First Years Away From Home: Letting Go and Staying Connected

Overview

Alcohol abuse is the leading cause of death and serious injury among college students, and students also experience significant harms from other types of substance misuse and risk behaviors. The proposed project is a randomized controlled trials that will test the protective effects of Letting Go and Staying Connected, a handbook for parents of students who are transitioning for the first time from home to college, the time when students are at greatest risk. The handbook encourages parent skill development and good management of their student's new independence, providing a clear framework to guide them in parenting at this stage. Targeted outcomes include reduction of substance use and risk behaviors. The primary hypothesis is that students who are in one of the two handbook conditions with their parents will report lower substance use and risk behaviors in the two years after college entry.

This is a randomized controlled trial (RCT) with two experimental conditions (Handbook and Handbook+) and a control. Each condition will be composed of 450 student/parent dyads, about half in Cohort 1 and half in Cohort 2. Participants will be recruited from the incoming freshman classes at Washington State University in 2017 and 2018 by the Social Development Research Group (SDRG) at University of Washington. Parents in both experimental conditions will receive a handbook in the summer before school starts. The handbook contains suggested conversation starters, checklists, and activities to be completed by the parent/student dyad. Activities are based in theory of social and emotional development and designed to address risk and protective factors for this age group as they transition to living independently from parents. Handbook plus parents will additionally receive text messages periodically throughout the fall and spring semesters, some of which are designed to let them know about major university events (e.g. home football games) which are known to be high-risk periods. Each cohort will be followed for two years. Parents complete surveys at baseline (spring before students starts) and in fall semester. Students complete surveys at baseline and each semester for their first two years at school. Specific Aims of the project are as follows: Aim 1: Test the efficacy of Letting Go and Staying Connected, a theoretically guided, developmentally targeted, and empirically supported handbook intervention for parents of students transitioning to college. The investigators hypothesize that students in the Handbook conditions will engage in fewer risk behaviors, including initiation, use, and abuse of substances (alcohol, marijuana, opiates, and prescription stimulants) and HIV/sex risk behaviors. As a result, Handbook students will also report fewer physical, emotional, social, and academic harms. The investigators will test a control condition vs. a "Handbook Only" and a "Handbook Plus" condition (an enhanced version of the intervention that adds strategically-timed Short Message Service and/or emails designed to increase utilization of the intervention). Aim 2: Test theoretically specified mediational pathways between Handbook utilization and the expected intervention changes in risk behaviors/associated harms. The investigators expect that the effect of the Handbook on student outcomes will be mediated by both parent and student behaviors, as described in the following path: 1) Parent engagement in and use of the intervention 2) Parent practices in family management, autonomy support, and emotional support 3) Students' perceived rewards for healthy behaviors 4) Healthy connections with parents and peers 5) Healthy beliefs and clear standards leading to 6) Reduced substance use and sex risk behaviors; increased academic performance. Aim 3: Identify specific attributes of parent-student communication that predict and covary with student substance use and HIV/sex risk behaviors. Early in their first semester, the investigators will survey students intensively over 2 periods of 7 days to learn about their daily communication with parents and their daily risk behaviors. Because little is known about parent-student communication at this developmental stage, the investigators will conduct descriptive and exploratory analyses on the content, emotional tone, timing, duration, mode, and frequency of parents' communication with their students and examine how those aspects of communication relate to student risk behaviors and associated harms at both a between-person and a within-person level. The SDRG has extensive experience in recruitment, data collection, data management, and data security. The investigators' detailed Data Safety Monitoring Plan contains information about specific procedures used to safeguard participant information. The investigators will recruit underrepresented students in proportion to Washington State University (WSU) student percentages. SDRG also has extensive experience in retaining study participants, in some studies for more than 20 years. Data cleaning. The investigators will conduct tests for out-of-range values, consistency across measurement points, normality, and influential outliers. The investigators will address any issues of bias related to outliers using accepted techniques (e.g. analyses robust to violation of assumptions). Counts of dichotomous substance use variables are frequently skewed; the investigators will use log transformations of variables and appropriate analyses (e.g. survival analysis or Poisson regressions) as needed. Selection and Attrition. Randomized selection of students for invitation into the study should ensure a representative sample of the student population, and numerous studies have demonstrated the ability of the Survey Research Division at the SDRG to retain over 90% of respondents over more than 10 years. However, the investigators will test for differential participation and attrition and for the likelihood of biased outcome estimates; if necessary, the investigators will model those effects and employ correction techniques. Variability in Implementation and Dosage. Particularly in the case of a