We appreciate the comments of Dowman and colleagues in reference to our study with orlistat in nonalcoholic steatohepatitis.1 Regarding study design, specifically with respect to the lack of a calculation for power, this study was designed as a pilot trial and not necessarily intended to definitively answer the question of whether orlistat therapy alone is efficacious. However, we have subsequently performed a retrospective power analysis using the values for weight loss reported in the article. The independent variables are treatment (control, orlistat) and time (before, after treatment). The dependent variable is body weight. The mean ± standard deviation (SD) of body weight before treatment is 225 ± 42 pounds. A 10% weight loss will be 22.5 pounds, an effect size of 0.54 SD. Four two-tailed, post hoc tests (two between and two within groups) are appropriate for this analysis, so a Bonferroni correction of P = 0.05/4 ≃ 0.01 was used. With SPSS Sample Power, version 2.0, we obtained a sample size estimate of 82 subjects per group for a power of 80% with a level of confidence of 95%. We concur that a larger sample size than was used in the article is needed to detect a 10% difference in weight loss between groups. With 23 subjects in one group and 18 in the other, we had the power to detect a 1.13 SD effect size or 25.4 pound difference in weight loss between groups. Presented as descriptive statistics, body weight in pounds was 226 (95% confidence interval [CI]: 153–299) in the orlistat group before treatment, 224 (95% CI: 132–316) in the control group before treatment, 208 (95% CI: 132–284) in the orlistat group after treatment, and 210 (95% CI: 120–300) in the control group after treatment. Regarding analysis, as previously noted in our article, there were no significant differences in histopathology between the two groups, so further case-by-case analysis was not thought to be helpful in this pilot trial. Subsequently, when no treatment effect was observed, the data were pooled and categorized by weight loss. The observed weight losses were 6.0% in the control group and 8.3% in the orlistat group. The 5% and 9% cut-points bracket the observed weight loss values and split the group sizes by a 2:3 ratio. Power analyses were not performed. Power is not an issue when statistically significant differences are found. However, increasing the number of hypothesis tests increases the probability of a Type I error. The results of Spearman correlation analysis between weight change ≥ 5% and changes in dependent variables were the quantitative insulin-sensitivity check index (QUICKI; r = −0.327, P = 0.039, n = 40) and steatosis (r = −0.388, P = 0.013, n = 40). The results of Spearman correlation analysis for weight change ≥ 9% were QUICKI (r = −0.385, P = 0.014, n = 40), steatosis (r = −0.453, P = 0.003, n = 40), ballooning (r = −0.326, P = 0.040, n = 40), inflammation (r = −0.321, P = 0.044, n = 40), adiponectin (r = 0.461, P = 0.002, n = 41), and the NAFLD activity score (NAS; r = −0.419, P = 0.007, n = 40). We agree that larger studies are needed to fully delineate the efficacy of orlistat in the treatment of nonalcoholic steatohepatitis. However, quantification of a specific weight loss goal that is associated with clinical and histopathologic improvement is also important. We thank Dowman and colleagues for giving us the opportunity to present the descriptive and nonparametric results. Stephen Harrison*, Elizabeth Brunt , Brent Neuschwander-Tetri , * Gastroenterology Service, Department of Gastroenterology, Brooke Army Medical Center, Fort Sam Houston, TX, Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, MO, Division of Gastroenterology and Hepatology, Saint Louis University, St. Louis, MO.
Read full abstract