05. Logistic models with covariates were assumed to have at least as much
power as the Fisher exact test and so, for the a priori analyses, the 85% power was seen as a lower bound. An interim analysis was prospectively planned and executed by an independent interim analysis committee when approximately 85 patients per group had completed at least 4 weeks of treatment. No a priori stopping rules were developed, however, a prospective charter allowed the interim analysis committee to discontinue Trichostatin A one or more arms based on safety, tolerability, lack of efficacy, or business considerations. Randomization would continue until approximately 170 patients were enrolled in each of the remaining groups. The primary end point of clinical response was analyzed according to the final statistical analysis plan prospectively implemented before database lock and unblinding. Clinical response at weeks 4 and 12 was analyzed RO4929097 clinical trial via a logistic regression using a generalized linear mixed effects model (GLMM) with center as a random effect and baseline WAP and stool consistency scores as covariates. Patients with <5 diary entries within week 4 or week 12 were categorized as nonresponders for that week. No imputation of data was performed if a diary entry was missed. Odds ratios from the logistic regressions were used to determine
statistical significance of treatment effects as compared
with placebo. The end points of bowel movement frequency, urgency episodes, and incontinence were modeled using a GLMM with fixed effects of treatment, time, and the treatment by time interaction; respective baseline frequencies were fitted in the model as baseline covariates. Additionally, a random effect was fit with patients as sampling units to account for repeated measurements of each outcome. Because outcomes were counts of events and likely non-normally distributed, Aspartate the GLMMs were fit assuming a Poisson response distribution and a natural log link function.12 and 13 Other secondary and exploratory end points were modeled with similar GLMMs. IBS Global Symptom score was modeled as a normal response distribution (identity link) with fixed and random effects similar to count data, with the exception that the baseline covariate was not included because of not having collected a baseline assessment. IBS-SSS, IBS-QOL, and EQ-5D scores were modeled with fixed effects of treatment, time, and the treatment by time interaction, the baseline score, and a random effect to account for repeated measurements. The models for IBS-QOL and EQ-5D assumed a normal distribution and identity link. IBS-adequate relief was modeled like the IBS Global Symptom score (ie, with no baseline covariate), but assuming a binary distribution and logit link function.