Topic Modeling of Nursing Documentation in Hemodialysis Units: A Mixed-Methods Study of Nursing Surveillance Activities.

AIM AND

OBJECTIVES

This study aimed to characterize documentation patterns in hemodialysis nursing records using integrated text mining and nursing surveillance analysis, focusing on surveillance priorities and documentation imbalances.

BACKGROUND

Hemodialysis patients require continuous nursing surveillance due to frequent intradialytic complications; however, systematic analyses of nursing documentation in this setting remain limited.

DESIGN

A mixed-methods study integrating quantitative text mining and structured analysis of nursing surveillance activities.

METHODS

Free-text nursing records (300,321 entries) generated during maintenance hemodialysis sessions for 188 adult patients at a tertiary hospital in South Korea (July 2020-June 2025) were analyzed. Keyword frequency analysis and latent Dirichlet allocation topic modeling were performed after preprocessing, resulting in an eight-topic model. Nursing surveillance activities were mapped to the 16-item Korean Nursing Surveillance Scale through sentence-level classification by two independent researchers.

RESULTS

The top 30 keywords accounted for 75.3% of all extracted terms, indicating highly repetitive documentation. Terms related to dialysis procedures, vascular access, blood pressure, vital signs, bleeding, falls, pain, dizziness, and hypotension were frequently documented, whereas education-related terms were less frequently documented. Topic modeling identified three overarching domains: the hemodialysis treatment process, clinical assessment and monitoring, and safety and risk management. Surveillance item analysis demonstrated high documentation frequencies for vascular access, vital signs, bleeding, and hypotension or dizziness, with low documentation of patient self-management and education.

CONCLUSION

Hemodialysis nursing documentation primarily emphasizes physiological surveillance and safety management, with relatively limited documentation of patient education and self-management support.

IMPLICATIONS FOR NURSE LEADERS: These findings suggest that optimizing electronic nursing record templates, along with incorporating structured prompts and natural language processing approaches, may support more balanced documentation and data-driven decision-making.