Refining Predictive Models for Neglected and Emerging Infectious Diseases

完善被忽视和新出现的传染病的预测模型

• 批准号: 10494778
• 负责人: Ye Shen
• 金额: $ 37.75万
• 依托单位: UNIVERSITY OF GEORGIA
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-09-21 至 2027-07-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10494778
• 关键词: Adopted; COVID-19; Collection; Communicable Diseases; Data; Data Analyses; Data Collection; Data Sources; Databases; Development; Emerging Communicable Diseases; Epidemic; Epidemiology; Equilibrium; Foundations; Future; Human; Intervention; Learning; Measures; Methodology; Modeling; Performance; Play; Research; Research Activity; Risk; Role; Schistosomiasis; Time; Update; Vaccines; Validation; base; climate data; computerized data processing; cost effective; data handling; disorder control; disorder prevention; epidemiologic data; epidemiology study; interest; machine learning method; method development; neglect; outcome prediction; prediction algorithm; predictive modeling; prognostic; seasonal influenza; statistical and machine learning; tool; wearable device

PROJECT SUMMARY Predictive models play an essential role in disease prevention and control. Recent advances in scientific research have allowed more thorough and in-depth data collection from epidemiological studies (e.g., GPS data, climate data, wearable device data). However, due to the many variables collected and the relatively short time frame for epidemiological data collection during some of the epidemics, missing information is unavoidable, and subsequent updates of the database may be necessary. How to incorporate data with partial information, i.e., with missingness, and predictors measured dynamically over time, into existing models to perform more accurate and efficient predictions remains a challenge. Recently, the PI and his team have developed predictive models for various purposes among several neglected and emerging infectious diseases, including schistosomiasis, COVID-19, and human seasonal influenza. While conducting these studies, we identified several practical issues prohibiting a broader implementation of the proposed models, such as missing data and a lack of adaptive mechanisms based on dynamic inflows of predictors. Existing models adopting the complete data analysis approach will significantly reduce the statistical power and cause potential bias. Moreover, predictive models applied in epidemiological infectious disease studies often rely on historical data collected up to a time point without taking into consideration of future data inputs. Meanwhile, the development in statistical and machine learning methods laid the foundation for new dynamic predictive models based on trajectory data, with recent progress in functional concurrent regression and incremental learning. However, these methodological advances have been poorly integrated into field applications. Even in recent COVID-19 research where advanced dynamic models have been developed, balancing the data flow and prediction window has not been well studied. In addition, existing models often require a large amount of variable collection, so a practical two-stage approach allowing limited data collection early on can be more time- and cost-effective. In this MIRA proposal, we aim at refining predictive models for several neglected and emerging infectious diseases. Specifically, three coherent projects with distinct research activities will be pursued, which include: 1) refining hotspot prediction models for schistosomiasis interventions; 2) development and validation of prognostic risk models for COVID-19 in the US, with methods development on missing data handling and functional regression for dynamic prediction; 3) development and validation of a vaccine benefits score for human seasonal influenza. The refined models are expected to be accompanied by new and more general predictive algorithms involving missing data processing and dynamic prediction mechanisms to enhance model performance and adaptability. The methodological development from this proposal will also inform other epidemiological studies with similar challenges and have a broader long-term impact beyond the scope of the infectious diseases covered in the currently proposed projects.

项目摘要 预测模型在疾病预防和控制中起着至关重要的作用。科学研究的最新进展 研究已经允许从流行病学研究中更彻底和深入地收集数据（例如，GPS 数据、气候数据、可穿戴设备数据）。然而，由于收集的变量很多， 在某些流行病期间，收集流行病学数据的时限很短， 这是不可避免的，随后可能需要更新数据库。如何将数据与部分 信息，即，随着时间的推移动态测量的预测因子， 进行更准确和有效的预测仍然是一个挑战。最近PI和他的团队 在几种被忽视的和新出现的传染病中开发了用于各种目的的预测模型， 包括血吸虫病、COVID-19和人类季节性流感。在进行这些研究时，我们 确定了妨碍更广泛地实施拟议模式的若干实际问题，例如 缺少数据和缺乏基于预测因素动态流入的适应机制。现有模型 采用完整的数据分析方法将大大降低统计功效， bias.此外，流行病学传染病研究中应用的预测模型往往依赖于历史数据， 在不考虑未来数据输入的情况下，收集到某个时间点的数据。同时 统计和机器学习方法的发展为新的动态预测奠定了基础 模型的基础上的轨迹数据，最近的进展，功能并发回归和增量 学习然而，这些方法上的进步并没有很好地融入实地应用。即使在 最近的COVID-19研究开发了先进的动态模型，平衡了数据流 而预测窗口的研究还不够深入。此外，现有的模型通常需要大量的 变量收集，因此允许早期有限数据收集的实用两阶段方法可能更多 时间和成本效益。在这个MIRA提案中，我们的目标是改进几个被忽视的预测模型， 新出现的传染病。具体而言，将开展三个具有不同研究活动的连贯项目， 继续进行，包括：1）完善血吸虫病干预的热点预测模型; 2）开发 在美国对COVID-19的预后风险模型进行验证，并对缺失数据进行方法开发 动态预测的处理和函数回归; 3）疫苗效益的开发和验证 人类季节性流感评分。改进后的模型预计将伴随着新的和更多的 涉及丢失数据处理和动态预测机制的通用预测算法， 增强模型性能和适应性。本建议的方法学发展还将 为具有类似挑战的其他流行病学研究提供信息，并产生更广泛的长期影响， 目前拟议项目涵盖的传染病范围。