Application of survival methods for the analysis of adverse event data.
The concept of collecting Adverse Events (AEs) arose with the advent of the
Thalidomide incident. Prior to this the development and marketing of drugs was
not regulated in any way. It was the teterogenic effects which raised people's
awareness of the damage prescription drugs could cause. This thesis will begin
by describing the background to the foundation of the Committee for the Safety of
Medicines (CSM) and how AEs are collected today.
This thesis will investigate survival analysis, discriminant analysis and logistic
regression to identify prognostic indicators. These indicators will be developed to
build, assess and compare predictor models produced to see if the factors
identified are similar amongst the methodologies used and if so are the
background assumptions valid in this case. ROC analysis will be used to classify
the prognostic indices produced by a valid cut-off point, in many medical
applications the emphasis is on creating the index - the cut-off points are chosen
by clinical judgement. Here ROC analysis is used to give a statistical background
to the decision. In addition neural networks will be investigated and compared to
the other models.
Two sets of data are explored within the thesis, firstly data from a Phase III clinical trial used to assess the efficacy and safety of a new drug used to repress the
advance of Alzheimer's disease where AEs are collected routinely and secondly
data from a drug monitoring system used by the Department of Rheumatology at
the Haywood Hospital to identify patients likely to require a change in their
medication based on their blood results.