent work, such as that by Shah & Mishra, (2024), has attempted to review the advancements of this incredibly quickly evolving technology in the field of occupational hygiene. However, as soon as research such as this is published, improvements to the systems and models quickly make it feel outdated.

The following paragraphs aim to explain the core concepts that underpin most AI applications (Figure 1) and the potential application in the field of OH. The author will lay out how change is already underfoot in this field and how developments in this technology may increase the reach of our discipline by promoting wider adoption of algorithms for exposure assessment and control.

Figure 1: Basic principles of AI

Machine Learning: Enhancing Exposure Assessment and Risk Prediction

Machine learning (ML) is a fundamental part of AI that focuses on algorithms that can process data sets containing millions of data points and learn to perform operational or analytical tasks without explicit instructions, gradually improving their accuracy through experience. The algorithms are trained to identify patterns and make predictions based on data. Machine learning is heavily hardware-driven and relies on specialised hardware like Graphics Processing Units (GPUs) to handle massive parallel processing for training and use.

Research carried out by Stanford University, as part of its AI index report for 2025, suggests that machine learning hardware performance has grown 43% annually, doubling every 1.9 years. Meanwhile, hardware costs are dropping 30% per year, while energy efficiency has increased by 40% annually (Institute for Human-Centered AI, Stanford University, 2025). With these rapid increases in performance and dramatic cost reductions, the technology is more likely to be utilised by broader groups and to be useful for more unique datasets, such as those found in the field of occupational hygiene.

Within our field, ML can be used to analyse historical exposure records, environmental or wearable sensor data, and operational variables to predict when and where hazardous exposures are likely to occur. For example, ML models could forecast airborne contaminant levels in specific work zones or identify high-risk job roles based on past incident data. This enables more proactive risk management, allowing hygienists to implement controls before exposures reach harmful levels
(Shah & Mishra, 2024). Upon reviewing the research, it becomes clear that early AI models have been utilised in the field of occupational hygiene for decades.

Developed in the 1990s, Estimation and Assessment of Substance Exposure (EASE) is an artificial intelligence program developed by the UK’s Health and Safety Executive to assess exposure. EASE computes estimated airborne concentrations based on a substance’s vapour pressure and the types of controls in the work area and was intended only to make broad predictions of exposure from occupational environments. Many validation studies have been carried out over the years, such as those by Johnston, et al. (2005) and Creely, et al. (2005) which reached similar conclusions: EASE is not a substitute for actual exposure monitoring.

This early tool offers a glimpse of what is possible, and with the rapid increase in performance, there is real potential to enhance exposure assessment and risk prediction using AI models. There have been a few updates to the model, including a substantial one (V2.0) in the early 2000s; however, it does not appear to have kept up with the rapidly developing technology that is machine learning. Instead, the concepts of EASE were used to build deterministic tools (not using AI) such as the ECETOC TRA (the European Centre for Ecotoxicology and Toxicology of Chemicals Targeted Risk Assessment). This is a consumer tool developed as a screening level tool to support industry compliance with the European Union’s Registration, Evaluation, Authorization, and Restriction of Chemicals (REACH) regulation (Zaleski, et al., 2023). Both EASE and these traditional deterministic tools have been shown to overestimate exposure by factors of 10–14 due to conservative assumptions (Creely, et al., 2005). This is no great surprise given the limitations of deterministic models.

Machine learning can add value and improve accuracy when utilised in models such as TREXMO+, which combines outputs from existing models with measured exposure data to reduce bias and improve exposure prediction accuracy. A study by Savic, et al. (2020), found that TREXMO+ performed better regarding bias, accuracy, and its correlation with measurements. The most important outcome was that the model’s predictions differ only by a factor between 2 and 3 in comparison with corresponding exposure measurements, relatively smaller than the factors of the traditional REACH models. The impact of these ML-driven models should also benefit from the rapid performance growth referred to earlier in this section as the inclusion of ML technology makes it a dynamic system (unlike deterministic tools). When new measurement data becomes available, the model will update and improve its accuracy over time.

