In the past few years, mathematical models have been developed to predict weight loss based on energy intake and energy expenditure data. However, to our knowledge, these sophisticated differential equation models have not been used to calculate statistical power for long-term weight loss trials in humans. We therefore used these models to determine how many participants need to be enrolled to detect various levels of daily energy imbalance between treatment groups.
For our modeling, we used data from the Diabetes Prevention Program (DPP), one of the largest NIH-funded weight loss trials. Using NIH’s Body Weight Planner and Pennington Biomedical Research Center’s Weight Loss Predictor, we modeled weight loss as a function of the degree of daily energy imbalance (e.g., 200 kcal/day). The statistical software G*Power was then used to calculate the sample sizes required to detect between-group differences in daily energy imbalance. Finally, we used an SQL query of the AACT database for ClinicalTrials.gov to identify all weight-related studies and then determined what fraction meet or exceed the derived sample sizes. Trials were included in the analysis if they were limited to adults and reported weight or BMI as one of the primary outcomes.
Sample sizes of 100, 172, and 400 participants are needed to detect between-group differences in energy balance of 200, 150, and 100 kcal/day, respectively, for a 6-month, two-arm trial. Only 49%, 31%, and 14% of weight loss studies registered in ClinicalTrials.gov have sufficient power to detect differences of 200, 150, and 100 kcal/day between treatment groups.
We conclude that a majority of weight loss studies are underpowered to detect weight loss differences less than or equal to 200 kcal/day. Therefore, previous weight loss studies may have failed to detect clinically meaningful differences between competing weight loss interventions. Future weight-related studies need to be designed with significantly higher sample sizes.