The following baseline factors were assessed for their relationship to lost to follow-up (attrition) with logit link exchangeable correlation GEE models: intervention group, racial/ethnic group, year in high school, find FAQ wave of study, cigarette use in the prior 30 days, cigar use in the prior 30 days, dip/chew use in the prior 30 days, perceiving tobacco use to cause moderate/great harm, being a Future Farmers of America member, being an athlete (rodeo, football, baseball, track, soccer, basketball, and other), month of enrollment, Tobacco Use Prevention Education program at school, continuation school, school Academic Performance Index score, and school area poverty status. A multivariable lost to follow-up GEE model was developed from the individual factors with p �� .
05 with backward elimination of factors with p > .10 for a final model. We used logit link exchangeable correlation GEE models accounting for clustering within schools to test the hypotheses of no association between intervention group and ST cessation among baseline ST users since randomization provides the basis for valid inference. Odds ratios (ORs) and 95% CIs were calculated. Subgroup analyses assessed Baseline Factor �� Group interactions in GEE models; since the baseline data (Gansky et al., 2008) showed Smoking �� Race/Ethnicity differences in ST use, examining Baseline Smoking �� Intervention Group was of particular interest; other subgroup analyses assessed included continuation school, baseline first ST use within 30 min of waking, baseline number of ST uses per week, and phone/mail follow-up.
Analyses used participants with both baseline and 1-year follow-up tobacco use data. However, multiple imputation (MI) with 20 imputations and Markov chain Monte Carlo was used to impute missing data with the factors relating to loss-to-follow-up, baseline smoking status, intervention group, ST use at follow-up, and Baseline Smoking �� Intervention Group. GEE assumes that missing data are missing completely at random, while MI makes the less stringent assumption of data being missing at random, providing more robust inferences. Initiating a new form of tobacco was Dacomitinib assessed by comparing the percent of baseline exclusive ST users who reported using smoked tobacco at the 1-year follow-up with that of baseline exclusive smoked tobacco users who reported using ST at follow-up with exact binomial 95% CIs; nonoverlapping CIs would indicate a significant difference. Results Twenty-one of 29 rural counties (72%), 41 of 73 randomly selected high schools (56%; of the 217 eligible schools), and 4,731 of the 9,391 eligible students (50%) participated in the study. At 1-year follow-up, 3,072 students participated for 65% retention.