TRAJECTORIES OF SMOKING AND ALCOHOL USE BY COLLEGE FRESHMEN
Chair: Richard Clayton
TRAJECTORIES OF SMOKING AND ALCOHOL USE BY COLLEGE FRESHMEN. Richard Clayton1, Craig Colder2, Brian Flay3, Lisa Dierker4, Members Of Tern5, 1University of Kentucky, Lexington, KY United States; 2University at Buffalo, Buffalo, NY United States; 3University of Illinois at Chicago, Chicago, IL United States; 4Wesleyan University, Middletown, CT United States; 5Tobacco Etiology Research Network, Lexington, KY United States
Although the initiation of cigarette and other alcohol use typically occur prior to age 18, there is evidence for considerable change in smoking and alcohol use behavior after this age. College may be a particularly important period to study smoking and alcohol use because it is a time when adolescents transition into a new social context where substance use is the norm. Some of the goals of the Tobacco Etiology Research Network (TERN) study of college freshmen were to: 1) obtain a detailed assessment of daily cigarette and alcohol use; 2) identify trajectories of smoking behavior and dependence, and 3) examine dynamic relationships between the use of tobacco and alcohol. Using a longitudinal design, daily assessments of cigarette smoking and alcohol drinking were collected during the entire first year of college for a large cohort of freshman (N=912) during the 2002-03 academic year (35 weeks). All 912 students had smoked at least a puff prior to entering the study, and 45% had smoked in the month prior to the study. Of the total sample, 636 students reported smoking one or more cigarettes during the freshman year. Each of the presentations will use data from subsamples of these 636 students to demonstrate 3 different views trajectories among these data. Craig Colder and colleagues use hierarchical linear modeling on data from 488 students who smoked at least 3 cigarettes during the year to model the daily, weekly, semester and year-long trajectory of smoking behavior. Brian Flay and collegues use growth mixture modeling of data from 193 subjects who reported smoking on at least 15% of the days during freshman year to model 7 different trajectories of smoking. Lisa Dierker and colleagues use bivariate time series analysis on data from 225 students who reported smoking or 10 or more occasions during the year to characterize detailed within-person smoking and drinking patterns. Finally, Richard Clayton, the chair of TERN, will discuss the implications of the reported results for future research, prevention and treatment.
THE PROXIMAL ASSOCIATION BETWEEN SMOKING AND ALCOHOL USE AMONG COLLEGE FRESHMAN. Lisa Dierker1, Elizabeth Lloyd-Richardson2, Marilyn Stolar3, Brian Flay4, Stephen Tiffany5, Linda Collins6, Mark Nichter7, Mimi Nichter7, Richard Clayton8, Members Of Tern8, 1Wesleyan University, Middletown, CT United States; 2Brown School of Medicine, Providence, RI United States; 3Yale University, New Haven, CT United States; 4University of Illinois at Chicago, Chicago, IL United States; 5University of Utah, School of Medicine, Salt Lake City, UT United States; 6Pennsylvania State University, University Park, PA United States; 7Univeristy of Arizona*, Tucson, AZ United States; 8University of Kentucky, Lexington, KY United States
Objective: This study was undertaken to evaluate the association between patterns of day-to-day smoking and drinking among college freshman. Using 210 days of weekly time-line follow-back diary data collected once a week using a web-driven survey, the authors examined the within-person relationships between smoking and drinking. Method: Participants were selected for the study based on at least some previous exposure to cigarettes and were asked to record their daily cigarette smoking and alcohol consumption across their entire freshman year. Reports of daily smoking and drinking were analyzed via bivariate time series procedures. Patterns were considered both within and across level of substance use. Results: Findings revealed a high degree of significant cross-correlations between smoking and drinking in which the amount of use of one substance could be predicted by past use of the other. For the majority of participants, smoking and drinking were positively associated with the alternate behavior within day as well as on past and future days. This association was most common at Lag 0 with 86% of the sample showing a significant cross-correlation between smoking and alcohol use within day. Notably, the majority of participants showed a relationship between smoking and drinking whether smoking precedes or follows drinking. Conclusions: Through day-to-day evaluation of naturalistic substance use, the present study provided unique insights into the co-occurrence and predictive relationships between smoking and drinking during the critical transition to college and across the first academic year. The use of bivariate time series methods in the consideration of cross-associations between smoking and drinking showed that the most common pattern of prediction was bidirectional. Future research is needed to establish the specific factors (i.e. third variables) and related mechanisms that may drive both behaviors.