One of the main limitations of ML is the size and availability of data sets. This is likely to be a significant factor in the field of occupational hygiene due to the relatively low number of exposure measurements taken per SEG (similarly exposed group) and the availability of that data due to health data privacy. Much of the measurement data is owned by businesses that may not make that data publicly available. These limitations may be overcome by advances in other areas of AI and by collaborative efforts to share data and improve these tools.

Deep Learning: Automating Complex Pattern Recognition

Deep learning, a subset of machine learning, uses neural networks with multiple layers to model complex relationships in data. It is particularly effective in processing unstructured data such as audio, video, and sensor signals. Deep learning has emerged as one of the most powerful tools in AI and has seen remarkable advances in both its theoretical underpinnings and practical applications in recent years. In healthcare, for example, deep learning models are used for early disease diagnosis, drug discovery, and personalized medicine (Shah & Mishra, 2024). In the automotive industry, deep learning plays a pivotal role in the development of self-driving cars. The ability of deep learning models to improve with more data and computational power has positioned them at the forefront of AI research and applications.

Examples of deep learnings use in the field of occupational hygiene include automated interpretation of data from wearable devices, detecting anomalies in physiological signals (e.g., heart rate variability under heat stress), or classifying noise patterns in industrial environments. Of these examples, a considerable amount of effort has gone into wearable technology, which will allow for continuous health monitoring and early detection of hazardous conditions that may not be evident through traditional methods. (Hermoza, et al., 2025)

The FDA (Food and Drug Administration, US) authorised its first AI-enabled medical device in 1995. For the next two decades, annual approvals remained in the single digits. In 2015 alone, six AI medical devices were approved. Since then, the number of yearly approvals has surged, peaking at 223 in 2023 (US Food and Drugs Administration, 2025)

The ability of these devices to monitor a broad range of employee performance and health metrics across diverse use-case environments means that the variability in how workers react to or metabolise certain hazards can be captured. This is normally a limiting factor when discussing the application of exposure limits (Shum, et al., 2024).

One example of a device currently on the market is the SoterCoach wearable. These devices use machine learning to analyse motion data from small sensors. The software can identify patterns of unsafe movements and classify posture risks in real time. ML is utilised as they continuously learn from aggregated user data to improve accuracy and provide instant feedback to prevent musculoskeletal (MSK) injuries. This is known as haptic feedback. The company website claims that SoterCoach is trusted by leading teams at Coca Cola, bp and Travis Perkins, among other large corporations in various sectors (Soter, 2025). Two recent papers have looked into the accuracy of commercial wearable systems (e.g., StrongArm Fuse, SoterCoach, Kinetic Reflex) that use a single inertia measurement unit (IMU) mounted on the trunk or waist to provide useful insight and/or feedback on high-risk postures. Savic, et al. (2021), found that single trunk IMU systems can provide moderate accuracy for estimating time-series lumbar moments but tend to perform worse during higher musculoskeletal loading, which are often the instances of highest ergonomic interest. While Nurse, et al. (2023), found that using only trunk motion data from an IMU (like what Soter wearables typically use) provides low to moderate accuracy in estimating low back disorder (LBD) risk. However, combining trunk IMU data with under – foot force measurements (e.g., pressure insoles) dramatically improves accuracy. Both studies focus on the risk assessment part of the
algorithm and don’t interrogate the benefits machine learning really brings to the system. In these
new AI driven wearables, machine learning personalises ergonomic coaching by learning each worker’s movement patterns and reinforcing safer habits over time using haptic feedback. They
also use predictive analytics to identify high – risk tasks or individuals, enabling proactive interventions before injuries occur. Unfortunately, the research into the reinforcing of healthy habits and coaching benefits from these devices is currently very limited.

A further subset of this is computer vison, used to interpret visual data from cameras and sensors. In the workplace, it can be used to monitor compliance with safety protocols, such as PPE (personal protective equipment) usage or safe lifting techniques. An excellent example of this is the use of thermal cameras to detect heat stress in workers by allowing employers to continuously monitor their employees body temperature with high accuracy (Khorshid & Song, 2025).

Singh, et al. (2020), showed that using computer vison in healthcare settings was equivalent to human observations in detecting hand hygiene dispenser use. This can be used in similar ways for PPE use, confined space entry monitoring, and chemical spills. Again, research in uses of computer vision for occupational health and safety is currently very limited.

