Smoking Patterns of US Mothers and Their Childrens' Smoking
Smoking Patterns of US Mothers and Their Childrens' Smoking
We study intergenerational transmission of smoking using data from a large populationrepresentative survey of US mother-child dyads and by employing mixture latent trajectory analyses and multinomial logistic regression models.
Data were from the Children and Young Adults of the National Longitudinal Survey of Youth 1979 cohort (NLSY79-CYA), a public use panel survey of all offspring of women in a population-representative cohort (NLSY79) commissioned by the US Bureau of Labor Statistics. The NLSY79-CYA used a biennial, cohort-sequential design in which all children born to NLSY79 women by 1986 as well as all subsequent children born after 1986 have been followed. The NLSY79-CYA thus includes multiple birth cohorts and children per mother. We selected respondents aged 14 to 25 years observed at any of the biennial surveys between 1994 and 2006 (i.e., birth cohorts 1970–1992). The NLSY79-CYA yearly completion rates range from 83.0% to 88.4%. By 2006, 6643 youths aged 14 years and older were eligible for the NLSY79-CYA and had been located for at least1 interview between 1994 and 2006. From this sample, 6349 youths responded to questions about cigarette smoking at least once.
The youth smoking trajectory is characterized by using latent class analysis with a set of dichotomous reports of cigarette smoking in the past 30 days constructed for each respondent from biennial survey assessments via computer-assisted personal interviewing. Because the set of smoking in the past 30 days for each respondent is truncated (with the earliest possible report at 14 years), we validated trajectories using a retrospective assessment of the respondent's report of the age they first smoked cigarettes and most recent report of ever smoked cigarettes from the biennial computer-assisted personal interview.
We constructed several variables to describe maternal smoking patterns before, during, and after the pregnancy and birth of the respondent. We created mother ever smoked daily as a dichotomous indicator for any maternal report of daily smoking in the NLSY79 substance use history supplements taken in 1992, 1994, and 1998. We created mother smoked during pregnancy as a categorical indicator for mother's reported cigarette consumption (did not smoke, < 1 pack/day, or a combination of 1–2 packs/day and 2 or more packs/day) from the NLSY79 birth history taken within1year of birth for this study's sample. Because of notable item nonresponse (n=1792), we used an identical retrospective question in the 2004 NLSY79-CYA to confirm reliability across the 2 assessments and to fill nonresponses. We created mother ever smoked daily by smoked during pregnancy to distinguish between respondents whose mother never smoked daily or during pregnancy, smoked daily but not during pregnancy, smoked daily and smoked during pregnancy less than 1 pack per day, and smoked daily and smoked during pregnancy 1 or more packs per day. We excluded from the sample respondents whose mother never smoked daily but smoked less than 1 pack per day during pregnancy (n=32). We created mother's smoking history to distinguish the full pattern of prepregnancy, prenatal, and postnatal exposures. This variable extends the previous composite variable by addressing the timing of initiation and cessation of daily smoking (reported and updated in the 3 NLSY79 substance use supplements) in relationship to the youth's date of birth. The 6 exposure categories are depicted in Figure 1: never smoked daily or during pregnancy (45.2%); quit daily before birth of child and did not smoke during pregnancy (7.4%); did not smoke during pregnancy but relapsed to daily smoking (10.0%); did not smoke during pregnancy, relapsed, but then quit daily smoking (6.6%); smoked any cigarettes during pregnancy and smoked daily but quit after birth (6.7%); and smoked any cigarettes during pregnancy and smoked daily after birth (24.2%).
(Enlarge Image)
Figure 1.
Distribution of prepregnancy, prenatal, and postnatal maternal smoking patterns experienced by US youths: NLSY79-CYA, 1994–2006.
The variables we used to control for sample design and selection into maternal smoking exposure groups entail youth sociodemographics (age at baseline, age at first smoking assessment, gender, and race/ethnicity), maternal sociodemographics (age at child's birth, and educational attainment and marital status when the child was aged 14 years), and maternal proclivity to health or risk behavior (breastfed, prenatal care, and a score of the mother's endorsement in 1980 of 12 adolescent delinquency behaviors from the NLSY79-modified Self-Reported Delinquency Interview). Summary statistics and item nonresponse are available in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org.
