Occupational Solar UV Exposure and Skin Cancer: A Quantitative Framework for Individual Causal Attribution in Dermatology Practice

Abstract

Background

Occupational solar ultraviolet radiation (UVR) exposure is a major risk factor for keratinocyte carcinomas in outdoor workers. Individual causal attribution for medico-legal purposes requires partitioning risk contributions between occupational exposure and individual susceptibility factors—a question distinct from population-level burden estimation. This study is primarily a quantitative/methodological application of FAIR to occupational UV-related skin cancer.

Methods

This methodological framework study applies the FAIR approach (Filtering evidence, Association quantification, Individual partitioning, Reasoned interpretation) to occupational solar UVR and skin cancer. Meta-analyses (1995–2025) were reviewed and adjusted for phototype, sex, and immune status. Individual attribution was estimated using game-theory–based decomposition (Shapley values) and stochastic Monte Carlo simulation to propagate uncertainty. Preventive scenarios incorporated sector-specific compliance and long-term economic modeling.

Results

Mean occupational causal shares were 35% for squamous cell carcinoma (SCC), 25% for basal cell carcinoma (BCC), 10% for melanoma, and 50% for actinic keratosis. The probability of exceeding the 20% medico-legal attribution threshold reached 85% for SCC and 75% for BCC. Prevention reduced expected cases by 10-15% and healthcare costs by 8-12%.

Conclusions

The FAIR framework provides a transparent, reproducible approach for integrating epidemiological evidence with individual attribution. These quantitative thresholds support the clinical and medico-legal recognition of work-related skin cancers.

1. Introduction

Occupational dermatology encompasses a broad spectrum of work-related skin diseases, ranging from irritant and allergic contact dermatitis to chronic UV-induced skin damage and keratinocyte carcinomas. Despite their different clinical manifestations, these conditions share a common challenge in medico-legal practice: the need to quantify the contribution of occupational exposures relative to individual susceptibility factors. Occupational exposure to solar ultraviolet radiation (UVR) represents one of the most pervasive yet under-recognized carcinogenic risks worldwide. Outdoor workers, such as those employed in agriculture, construction, maritime trades, and transportation, are chronically exposed to high cumulative doses of solar UVR due to prolonged outdoor activity and limited opportunities for shade or protective equipment use [1,2]. In 2019, approximately 1.6 billion workers, representing nearly 28% of the global workforce, were occupationally exposed to solar UVR [3]. According to the WHO-ILO joint estimates, this exposure accounts for approximately one in three deaths from non-melanoma skin cancer (NMSC), making it one of the leading occupational carcinogenic risk factors globally [3,4]. Beyond keratinocyte carcinomas and melanoma, occupational solar UV exposure also contributes to a spectrum of chronic dermatoses, including actinic damage and field cancerization in outdoor workers, which are central concerns in occupational dermatology practice.

The carcinogenicity of solar UVR is well established. The International Agency for Research on Cancer classifies solar radiation as a Group 1 human carcinogen [5]. Multiple meta-analyses have demonstrated significantly increased risks of NMSC, particularly cutaneous squamous cell carcinoma (cSCC) and basal cell carcinoma (BCC) - among outdoor workers compared with indoor workers, with relative risks typically ranging between 1.4 and 2.2 [[6], [7], [8]]. In contrast, the association between chronic occupational sun exposure and malignant melanoma remains less clear. Some systematic reviews suggest that continuous occupational exposure may not increase melanoma risk to the same extent as intermittent high-intensity recreational exposure [9].

Despite robust epidemiological evidence, occupational recognition and compensation for solar UVR-induced skin cancers remain inconsistent across jurisdictions. Some countries, including Germany and Australia, formally recognize NMSC caused by occupational solar UV exposure as a compensable occupational disease, whereas others rely on case-by-case adjudication [10]. Comparative analyses have highlighted substantial variation in prevention strategies, compensation frameworks, and legal causation thresholds across high-income countries [11].