One further example, and perhaps the most promising from a change in OH practice, is advancements AI brings to chemical research. Platforms like Google DeepMind’s ‘AlphaFold’ leverage machine learning to predict molecular interactions with unprecedented accuracy. This enables chemists to design new compounds and refine existing ones for increased selectivity and efficiency (AlphaFold, 2025). In the future we will be able to quickly design alternative chemicals with enhanced favourable properties and reduced health impacts. Most of the research is currently focused on advancements in drug and protein/amino acid sequencing but will certainly be a topic for further research.

Natural Language Processing: Extracting Insights from Textual Data

Natural Language Processing (NLP) enables AI systems to understand and analyse human language both written and spoken. It bridges the gap between human communication and computer analysis. It is often used for extracting key data points for articles, research, reports etc. It also powers chatbots and allows them to understand and answer questions.

In occupational hygiene, NLP can be used to process large volumes of exposure reports, inspection records, and regulatory documents. For instance, NLP tools can identify recurring themes in reports, such as common causes of exposure or frequently cited recommendations. A systematic review by Khairuddin, et al. (2022), analysed text mining and NLP models to classify accident types and predict causal factors from occupational injury reports. They concluded that the models were potentially proven in analysis of occupational injury data and predicting the occupational injury. This study is limited by the size of the dataset, a common theme in the studies cited, as businesses are unlikely to share full details of their injury data. However, the potential for occupational hygiene should be clear, there will be mountains of specific injury data such as MSK injuries or chemical burns within businesses or industries.

These are likely distributed between different systems and ways of recording, but NLP can
bring these together and provide useful insights into high – risk tasks, operations or individuals.

In addition to the synthesis of technical data, NLP analyses social media and stakeholder feedback to gauge sentiment on policies, including health policy. This can be in the form of policy acceptance strategies and communication campaigns for new regulations. Jerfy, et al. (2024), demonstrate how NLP-driven sentiment analysis shapes public health policy decisions, a principle increasingly applied in occupational hygiene. One powerful example of this is the use of social media data used to develop pandemic prediction models based on reported symptoms. Currently, there is very little research on how best to utilise this and whether positive impacts have been recorded in the occupational hygiene community.

Large Language Models and Generative AI: Supporting Decision-Making and Training

Large Language Models (LLMs) can generate human-like text, images and videos and answer complex queries. Popular examples include OpenAI’s GPT, Google’s Gemini and Anthropic’ s Claud. LLMs are some of the most user friendly and popular AI systems used by millions daily.

In occupational hygiene, LLMs can be used to develop interactive training tools and generate communication material that can be tailored to different literacy levels or even languages. Generative AI can also assist in drafting methodologies, summarising technical documents, or creating scenario-based simulations for emergency response training (Shah & Mishra, 2024).

Businesses now build in-house chatbots which allow for greater sharing of information and enables users to quickly look back at similar incidents or regulatory issues. This utilises NLP in Large Language Models by allowing users to express their needs or questions naturally and clearly by speaking or typing. They’re easy to use and have the most detailed responses to questions (Kasthuri & Balji, 2023). In the future, it may be possible for AI to simulate expert consultations for small organizations lacking in-house expertise.

A study by Padovan, et al. (2024), compared responses to complex occupational health questions from ChatGPT and human experts (physicians). Results showed AI-generated answers were comparable to human experts when legislative context was provided, suggesting potential for regulatory and compliance support in occupational hygiene contexts. However, the study also asked for user feedback, which was mostly in favour of the answers generated by the humans. A clear example that user acceptance is still in its infancy especially in a healthcare setting. This is an ongoing area of development, with recent studies undertaken demonstrating ways to increase trust by improving the way these systems use and understand the nuances of human language (Ye, et al., 2023).

Drawbacks & Ethical Considerations

The incredible advances in artificial intelligence over recent years has generated a lot of interest and potential in the field of occupational hygiene, but this has also raised a number of concerns. Based on the research presented in this essay, the most frequently cited concerns are over accuracy, how to ensure privacy and data security, how to combat discrimination or bias and psychosocial stresses including job displacement. Furthermore, developing and implementing AI systems can be expensive and may require significant investment in hardware, software, and training.

