Statistical modelling for the prognostic classification of patients with pancreatic cancer for optimisation of treatment allocation

摘要: Pancreatic cancer is a common cause of death and difficult to diagnose treat. A prognostic index can be used in clinical practice predict survival. Thirty six factor studies were identified but size statistical methods inappropriate. Continuous variables are often simplified incorrectly i) assuming linear relationships between predictors log-hazard or ii) using dichotomisation. Non-linearity addressed for the first time this disease site restricted cubic spline fractional polynomial functions. Multivariable models containing non-linear transformations gave substantially better fit. Important effects some covariates unrecognised under simplistic assumptions. The fitted functions generated by two similar. direct comparison these strategies was based on assessing difference AIC values calculating sampling distribution multiple bootstrap resamples. Model validation also suggested minimal over-fitting with reproducible information when external data. This thesis provides validated tool advanced pancreatic developed appropriate methodology. Risk-sets model could help clinicians target treatments patients more appropriately have an impact future trial design analysis.