Individual susceptibility further complicates causal attribution. Skin phototype and latitude strongly influence UV-related cancer risk, with fair-skinned individuals (phototypes I-II) exhibiting two-to four-fold higher risks compared with darker skin types [12]. Sex-related differences in occupational exposure patterns have also been described, with male outdoor workers generally demonstrating higher incidence rates of keratinocyte carcinomas [13]. Furthermore, immunosuppressed individuals, particularly organ transplant recipients, have dramatically increased risks of NMSC, often exceeding those of the general population by 50- to 100-fold [14].

Prevention of occupational UV-related skin cancer is feasible but remains incompletely implemented. Protective strategies, including sunscreen use, protective clothing, shade provision, and work-schedule adaptation, have demonstrated effectiveness in reducing occupational UV exposure [15]. Economic analyses suggest that large-scale UV protection programs for outdoor workers are cost-effective and may substantially reduce healthcare expenditures associated with keratinocyte carcinomas [16]. Climate change may further amplify occupational UV risks through increased UV indices, prolonged outdoor work seasons, and changing environmental exposure patterns [17].

Given this complex interaction between environmental exposure, individual susceptibility, and preventive interventions, traditional epidemiological measures such as population-attributable fractions are insufficient for individual medico-legal attribution. They estimate population burden but do not allow partitioning of causal contributions between interacting determinants. To address this gap, the FAIR framework (Filtering evidence, Association quantification, Individual partitioning, Reasoned interpretation) was developed to provide a structured epidemiological approach for individual causal attribution in occupational disease claims [18]. However, practical tools translating this framework into concrete thresholds usable by dermatologists in daily occupational skin cancer assessments are still lacking.

In this study, we apply the FAIR framework to occupational solar UV exposure and skin cancer risk among outdoor workers. Our objectives are to: (i) quantify epidemiological associations between occupational solar UV exposure and keratinocyte carcinomas and melanoma; (ii) estimate individual attributable risk using stochastic Monte Carlo propagation of uncertainty; (iii) partition causal contributions between occupational exposure and individual susceptibility factors; and (iv) evaluate the implications for medico-legal attribution and occupational prevention policies. For clarity, FAIR-based attribution answers a different question than the classical population-attributable fraction (PAF). PAF asks “what proportion of all cases in a population is due to occupational UV exposure?“, whereas FAIR asks, “what proportion of this particular patient's excess risk is due to occupational UV exposure, given their individual risk profile?“. In this sense, PAF supports public health prioritisation, while FAIR supports case-by-case clinical and medico-legal decisions.

By integrating deterministic attribution, stochastic uncertainty propagation, and preventive modelling, the FAIR framework provides a reproducible and transparent approach linking epidemiological burden, individual causation, and occupational prevention. This study is primarily a methodological and modelling application of the FAIR framework to occupational solar UV-related skin cancer, rather than a conventional observational epidemiological study.

2. Methods

2.1. Filtering evidence

A systematic review was conducted covering the period 1995–2025, targeting meta-analyses, pooled cohort studies, and large population-based investigations reporting adjusted relative risks (RRs) or hazard ratios (HRs) for outdoor workers. Studies were stratified by lesion type (squamous cell carcinoma [SCC], basal cell carcinoma [BCC], melanoma, actinic keratosis [AK]), skin phototype, sex, and immune status.

Methodological quality was assessed using the Office of Health Assessment and Translation (OHAT) risk of bias tool [19], and publication bias was screened using Egger regression tests [20] (p > 0.1 as inclusion criterion).

Baseline incidence rates (p₀, range = 8-387 per 100,000 person-years) were extracted from national cancer registries including AIHW, SEER, IARC, ECDC, and NII and stratified by phototype, sex, and latitude to capture geographic variability [21]. When multiple registries were available, weighted means were calculated using population denominators.