Regarding accuracy, complex reasoning remains a problem in 2025. Even though the addition of mechanisms such as chain-of-thought reasoning (showing your workings) has significantly improved the performance and accountability of LLMs, these systems still cannot reliably solve problems for which provably correct solutions can be found using logical reasoning, such as arithmetic and planning, especially on instances larger than those they were trained on. This has a significant impact on the trustworthiness of these systems and their suitability in high-risk applications (Institute for Human-Centered AI, Stanford University, 2025).

AI systems rely on high-quality data to be most effective and accurate. If the data is incomplete, outdated, or inaccurate it can significantly impact the system’s performance and, in the cases described above, could lead to increased health and safety risks (Steimers & Schneider, 2022). AI is also susceptible to amplifying bias if it is trained on biased data; therefore the data must be representative. Historic exposure data within occupational hygiene is likely to be an excellent data set for AI, as it generally follows established methods that have not changed for decades. Furthermore, recording similar exposure groups and using data predominantly from industries that have not changed much should ensure that viable datasets are built without bias.

A further critical ethical issue arises from privacy and data security, given that AI systems rely on extensive datasets containing personal information, such as wearable devices and sensors (Santoni de Sio, 2024). This is even more relevant in the field of occupational hygiene, where health data may be at stake. It is therefore essential for concepts and approaches utilising AI to guarantee the ethical and secure collection, utilisation, and storage of this data.

Some suggest that integrating AI into occupational hygiene could negatively impact workers’ mental health, as workers may feel a loss of control in an environment monitored by AI systems and experience isolation/disconnect from human colleagues. Moreover, these changes in working conditions may entail new and serious health risks for employees (Howard & Schulte, 2024). For female employees in particular, the increasing use of AI and the associated demand for greater working-time flexibility is likely to be a major challenge and might even become an employment risk if adequate and flexible childcare facilities are not available (Howard & Schulte, 2024). This research raises major ethical considerations.

Conclusions

The rapid development and application of artificial intelligence is already changing practice, research, and policy development in industrial hygiene, as explored in this essay. AI driven tools are harnessing the power of machine learning models with deep learning neural networks, brought to life with natural language processing in large language models for exposure assessment and enhancing the ability of occupational hygienists to detect hazards, predict risks, and implement more effective controls.

AI has the potential to influence controls from the very top of the hierarchy of control, by elimination or substitution of chemicals using ML to identify safer alternatives quickly and accurately. This is probably the most striking example of how occupational hygiene practice
might be made more efficient by AI, but the technology is likely to influence every aspect of
the discipline over time.

However, integrating AI into occupational hygiene is not without its challenges. Concerns over accuracy, privacy and data security, combating discrimination or bias, and psychosocial stresses must be addressed through a robust ethics framework and transparent policy. Work on this is happening in areas such as government policy but there are fears that it is not keeping up.

There are many more examples of AI applications in development or even fully realised within businesses with the aim of good health risk management, but these have not yet been proven or researched due to the lagging of research studies. Indeed, the pace of technological advancement is outstripping the current body of research, particularly in occupational hygiene, highlighting the need for ongoing critical evaluation and interdisciplinary collaboration.

In summary, AI should not be seen as a teleportation device to travel from the start to the end of the process. Instead, it should be used as a map that an industrial hygienist learns to read. It should complement and not replace the expertise and judgment of occupational hygienists, but it should be embraced and included as part of the toolkit.

References

AlphaFold, 2025. AlphaFold Protein Structure Database. [Online] Available at: https://alphafold.ebi.ac.uk/ [Accessed 12 November 2025].

Carbonero, F. et al., 2023. The impact of artificial intelligence on labor markets in developing countries: a new method with an illustration for Lao PDR and urban Viet Nam. Journal of evolutionary economics, pp. 1-30.

Creely, K. S. et al., 2005. Evaluation and Further Development of EASE Model 2.0. he Annals of Occupational Hygiene, 49(2), pp. 135 -145.

Hermoza, S., Morales, H. & Segovia, S., 2025. Deep learning in occupational safety and health and its contribution to workplace risk management: A systematic review. Zenodo.

Howard, J. & Schulte, P., 2024. Managing workplace AI risks and the future of work. American Journal of Industrial Medcine, Volume 67, pp. 980-993.

