Work

147 Two-stage clustering of temporal network metadata for classification of postoperative periods for renal transplant paediatric patients using healthcare professional-patient interactions: a cohort study

Year: 2025

Type: conference-abstract

Abstract: Aim Characterising postoperative periods after major procedures for paediatric patients can be challenging due to the diversity of the patients and complexity of temporal data. Our hypothesis is that the use of temporal network techniques based on Healthcare Professional (HCP) patient interactions can provide useful additional knowledge about this critical period.Method A retrospective cohort study was conducted analysing patient postoperative periods following renal transplantation between May 2019 and June 2024. Data from the hospital Electronic Patient Record system for HCP interactions with patients was used. Temporal networks for each day of admission within sessions (defined by whether patients were in the Intensive Care Unit or standard hospital ward) were constructed. First stage clustering categorized days of care using k-means, with network metadata providing fourteen features, for three to ten clusters. Second stage clustering categorized the postoperative periods, using days up to day twenty-one as the features, categorized by the results of stage one, for three to ten clusters. Silhouette coefficient was used to determine optimum cluster number and feature importance was calculated using XGBoost.Results From 144 patients 2,300 networks were constructed. Stage one clustering resulted in five clusters with main contributing features being: number of clusters (within the network), number of edges, closeness centralization and diameter. Stage two clustering of postoperative periods identified three groups of patients: cluster one (57.1%) five to 10 days, cluster two (35.1%) from nine to 16 days and cluster three (7.8%) 14 to 21 days. The main contributing factors being the key decision days, were days nine, 11, 17 and day six.Conclusion Two-stage clustering of temporal network metadata provides a novel approach to identification and classification of in-patient postoperative periods for complex procedures. HCP patient interactions data provides deeper understanding postoperative stay, which in turn impacts clinical operations and discharge planning.Acknowledgements for Funding or Support John Booth is Funded by CIRP via GOSH Children’s Charity Award VS0618. The study and researchers were part funded through the NIHR GOSH Biomedical Research Centre.

Source: BMJ Paediatrics Open

DOI: 10.1136/bmjpo-2025-gosh.79