Statistical Modelling for the Prognostic Classification of Patients with Pancreatic Cancer for Optimisation of Treatment Allocation [PhD Thesis]

  1. Lookup NU author(s)
  2. Dr Deborah Stocken
Author(s)Stocken DD
Publication type Report
TypePhD Thesis
Series Title
Year2010
Date01-03-2010
Pages171
Full text is not currently available for this publication.
Pancreatic cancer is a common cause of cancer death and is difficult to diagnose and treat. A prognostic index is a tool that can be used in clinical practice to predict survival. Thirty six prognostic factor studies were identified but the size and statistical methods were inappropriate. Valid statistical analyses are essential to make best use of data and optimise clinical application. Continuous variables are often simplified incorrectly by i) assuming linear relationships between predictors and log-hazard or ii) using dichotomisation. Non-linearity is addressed for the first time in this disease site using restricted cubic spline and fractional polynomial functions, ideal for smooth curved relationships. Multivariable models containing non-linear transformations gave a substantially better fit. Important effects of some covariates were unrecognised under simplistic assumptions. The fitted functions generated by the two methods were similar. A direct comparison of these strategies was based on a novel approach assessing the difference in the AIC values by calculating a sampling distribution in multiple bootstrap resamples. Model validation is also addressed for the first time in this disease and suggested minimal over-fitting with reproducible prognostic information when fitted to external data. This thesis provides the first validated prognostic tool in advanced pancreatic cancer developed using appropriate statistical methodology. Four risk-sets identified by the model could help clinicians target treatments to patients more appropriately and have an impact on future trial design and analysis.
InstitutionUniversity of Birmingham
Place PublishedBirmingham
URLhttp://etheses.bham.ac.uk/1303/1/Stocken10PhD.pdf
NotesA thesis submitted to The University of Birmingham for the degree of Doctor of Philosophy.
ActionsLink to this publication