THE NATURAL HISTORY OF COLLEGE FRESHMAN SMOKING. Colder Craig1, Donald Hedeker2, Brian Flaherty3, Brian Flay2, Linda Collins3, Elizabeth Loyd-Richardson4, Richard Clayton5, Members Of Tern6, 1SUNY at Buffalo, Buffalo, NY United States; 2University of Illinois at Chicago, Chicago, IL United States; 3Pennsylvania State University, University Park, PA United States; 4Brown University, Providence, RI United States; 5University of Kentucky, Lexington, KY United States; 6Tobacco Etiology Research Network, Lexington, KY United States
Objectives: Prior research suggests that many students who abstained from smoking in high school are likely to experiment with cigarettes when they get to college, and those who were light occasional smokers in high school are likely to become more frequent heavy smokers in college. Little is known about the longitudinal course of smoking during the first year of college, and this is a notable gap in the literature because what we know about smoking prior to age 18 may not apply to college smoking. The goal of this paper was to characterize the natural history of smoking during the freshman year of college, a period of substantial transition in multiple domains. Method: We used a longitudinal design and gathered daily assessments of smoking in a large sample (N=488) of college students during their freshman year. This micro-level assessment allowed us to not only model individual differences in trajectories of smoking during an entire academic year, but also to examine temporal variability across days, weeks, and semesters. Random effects binomial regressions were used to model trajectories of smoking. Results: Findings suggested a weekly cycle of smoking such that the probability of smoking was much higher on weekends (Friday and Saturday) than on weekdays. Laid over this weekly cycle was an overall trend for smoking to decline over the course of the year. Substantial individual variability in how smoking changed over time was observed. Conclusions: These findings provide new insights into college smoking, and suggest that the beginning of freshman year, particularly weekends, is a period of risk for tobacco use. This experimentation with tobacco declines over the course of the academic year for most students, but the significant individual variability observed in this study suggests that the longitudinal trajectories vary from student to student. That is, smoking likely declines, increases, or remains stable for different groups of students. Overall, our findings suggest that smoking among freshman college students is of concern, and an important direction for future research will be to identify risk and protective factors that are associated with smoking trajectories during this period.
TRAJECTORIES OF SMOKING DURING THE COLLEGE FRESHMAN YEAR. Brian Flay1, Eisuke Segawa1, Donald Hedeker1, Colder Craig2, Members Of Tern3, 1University of Illinois at Chicago, Chicago, IL United States; 2SUNY at Buffalo, Buffalo, NY United States; 3Tobacco Etiology Reseach Network, Lexington, KY United States
Objectives: 1). To classify students into a manageable number of meaningful trajectories that describe changes in their smoking behavior during freshman year, and 2). To identify classes of students who mainly increase or decrease their level of smoking during freshman year. Methods: We identified 193 students who reported smoking or not on at least 90 days during the 9 months of the academic year and for at least 7 days every month, and who smoked on at least 15% of the days for which they filed reports (or on 40% of days in any one month). We aggregated the daily data to monthly (mean numbers of cigarettes smoked per day) and conducted latent growth mixture model (LGMM) analyses of the monthly data. Some students increased their smoking levels during the year, some decreased and some did not change. We specified a common random intercept in order to give priority to change (both linear and quadratic), our major interest, rather than the average level of smoking (intercept). Results: The best fitting model was one consisting of 7 classes. Two classes consisted of students who increased their levels of smoking during the year: one from an average of 4.2 to an average of 8.2 cigarettes per day, and one from 0.5 to 2.2 cigarettes per day. Four classes consisted of students who decreased their smoking during the year: one from 12 to 3.2 cigarettes per day, one from 6.3 to 0.9 by mid-year and then back up to 1.4, one from 4.5 to 3.2, and one from 2.8 to 0.7. One class was essentially flat, beginning and ending at about 0.7 cigarettes per day with a dip to about 0.4 at midyear. Variability about these trajectories that differed by class will be described and explained. Conclusions: This study provides unique insights into the variability in levels of smoking during the course of the college freshman experience. The use of latent growth mixture modeling enables the separation of meaningfully different trajectories of smoking during a year of transition from adolescence to adulthood in a college environment. Contrary to prior reports, most of these students did not increase their level of smoking during the year. Of those that did, only a small proportion showed major increases. Among the majority of students whose level of smoking decreased during the year, most decreased by only small amounts, and most did not quit. Future research will investigate the predictors of class membership, and the long-term trends of each.