We first characterized youth smoking trajectories for youths aged 14 to 25 years using mixture latent trajectory analysis (LTA) to identify those with similar age patterns of smoking in the past 30 days. LTA uses statistical evidence rather than a priori assumptions to characterize trajectories. By using the multiple assessments of smoking in the past 30 days over the biennial surveys (1994–2006) to describe the smoking trajectory rather than 1 or more retrospective measures of a respondent's smoking history, our analysis also had the advantages of a time-sampling approach with reduced measurement error for the trajectory of smoking entries and exits.
Our LTA statistical model entailed parameters for class membership probabilities and class-specific variable endorsement probabilities for smoking at each age and thus did not impose any functional form on the age trajectory of smoking in the past 30 days within each of the smoking trajectory classes. Consequently, the model accommodated dynamic age patterns of entry, exit, and even relapse in smoking in the past 30 days. We used full information maximum likelihood estimation (which estimates model parameters in the presence of missing data) to appropriately model the biennial and cohort structure of the sample smoking in the past 30 days response patterns whereby 35% of the sample was surveyed more than 4 times, 23% 3 times, 23% twice, and 19% once (n=6349). We assessed model fit using the Akaike information criterion and sample size-adjusted Bayes information criterion. We estimated respondents' posterior probability of membership for each smoking trajectory class and assigned them the smoking trajectory class with the highest posterior probability.
We then considered the relationship between youth smoking trajectories and maternal smoking patterns. The sample entails no missing data on the maternal smoking pattern or variables used to control for selection (n=5027). We first assessed the bivariate relationship between maternal and youth smoking using descriptive statistics and unadjusted multinomial logistic regression models (Table 1). Then, we reported the adjusted odds ratios (AORs) for how prenatal and postnatal exposure to maternal smoking patterns influenced the odds of each youth smoking trajectory, adjusting for the variables we used to control for selection (Table 2). We tested gender differences in the relationship between youth and maternal smoking reported in Table 1 and Table 2 using interaction terms. We found no statistically significant gender differences (findings not shown).
All analyses used weights and corrected standard errors to address the complex sampling structure of the NLSY79-CYA and inclusion of siblings. We implemented LTA using Mplus 5.21 (Muthén and Muthén, Los Angeles, CA) and the multinomial logistic regression models using Stata/MP11(StataCorp LP, College Station, TX).
Methods
We study intergenerational transmission of smoking using data from a large populationrepresentative survey of US mother-child dyads and by employing mixture latent trajectory analyses and multinomial logistic regression models.
Data
Data were from the Children and Young Adults of the National Longitudinal Survey of Youth 1979 cohort (NLSY79-CYA), a public use panel survey of all offspring of women in a population-representative cohort (NLSY79) commissioned by the US Bureau of Labor Statistics. The NLSY79-CYA used a biennial, cohort-sequential design in which all children born to NLSY79 women by 1986 as well as all subsequent children born after 1986 have been followed. The NLSY79-CYA thus includes multiple birth cohorts and children per mother. We selected respondents aged 14 to 25 years observed at any of the biennial surveys between 1994 and 2006 (i.e., birth cohorts 1970–1992). The NLSY79-CYA yearly completion rates range from 83.0% to 88.4%. By 2006, 6643 youths aged 14 years and older were eligible for the NLSY79-CYA and had been located for at least1 interview between 1994 and 2006. From this sample, 6349 youths responded to questions about cigarette smoking at least once.
The youth smoking trajectory is characterized by using latent class analysis with a set of dichotomous reports of cigarette smoking in the past 30 days constructed for each respondent from biennial survey assessments via computer-assisted personal interviewing. Because the set of smoking in the past 30 days for each respondent is truncated (with the earliest possible report at 14 years), we validated trajectories using a retrospective assessment of the respondent's report of the age they first smoked cigarettes and most recent report of ever smoked cigarettes from the biennial computer-assisted personal interview.
