Diabetes
Methodological approaches for addressing medication usage in studying exposure to volatile organic compounds and HbA1c among oil-spill cleanup workers Hanna Jardel* Hanna Jardel Alex Keil Chantel Martin David Richardson Patricia Stewart Mark Stenzel Larry Engel Dale Sandler
Background Oil spill cleanup workers are exposed to volatile organic compounds including benzene, toluene, ethylbenzene, xylenes, and hexane (BTEX-H). BTEX-H may disrupt glycemic regulation processes, detectable with high glycosylated hemoglobin (HbA1c) concentrations used to diagnose diabetes mellitus. This study aims to characterize how BTEX-H exposure among oil-spill cleanup workers affects HbA1c change over time. However, this change can be obscured in participants who use HbA1c-lowering medications.
Methods Data are from the Gulf Long-term Follow-up Study– a prospective cohort of workers who participated in oil spill cleanup following the 2010 Deepwater Horizon disaster. Participants provided interview data, biological specimens, and anthropometric and clinical measurements at two time points (enrollment, 2011-2013 and follow-up exam, 2014-2016).
Participants could have started HbA1c-lowering medication before either time point and continued through followup. Excluding medication users would obscure trends in HbA1c levels; we sought analytic solutions that would use all participants’ data and censor HbA1c values under treatment. We evaluated two approaches: a) g-computation to predict HbA1c values, absent treatment, and linear regression and b) Tobit regression with inverse probability weighting (IPW), both after imputation to address covariate missingness. Both approaches assess the impact of quantile increases in exposure to the BTEX-H mixture and each individual chemical in relation to change in HbA1c over time.
Discussion We will present results of both strategies and discuss their strengths and limitations. Compared to Tobit regression with IPW, G-computation is computationally demanding; initial work suggests that it yields more stable point estimates. The lack of prior research addressing HbA1c-lowering medication usage in analyses necessitates formulation of strategies suitable to the available data, research question, and available computational power.