Data were obtained from 644 participants of the TODAY
study at baseline, 616 participants at the 6-month assessment,
and 543 participants at the 24-month assessment (Table I).2
At baseline, participants were on average 14.2 years old (SD
2.1; range 10-18); 64.27% female; 22% Non-Hispanic
white, 43% Hispanic, and 35% Non-Hispanic black. Mean
BMI-for-age percentile was 97.8 (SD 3.2; range 70.4-100.0)
with the majority classified as obese (n = 576; 89.4%) and a
much smaller percentage as overweight (n = 60; 9.3%) with
only 1.3% (n = 8) in the normal-weight category. Mean
HbA1c was 6.05% (SD 0.84; range 4.3%-10.9%), and all
participants were on oral medications and not insulin at
baseline. No difference existed between participants
included and not included in the analysis at 6 months. At
24 months, those included in the analysis were on average
younger (13.8 2.0 years old vs 14.7 2.0 years;
P < .0001) and engaged in less-risky behavior at baseline
(9.8% smoke and drink, 3.3% smoke only, 16.3% drink
only vs 13.6 smoke and drink, 6.8% smoke only, 21.4%
drink only; P = .0138) compared with those not in the
analysis.
For establishing prevalence estimates in this cohort, we
found that at baseline 10.4% had ever tried both smoking cigarettes
and drinking alcohol, which increased over time
(13.5% at 6 months, 17.9% at 24 months). Percent trying
smoking only was similar across time (ranging 3.9%-4.1%),
and percent trying only alcohol increased (17.1% at baseline,
20.8% at 6 months, 26.2% at 24 months). The majority reported
not ever trying smoking cigarettes or drinking alcohol
at baseline (68.6%), 6 months (61.7%), and 24 months
(52.1%).
Analysis of covariates was undertaken in which we examined
smoking only, drinking only, and a composite variable
of both smoking and drinking (Tables II-IV; Tables II and
III available at www.jpeds.com). In general, the modeling
results did not differ among the 3 methods for grouping
outcomes, but exceptions are noted in Results and
Discussion sections. Cross-sectional analyses of associations
for health risk behaviors via the use of univariate (ie,
unadjusted) models were conducted for youth
psychosocial, parent/family, environmental, and biological
factors. P values for the association of independent factors
separately with health risk behaviors were examined at each
time point. Of all of the variables, only 3 (household
income, perceived personal safety, and HbA1c) were not
statistically significant at any time point (P < .05) and were
removed from further analyses.
Multivariate modeling indicated that age was significantly
related to risk behaviors at every time period in the positive
direction in all of the models (composite, smoking, and
drinking) (P < .001), ie, older participants were more likely
to report trying risk behaviors. Males were significantly