Assessing smoking, like assessing and quantifying many other addictive behaviors, often suffers from recall bias, deliberate misreporting and even non-disclosure. This problem is particularly prevalent during pregnancy. For that reason, studies of effects of prenatal exposure to cigarettes frequently acquire both self- report and biologic assays (such as cotinine levels in urine or blood) of maternal smoking. However, although biological assays have been used to assert non-disclosure of smoking (simple """"""""smoker"""""""" versus """"""""non-smoker"""""""" status), little attention has been paid to using actual biological assays to calibrate self-reported measures. Methods for combining information from both self-report and bioassays would enhance the precision of smoking exposure measurement, and yield better understanding of the effects of smoking on a variety of physiological and psychological outcomes. For example, in pregnancy smoking studies, enhancing the quality of smoking exposure measurement could substantially advance the studies of teratologic effects of exposure on both physiological and behavioral development of offspring, as well as shed more light onto addictive behavior patterns in general. Recently, Dukic et al. (2007) have devised a method to combine self-report with biological assays (such as urine, serum or saliva cotinine), in order to extract a combined (and calibrated) measure of smoking exposure, based on the known metabolic models for decay of cotinine. Our findings thus far highlight the usefulness of using combined biological and self-reported measures as a predictor of child behavior outcomes. However, they also reveal some marked limitations of current designs for collection of biological measures of exposure in longitudinal studies, and the need for the development of better research protocols for assessing self-disclosure of smoking behavior. In the proposed research, some of the main limitation built into the original models will be alleviated in two main ways: 1) by proposing more realistic Bayesian models which incorporate metabolic and time variation, and information from other samples, and 2) by solving for optimal schemes for collection of biological samples. These results can have marked effects on research protocols designed to study effects of complex smoking behavior.
The aims of the work in this project are thus three-fold: (R21-1) to obtain better models for smoking exposure that rely on both biological assays and self-report measures;(R21-2) use these models to recommend better data collection schedules for future studies and (R33-1) to validate and improve these models by predicting outcomes suggested to be related to smoking exposure (youth problem behavior and nicotine addiction), while accounting for genetic variability, in other prenatal and youth behavior datasets. It is important however to note that although this project focuses on smoking during pregnancy, the methodological advances developed would be directly applicable to non-pregnant smoker populations as well.

Public Health Relevance

Though studies of effects of prenatal exposure to cigarettes frequently acquire both self-report and biologic assays (such as cotinine levels in urine or blood) of maternal smoking, little attention has been paid to methods for combining information from both sources in order to enhance the precision of exposure measurement. Both measures have their own source of bias -- single bioassay measure alone cannot reflect intricate metabolic mechanism over time, while self-report is subject to reporting, topographic, and metabolic biases - and information. This project is proposing to devise new Bayesian statistical models for prediction and calibration of smoking exposure measure, derived from combined biological and self-report information which can reflect metabolic differences among women and during pregnancy.

National Institute of Health (NIH)
National Institute on Drug Abuse (NIDA)
Exploratory/Developmental Grants (R21)
Project #
Application #
Study Section
Special Emphasis Panel (ZDA1-NXR-B (02))
Program Officer
Wanke, Kay
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Colorado at Boulder
Biostatistics & Other Math Sci
Schools of Arts and Sciences
United States
Zip Code
Hagar, Yolanda; Dignam, James J; Dukic, Vanja (2017) Flexible modeling of the hazard rate and treatment effects in long-term survival studies. Stat Methods Med Res 26:2455-2480
Land, Thomas G; Landau, Anna S; Manning, Susan E et al. (2012) Who underreports smoking on birth records: a Monte Carlo predictive model with validation. PLoS One 7:e34853
Fang, Hua; Dukic, Vanja; Pickett, Kate E et al. (2012) Detecting graded exposure effects: a report on an East Boston pregnancy cohort. Nicotine Tob Res 14:1115-20