Analysis of weight loss after bariatric surgery using mixed-effects linear modeling

Abstract

Background: The standard analysis of bariatric surgery weight outcomes data (using t tests) is well known. However, these uncontrolled comparisons may yield misleading results and limit the range of research questions. The aim of the study was to develop a valid approach to the longitudinal analysis of weight loss outcomes after bariatric surgery using multivariable mixed models. This study has a multi-institutional setting.

Methods: We developed a mixed-effects model to examine weight after gastric bypass surgery while controlling for several independent variables: gender, anastomotic technique, age, race, initial weight, height, and institution. We contrasted this approach with traditional uncontrolled analyses using percent excess weight loss (%EWL).

Results: One thousand one hundred sixty-eight gastric bypass procedures were performed between 2000 and 2006. The average %EWL at 1, 2, and 3 years was 71%, 79%, and 76%, respectively. Using weight as the outcome variable, initial weight and gender were the only independent predictors of outcome (p<0.001). %EWL was substantially less accurate than weight as an outcome measure in multivariable modeling. Including initial weight and height as separate independent variables yielded a more accurate model than using initial body mass index. In a traditional uncontrolled analysis, average %EWL was higher in women than men. However, average weight loss was lower, not higher, in women (p<0.001) in our multivariable mixed model. Height, surgical technique, race and age did not independently predict weight loss.

Conclusions: Multivariable mixed models provide more accurate analyses of weight loss surgery than traditional methods and should be used in studies that examine repeated measurements.