SCH: Statistical Foundation and Predictive Modeling for Personalized Diabetes Management: Continuous Glucose Monitoring (CGM), Electronic Health Records (EHR), and Biobanks

动态血糖监测（CGM）允许在整个24小时周期内进行接近连续的血糖测量。临床试验表明，CGM的使用可以改善糖尿病患者的血糖控制，如减少严重的低血糖。传感器技术的快速发展、易用性和扩大的报销范围极大地促进了CGM的使用。然而，尽管CGM采用率增加，但常规临床实践中CGM数据的成功利用率仍然很低。CGM设备产生的大量数据以及CGM总结报告对患者长期受益的明确转化价值的缺乏阻碍了其采用。这严重限制了CGM在个性化糖尿病护理方面的全部潜力。在该项目中，通过将CGM数据与患者的健康数据（包括药物、实验室测量和合并症）相结合，PI建议开发一套新的数据科学方法，用于构建强大且可解释的预测模型，以早期检测和预防短期和长期不良糖尿病结局。糖尿病疾病异质性、风险因素轨迹和现代器械中的数据不确定性将在CGM数据建模中考虑。高度可扩展的算法将进一步提高CGM的临床价值。该项目将为糖尿病患者及其医生提供更智能和自动化的帮助，以实现最佳的糖尿病管理。通过利用与临床医生和行业合作伙伴的合作研究，该项目有可能大幅推进可穿戴健康传感器领域。该项目还将为下一代统计学家和数据科学家的专业发展提供许多跨学科的机会，让参与的学员接触到最先进的智能健康数据科学技术。该项目还将积极促进STEM领域更加多样化和包容性的气候。 虽然CGM可以捕获动态血糖谱并在临床实践中发挥越来越大的作用，但它们的测量高度依赖于环境和行为因素，并且容易出现测量误差。作为CGM数据的补充，电子健康记录（EHR）和生物库提供了额外的信息来量化健康状况，疾病进展和相关的随时间变化的风险因素。这些大规模、多模式的数据来源使前瞻性研究能够详细收集长期时间依赖性暴露信息，以评估影响疾病发作的风险因素。到目前为止，由于计算成本过高，还没有统计方法可以同时分析传感器数据和大规模实时疾病发作。该项目的具体目标包括CGM轨迹的稳健联合建模、随时间变化的风险因素、大规模事件发生时间数据以及大规模复发事件。此外，将开发新的方法，用于动态预测不良结局，包括CGM轨迹、患者病史和基因组生物标志物。开发的工具和算法将以公开可用的软件以及手机应用程序的形式实施，以促进CGM的使用。该项目的成果预计将对我们社会的健康和福祉产生深远的影响。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Continuous glucose monitoring (CGM) allows near-continuous glucose measures throughout the 24-hour cycle. Clinical trials have shown CGM use can improve glycemic control of diabetes patients, such as reducing severe hypoglycemia. The rapid advances in sensor technology, ease of use, and expanded reimbursement dramatically promote CGM usage. However, despite increased CGM adoption, successful utilization of CGM data in routine clinical practice still remains low. Enormous amounts of data produced by CGM device and a lack of clear translational value of CGM summary reports for patients’ long-term benefits hinders its adoption. This severely limits realizing CGM’s full potential for personalized diabetes care. In this project, by integrating CGM data with patients’ health data, including medications, lab measures, and comorbidity conditions, PIs propose to develop a set of new data science methods for building robust and interpretable predictive models for early detection and prevention of short-term and long-term adverse diabetes outcomes. Diabetes disease heterogeneity, risk factor trajectories, and data uncertainty in modern devices will be considered in modeling CGM data. Highly scalable algorithms will further enhance the clinical value of CGM. The project will facilitate more intelligent and automated assistance for diabetes patients and their physicians to achieve optimal diabetes management. By harnessing the collaborative research with clinicians and industry partners, the project has a potential to substantially advance the field of wearable health sensors. The project will also provide numerous interdisciplinary opportunities for professional development of the next generation of statisticians and data scientists, by exposing the involved mentees to state-of-the-art data science techniques for smart health. The project will also actively promote more diverse and inclusive climate in STEM. While CGM captures the dynamic glucose profile and plays an increasing role in clinical practice, their measurements are highly dependent on environmental and behavioral factors and subject to measurement errors. Supplementary to CGM data, electronic health records (EHRs) and biobanks offer additional information to quantify health conditions, disease progression, and the associated time-varying risk factors. These large-scale, multimodal data sources enable prospective studies with a detailed collection of long-term time-dependent exposure information to assess risk factors influencing disease onset. To date, there are no statistical methods that can simultaneously analyze sensor data and disease onset at scale in real-time due to forbidding computational costs. The specific thrusts of the project include robust joint modeling of CGM trajectory, time-varying risk factors, time-to-event data at scale; and recurrent events at scale. Furthermore, the new methodology will be developed for dynamic prediction of adverse outcomes incorporating CGM trajectories, patients’ medical history, and genomic biomarkers. The developed tools and algorithms will be implemented in a form of publicly available software as well as cell phone applications to facilitate the CGM usage. The results of the project are expected to have a profound impact onto health and wellbeing of our society.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.