Abstract Rationale Tobacco smoke exposure induces widespread biologically and clinically relevant epigenetic changes. Self-report and cotinine measurements are inexpensive, first-line tools for assessing exposure type and duration,but are limited by recall bias and cotinine’s short half-life. We aimed to develop a DNA methylation (DNAm)-based model that complements existing clinical tools in improving the identification of tobacco smoke exposure in children and adolescents. Methods We analyzed 670 participants aged 8-21 years with and without asthma from the Genes-environments and Admixture in Latino Asthmatics (GALA II) and Study of African Americans, Asthma, Genes, and Environments (SAGE II) cohorts, each with paired plasma cotinine, self-report tobacco exposure and whole-blood DNAm data. Tobacco exposure was assessed via self- or parent-report and plasma cotinine levels at enrollment. Participants reporting no exposure and with cotinine ≤ 1 ng/mL were classified as unexposed (n = 306), while those reporting any exposure with cotinine 1 ng/mL were classified as exposed (n = 112).Logistic regression with Least Absolute Shrinkage and Selection Operator (LASSO) models were trained on the n = 418 participants to predict exposure in the remaining n = 268 participants using 1,460 DNAm sites previously associated with tobacco smoke exposure. DNAm M-values were adjusted for age, age2, and asthma status, and models were adjusted for sex, ancestry (African and Native American), 10 residual cell composition components, and socioeconomic status. Results A total of 567 DNAm sites were selected from GALA II and SAGE II combined. The LASSO model achieved an area under the ROC curve (AUC) of 0.84 in the training subset (n = 418). Among participants with cotinine levels ≤ 1 ng/mL (n = 306), model-predicted exposure probabilities were significantly higher for those reporting exposure (n = 252) than for those reporting none (p = 1.0 × 10-12), with the DNAm classifier showing moderate discrimination (AUC = 0.67). Among those with undetectable cotinine (0 ng/mL), the predicted probabilities remained higher for self-reportedly exposed participants(n = 58) (p = 0.01; AUC = 0.61). Conclusions A DNAm-based classifier trained on paired cotinine and DNAm data effectively identified tobacco exposure, including latent or misclassified cases undetected by self-reports or cotinine alone. By quantifying epigenetic signatures of exposure, DNAm-based tools may augment traditional clinical and biochemical assessments, refine exposure classification, and advance precision medicine approaches for pediatric respiratory health. This abstract is funded by: NIH/NHLBI
Oseguera et al. (Fri,) studied this question.