2.2. Association quantification

Meta-analytic estimates were parameterized as log-normal distributions to reflect multiplicative uncertainty in risk estimates:

• •SCC: RR = 1.9 (95 % CI 1.6-2.2)
• •BCC: RR = 1.35 (95 % CI 1.2-1.5)
• •Melanoma: RR = 1.15 (95 % CI 0.9-1.3)
• •AK: RR = 3.1 (95 % CI 2.4-3.8)

Baseline incidence (p₀) was modeled using a Gamma (shape, rate) distribution to represent count-based stochastic variability.

The main epidemiological inputs by lesion type, including relative risks, baseline incidence sources and major stratifications, are summarised in Table 1.

Table 1.

Key epidemiological inputs for occupational solar UV–related skin cancer by lesion type

Lesion type	Main occupational RR (95% CI)	Baseline incidence p0 (per 100,000)	Key RR sources	Registries	Stratifications
SCC	1.90 (1.60–2.20)	∼20–200∗	Loney 2021 [7]; Schmitt 2011 [8]; Lee 2024 [6]	AIHW, SEER, IARC, NII	Phototype, sex, latitude
BCC	1.35 (1.20–1.50)	∼50–300∗	Loney 2021 [7]; Lee 2024 [6]	AIHW, SEER, IARC, NII	Phototype, sex, latitude
Melanoma	1.15 (0.90–1.30)	∼8–40∗	Maduka 2023 [9]; WHO–ILO [4,5]	AIHW, SEER, IARC	Phototype, sex, latitude
AK	3.10 (2.40–3.80)	Variable†	Modenese 2018 [2]	Regional registries	Phototype, sex, latitude

^∗Derived from pooled registry ranges.

^†Not systematically recorded in cancer registries.

Example:

If mean incidence = 0.00387 (387 cases per 100,000 person-years) and variance = (0.00038), then:

• -shape = mean²/variance = 103.6
• -rate = mean/variance = 26,776
• -Gamma (shape = 103.6, rate ≈ 26,800)

This formulation allows direct sampling of p₀ in the Monte Carlo component while preserving epidemiological plausibility.

2.3. Individual partitioning

Deterministic FAIR attribution applies Shapley decomposition across the set of causal factors

$$N=\left\{\text{UV},\text{phototype},\text{immunosuppression},\text{sex},\text{sector}\right\}$$

The Shapley value provides a fair allocation of the marginal contribution of each factor across all possible permutations of interacting determinants. For each factor $i\in N$, its marginal contribution ${\mathrm{\varphi }}_i$ is defined as:

$${\mathrm{\varphi }}_i=\sum _{S\subseteq N\setminus \left\{i\right\}}\frac{\left|S\right|!\left(n-\left|S\right|-1\right)!}{n!}\left[v\left(S\cup \left\{i\right\}\right)-v\left(S\right)\right]$$ (1)

where: $S$ is any subset (coalition) of factors not containing i, $n=\left|N\right|$ is the total number of causal factors, $v\left(S\right)$ is the coalition value function

Intuitively, Shapley values average each factor's contribution across all possible combinations of risk factors, ensuring no single determinant is over- or under-weighted. For the purposes of causal attribution, the coalition value function v(S) was defined as the excess risk attributable to a given subset of determinants S. Under the multiplicative risk assumption commonly applied in epidemiology, the relative risk for a coalition was computed as the product of individual relative risks (${RR}_s=\prod RRᵢ)$. The coalition value was therefore expressed as the excess absolute risk relative to baseline incidence:

$$v\left(S\right)=p^0\times \left({RR}_s-1\right)$$

This formulation ensures that the total attributable excess risk remains bounded and allows direct comparison with medico-legal thresholds. Shapley allocation then partitions the total excess risk v(N) among all determinants by averaging their marginal contributions across all possible causal orderings.

2.4. Monte Carlo

The stochastic extension propagated uncertainty in key parameters through 10,000 Monte Carlo iterations.

