Using DNA methylation to validate an electronic medical record phenotype for smoking

Area of Science:

• Genetics
• Bioinformatics
• Public Health

Background:

• Accurate phenotyping of smoking status from electronic medical records (EMR) is essential for genetic discovery.
• Established DNA methylation sites serve as a reliable criterion standard for smoking behavior.

Purpose of the Study:

• To compare three candidate EMR smoking metrics (most recent, modal, trajectories) for phenotyping smoking status.
• To evaluate the validity of these EMR metrics against DNA methylation data.

Main Methods:

• Utilized longitudinal EMR text notes from the Veterans Aging Cohort Study (VACS).
• Employed a validated algorithm to categorize smoking status (current, past, never).
• Compared EMR phenotypes using descriptive statistics, Spearman correlations, logistic regression, and AUC analyses with DNA methylation data from VACS Biomarker Cohort (VACS-BC).

Main Results:

• EMR smoking phenotypes varied by metric: current (33-53%), past (16-24%), never (24-33%).
• Modal and trajectory metrics showed the highest association (rho=0.89).
• All EMR metrics associated with DNA methylation sites (cg05575921, cg03636183), with trajectories showing the strongest association (p<0.001).

Conclusions:

• Longitudinal EMR smoking data, particularly when summarized as statistical trajectories, provide a valid approach for smoking phenotyping.
• The trajectory-based EMR smoking phenotype enhances validity for genetic studies.