通过机器学习揭示减肥效果的关键因素。

Uncovering key factors in weight loss effectiveness through machine learning.

作者信息

Yang Hui-Wen, De la Peña-Armada Rocío, Sun Haoqi, Peng Yu-Qi, Lo Men-Tzung, Scheer Frank A J L, Hu Kun, Garaulet Marta

机构信息

Medical Biodynamics Program, Division of Sleep and Circadian Disorders, Brigham and Women's Hospital, Boston, MA, USA.

Division of Sleep Medicine, Harvard Medical School, Boston, MA, USA.

出版信息

Int J Obes (Lond). 2025 May 6. doi: 10.1038/s41366-025-01766-w.

DOI:10.1038/s41366-025-01766-w

PMID:40328924

Abstract

BACKGROUND/OBJECTIVES: One of the main challenges in weight loss is the dramatic interindividual variability in response to treatment. We aim to systematically identify factors relevant to weight loss effectiveness using machine learning (ML).

SUBJECTS/METHODS: We studied 1810 participants in the ONTIME program, which is based on cognitive-behavioral therapy for obesity (CBT-OB). We assessed 138 variables representing participants' characteristics, clinical history, metabolic status, dietary intake, physical activity, sleep habits, chronotype, emotional eating, and social and environmental barriers to losing weight. We used XGBoost (extreme gradient boosting) to predict treatment response and SHAP (SHapley Additive exPlanations) to identify the most relevant factors for weight loss effectiveness.

RESULTS

The total weight loss was 8.45% of the initial weight, the rate of weight loss was 543 g/wk., and attrition was 33%. Treatment duration (mean ± SD: 14.33 ± 8.61 weeks) and initial BMI (28.9 ± 3.33) were crucial factors for all three outcomes. The lack of motivation emerged as the most significant barrier to total weight loss and also influenced the rate of weight loss and attrition. Participants who maintained their motivation lost 1.4% more of their initial body weight than those who lost motivation during treatment (P < 0.0001). The second and third critical factors for decreased total weight loss were lower "self-monitoring" and "eating habits during treatment" (particularly higher snacking). Higher physical activity was a key variable for the greater rate of weight loss.

CONCLUSIONS

Machine learning analysis revealed key modifiable lifestyle factors during treatment, highlighting avenues for targeted interventions in future weight loss programs. Specifically, interventions should prioritize strategies to sustain motivation, address snacking behaviors, and enhance self-monitoring techniques. Further research is warranted to evaluate the efficacy of these strategies in improving weight loss outcomes.

TRIAL REGISTRATION

clinicaltrials.gov: NCT02829619.

摘要

背景/目的：体重减轻的主要挑战之一是个体对治疗的反应存在巨大差异。我们旨在使用机器学习（ML）系统地识别与体重减轻效果相关的因素。

受试者/方法：我们研究了ONTIME项目中的1810名参与者，该项目基于肥胖认知行为疗法（CBT-OB）。我们评估了138个变量，这些变量代表参与者的特征、临床病史、代谢状况、饮食摄入、身体活动、睡眠习惯、昼夜节律类型、情绪化进食以及减肥的社会和环境障碍。我们使用XGBoost（极端梯度提升）来预测治疗反应，并使用SHAP（SHapley值加法解释）来识别对体重减轻效果最相关的因素。

结果

总体重减轻为初始体重的8.45%，体重减轻速率为543克/周，失访率为33%。治疗持续时间（平均±标准差：14.33±8.61周）和初始体重指数（28.9±3.33）是所有这三个结果的关键因素。缺乏动力是总体重减轻的最显著障碍，并且也影响体重减轻速率和失访率。保持动力的参与者比在治疗期间失去动力的参与者多减轻了1.4%的初始体重（P<0.0001）。总体重减轻减少的第二和第三个关键因素是较低的“自我监测”和“治疗期间的饮食习惯”（特别是较高的零食摄入量）。较高的身体活动是体重减轻速率更高的关键变量。