Parameters were sampled as follows:

• •RR distributions: log-normal distributions parameterized using meta-analytic point estimates and 95% confidence intervals;
• •Baseline risk (p₀): Gamma(shape = k, rate = β);
• •Inter-factor correlations (r): truncated normal distribution N(0.45, 0.15²) bounded between 0.3 and 0.6;

Simulation outputs included the distribution of individual attributable risk (RAI) and the probability of exceeding medico-legal thresholds:

• •P(RAI ≥20 %) = “substantial contribution”;
• •P(RAI >50 %) = “more likely than not.”

Convergence diagnostics confirmed model stability (potential scale reduction factor Gelman–Rubin $\hat{\mathrm{R}}$ < 1.05 for all parameters).

All simulations were implemented in R (version 4.4) using custom scripts for Monte Carlo sampling and Shapley decomposition.

2.5. Correlation and multicollinearity diagnostics

To verify model stability, multicollinearity among predictors (UV, phototype, immunosuppression, sex, sector) was assessed using the condition number (κ) of the 5 × 5 correlation matrix:

$$\mathrm{\kappa }=\frac{{\lambda }_{\mathit{max}}}{{\lambda }_{\mathit{min}}}=24$$ (2)

Eigenvalue decomposition (R function eigen) confirmed acceptable collinearity (κ < 30). Pairwise correlations were inspected to ensure independence assumptions were not violated.

Model performance was assessed using internal validation procedures including cross-validation and bootstrap resampling.

2.6. Comparison with population-attributable fraction (PAF)

For contextualization, Levin's PAF was computed using:

$$\text{PAF}=\frac{P_{\exp }\left(RR-1\right)}{1+P_{\exp }\left(RR-1\right)}$$ (3)

where P_exp is the prevalence of occupational UV exposure derived from CAREX data.

While PAF estimates the proportion of cases in a population attributable to occupational UV exposure, FAIR estimates the proportion of an individual patient's excess risk attributable to occupational UV exposure given their specific risk profile. PAF supports preventive policy planning, whereas FAIR informs case-by-case medico-legal adjudication. Both approaches are complementary but address fundamentally different levels of inference.

2.7. Preventive and economic simulation

Preventive scenarios modeled reductions in occupational relative risks (10–40%) reflecting heterogeneity in compliance across occupational sectors. Compliance assumptions were informed by published intervention studies and policy analyses on UV protection among outdoor workers [10,11,15]. These preventive scenarios are intended as illustrative policy simulations rather than precise predictive models. The ranges of compliance and relative risk reduction were derived from published intervention studies and occupational health policy analyses, reflecting variability across sectors.

We assumed a compliance rate of 0.40 with a 10–20% relative risk reduction in agriculture, 0.55 with a 20–30% reduction in construction, 0.50 with a 15–25% reduction in maritime settings, and 0.60 with a 30–40% reduction in tourism-related outdoor occupations. These values reflect sector-specific differences in sun protection practices, regulatory frameworks, and occupational exposure patterns [10,11,16]. Economic outcomes were evaluated via Net Present Value (NPV) over 30 years, applying a 3% annual discount rate:

$$\text{NPV}=\sum _{t=0}^{30}\frac{Case{s}_t\times \mathit{Cos}{t}_t}{{\left(1.03\right)}^t}$$ (4)

Costs included direct medical expenditures (treatment and follow-up) and indirect productivity losses (0.5 × GDP per capita). Preventive program costs were estimated at €200 per worker per year.

2.8. Directed acyclic graph (DAG) causal structure

A causal graph was specified to represent pathways and potential confounders:

$$JobSector\to UVExposure\to SkinCancer$$

$$\downarrow \downarrow$$

$$Phototype\to BaselineRisk\left(p_0\right)$$

$$\downarrow$$

$$Immunosuppression\to Progression$$

This DAG clarifies causal directionality, distinguishes mediators (exposure intensity, phototype) from effect modifiers (immunosuppression), and supports transparency in factor inclusion for Shapley decomposition and Monte Carlo simulation.

