Causal relationship between two occupational carcinogens and lung cancer

This directed acyclic graph is used to inform the estimation of the causal effect of lifetime cumulative exposure to asbestos (main exposure) on lung cancer (outcome) in the context of individual participant data (IPD) meta-analysis of the SYNERGY pooled case-control study [1]. Co-exposure to other occupational lung carcinogens is represented with respiratory cristaline silica (RCS) as a motivating example. True exposure is unobserved (X_asb_true and X_rcs_true) and thus estimated with the job exposure matrix SYN-JEM (asbestos_cum0 and silica_cum0), resulting in loss of individual variation (Epsilon), and exposure error terms (U) that are independent of outcome status (Berkson-type error) [2]. The minimal set of adjustment for this model consists of age (agegroup), sex, smoking status and time since quitting (time_quit), lifetime smoking intensity (packyears), and study (study_name). Study is considered a proxy for other study-level covariates and differences in baseline lung cancer risk. Occupational lung carcinogens are assumed to be related through the common cause industry which serves as a proxy for shared working environment conditions. This backdoor pathway was intentionally not closed as co-exposure to occupational lung carcinogens is frequent [3]. Estimating conditional effects by setting such co-exposures to zero could violate positivity (if certain covariate combinations never occur for a given job title) and consistency (counterfactual scenarios such as high asbestos exposure with zero RCS are implausible in some industries). The model assumes monotonicity, implying that asbestos cannot prevent lung cancer through any causal mechanism, and no-interference (only individual exposure is relevant for the effect and not affected by others' exposures). Lastly, both age and sex are matching factors in the individual case-control studies, thus potentially causing selection bias (not shown in this DAG for simplicity). Adjustment for these matching variables through unconditional logistic regression results in unbiased estimates under the rare disease assumption [4].

Additional details and documentation on this analysis are available at: https://github.com/UtrechtUniversity/PoC-Asbestos

References:
1 International Agency for Research on Cancer (IARC). About SYNERGY. [cited 2024 Oct 17]. SYNERGY: Home. Available from: https://synergy.iarc.who.int/about
2 Armstrong BG. Effect of measurement error on epidemiological studies of environmental and occupational exposures. Occupational and Environmental Medicine 1998; 55: 651–656.
3 Olsson A, Bouaoun L, Schüz J, et al. Lung Cancer Risks Associated with Occupational Exposure to Pairs of Five Lung Carcinogens: Results from a Pooled Analysis of Case-Control Studies (SYNERGY). Environmental Health Perspectives 2024; 132: 17005.
4 Kamenetsky ME, Keil AP. (Re-)match: adjusting for matching factors in case-control studies may be unnecessary or insufficient. American Journal of Epidemiology 2025; kwaf116. doi: 10.1093/aje/kwaf116