Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.643060
Title: Analysis of patient-safety related data using statistical modeling
Author: Deng, Lisha
Awarding Body: Lancaster University
Current Institution: Lancaster University
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Abstract:
To improve the quality of healthcare service, in particular reducing unintended harm to patients during the delivery of the service, patient safety study has become an important topic since the 1990s. This thesis aims to make a contribution to the patient safety research through statistical modelling based on the analysis of incident reports. Analysis of incident report-based data can use time series methods of count data or point process methods. However, strictly speaking, point process models using exact data should be used, because estimates using point process methods will lead t.o more efficient estimates than using interval-censored count data which discarded information, in particular when the underlying intensity driven the process is very wiggly. We have provided a theoretical analysis using Poisson process as an example to illustrate the efficiency loss in Chapter 5. The thesis also illustrated four case studies related to patient safety data. Safety incident report study and Ventilator-Associated-Pneumonia (VAP) study used time series methods, in particular Poisson log-linear model, to study what factors influence the trends of incidence and whet.her t.he rates of incidence differs amongst different hospital sites. Methicillin-Resistant Staphylococcus aurens (IVIRSA) and Campylobacteriosis study used point process methods, in particular Poisson process models, to study the trends of the incident rates, and what. population groups have higher risk rate and whether the rates of incidence differ amongst hospitals. However, we assumed that the counts/ infections occurred independently, which might be unrealistic for time series/ infectious disease data sometimes, if dependence such as cross infections cannot be neglected. Therefore , we proposed a new method in estimating parameters of the Log-Gaussian Cox process which is often used for clustered events. The method uses importance sampling in conjunction with non-parametric intensity estimation. This method is computationally easier than the Markov Chain Monte Carlo (MCMC) approach. It also appears to be more efficient than the minimum contrast estimating method using the K-function and the pair correlation g-function in the simulation study when the intensity function is smooth.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.643060  DOI: Not available
Share: