Skip to main content
ARS Home » Northeast Area » Boston, Massachusetts » Jean Mayer Human Nutrition Research Center On Aging » Research » Publications at this Location » Publication #359155
Title: Machine learning approach to identify optimal criteria for successful weight loss maintenance on the basis of cardiometabolic risk factors at year 4 of the Look AHEAD Trial

Author
item BERGER, SAMANTHA - Jean Mayer Human Nutrition Research Center On Aging At Tufts University
item HUGGINS, GORDON - Tufts University
item MCCAFFERY, JEANNE - Brown University
item JACQUES, PAUL - Jean Mayer Human Nutrition Research Center On Aging At Tufts University
item LICHTENSTEIN, ALICE - Jean Mayer Human Nutrition Research Center On Aging At Tufts University

Submitted to: Journal of Federation of American Societies for Experimental Biology
Publication Type: Abstract Only
Publication Acceptance Date: 4/1/2017
Publication Date: 4/1/2017
Citation: Berger, S., Huggins, G.S., McCaffery, J.M., Jacques, P.F., Lichtenstein, A.H. 2017. Machine learning approach to identify optimal criteria for successful weight loss maintenance on the basis of cardiometabolic risk factors at year 4 of the Look AHEAD Trial [abstract]. Journal of Federation of American Societies for Experimental Biology. 31(1):643.14.

Interpretive Summary:

Technical Abstract: Objective: To date, different criteria have been used to dichotomize individuals into successful weight loss maintenance and regain groups to assess factors associated with successful weight loss maintenance/regain. Within this context the amount of weight regain that can occur before benefits of weight loss are reversed is not well established. Our goal was to identify an optimal percentage cut point of successful weight loss maintenance by modeling different weight change percentages on the basis of cardiometabolic risk factors. Methods: We used publically available data from the Action for Health in Diabetes (Look AHEAD) trial, a randomized lifestyle weight loss intervention in overweight/obese individuals with type 2 diabetes. Analyzed was the subgroup of individuals in the intensive lifestyle intervention (ILI) who lost >/=3% of initial weight by end of the first year (n=1791). We used classification trees through machine learning to identify the optimal cut point to predict cardiometabolic risk at year 4 using the following % of weight change: final weight as % of initial weight, final weight as % of post-loss weight, and weight change in maintenance as % of initial weight loss. Cardiometabolic risk was defined using individual components of NHLBI criteria to diagnose metabolic syndrome. A separate classification tree was generated for each risk factor. One cut point was generated for each percentage by averaging results from each risk factor. Performance of each cut point was assessed based on % of individuals with higher cardiometabolic risk captured in the regain group. We controlled for age, sex, race, baseline body mass index, initial weight loss, baseline values of each risk factor and medication use. Results: When examining final weight as % of initial weight, the average optimized cut point to predict risk factors was -9.3% +/- 4.5%, indicating that staying at least 9.3% below initial weight is associated with lower cardiometabolic risk. Average risk accounted for by using this cut point was 73.9% +/- 16.8%. When examining final weight as % of post-loss weight, the average optimal cut point was -1.5% +/- 1.4%. Average risk accounted for using this cut point was 78.4% +/- 5.3%. When examining weight change in maintenance as % of initial loss, the optimal cut point was 48.3% +/- 23.8%, so maintaining >/= 48% of weight lost during the intervention was associated with lower cardiometabolic risk. Average risk accounted for using this cut point was 55.7% +/- 14.0%. Conclusions: When dichotomizing successful weight loss maintenance using final weight as % of initial weight, final weight as % of post-loss weight or weight change in maintenance as % of initial loss, maintaining >/= 10%, 1.5% and 50%, respectively, optimizes the amount of regain that can occur before benefits of weight loss are undone.