Center for Weight, Eating, and Lifestyle Sciences (WELL Center), Drexel University, Philadelphia, USA.
Weight Control & Diabetes Research Center, Warren Alpert Medical School of Brown University, Providence, USA.
Transl Behav Med. 2019 Nov 25;9(6):989-1001. doi: 10.1093/tbm/ibz137.
Individual instances of nonadherence to reduced calorie dietary prescriptions, that is, dietary lapses, represent a key challenge for weight management. Just-in-time adaptive interventions (JITAIs), which collect and analyze data in real time to deliver tailored interventions during moments of need, may be well suited to promote weight loss by preventing dietary lapses. We developed OnTrack (OT), a smartphone application (app) that collects data on lapses and triggers of lapse, uses a continuously improving machine learning model to predict lapse risk, and delivers tailored interventions when risk is elevated. The current study evaluated the efficacy of OT against an active control in facilitating weight loss. Participants (N = 181) with overweight/obesity (MBMI = 34.32; 85.1% female; 73.5% White) were randomized to receive either the WW (formerly Weight Watchers) Beyond the Scale (BTS) digital program alone or WW plus OnTrack (WW + OT) for 10 weeks. In an unplanned, natural experiment, the WW program changed mid-way through the trial from BTS to a more flexible one, Freestyle (FS). A general linear model revealed a treatment condition × diet plan interaction (F[1, 173] = 9.68, p = .002) such that OT demonstrated greater efficacy only among those receiving BTS (weight loss MWW + OT = 4.7%, standard error [SE] = .55 versus MWW = 2.6%, SE = .80). Compared to FS, BTS WW + OT participants also reported considerably higher satisfaction with the intervention, engagement was higher, and algorithm accuracy was superior. Overall, results offer qualified support for OT and generally for machine learning-powered JITAIs that facilitate weight loss by predicting and preventing dietary lapses.
个体对低热量饮食处方的不遵守，即饮食失误，是体重管理的一个关键挑战。实时自适应干预（JITAIs）可以实时收集和分析数据，在需要的时候提供定制化的干预措施，可能非常适合通过预防饮食失误来促进减肥。我们开发了 OnTrack（OT），一种收集饮食失误和失误触发数据的智能手机应用程序（app），使用不断改进的机器学习模型来预测失误风险，并在风险升高时提供定制化的干预措施。本研究评估了 OT 相对于活跃对照组在促进减肥方面的效果。参与者（N=181）超重/肥胖（MBMI=34.32；85.1%女性；73.5%白人）被随机分配接受 WW（前身为 Weight Watchers）Beyond the Scale（BTS）数字计划单独治疗或 WW 加 OnTrack（WW+OT）治疗 10 周。在一项未计划的自然实验中，WW 计划在试验中途从 BTS 更改为更灵活的 Freestyle（FS）。一般线性模型显示治疗条件与饮食计划的相互作用（F[1,173]=9.68，p=0.002），表明只有接受 BTS 治疗的参与者才能看到 OT 的更大效果（体重减轻 MWW+OT=4.7%，标准误差[SE]=0.55 与 MWW=2.6%，SE=0.80）。与 FS 相比，BTS WW+OT 参与者还报告了对干预措施的满意度明显更高，参与度更高，算法准确性更高。总体而言，结果为 OT 提供了有条件的支持，为机器学习驱动的 JITAIs 提供了普遍支持，这些干预措施通过预测和预防饮食失误来促进减肥。