self-administered intervention, implementation and dosage are not distributed randomly -- for example, parents in the intervention condition who are particularly motivated or compliant may do all activities, whereas others may pick and choose. The non-random distribution of intervention components across intervention participants confounds the ability to draw causal inferences about program effects and obscures the mechanisms of program effects by treating implementation as a binary event. In addition to intent-to-treat analyses the investigators will employ recent methodological advances that allow for estimation, modeling, and control of selection effects in implementation through use of propensity scoring and inverse probability weighting. Missing Data. When respondents skip items or miss one or more data collection points, missing data may introduce bias and increase the likelihood of Type I errors. The investigators' primary analytic approaches use Full Information Maximum Likelihood (FIML)(38) to manage missing data. In case of excessive missingness, the investigators will use multiple imputation . Family-wise Error. The investigators will use summary variables of risk behaviors/harms, but the investigators will also conduct separate analyses for each risk outcome to provide a finer-grained look at individual substances, increasing the likelihood of a Type 1 error. Responses to substance use items are also likely to be highly correlated; in such cases standard Bonferroni corrections tend to be overly conservative with a higher likelihood of Type 2 error. Resampling procedures (e.g. Statistical Analysis Software (SAS) Procedure MULTTEST) are more powerful than the Bonferroni test and can be applied to categorical, continuous, and censored data tests, mixed method analyses of between-subjects effects and longitudinal within-subject tests(15,148). A global test statistic will be used when appropriate. Analyses: Aim 1: To measure intervention impact on change across time the investigators will model latent growth curves (LGC) for each of the hypothesized outcomes. First, unconditional models will be estimated to confirm the expected escalation across 5 waves of data. The investigators expect a two factor model: an intercept factor set to pre-test levels, and a linear slope factor estimated with factor loadings set to indicate pre-test as intercept and non-equal intervals between surveys. A non-linear slope will be estimated using quadratic forms for loadings, however the investigators expect primarily linearly increases over all 5 time points. A covariance is typically estimated between the intercept and slope factors to control for pre-test levels of the outcome. Control variables will be added in the next step. Finally, intervention group assignment (in the form of 2 dummy variables) will be added as predictors of the mean and variance of the slope factor (random assignment should preclude group effects on the intercept set at pre-test). The investigators expect these parameters to reflect significant positive slope means and variances in the unconditional model and the minimally conditional model. In the final models, intervention dummy variables are hypothesized to have significant negative effects on slope mean and variance compared to the control condition indicating both intervention conditions reduced the rate of escalation in risk behaviors and consequences and also reduced the variability in these outcomes. By manipulating the coding of dummy variables the investigators can then test the comparison between H and H+ with the expectation that H+ parameters will be significantly more strongly negative than those for H. Aim 2: Preliminary intent-to-treat analyses (ANCOVA) will be conducted on mediators at each follow up as described for outcomes above. Then LGC analyses described above will be extended to include mediating variables. However, prior to estimating mediation, each mediator will be modeled over time in a LGC model to determine its shape (intercept, linear slope, and quadratic slope) conditioned on control variables as described for outcomes in Aim 1. In the case of mediators the slope factors are less predictable, although in general these positive parenting and parent-student relationship factors are expected to decline over time. Once appropriate models of change have been determined, concurrent/parallel growth processes in mediators and outcomes will be estimated. Parallel process Longitudinal Growth Modeling (LGM) is a technique that has been used to examine interrelationships among various developmental outcomes, including co-occurring drug use and delinquency, in adolescents and young adults and for testing the mechanisms of preventive intervention effects. In the final step the intervention group status dummy variables will be added to the parallel process models. Aim 3: Given the exploratory nature of this aim, there are no hypotheses. Note that changes in communication are not posited to mediate intervention effects. However, if significant relationships are found between intervention status and communication measures, exploratory analyses may be conducted on the control group only to better reflect etiological relationships un-confounded by possible intervention influences. Preliminary analyses will focus on describing levels and variability in content, emotional tone, timing, duration, mode, and frequency and developing appropriate summary scores to best reflect these factors based on reports from the Time Line Follow Back surveys. Latent class analyses (LCA) with maximum likelihood estimation and robust standard errors will be used to explore parent-student communication patterns using those summary scores. The investigators will derive empirically based classes of participants based on content, emotional tone, timing, duration, mode, and frequency. This will involve estimating a series of models to determine the number of unique classes that best fit the data by comparing the Bayesian Information Criterion, the Lo-Mendell-Rubin adjusted Likelihood Ratio test and the Parametric Bootstrapped Likelihood Ratio test across models and selecting the most parsimonious model with the best fit.