Modelling the Recovery Process After Severe Head Injury

Murray, Lilian S (1988) Modelling the Recovery Process After Severe Head Injury. PhD thesis, University of Glasgow.

Abstract

The problem considered in this thesis is the prediction of the quality of survival after severe head injury. A model of the recovery trend of the patient through time is derived and this model is used to predict ultimate outcome. Chapter 1 introduces the problem of prognosis in clinical decision making, and in particular, its importance in the context of severe head injuries. It identifies the need for a new statistical approach to this problem. Chapter 2 describes the development of the Head Injury Study data bank from the initial stages when terminology needed to be carefully defined to the present day. It gives a detailed description of the Glasgow Coma and Outcome Scales. The data collection methods are described along with the problems encountered in establishing a reliable data bank. Suggestions are given to minimise these problems. In Chapter 3 discriminant analysis is introduced and its terminology defined. The factors involved in variable selection, the problem of missing data and the assessment of the performance of a discriminant rule are discussed in general terms. Two major studies are described where the prediction of outcome after severe head injury is made using information from the Head Injury Study data bank: first the early work using an independence model, and then a comparative study which was carried out to assess the relative merits of different discrimination techniques. Chapter 3 finishes by illustrating that, while these methods are successful in the prediction of death or survival, a new approach is required to predict the quality of survival. Chapter 4 contains the work involved in modelling the recovery trend of the survivors. This is done by modelling the coma score through time. The first order autoregressive model which was initially adopted is described along with the modifications required to give an adequate decription of the data. Ways of reducing the number of parameters which need to be estimated are considered, as well as the effect of using a pseudo maximum likelihood approach to reduce the computation involved in obtaining the parameter estimates. Three methods which adequately model the recovery trend are obtained. Chapter 5 examines the performance of these methods by assessing their ability to predict the quality of survival. This assessment is based on the classification matrices and three separation measures (the error rate, average logarithmic score and average quadratic scores). How performance is affected by different priors and the 'jack-knife' technique is examined. The performance of the models incorporating trend is compared with that of other available models. Age is shown to have a substantial effect on the prediction of prognosis. In Chapter 6, age is incorporated into the models considered in Chapter 5 and the performance is re-assessed. Chapter 7 discusses the possible clinical reasons for the general lack of success of the methods considered in Chapter 5 and Chapter 6. The use of the verbal component of the coma scale is considered, and alternative data which may be useful to predict the quality of survival are discussed. Recommendations are made for future work, the importance of the quality of the information collected is stressed, and the vital role which simple statistical techniques have to play is emphasised.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Adviser: G M Teasdale
• Keywords: Biostatistics, Neurosciences
• Date of Award: 1988
• Depositing User: Enlighten Team
• Unique ID: glathesis:1988-76495
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 19 Nov 2019 14:16
• Last Modified: 19 Nov 2019 14:16
• URI: https://theses.gla.ac.uk/id/eprint/76495