OPTION 1: MEAS TASK AREA A DIVERSITY SUPPLEMENT - SUPPLEMENTAL FUNDING, JANUARY 6, 2021-JANUARY 5, 2022.

本附录的第一个目标是确定一种使用营养信息来识别可理解食物的方法 预防心脏病的模式。我们将考虑将饮食与三条已建立的生物学途径联系起来 心血管疾病包括血压、体重指数和胆固醇。我们将比较各种监督和非监督的情况 方法构建可解释的饮食模式以预测这些心脏病危险因素中的每一个。使用的分析 MESA数据将得到模拟研究的补充，以展示在更广泛的数据范围内的表现。我们 我怀疑Lasso和稀疏多数据块偏最小二乘(SMBPLS)将执行得最好，因为它们都 结合变量选择和使用结果信息。我们将其与主成分分析进行比较 (PCA)，因为它是营养科学中使用的一种常见技术，可以为我们的比较奠定基础。 第二个目标是改进当前的统计方法，我们专注于具体扩展SMBPLS。而当 SMBPLS有助于分析饮食与心血管疾病的三条途径(血压、BMI和 胆固醇)，这种分析方法仅限于连续的线性终点。存在其他相关的端点，这些端点将 有趣的探索，例如冠状动脉钙化(CAC)评分和发生心血管事件的时间。尤其是， CAC与饮食之间的关系尚未完全确定，因此这可能是一项有价值的科学贡献。 尽管CAC得分是连续的，但它是异常分布的，因此它被分析为两个结果(无与 任何)在过去。因此，第二个目标的重点将是扩展SMBPLS，以便它能够进行更多分析 泛化端点。

The first goal of this supplement is to determine a way to use nutritional information to identify understandable food patterns that protect against heart disease. Three established biological pathways that we will consider relating diet to CVD are blood pressure, BMI, and cholesterol. We will compare how well various supervised and unsupervised methods construct interpretable diet patterns to predict each of these heart disease risk factors. The analyses with MESA data will be supplemented with simulation studies to demonstrate performance on a broader range of data. We suspect that Lasso and Sparse Multi-Block Partial Least Squares (SMBPLS) will perform the best, as they both incorporate variable selection and use outcome information. We compare these to Principal Component Analysis (PCA), as it is a common technique used in nutritional science and can ground our comparison for that audience. The second goal is to improve upon current statistical methods, and we focus on extending SMBPLS specifically. While SMBPLS is useful to analyze the relationship between diet and our three CVD pathways (blood pressure, BMI, and cholesterol), this analysis method is limited to continuous linear endpoints. Other relevant endpoints exist that would be interesting to explore, such as coronary artery calcium (CAC) score and time to CVD event. In particular, the relationship between CAC and diet has not been well established, so this could be a valuable scientific contribution. Although CAC score is continuous, it is unusually distributed, so it has been analyzed as a binary outcome (none vs. any) in the past. Thus, the focus of the second goal would be to extend SMBPLS so that it is equipped to analyze more generalized endpoints.