CONCURRENT 9, METHODS, Grouped papers
Measurement Biases in Selection and Estimation
Chair: C. Hendricks Brown
MEASUREMENT BIAS ACROSS HISPANICS AND NON-HISPANIC CAUCASIANS ON A STANDARDIZED MEASURE OF ALCOHOL ABUSE: EMPIRICALLY EVALUATING THE IMPACT OF OBSERVED MEASUREMENT BIAS. Adam Carle1, 1U.S. Census Bureau, Washington, DC United States
Measurement bias, a form of non-sampling error, occurs when individuals equivalent on a construct being measured, but from different groups, do not have equal probabilities of observed responses. Confirmatory factor analysis (CFA) is a commonly used quantitative model to examine bias. In the model, equations specify the relation between observed responses and the latent variable of interest. Measurement bias exists when the relevant parameters in these equations differ significantly across the groups. However, given the number of parameters included in the model, it is reasonable to expect that some parameters will differ significantly. Likewise, in large scale epidemiological research, it will often be the case that sufficient power will be present to identify even small differences as statistically significant. Partial measurement invariance refers to the case when some parameters are equivalent, while others are not. Under partial invariance, the possibility exists that statistically significant differences are not large enough to impact observed scores in a meaningful way. Unfortunately, no standard method exists and few guidelines are available to empirically evaluate the impact of partial measurement invariance. Using data from the National Longitudinal Alcohol Epidemiological Survey (NLAES), a nationally representative household survey of 42,692 adults, the current paper extends a recently proposed method to evaluate the impact of measurement bias. By studying group differences in selection accuracy as a function of measurement bias, the technique quantifies the impact of measurement bias on a standardized measure of alcohol abuse across Hispanics and non-Hispanic Caucasians. The general mathematical model is extended to CFA for ordered-categorical measures and the analytical approach is discussed with real data.
INDIVIDUALIZED RISK ESTIMATION AND THE NATURE OF PREVENTION. Christine Holmberg1, Mark Parascandola1, 1National Cancer Institute, Bethesda, MD United States
In recent years there has been a tremendous increase in the development and availability of “individualized” risk models that purport to provide individuals with a quantitative estimate of future disease risk. The development of these models has been aided by both the explosion of genetic information about cancer susceptibility and a growing body of epidemiologic data about a range of host and environmental risk factors, demonstrating that cancer risk does differ substantially between individuals. Risk models are being made available to the general public through websites, advertising, and the media as well as in clinical settings.
Risk models hold substantial promise for the practice of cancer prevention by helping to identify high risk populations and subsequently guide decisions about surveillance, possible treatments or further genetic or other testing. However, they have also come under criticism for creating new categories of disease-free but `at risk´ individuals. In order to assess the potential risks and benefits of these models for the practice of cancer prevention, we analyzed the debate over the interpretation of risk estimates. In particular, we focused on the Gail model for breast cancer risk assessment, which is the most widely used model for “individualized” cancer risk estimation.
Critics of the Gail model have argued that while it may appear to predict well at the population level, it does a poor job of predicting individual outcomes and identifying those individuals who are most likely to develop cancer. Moreover, some have claimed that risk prediction models do not really apply to individuals at all because they are based on population-level epidemiologic data. Additionally, there is a mounting critique which argues that the model primarily serves to create large populations of “at risk” individuals eligible for medical treatment rather than contributing to identifying causes of disease and primary prevention strategies.
While debate about the strengths and weaknesses of the Gail model initially appears to be a debate about methodology, we argue that, in fact, it is driven by a tension between two very different views of what prevention is. Does the practice of cancer prevention consist primarily of expanding the category of `diseased´ individuals, creating `at risk´ groups in need of medical intervention, or is it aimed at identifying causes so that they can be removed? It is important for prevention scientists to understand the underlying concerns driving this debate and its relevance for resource allocation and priority setting in cancer prevention.
A COMPARISON OF APPROACHES TO EXAMINE INTERVENTION EFFECTS OF THE OLWEUS BULLYING PREVENTION PROGRAM. Torbjorn Torsheim1, Dan Olweus1, 1University of Bergen, Bergen, Norway
The paper compares different approaches to examining the impact of a systematic intervention that is delivered to individuals in clusters. The empirical data for the paper originate from a study on intervention with the Olweus Bullying Prevention Program (OBPP). During the academic year of 1997/1998, the OBPP was implemented in 14 elementary and junior high schools, with 16 schools serving as comparison schools. Levels of bully/victim problems were measured in May 1997(T1), and in May 1998 (T2) after eight months of intervention. Hierarchical linear modelling was used to contrast four ways to estimate the effects of the intervention: (a ) a hierarchical linear four-level model using measurement occasion as the lowest level, and individuals, classrooms, and schools as higher-level units; (b) three-level 'residual-change' approach measuring group differences at T2, controlling for individual T1 status; (c) a three-level model on the raw change scores, and (d) a selection cohort approach comparing T2 measurements with age-equivalent groups measured at T1 as a baseline. Preliminary results indicated that all four approaches, although resting on partly different principles, arrived at roughly similar effect estimates for the intervention. However, the standard error for the treatment effect estimate was comparatively smaller for the selection cohort approach. Advantages and shortcomings of the four approaches will be discussed with a particular focus on flexibility, statistical power, and user friendliness.
