Akbarov, A, Williams, R, Brown, B, Mamas, M, Peek, N, Buchan, I and Sperrin, M (2015) A Two-stage Dynamic Model to Enable Updating of Clinical Risk Prediction from Longitudinal Health Record Data: illustrated with Kidney Function. In: Medinfo 2015: Proceedings of the 15th World Congress on Health and Biomedical Informatics. Studies in Health Technology and Informatics, 216 . IOS Press, pp. 696-700. ISBN 9781614995647

[thumbnail of SHTI216-0696.pdf]
SHTI216-0696.pdf - Published Version
Available under License Creative Commons Attribution Non-commercial.

Download (266kB) | Preview


We demonstrate the use of electronic records and repeated measures of risk factors therein, to enable deeper understanding of the relationship between the full longitudinal trajectory of risk factors and outcomes. To illustrate, dynamic mixed effect modelling is used to summarise the level, trend and monitoring intensity of kidney function. The output from this model then forms covariates for a recurrent event Cox proportional hazards model for predicting adverse events (AE). Using data from Salford, UK, our multivariate model finds that steeper declines in kidney function raise the hazard of AE (HR:1.13, 95% CI (1.05, 1.22)). There is a non-proportional relationship between the hazard of AE and the monitoring intensity of kidney function. Neither of these variables would be present in a classical risk prediction model. This work illustrates the potential of using the full longitudinal profile of risk factors, rather than just their level. There is an opportunity for deep statistical learning leading to rich clinical insight using longitudinal signals in electronic data.

Item Type: Book Section
Subjects: R Medicine > RC Internal medicine > RC870 Diseases of the genitourinary system. Urology
Divisions: Faculty of Medicine and Health Sciences > Institute for Science and Technology in Medicine
Depositing User: Mr Scott McGowan
Date Deposited: 21 Aug 2015 08:45
Last Modified: 22 Jun 2018 13:28
URI: https://eprints.keele.ac.uk/id/eprint/836

Actions (login required)

View Item
View Item