To the Editor:

We read with interest the article by Harrison et al.1 on the use of Orlistat in overweight patients with nonalcoholic steatohepatitis (NASH). Their prospective randomized trial of 41 patients concluded that improvement in liver histology was not significantly different after 36 weeks of treatment with Orlistat/vitamin E compared to vitamin E alone. The mean weight loss between the two groups was higher in the Orlistat group but did not reach statistical significance (8.3% versus 6.0%). The authors therefore reanalyzed the data to compare those who lost < or ≥ 5% body weight or < or ≥ 9% body weight, respectively. Using these stratifications, they demonstrated that weight loss of 5% was associated with improvement in steatosis but not NAFLD activity score (NAS), whereas weight loss of 9% was also associated with an improvement in NAS.

Regarding design, there are no power calculations presented in the methods section, so there is no way of assessing if the trial was adequately powered to detect differences between the treatment groups. Also, three primary endpoints are listed; does this imply the study was powered for all three primary outcomes? We estimate that to detect a 10% reduction in weight could require up to 60 patients per group and question whether the study presented was adequately powered. The authors' data demonstrates greater loss of weight in the Orlistat group, but this did not reach statistical significance and could be due to power. With respect to the subgroup analysis, this does not appear to have been powered at the outset, and as such, descriptive statistics only should be presented.

Regarding analysis, it would be helpful if the authors could clarify why the percent weight loss is categorized into two further analyses: <5%, >5% and <9%, >9% and if these were preplanned or data-driven cutpoints. As a continuous measure, the percent weight loss should be analyzed as a continuous variable since loss of information and bias are introduced by dichotomizing variables. The statistical analyses presented use a mix of parametric and nonparametric approaches; given the small size of the study, then nonparametric statistics are more appropriate (Spearman's correlation, Mann-Whitney tests).

No significant difference was observed in the histological improvements seen between the two groups at 36 weeks; however, a detailed histopathological description was only provided for liver biopsies at the start of the study. We would be interested to see a more detailed description of the histopathological changes at follow-up of the two treatment groups.

We know that the degree of weight loss achieved with Orlistat is variable. Orlistat may provide additional histological improvement compared with a similar degree of diet-induced weight loss, but only in the subset of patients who achieve >5%–10% weight loss. Indeed, previous studies have suggested that Orlistat can improve hepatic steatosis beyond its effect on weight reduction,2 and exerts additional beneficial effects on inflammation and fibrosis.3

We feel it may be premature at this stage to discount a role for Orlistat in NASH. Given the magnitude of NASH in the West, there remains a pressing need for larger studies which are powered to detect differences in weight loss/NAS score.


  1. Top of page

Joanna K. Dowman M.D.*, Deborah Stocken C.Stat.†, Jeremy W. Tomlinson M.D., Ph.D.‡, Philip Newsome M.D., Ph.D.§, * Wellcome Clinical Research Fellow in Hepatology, Institute of Biomedical Research, Edgbaston, Birmingham, UK, † Senior Biostatistician, Cancer Research UK Clinical Trials Unit, Edgbaston, Birmingham, UK, ‡ Wellcome Clinician Scientist and Honorary Consultant Physician, School of Clinical and Experimental Medicine, Institute of Biomedical Research, University of Birmingham, Edgbaston, Birmingham, UK, § Senior Lecturer in Hepatology and Honorary Consultant Physician, Centre for Liver Research, Institute of Biomedical Research, Edgbaston, Birmingham, UK.