Additional methodological details and extended simulation results including sensitivity analyses varying compliance rates and risk reduction magnitudes across their plausible ranges are provided in the Supplementary Material.

2.9. Statistics calculation

All statistical computations and modelling were conducted using the R statistical environment (Posit Cloud). The calculations originally performed in the NII environment were independently reproduced and verified at the Department of Public Health, North Dakota State University (NDSU), to confirm the robustness of the results.

2.10. Analytical workflow

To improve transparency and reproducibility, the FAIR analytical pipeline was structured into five sequential steps:

• 1.Evidence filtering using systematic review and OHAT risk-of-bias assessment;
• 2.Association quantification through extraction and parameterisation of relative risks (RR) and baseline incidence (p₀);
• 3.Individual partitioning using Shapley value decomposition to allocate excess risk across determinants;
• 4.Stochastic propagation of uncertainty via Monte Carlo simulation (10,000 iterations);
• 5.Derivation of outputs including individual attributable risk (RAI) and probability of exceeding medico-legal thresholds (PRAI).

The full FAIR analytical pipeline is explicitly summarised in Fig. 1, which provides a stepwise overview from evidence filtering and parameterisation to stochastic simulation and derivation of individual attribution metrics (RAI and PRAI).

Fig. 1.

FAIR analytical workflow for individual causal attribution.

2.11. Interpretation of FAIR outputs in practice

In clinical and medico-legal contexts, the individual attributable risk (RAI) represents the estimated proportion of excess risk attributable to occupational exposure. The probability metrics (PRAI) quantify the likelihood that this contribution exceeds legally relevant thresholds (≥20% or >50%). These outputs should be interpreted alongside clinical judgement, exposure history, and individual susceptibility factors.

3. Results

3.1. Deterministic FAIR attribution

The deterministic FAIR model estimated mean occupational causal shares consistent with previously reported epidemiological risk gradients.

• •35% for squamous cell carcinoma (SCC),
• •25% for basal cell carcinoma (BCC),
• •10% for malignant melanoma, and
• •50% for actinic keratosis (AK).

These values represent the expected proportion of individual attributable risk (RAI) derived from the Shapley partitioning of multifactorial risk among UV exposure, phototype, immunosuppression, sex, and occupational sector. The ranking of causal shares was consistent across sensitivity scenarios, confirming that keratinocyte carcinomas (SCC > BCC) have markedly higher occupational attribution than melanoma.

3.2. Stochastic FAIR simulation

Monte Carlo propagation of uncertainty across 10,000 iterations yielded stable RAI distributions with narrow 90% credible intervals (±7-10 %) for all endpoints (Fig. 2).

Fig. 2.

Monte Carlo simulated distribution of occupational attributable risk (RAI) for major UV-related skin cancers based on 10,000 iterations of the FAIR stochastic model. Boxes represent the interquartile range, horizontal lines the median, and whiskers the simulated distribution range.

The resulting probabilities of exceeding medico-legal attribution thresholds were:

• •P(RAI ≥20%) = 0.85 for SCC,
• •P(RAI ≥20%) = 0.75 for BCC,
• •P(RAI ≥20%) = 0.45 for melanoma.

At the higher legal threshold (“more likely than not”, >50 % attribution), the exceedance probability varied geographically from 0.60 in Northern latitudes (e.g., Northern Europe, Canada) to 0.30 in lower-latitude regions (e.g., South and East Asia), reflecting differences in baseline risk (p₀) and phototype distributions.

Across all lesion types, the Monte Carlo posterior mean closely matched deterministic FAIR outputs (Δ < 1.5 %), confirming internal model coherence between the deterministic and stochastic components (Fig. 2).

A concise overview of mean individual attributable risks (RAI) and the probabilities of exceeding medico-legal thresholds (PRAI20 and PRAI50) for each lesion type is presented in Table 2.

