Use this URL to cite or link to this record in EThOS:
Title: Prediction models for mortality in tuberculous meningitis
Author: Thao, Le Thi Phuong
ISNI:       0000 0004 7966 3471
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Tuberculous meningitis (TBM) is the most severe manifestation of tuberculosis, killing or disabling nearly half of all sufferers despite proper antituberculosis treatment. The British Medical Research Council (MRC) grade, which was first introduced in 1948, is the most widely used grading system for TBM severity. Despite its age and its lack of statistical derivation, improved and robust prediction models for poor outcomes in TBM based on large cohort studies and rigorous statistical methodology are still lacking. In addition, no studies have exploited repeated measurements to update the risk prediction. The purpose of my DPhil thesis is to develop prognostic models for mortality in adult TBM patients and to address statistical issues that may arise during the modelling process. Firstly, I present prognostic models for 9-month mortality based on patient characteristics at diagnosis. Separate models were constructed for patients with and without HIV coinfection. The models were developed and validated based on a large pooled database of 748 HIV-infected and 951 HIV-uninfected subjects. This is the largest database of prospectively characterized TBM patients to date. The final models showed good performance and are clearly superior to the MRC grading system. However, predictions may be improved and updated once new information on the disease progress becomes available during follow-up. Therefore, in the second part of my thesis, I present prognostic models that dynamically predict death using time-updated Glasgow coma score and plasma sodium measurements, together with patient baseline characteristics. The models were carefully developed and validated based on 518 HIV-infected and 550 HIV-uninfected patients respectively. They outperformed the baseline models. I implemented the baseline models and the dynamic models in web-based calculators to facilitate their application in practice. These prognostic models can be used to assist clinicians in decision-making regarding treatment and patient management. Lastly, I describe and compare strategies for incorporating variable selection with multiply imputed data in the procedure of building a prediction model. I evaluated the performance of the methods on simulated data, and on the same data set of 951 HIV-uninfected patients that I used in the first part. I provide insights on the strengths and weaknesses of the pragmatic approaches that are commonly used.
Supervisor: Geskus, Ronald ; Wolbers, Marcel ; Thwaites, Guy Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available