CONCURRENT 10, DISSEMINATION, Organized symposia
ISSUES IN A POPULATION-LEVEL APPROACH TO STRENGTHENING PARENTING
Chair: Ron Prinz
ISSUES IN A POPULATION-LEVEL APPROACH TO STRENGTHENING PARENTING. Ron Prinz1, 1University of South Carolina, Columbia, SC United States
This symposium examines contemporary issues in the implementation and evaluation of large-scale population interventions aimed at strengthening parenting. In the field of prevention, parenting as the main focus rather than as a secondary one is a relatively unique approach, and addressing parenting at a population level even more unusual. This symposium is based on the premise that changes in parenting practices often are (or need to be) central to prevention for a number of youth areas including substance abuse, antisocial behavior and violence, child maltreatment, academic failure, depression and suicide, and risky sexual behavior. The presenters are all actively involved in large-scale population level projects related to parenting. Sanders and Calam describe the creation and evaluation of an innovative British television series on positive parenting and discuss the implications of population-based media strategies in prevention. Prinz and Sanders report on preliminary developments in the U.S. Triple P System Population Trial, with particular emphasis on quality/fidelity of training, initial penetration of the population, and issues related to multi-disciplinary dissemination. Embry describes the statewide effort called the Wyoming Parenting Initiative, which draws on multiple agencies, public and private sectors, parent advocacy collaborators, and flexible dissemination strategies. The discussant, Elizabeth Robertson, will discuss the presentations with respect to implications for prevention science, policymaking, and societal impact.
A CONTROLLED EVALUATION OF THE CHILD AND PARENT EFFECTS OF A TELEVISION SERIES ON POSITIVE PARENTING. Matthew Sanders1, Rachel Calam2, 1University of Queensland, Brisbane, Queensland Australia; 2University of Manchester, Manchester, United Kingdom
The first level of the Triple P positive parenting system involves the implementation of a comprehensive media strategy to normalise and de-stigmatize parenting. Such a strategy can include use of both print and electronic media. This paper describes a randomised controlled trial evaluating the effects of a six episode documentary series on parenting “Driving Mum and Dad Mad”, being shown on a major network in Britain. It involves 2000 parents being pre-assessed via a web-based survey on a series of self report measures of child behavior, parenting and parental adjustment. Parents were randomly assigned to either a television-alone condition or to television plus a structured self-directed program. We hypothesized that although both conditions would experience an improvement in parental self efficacy and lower levels of coercive parenting, parents in the enhanced TV condition would demonstrate greater improvements in child behavior, and parenting skills. The series captured the experiences of five families of children with severe behavior problems as their parents underwent an 8 session group behavioral family intervention program (Group Triple P). The television series captured daily interactions of children and parents over the course of intervention and the group process parents participated in as they learned positive parenting and discipline techniques. Results will be reported focusing on the reach of the program, the impact of the intervention on participating parents and families, the effects on children´s behaviors, and the interactions of parents viewing the series. Analyses examining socio demographic and pre-treatment characteristics of families as predictors, mediators and moderators of intervention effects will be presented. Implications for public health approaches to strengthening parenting skills will be highlighted.
POPULATION ROLLOUT IN THE U.S. TRIPLE P SYSTEM TRIAL: PRELIMINARY DEVELOPMENTS. Ron Prinz1, Matthew Sanders2, 1University of South Carolina, Columbia, SC United States; 2University of Queensland, Brisbane, Queensland Australia
The area of child-maltreatment prevention tends still to operate at individual and clinical levels rather than truly adopting a public-health population approach. One exception is the U.S. Triple P (Positive Parenting Program) System Population Trial. This trial is systematically testing the reduction of risk for child maltreatment and the strengthening of parenting at a population level through the implementation of the multi-level Triple P system. Eighteen medium-sized counties in South Carolina have been randomized to either Triple P or to a usual-services comparison condition. At this stage in the trial, over 500 practitioners across many disciplines and settings have acquired Triple P training. Given that the Triple P system has been imported to the U.S. from Australia, data will be presented on the continuity of quality/fidelity of training, for example with respect to the acquisition of consultation skills and the degree of practitioner satisfaction with training. Preliminary data on population penetration 18 months into the trial will be presented. Issues to be discussed include: (a) the synergy and integration of a multi-level, multi-setting approach to prevention; (b) challenges in broad dissemination while maintaining program integrity; (c) the invoking of a self-regulatory framework with parents, practitioner, and training staff; and, (d) considerations for population-level indicators.