Table 2.

Summary of FAIR-based individual attributable risk (RAI) and medico-legal exceedance probabilities (PRAI) for occupational solar UV–related skin cancers (10,000 Monte Carlo iterations)

Lesion type	Mean RAI % (90% CI)	PRAI ≥20%	PRAI >50%	Phototype gradient (RAI %)	Latitude gradient (PRAI >50%)
SCC	35 (28–43)	0.85	0.42∗	42 (I) → 18 (IV) → 7 (VI)	0.60 (45–60°N) vs 0.30 (25–35°N)
BCC	25 (19–32)	0.75	0.28	∼30 (I) → 15 (IV) → 5 (VI)†	0.50 vs 0.18
Melanoma	10 (5–16)	0.45	0.08	∼15 (I) → 8 (IV) → 3 (VI)†	0.15 vs 0.05
AK	50 (42–58)	0.95	0.68	∼60 (I) → 30 (IV) → 10 (VI)†	0.75 vs 0.55

^∗Overall mean across latitudes.

^†Approximate values derived from model gradients.

3.3. Sensitivity analysis

The tornado diagram revealed that baseline incidence (p₀) contributed the largest share of total variance (68%), followed by RR uncertainty (18%), inter-factor correlation (10%), and preventive compliance (4%). Notably, preventive compliance contributed only 4% of total variance, indicating that the main attribution results are robust to the assumed compliance ranges. Subgroup simulations by skin phototype showed graded decreases in mean RAI from 42% (type I) to 7% (type VI), consistent with phenotypic protection gradients. Sensitivity analyses confirmed that no single parameter shift exceeded ±10% of total attributable risk, indicating overall stability of the attribution model.

3.4. Economic and preventive impact simulation

Under a baseline compliance of 60%, preventive interventions (sunscreen use, shade, scheduling) reduced expected occupational skin cancer cases by 10–15% and annual healthcare costs by 8–12%.

Projected 30-year net present value (NPV) analyses indicated that cumulative direct and indirect cost savings could reach €1.2–1.8 billion in high-exposure sectors (construction, agriculture, maritime). Preventive elasticity - defined as the percentage decrease in disease burden per 1% increase in compliance - averaged 0.25 (SCC) and 0.18 (BCC), supporting the cost-effectiveness of integrated UV-safety policies.

3.5. Internal validation

Model reliability was confirmed through cross-validation (70% training, 30% test), yielding a mean absolute error (MAE) of 3.2% between predicted and simulated RAI values.

Bootstrap resampling (n = 1,000) produced a median R² = 0.94 (95 % CI 0.92-0.96), demonstrating excellent internal consistency and predictive stability.

Posterior predictive checks indicated no systematic bias across strata of phototype, sex, or immunosuppression (p > 0.10, Kolmogorov-Smirnov test).

3.6. Case illustration

For clinical illustration, a phototype II male construction worker with 25 years of outdoor exposure yielded:

• •Deterministic RAI = 38%
• •Monte Carlo RAI = 37.5% (90 % CI 28-46 %),
• •P(RAI ≥20 %) = 0.89.

This exceeds the “substantial contribution” threshold (≥20%) adopted by French, Israeli, and Australian compensation systems, thus meeting medico-legal criteria for occupational causation.

Stratified analyses demonstrated a gradient in attributable risk across skin phototypes, decreasing from 42% in phototype I to 7% in phototype VI.

In practical terms, this allows the dermatologist to justify, in a transparent and quantitative way, that occupational solar UV exposure made a major contribution to this patient's cancer, beyond background and personal risk factors.

A second illustrative scenario was considered for a phototype IV individual with intermittent outdoor exposure and melanoma. The deterministic RAI was estimated at 12%, with a Monte Carlo estimate of 11% (90% CI 4–18%), and P(RAI ≥20%) = 0.28. This scenario falls below the medico-legal threshold for substantial contribution, illustrating the lower occupational attribution typically observed for melanoma in darker phototypes.

