OBJECTIVES: Majority of research on social determinants of health factors is focused on the use of structured data for this analysis. Yet, social work notes provide a more open-ended overview because they contain information that is otherwise not recorded as structured fields. This study, to our knowledge, is the first attempt in performing topic modeling on social work notes recorded in a large healthcare system in the US to explore different topics of discussions that can be used for understanding the influence of social factors on health outcomes.

METHODS: We first retrieved the social work notes from the UCSF de-identified Caboodle Data Warehouse between 2012 and 2021. Notes that are duplicated or extremely short were excluded. Latent Dirichlet Allocation was used to identify the topics in individual social work notes. Topic coherence metric and Jaccard distance were implemented to decide the optimal clustering results.

RESULTS: The study included 0.95 million social worker notes. The topic modeling analysis identified topics related to social determinants of health risk factors including treatments, social support, mental health, family relationship, living condition, finance, abuse assessment. Among them, social determinants of health risk factors including social support, family relationship, living condition, finance, abuse assessment are mostly absent in structured fields. Moreover, topics present in the social work notes vary according to the sub-categories: telephone encounter notes contain more diverse information about education and consultative services, community resources while progression notes contain unique information about clinical decision, invasive surgery and abuse assessment.

CONCLUSIONS: This study revealed that the social work notes contain rich and unique information about social determinants of health factors. Without using notes, it would be impossible to consider these factors in analyzing health outcomes. Investigating the methodology of utilizing the unstructured social work notes might make a critical impact in clinical outcomes.