Yarkiner, Zalihe (2015) Developing longitudinal models for monitoring chronic diseases in computerised general practice (GP) records : a case study in chronic kidney disease (CKD). (PhD thesis), Kingston University, .
Abstract
Analysis of longitudinal data is a rapidly growing field of statistical analysis, in response to the increasing availability of longitudinal datasets in many disciplines. Longitudinal studies are becoming more popular as they allow investigation of the same individuals over time, and where both within-individual and between-individual differences can be examined. Since the study of change over time is necessary in many areas, longitudinal studies and meaningful analysis of longitudinal data is essential. The health sector is one such area where longitudinal research is playing an increasingly important role. The aim of this research is to examine statistical methodologies for the analysis of longitudinal medical data, specifically General Practice (GP) records. All General Practices (GPs) in England and Wales are now computerized and routinely record detailed patient information, hence providing a rich longitudinal dataset. This research investigates new techniques and adaptations of existing methodologies to understand and explain patterns of change and the natural development and treatment of chronic diseases within routinely collected GP data. The data used here, although taken from a raw sample of 129 General Practice records, have been subjected to some cleaning and recoding in places, hence it should be considered as a secondary data source. Through out the data driven applications presented, different sub¬samples of the original dataset have been used. For the main part the full cleaned sample of 876951 patients is used where possible. Smaller samples ranging between 472 and 58675 patients are used depending on the outcome of interest and the availability of valid observations for the various applications employed. Mainly regression-based techniques, in two broad categories, were used to analyse the repeated measurements from each patient in our dataset. Firstly, linear and generalized mixed modelling approaches were used, whereas in the second phase of the project, the applications of semi-parametric and non-parametric approaches were investigated. The case study of particular interest in this research project is the incidence and progression of chronic kidney disease (CKD). There is a lack of knowledge and understanding of the natural history ofCKD and its progression over time. This project aims to address these issues. The advanced statistical models used in this research quantify how kidney function, assessed using estimated Glomerular Filtration Rate (eGFR), changes with respect to time and how other factors, including other related medical conditions (known as co-morbidities of CKD), affect kidney function and its change over time. The techniques and approaches used in this study are motivated by mixed model designs. The decline of kidney function as time progresses for typical CKD patients is observed to be non-linear. The type of nonlinear mixed models developed in this project do not assume that the decline of eGFR over time is linear, and hence are better able to model the progression of CKD than more traditional linear models. As a consequence, the proportion of the total variation in the outcome that can be explained by considering the patient level factors is tripled through the use of these non-linear models, showing they have much greater explanatory power than previous, simpler statistical models. The disease under study is Chronic Kidney Disease (CKD) but the methodologies should also be applicable other chronic, progressive diseases.
Actions (Repository Editors)
 |
Item Control Page |