Social determinants of health: the need for data science methods and capacity.
Social determinants of health: the need for data science methods and capacity.
复制标题
健康的社会决定因素：对数据科学方法和能力的需求。
DOI：
10.1016/s2589-7500(24)00022-0
复制
发表时间：
2024
期刊：
影响因子：
--
通讯作者：
Hogan,Joseph
中科院分区：
文献类型：
--
作者：
Chunara,Rumi;Gjonaj,Jessica;Immaculate,Eileen;Wanga,Iris;Alaro,James;Scott-Sheldon,LoriAJ;Mangeni,Judith;Mwangi,Ann;Vedanthan,Rajesh;Hogan,Joseph
健康的社会决定因素包括影响人们出生、成长、工作、生活和衰老环境的力量、系统和条件。例如，种族主义和气候变化通过结构性因素影响生活质量和其他健康结果，这些结构性因素包括经济政策、社会规范和其他影响环境从而影响人们行为的因素。实际上，减轻卫生不公平现象需要注意其根本原因和邻近因素。与单独的医疗支出或生活方式选择相比，SDoH对生活质量和其他健康结果的影响更大。1,2因此，我们需要为分析模型（例如，用于评估暴露或干预措施对健康结果的影响的模型）提供信息，以解释、分析和实施对SDoH的干预措施，如绿地改善。在这篇评论中，我们强调了测量和分析SDoH的三个挑战，数据科学是一套跨学科的技能，通过负责任和有效地使用数据来做出判断和决策。本文简要介绍并详细阐述了所列出的三个挑战，包括数据科学方法的明确示例：适当地在多个层面捕获感兴趣的暴露所需的数据并不总是可用的，也不容易测量；与合并症等生物医学决定因素相比，SDoH对个人健康结果的影响较小；与健康结果相关的远端SDoH放置需要很长时间来观察其影响（在某些情况下需要几十年或几代人）。图中显示了SDoH在个人、社区、社会和政策层面的复杂相互作用，以及这些决定因素如何影响健康结果。第一个挑战是，在多个层面（如社会、政策等）适当地捕捉利益暴露所必需的数据并不总是可用或易于衡量。例如，卫生框架已得到加强，纳入了各种形式的歧视（如种族主义）4，但这些现象难以量化。以前的研究使用了可观察到的但不完美的歧视代理，例如社会地位（例如，通过使用种族来指代个人或群体的位置或地位，而不是种族主义的作用）或变量（例如，人口隔离），这些变量可能与所有个人的生活经验都不相关。数据科学方法可用于从没有指定格式的非结构化数据（例如，诊所记录）和其他不容易捕获或在现有数据库中容易获得的概念中创建区分措施。例如，自然语言处理和机器学习以及社交媒体数据的精确地理定位已被用于确定歧视性气候——结构性歧视的一个方面，体现在一个特定的地方。6 .制定这类措施，通过补充现有的数据类型，如调查中发现的歧视、人口普查中的隔离，确保更好地了解多种形式的歧视对健康的影响
Social determinants of health (SDoH) include forces, systems, and conditions that shape the environments in which people are born, grow, work, live, and age. Racism and climate change, for example, affect quality of life and other health outcomes through structural factors, including economic policies, social norms, and other factors that shape environments and consequently the behaviours of people. Indeed, mitigating health inequities requires attention to their root causes and adjacent factors. SDoH can have a greater effect on quality of life and other health outcomes than healthcare spendings or lifestyle choices alone. 1, 2 Accordingly, we need to inform analytical models (eg, those used to assess the effect of exposures or interventions on health outcomes) to account for, analyse, and implement interventions on SDoH, such as greenspace improvements. 3 In this Comment, we highlight three challenges to measuring and analysing SDoH for which data science—a cross-disciplinary set of skills to make judgements and decisions with data by using it responsibly and effectively—can be harnessed. The three challenges listed are briefly introduced and elaborated on, including clear examples of data science approaches to address: data necessary for capturing the exposure of interest at multiple levels appropriately are not always available nor easy to measure; SDoH are distal to individual health outcomes compared to biomedical determinants such as comorbidities; and the distal placement of SDoH in relation to health outcomes results in requires long periods of time to observe their effect (in some cases over decades or generations). The complex interplay of SDoH at individual, community, societal, and policy levels, and how these determinants operate to affect health outcomes is shown in the figure. The first challenge is that data necessary for capturing the exposure of interest at multiple levels (eg, social, policy, etc) appropriately are not always available or easy to measure. For example, frameworks of health have been enhanced to include forms of discrimination (eg, racism), 4 but these phenomena are difficult to quantify. Previous work makes use of observable but imperfect proxies of discrimination, such as social position (eg, refering to an individual’s or a group’s place or status by using race, instead of the acting force of racism) or variables (eg, population segregation) that might not be at scales relevant to the lived experience of all individuals. 5 Data science methods can be used to create measures of discrimination from unstructured data without a specified format (eg, clinic notes) and other concepts not easily captured or readily available in existing databases. For example, natural language processing and machine learning along with precise geolocation of social media data have been used to ascertain discriminatory climate—one aspect of structural discrimination, embodied by a particular place. 6 Creating such measures assures improved understanding of the effects of multiple forms of discrimination on health by complementing existing data types, such as perceived discrimination from surveys, segregation from census