Sas proc surveylogistic

Logistic regression was conducted to determine whether tobacco website exposure was associated with tobacco product use and susceptibility. Separate models were run for the six outcome variables: lifetime cigarette use, lifetime e-cigarette use, past-month cigarette use, past-month e-cigarette use, cigarette susceptibility, and e-cigarette susceptibility. The analyses of lifetime and past-month use included all respondents, whereas the analyses of susceptibility included never-users only. All analyses were controlled for age group, sex, race, perceived access to tobacco products, past-month mental health symptoms, past-month alcohol use, and past-month marijuana use as potential confounders (measures are described by McDonell, Comtois, Voss, Morgan, & Ries, 2009 (link)). SAS PROC SURVEYLOGISTIC was used to apply the sampling weights.

Both univariate and multivariable methods were used to identify factors associated with receipt of guideline care for initial therapy for those with nonmetastatic. The χ² tests were used to determine univariate differences and multivariable logistic regression modeling was used to estimate significant independent predictors of use of guideline-discordant care, including variables that were significant at P < 0.10 from the univariate results. The results of the logistic regression analyses were expressed as adjusted odds ratios (ORs) with 95% confident limits (CL). All results were weighted based on the sampling fraction to provide results that represented the source population of the sampled cases using SAS Proc Survey-Logistic and SAS Proc SurveyFreq.

The authors conducted mixed linear (SAS PROC MIXED) and logistic models (SAS PROC GLIMMIX) to assess differences in outcomes between intervention and control churches. Two outcomes initially conceptualized as continuous variables violated normality assumptions and were dichotomized: FV consumption (≥5 cups/day vs <5 cups/day) and FV opportunities (almost all of the time versus less frequent). For one outcome (FV intake), where the intraclass correlation was 0, church clustering was accounted for with robust SEs (SAS PROC SURVEYLOGISTIC). Otherwise, variance components were estimated for church-level random effects. Missing participant-level data were not imputed.
Attendee age, gender, and education, variables related to PA and HE in the literature, were selected as covariates. Self-reported cancer history also differed by study condition and was added as a covariate. Attendee race almost completely overlapped with church race, which was unequally distributed between study conditions. Therefore, analyses controlled for predominant race of church.
Effect sizes (Cohen’s d for continuous outcomes; OR for dichotomous outcomes) were computed comparing intervention and control churches. Improbable values were set to missing prior to analyses (≥8 cups/day of fruits, ≥8 cups/day of vegetables, ≥480 minutes/day of moderate PA, ≥240 minutes/day of vigorous PA).