Machine Learning-Supported Efficient VTE Risk Assessment using Routinely Collected Electronic Health Record Data
- 1Imperial College London - St Mary's Campus
Abstract
ObjectiveVenous thromboembolism (VTE) is a leading cause of preventable inpatient mortality, while the real-world performance of mandated risk assessment and the potential for automating using electronic health record (EHR) data remain unclear. We evaluated completion, concordance with structured EHR data, clinical validity, and feasibility of EHR-supported VTE risk assessment assisted by machine learning.
Materials and MethodsWe analysed 577,904 admissions and 726,896 VTE assessment forms across five NHS hospitals between 2015 and 2025. Concordance and clinical validity were evaluated in 87,118 completed assessments from 2023 to 2024. Machine learning models compared clinician-recorded and EHR-derived risk factors in 74,372 encounters with an assessment completed within 14 hours of admission.
ResultsOverall completion was 96.7%, and timely completion improved from 47.4% in 2015 to 90.5% in 2024. Agreement between forms and EHR data was good for common risk factors (e.g. age [≥] 60 years: precision 0.95, recall 0.77), but low-prevalence variables were often under-documented in the forms (e.g. critical care admission: precision 0.20, recall 0.14). Despite these discrepancies, form-derived thrombosis risk was associated with increased VTE incidence (adjusted OR 3.31, 95% CI 2.81-3.90). Models using EHR-derived factors available within 14 hours achieved an AUROC of 0.709, compared with 0.704 for clinician-recorded variables.
DiscussionMuch of the information the risk assessment requires already exists in structured EHR data, and EHR-derived variables contained discriminative information comparable to that in clinician-recorded factors.
ConclusionRoutinely collected EHR data could support real-time pre-population of structured VTE assessments, thereby reducing documentation burden while preserving clinical validity.
Related
- Monitoring Emergency Medical Services Frequent Use: Protocol for a Mixed-Methods Study Using Routinely-Collected Data
- Measuring mild cognitive impairment using routinely collected health-care data: A conceptual framework and guidance for embedded pragmatic clinical trials.
- Identifying Mild Cognitive Impairment from Routinely Collected Physical, Functional and Psychosocial Measures in Community-Dwelling Older Adults: A Comparative Machine-Learning and Regression Modelling Study in the Northern Japanese ORANGE Registry.
- People living with multiple long-term conditions have different pathways of unscheduled care in hospital: findings from an analysis of routinely-collected clinical data
- Ensemble-based support vector machine classifiers as an efficient tool for quality assessment of beef fillets from electronic nose data
BibTeX
@misc{li2026machine,
title = {{Machine Learning-Supported Efficient VTE Risk Assessment using Routinely Collected Electronic Health Record Data}},
author = {Z. Li and E. Yagis and A. Riad and O. Windrath-Carr and M. Arribas and T. Sodiq and K. Goldsmith and B. Glampson and K. Flott and G. Haji and Z. Khan and C. Baker and E. K. Mayer},
howpublished = {medRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.18.26360687},
url = {https://www.medrxiv.org/content/10.64898/2026.08.18.26360687v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Machine Learning-Supported Efficient VTE Risk Assessment using Routinely Collected Electronic Health Record Data
AU - Z. Li
AU - E. Yagis
AU - A. Riad
AU - O. Windrath-Carr
AU - M. Arribas
AU - T. Sodiq
AU - K. Goldsmith
AU - B. Glampson
AU - K. Flott
AU - G. Haji
AU - Z. Khan
AU - C. Baker
AU - E. K. Mayer
PY - 2026
DA - 2026/08/18
PB - medRxiv
DO - 10.64898/2026.08.18.26360687
UR - https://www.medrxiv.org/content/10.64898/2026.08.18.26360687v2
AB - ObjectiveVenous thromboembolism (VTE) is a leading cause of preventable inpatient mortality, while the real-world performance of mandated risk assessment and the potential for automating using electronic health record (EHR) data remain unclear. We evaluated completion, concordance with structured EHR data, clinical validity, and feasibility of EHR-supported VTE risk assessment assisted by machine learning. Materials and MethodsWe analysed 577,904 admissions and 726,896 VTE assessment forms across five NHS hospitals between 2015 and 2025. Concordance and clinical validity were evaluated in 87,118 completed assessments from 2023 to 2024. Machine learning models compared clinician-recorded and EHR-derived risk factors in 74,372 encounters with an assessment completed within 14 hours of admission. ResultsOverall completion was 96.7%, and timely completion improved from 47.4% in 2015 to 90.5% in 2024. Agreement between forms and EHR data was good for common risk factors (e.g. age [≥] 60 years: precision 0.95, recall 0.77), but low-prevalence variables were often under-documented in the forms (e.g. critical care admission: precision 0.20, recall 0.14). Despite these discrepancies, form-derived thrombosis risk was associated with increased VTE incidence (adjusted OR 3.31, 95% CI 2.81-3.90). Models using EHR-derived factors available within 14 hours achieved an AUROC of 0.709, compared with 0.704 for clinician-recorded variables. DiscussionMuch of the information the risk assessment requires already exists in structured EHR data, and EHR-derived variables contained discriminative information comparable to that in clinician-recorded factors. ConclusionRoutinely collected EHR data could support real-time pre-population of structured VTE assessments, thereby reducing documentation burden while preserving clinical validity.
ER -