4. Discussion

The FAIR framework establishes a quantitative bridge between epidemiological burden and medico-legal reasoning by integrating deterministic causal attribution with stochastic uncertainty propagation. To our knowledge, this study represents one of the first attempts to combine epidemiological evidence synthesis, game-theory-based causal decomposition (Shapley values), and Monte Carlo uncertainty propagation within a single framework for occupational dermatology. The explicit formulation of the coalition value function v(S) ensures that multifactorial risk is partitioned among cofactors while preserving additivity, fairness, and interpretability, improving transparency compared with conventional regression-based attribution methods, which may obscure joint effects or mediation.

To facilitate practical interpretation, the implications of the FAIR framework are structured below according to key stakeholders.

4.1. Clinical and dermatology practice

The FAIR framework provides dermatologists with a structured approach to assess the contribution of occupational UV exposure in individual patients. By integrating lesion type, cumulative exposure, skin phototype, immunosuppression, and occupational sector, clinicians can derive an individual attributable risk (RAI) and communicate uncertainty explicitly through probabilistic outputs (PRAI). In practice, PRAI values allow clinicians to report, in quantitative terms, whether the contribution of occupational exposure is unlikely, compatible, or highly compatible with medico-legal thresholds (for example, "substantial contribution" at ≥20% or "more likely than not" at >50%). FAIR-based outputs should be interpreted alongside clinical history, exposure chronology, and lesion morphology, and should never replace a comprehensive dermatological assessment. This framework may also be extended to chronic UV-induced dermatoses such as actinic damage and field cancerization in outdoor workers, further strengthening its relevance for occupational dermatology.

4.2. Compensation systems and medico-legal adjudication

For compensation systems, FAIR offers a reproducible and auditable method aligned with legal thresholds. Numerical PRAI values can be incorporated into medico-legal reports as transparent summary indicators, while leaving final causation decisions to the adjudicator. Explicitly linking RAI and PRAI to existing legal standards of proof may reduce unexplained variability between experts and jurisdictions. The threshold of 20% for "substantial contribution" corresponds to standards used in several compensation systems, including those in Australia and parts of Europe, although alternative thresholds may apply depending on jurisdiction.

4.3. Prevention and occupational health policy

At the population level, FAIR-derived RAI distributions and PRAI gradients by phototype and latitude can help identify high-risk worker groups and prioritise sectors for UV-prevention programmes. The preventive simulation results suggest that even moderate improvements in compliance with sun-protection measures may yield measurable reductions in disease burden and healthcare costs, consistent with previous studies demonstrating the feasibility and cost-effectiveness of UV protection programs for outdoor workers. Sensitivity analyses indicated that variation in preventive assumptions contributed modestly to overall model variance (<5%), confirming the robustness of the main attribution results. These outputs complement traditional population-attributable fractions by connecting population burden with the probability that occupational exposure substantially contributes to disease in individual workers.

4.4. Consistency with epidemiological and biological evidence

The FAIR estimates are consistent with current knowledge. Keratinocyte cancers — particularly squamous cell carcinoma and actinic keratosis — show substantially higher occupational causal shares than melanoma, reflecting the cumulative-dose mechanism of solar UVR carcinogenesis [[1], [2], [3]]. In contrast, melanoma risk appears more strongly associated with intermittent high-intensity recreational exposure patterns rather than chronic occupational exposure. The observed phototype gradient is consistent with established biological mechanisms and supports the inclusion of host susceptibility factors in causal attribution models.