Institute for Human-Centered AI, Stanford University, 2025. The AI Index 2025 Annual Report, s.l.: Stanford University, Stanford.

Jerfy, A., Selden, O. & Rajesh, B., 2024. The Growing Impact of Natural Language Processing in Healthcare and Public Health. The Journal of Health Care Organization, Provision, and Financing.

Johnston, K. L., Phillips, M. L., Esmen, N. A. & Hall, T. A., 2005. Evaluation of an Artificial Intelligence Program for Estimating Occupational Exposures. The Annals of Occupational Hygiene, 49(2), p. 147–153.

Kasthuri, E. & Balji, S., 2023. Natural language processing and deep learning chatbot using long short term memory algorithm. Materials Today: Proceedings, 81(2), pp. 690-693.

Khairuddin, M. et al., 2022. Predicting occupational injury causal factors using text-based analytics: A systematic review. Frontiers in public health.

Khorshid, S. & Song, S., 2025. Exploring Intervention Strategies to Prevent Occupational Heat Stress. A Scoping Review. Safety and health at work, 16(2), pp. 134-144.

Matijevich, E. S., Volgyesi, P. & Zelik, K. E., 2021. A Promising Wearable Solution for the Practical and Accurate Monitoring of Low Back Loading in Manual Material Handling. Sensors, Volume 21, p. 340.

Niehaus, S., Hartwig, M., Rosen, P. H. & Wischniewski, S., 2022. An Occupational Safety and Health Perspective on Human in Control and AI. Frontiers in Artificial Intelligence.

Nurse, C. A., Elstub, L. J., Volgyesi, P. & Zelik, K. E., 2023. How Accurately Can Wearable Sensors Assess Low Back Disorder Risks during Material Handling? Exploring the Fundamental Capabilities and Limitations of Different Sensor Signals. Sensors, 23(4), p. 2064.

Padovan, M. et al., 2024. ChatGPT in Occupational Medicine: A Comparative Study with Human Experts. Bioengineering, 11(1), p. 57.

Pishgar, M. et al., 2021. A Novel Framework to Review Artificial Intelligence and Its Applications in Occupational Safety and Health. International journal of environmental research and public health, Volume 18.

Santoni de Sio, F., 2024. Artificial Intelligence and the Future of Work: Mapping the Ethical Issues. The Journal of Ethics, Volume 28, pp. 407-427.

Savic, N., Lee, G. E., Gasic, B. & Vernez, D., 2020. TREXMO plus: an advanced self-learning model for occupational exposure assessment. Journal of Exposure Science & Environmental Epidemiology, Volume 30, pp. 554-566.

Shah, I. A. & Mishra, S., 2024. Artificial intelligence in advancing occupational health and safety: an encapsulation of developments. Journal of Occupational Health, Volume 4, p. 66.

Shum, M., McLean, S. & Lu, E., 2024. What to wear? Occupational hygiene applications for wearables. Annals of Work Exposures and Health, 68(1), p. 1.

Singh, A. et al., 2020. Automatic detection of hand hygiene using computer vision technology. Journal of the American Medical Informatics Association, 27(8), pp. 1316-1320.

Soter, 2025. SoterCoach: AI-powered coaching to reduce workplace injuries. [Online] Available at: https://www.soter.com/sotercoach [Accessed 16 Novemeber 2025].

Steimers, A. & Schneider, M., 2022. Sources of Risk of AI Systems. International journal of environmental research and public health, 19(6), p. 3641.

Turing, A., 2009. Computing Machinery and Intelligence. s.l.:Springer, Dordrecht.

US Food and Drugs Administration, 2025. Remote or Wearable Patient Monitoring Devices EUAs. [Online] Available at: https://www.fda.gov/medical-devices/covid-19-emergency-use-authorizations-medical-devices/remote-or-wearable-patient-monitoring-devices-euas [Accessed 17 November 2025].

Ye, Y., You, H. & Du, J., 2023. Improved Trust in Human-Robot Collaboration With ChatGPT. IEEE Access.

Zaleski, R. T. et al., 2023. An assessment of the ECETOC TRA Consumer tool performance as a screening level tool. Journal of exposure science & environmental epidemiology, 33(6), pp. 980 – 993.