We constructed several variables to describe maternal smoking patterns before, during, and after the pregnancy and birth of the respondent. We created mother ever smoked daily as a dichotomous indicator for any maternal report of daily smoking in the NLSY79 substance use history supplements taken in 1992, 1994, and 1998. We created mother smoked during pregnancy as a categorical indicator for mother's reported cigarette consumption (did not smoke, < 1 pack/day, or a combination of 1–2 packs/day and 2 or more packs/day) from the NLSY79 birth history taken within1year of birth for this study's sample. Because of notable item nonresponse (n=1792), we used an identical retrospective question in the 2004 NLSY79-CYA to confirm reliability across the 2 assessments and to fill nonresponses. We created mother ever smoked daily by smoked during pregnancy to distinguish between respondents whose mother never smoked daily or during pregnancy, smoked daily but not during pregnancy, smoked daily and smoked during pregnancy less than 1 pack per day, and smoked daily and smoked during pregnancy 1 or more packs per day. We excluded from the sample respondents whose mother never smoked daily but smoked less than 1 pack per day during pregnancy (n=32). We created mother's smoking history to distinguish the full pattern of prepregnancy, prenatal, and postnatal exposures. This variable extends the previous composite variable by addressing the timing of initiation and cessation of daily smoking (reported and updated in the 3 NLSY79 substance use supplements) in relationship to the youth's date of birth. The 6 exposure categories are depicted in Figure 1: never smoked daily or during pregnancy (45.2%); quit daily before birth of child and did not smoke during pregnancy (7.4%); did not smoke during pregnancy but relapsed to daily smoking (10.0%); did not smoke during pregnancy, relapsed, but then quit daily smoking (6.6%); smoked any cigarettes during pregnancy and smoked daily but quit after birth (6.7%); and smoked any cigarettes during pregnancy and smoked daily after birth (24.2%).
(Enlarge Image)
Figure 1.
Distribution of prepregnancy, prenatal, and postnatal maternal smoking patterns experienced by US youths: NLSY79-CYA, 1994–2006.
The variables we used to control for sample design and selection into maternal smoking exposure groups entail youth sociodemographics (age at baseline, age at first smoking assessment, gender, and race/ethnicity), maternal sociodemographics (age at child's birth, and educational attainment and marital status when the child was aged 14 years), and maternal proclivity to health or risk behavior (breastfed, prenatal care, and a score of the mother's endorsement in 1980 of 12 adolescent delinquency behaviors from the NLSY79-modified Self-Reported Delinquency Interview). Summary statistics and item nonresponse are available in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org.
Statistical Procedures
We first characterized youth smoking trajectories for youths aged 14 to 25 years using mixture latent trajectory analysis (LTA) to identify those with similar age patterns of smoking in the past 30 days. LTA uses statistical evidence rather than a priori assumptions to characterize trajectories. By using the multiple assessments of smoking in the past 30 days over the biennial surveys (1994–2006) to describe the smoking trajectory rather than 1 or more retrospective measures of a respondent's smoking history, our analysis also had the advantages of a time-sampling approach with reduced measurement error for the trajectory of smoking entries and exits.
Our LTA statistical model entailed parameters for class membership probabilities and class-specific variable endorsement probabilities for smoking at each age and thus did not impose any functional form on the age trajectory of smoking in the past 30 days within each of the smoking trajectory classes. Consequently, the model accommodated dynamic age patterns of entry, exit, and even relapse in smoking in the past 30 days. We used full information maximum likelihood estimation (which estimates model parameters in the presence of missing data) to appropriately model the biennial and cohort structure of the sample smoking in the past 30 days response patterns whereby 35% of the sample was surveyed more than 4 times, 23% 3 times, 23% twice, and 19% once (n=6349). We assessed model fit using the Akaike information criterion and sample size-adjusted Bayes information criterion. We estimated respondents' posterior probability of membership for each smoking trajectory class and assigned them the smoking trajectory class with the highest posterior probability.
We then considered the relationship between youth smoking trajectories and maternal smoking patterns. The sample entails no missing data on the maternal smoking pattern or variables used to control for selection (n=5027). We first assessed the bivariate relationship between maternal and youth smoking using descriptive statistics and unadjusted multinomial logistic regression models (Table 1). Then, we reported the adjusted odds ratios (AORs) for how prenatal and postnatal exposure to maternal smoking patterns influenced the odds of each youth smoking trajectory, adjusting for the variables we used to control for selection (Table 2). We tested gender differences in the relationship between youth and maternal smoking reported in Table 1 and Table 2 using interaction terms. We found no statistically significant gender differences (findings not shown).
All analyses used weights and corrected standard errors to address the complex sampling structure of the NLSY79-CYA and inclusion of siblings. We implemented LTA using Mplus 5.21 (Muthén and Muthén, Los Angeles, CA) and the multinomial logistic regression models using Stata/MP11(StataCorp LP, College Station, TX).
Source...