Korvink Michael, Biondolillo Madeleine, Van Dijk Julie Willems, Banerjee Anjishnu, Simenz Christopher, Nelson David
ITS Data Science, Premier, Inc., Charlotte, NC, 28227, USA; Institute for Health and Equity, Medical College of Wisconsin, Milwaukee, WI, 53226, USA.
ITS Data Science, Premier, Inc., Charlotte, NC, 28227, USA.
Soc Sci Med. 2025 May;373:118025. doi: 10.1016/j.socscimed.2025.118025. Epub 2025 Mar 26.
Understanding social determinants of health (SDOH) as a complex system is necessary for designing effective public health interventions. Traditional expert-driven approaches to mapping SDOH relationships, when used in isolation, are susceptible to subjective biases, incomplete knowledge, and inconsistencies across different domains of expertise. Additionally, SDOH variables often contain overlapping information, making it difficult to isolate unique SDOH constructs. A data-driven approach integrating dimensionality reduction and causal discovery can provide a more objective framework for identifying and mapping SDOH factors within a causal system. The data-driven method may serve as a starting point to overcome potential research biases in the development of causal structures.
An observational study was conducted using census tract-level SDOH data from the 2020 Agency for Healthcare Research and Quality (AHRQ) database. Principal Component Analysis (PCA) was applied to derive latent constructs from 157 SDOH variables across 85,528 U.S. census tracts. The Greedy Equivalence Search (GES) algorithm was then used to identify dominant causal pathways between these constructs.
PCA-derived components explained substantial variance within each domain, with food access (71.1 %) and income (50.0 %) explaining the most within-domain variance. The causal graph revealed economic stability as a central determinant influencing education, employment, housing, and healthcare access. Education, access to care, and access to technology mediated many pathways.
Findings highlight the interconnected nature of SDOH, emphasizing financial stability as a foundational determinant. The role of digital equity in health outcomes is increasingly significant. The data-driven approach may serve as an important tool to support researchers in the mapping of SDOH causal structures.
This study demonstrates the utility of combining PCA and GES to uncover causal pathways among SDOH constructs. Developing causal systems using data-driven methods provides an enhanced method for conducting public health assessments, identify optimal intervention points, and informing policy development.
将健康的社会决定因素（SDOH）理解为一个复杂系统，对于设计有效的公共卫生干预措施至关重要。传统的由专家驱动的绘制SDOH关系的方法，若单独使用，容易受到主观偏见、知识不完整以及不同专业领域之间不一致性的影响。此外，SDOH变量通常包含重叠信息，使得难以分离出独特的SDOH结构。一种集成降维和因果发现的数据驱动方法，可以为在因果系统中识别和绘制SDOH因素提供一个更客观的框架。这种数据驱动方法可作为克服因果结构发展中潜在研究偏见的起点。
使用来自2020年医疗保健研究与质量局（AHRQ）数据库的普查区层面的SDOH数据进行了一项观察性研究。应用主成分分析（PCA）从美国85,528个普查区的157个SDOH变量中导出潜在结构。然后使用贪婪等价搜索（GES）算法来识别这些结构之间的主要因果路径。
PCA导出的成分解释了每个领域内的大量方差，食物获取（71.1%）和收入（50.0%）在领域内解释的方差最多。因果图显示经济稳定性是影响教育、就业、住房和医疗保健获取的核心决定因素。教育、医疗服务获取和技术获取介导了许多路径。
研究结果突出了SDOH的相互联系性质，强调金融稳定性是一个基础决定因素。数字公平在健康结果中的作用日益显著。数据驱动方法可作为支持研究人员绘制SDOH因果结构的重要工具。
本研究证明了结合PCA和GES来揭示SDOH结构之间因果路径的效用。使用数据驱动方法开发因果系统为进行公共卫生评估、确定最佳干预点和为政策制定提供信息提供了一种改进方法。
