Feasibility Assessment of Extracting Social Determinants of Health From Electronic Medical Records: A Multicenter Study Across Chinese Healthcare Institutions.

Background: Social determinants of health (SDoH) are pivotal in influencing health outcomes and disparities across various populations. Real-world data rich in SDoH information, such as electronic medical records (EMRs), can considerably enhance public health interventions. However, in Chinese medical practice, these non-clinical factors are often neglected, with many healthcare providers failing to recognize the importance of SDoH information in the improvement of patient care. The objective of this study is to explore the feasibility and effectiveness of extracting SDoH information from Chinese EMRs. Methods: We developed a China-specific SDoH classification framework by integrating findings from relevant research using real-world data. This framework was applied to over three million patient records from Chinese EMRs to examine the completeness and availability of SDoH-related attributes within relevant fields. We also developed a quantitative assessment framework for evaluating SDoH information in EMR fields. This two-dimensional evaluation system measures data completeness and availability using a three-level hierarchical scoring approach, progressing from basic to advanced criteria. Additionally, we analyzed variations in SDoH information extraction across different healthcare institutions. Results: Drawing on the literature and 2000 manually annotated EMRs, we established a standardized framework of SDoH factors, comprising 50 features tailored to the Chinese medical diagnostic and treatment environment. We analyzed over 5.6 million EMRs from 40 hospitals within the National Clinical Research Data Center and found that tables and fields in Chinese EMRs cover all six primary SDoH categories, encompassing 25 out of 50 specific attributes. However, data extraction feasibility was relatively poor, with only seven features being fully extractable, with most "social and community context" data missing. Our evaluation of 43 electronic health record fields containing SDoH data revealed significant disparities between completeness and availability metrics. The composite completeness score averaged 1.47 (95% confidence interval: 1.20-1.73) out of a maximum score of 3. Availability assessments demonstrated notably higher performance with a mean score of 2.14 (95% confidence interval: 1.83-2.44) out of a maximum score of 3. Conclusions: This research established a culturally adapted SDoH framework for China and demonstrated the feasibility of extracting SDoH attributes from Chinese EMRs. Although some SDoH information in EMRs still cannot be captured or requires more advanced data processing to be usable, Chinese EMRs contain a wealth of SDoH data, allowing us to use large existing EMR databases to support ongoing SDoH research. Natural language processing technology has a critical role in this process, underscoring the importance of medical informatics and current artificial intelligence techniques in medicine and public health. Our research lays a foundation for future SDoH studies in China, enabling more comprehensive research and encouraging the government and relevant agencies to focus on SDoH interventions.