Using Machine Learning to Develop Just-in-Time Adaptive Interventions for Smoking Cessation

项目总结 移动技术在提供高度创新的动态戒烟方面具有巨大的潜力 干预措施。手机传感器、可穿戴技术，以及生态等实时数据收集方法 瞬时评估(EMA)使收集丰富的环境和生理指标成为可能 位置、心率和情绪等数据。环境和情景提示，如渴望和接近 对于其他人来说，吸烟是那些试图戒烟的人失误的高度预测，这表明失误的风险是 以直接的、动态的影响为特征的。及时自适应等新兴战略 干预措施(JITAI)，旨在利用通过移动技术提供的量身定制的支持来防止吸烟 最需要它的时刻。尽管研究已经确定了吸烟减少的前因，但基于 从EMA数据观察到，研究一直无法利用上下文和 现有技术提供的环境数据。考虑到动态影响对失误的重要性 风险，迫切需要能够准确识别最高失误风险时刻的策略来改进 戒烟干预。最近的研究证明了机器学习在预测个体方面的效用 行为。机器学习是一种稳健的数据分析策略，可以产生高度准确的预测 模型来自大型数据集，并可以实时自动适应新数据。这样做的总体目标是 应用是使用有监督的机器学习方法来开发一种自动算法来量化 吸烟在个人层面上有失误的风险。具体来说，我们的目标是：1)应用有监督的机器学习 方法量化戒烟的个体化风险，2)评估其可行性和初步 提供由机器学习驱动的个性化、即时自适应干预的有效性 实时预测吸烟失误风险。拟议的研究和培训计划将在 俄克拉荷马大学健康科学中心(OUHSC)和斯蒂芬森癌症中心(SCC)。培训 将侧重于增加机器学习方法的知识，并开展和分析联合技术援助， 这将促进拟议项目的完成。拟议的研究结果有可能 减少参与者和传感器需要的数据量和频率，从而能够开发 不那么繁琐的干预。预计这些目标的完成将产生初步数据以供参考 自动、动态干预，充分利用移动技术的优势进行测量 实时的个人行为和环境背景。

PROJECT SUMMARY Mobile technology has enormous potential for delivering highly innovative, dynamic smoking cessation interventions. Phone sensors, wearable technology, and real time data collection methods such as ecological momentary assessment (EMA) have made it possible to collect a wealth of environmental and physiological data such as location, heart rate, and mood. Environmental and situational cues such as craving and proximity to others smoking are highly predictive of lapse among those trying to quit, suggesting that lapse risk is characterized by immediate, dynamic influences. Emerging strategies such as just-in-time adaptive interventions (JITAI), aim to prevent smoking lapse using tailored support delivered via mobile technology in the moments when it is most needed. Although research has identified antecedents of smoking lapse based on observations from EMA data, studies have been unable to utilize the full spectrum of contextual and environmental data available with current technology. Given the importance of dynamic influences on lapse risk, there is a critical need for strategies that accurately identify moments of highest lapse risk to improve cessation interventions. Recent research has demonstrated the utility of machine learning to predict individual behavior. Machine learning is a robust data analytic strategy that can produce highly accurate predictive models from large datasets and can automatically adapt to new data in real time. The overall objective of this application is to use supervised machine learning methods to develop an automated algorithm to quantify smoking lapse risk at the individual level. Specifically, we aim: 1) to apply supervised machine learning methods to quantify personalized risk of smoking lapse, and 2) to evaluate the feasibility and preliminary effectiveness of delivering a personalized, just-in-time adaptive intervention driven by machine learning prediction of smoking lapse risk in real time. The proposed research and training plan will take place at The University of Oklahoma Health Sciences Center (OUHSC) and the Stephenson Cancer Center (SCC). Training will focus on increasing knowledge of machine learning methodology, and the conduct and analysis of JITAIs, which will facilitate completion of the proposed project. Results of the proposed research have the potential to reduce the amount and frequency of data needed from participants and sensors, enabling the development of less burdensome interventions. It is expected that completion of these aims will yield preliminary data to inform an automated, dynamic intervention that fully utilizes the strengths of mobile technology for measuring individual behavior and environmental context in real time.