Drinking data were aggregated to the weekly level. The number of days/week of heavy drinking (i.e., ≥4 drinks in a day for women and ≥5 drinks in a day for men) and of abstinence were the primary outcomes. Generalized linear mixed models with a binomial distribution and logit link function were used to examine medication group differences in changes in these outcomes during treatment. The models included fixed effects for medication group, week, and the interaction between medication and week, and a random effect for intercept. “Week” was recoded by subtracting the number of weeks of the study (12 (link)) so that the test of treatment compared groups at the conclusion of the study, week 12, rather than at baseline. The interaction term tested for different rates of change in the outcome during the study.
Two sets of analyses of the number of heavy drinking and abstinent days were conducted. First, an intent-to-treat analysis included all 138 patients. In addition to examining changes in drinking over time, we conducted a responder analysis that examined the number of patients in each group with no heavy drinking days during the last four weeks of treatment, consistent with the approach recommended by the Food and Drug Administration (29 (link)). Second, we conducted a pharmacogenetic analysis that was limited to self-identified European-American patients (N=122) due to substantial population differences in rs2832407 allele frequency. Initially, we used the three-level genotype for rs2832407 by adding it to the linear mixed analysis. We then combined the AA and AC groups, comparing them with the CC group as a dichotomous genotype.
Timeline follow-back data were available for 92.4% (SD=22.7) of the 84 days of treatment [92.9% (SD=20.9) in topiramate patients and 91.9% (SD=24.5) in placebo patients]. To examine the impact of missing data, multiple imputation using Markov Chain Monte Carlo single chain, based on patient baseline characteristics and weekly drinking, was employed to create 10 imputed data sets. Models were re-run on the imputed data sets using SAS proc mianalyze.
Two sets of analyses of the number of heavy drinking and abstinent days were conducted. First, an intent-to-treat analysis included all 138 patients. In addition to examining changes in drinking over time, we conducted a responder analysis that examined the number of patients in each group with no heavy drinking days during the last four weeks of treatment, consistent with the approach recommended by the Food and Drug Administration (29 (link)). Second, we conducted a pharmacogenetic analysis that was limited to self-identified European-American patients (N=122) due to substantial population differences in rs2832407 allele frequency. Initially, we used the three-level genotype for rs2832407 by adding it to the linear mixed analysis. We then combined the AA and AC groups, comparing them with the CC group as a dichotomous genotype.
Timeline follow-back data were available for 92.4% (SD=22.7) of the 84 days of treatment [92.9% (SD=20.9) in topiramate patients and 91.9% (SD=24.5) in placebo patients]. To examine the impact of missing data, multiple imputation using Markov Chain Monte Carlo single chain, based on patient baseline characteristics and weekly drinking, was employed to create 10 imputed data sets. Models were re-run on the imputed data sets using SAS proc mianalyze.