Importantly, moderate uncertainty in meta-analytic relative risks did not substantially modify the ranking of causal attribution, demonstrating the stability of the FAIR framework under uncertainty propagation. Sensitivity analysis indicated that baseline incidence (p₀) accounted for the largest proportion of model variance (68%), followed by uncertainty in relative risk estimates (18%), highlighting the importance of accurate regional cancer registry data for reliable attribution estimates. The use of Gamma priors for baseline incidence improved calibration across heterogeneous registries, while the directed acyclic graph (DAG) specification clarified causal relationships and aligns with contemporary causal inference approaches in occupational epidemiology.

4.5. Future directions

Several avenues may further strengthen the framework. First, external validation using real-world occupational disease compensation datasets (e.g., Comcare in Australia, the French CRRMP network, the Israeli National Insurance Institute) would provide valuable evidence regarding predictive performance. Second, the use of personal UV dosimeters and wearable sensors may allow more accurate characterization of long-term occupational UV exposure. Third, integration of biological susceptibility markers — including genetic variants related to pigmentation or DNA repair — may improve modeling of inter-individual susceptibility.

The FAIR methodology is readily extendable to other occupational carcinogens, provided that key epidemiological inputs are available: (i) meta-analytic estimates of relative risk; (ii) baseline incidence from reliable registries; (iii) identification of major cofactors; and (iv) exposure characterisation. Priority exposure–disease pairs for future application include asbestos–mesothelioma, crystalline silica–lung cancer, diesel exhaust–lung cancer, and benzene–haematological malignancies, where robust epidemiological evidence and compensation relevance are well established.

4.6. Limitations and boundaries of appropriate use

Several limitations should be acknowledged. The model relies primarily on secondary epidemiological data from published studies and national cancer registries. Variability in registry completeness and case ascertainment may introduce geographic bias, particularly for keratinocyte cancers, which are incompletely recorded in some regions. Epidemiological evidence remains limited for darker skin phototypes, reflecting lower incidence and underrepresentation in existing datasets. Individual UV dosimetry data were not available, limiting dose-response precision, and phototype-specific relative risks remain insufficiently characterized. Expanding registry data from underrepresented regions — including parts of Africa and Southeast Asia—would improve global applicability.

The current parameterisation should not be applied without local calibration in settings where cancer registries lack coverage or quality control, where keratinocyte carcinomas are underreported, or where occupational UV exposure patterns differ markedly from those represented in the underlying studies (for example, tropical regions with very high background exposure or workforces with predominantly dark phototypes). In such contexts, recalibration of baseline incidence, exposure prevalence, and cofactor distributions using local data is essential.

FAIR-derived probabilities are intended strictly as decision-support tools and should not replace holistic clinical assessment, detailed occupational history taking, or legal judgement. Over-reliance on numerical thresholds without contextual interpretation may lead to misclassification. Potential misuse includes applying the model without appropriate regional adjustment, or interpreting probabilistic outputs as definitive proof of causation rather than as quantitative support within a broader medico-legal assessment.

Despite these limitations, internal validation demonstrated strong model stability and predictive consistency, supporting the methodological robustness of the FAIR framework.

5. Conclusion

The FAIR framework provides a transparent and reproducible approach for estimating the probability that occupational solar ultraviolet exposure substantially contributes to skin cancer development. By integrating deterministic causal attribution, stochastic uncertainty propagation, and preventive economic modeling, the method bridges the gap between epidemiological evidence, clinical causation assessment, and occupational compensation systems. Beyond solar UV exposure, the FAIR methodology offers a scalable framework for addressing causal attribution in other multifactorial occupational diseases. With further validation and refinement, such approaches may contribute to more consistent and equitable recognition of occupational health risks.

CRediT authorship contribution statement

Eytan Ellenberg: Writing – original draft, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Arieh Ingber: Writing – review & editing, Validation, Methodology. Akshaya S. Bhagavathula: Writing – review & editing, Methodology, Formal analysis, Data curation. Eldad Katorza: Writing – review & editing, Supervision. Yuval Ramot: Writing – review & editing, Validation, Methodology.

Funding

None.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Appendix A. Supplementary data

The following is the Supplementary data to this article: