Dynamic prediction of survival with clinical and
genomic data
Hans C. van Houwelingen
Dept. of Medical Statistics and Bioinformatics, Leiden University Medical
Center
An important clinical application of
biostatistics is the development of statistical models for the prognosis of
a patient at the moment of diagnosis. In cancer the usual way of giving a
prognosis is by means of the x-year survival probability, with x=1, 5 or 10,
for example. Traditionally, the prognosis is based on clinical information
at the start of the treatment, like age, gender, size of the tumor, tumor
stage etc. In the last decade new types of genomic information have become
available like micro-array gene expression and proteomic mass spectrometry
data. The problem with this new type of data is its abundance. Micro-arrays
can measure the expression of tens of thousands of genes, for example.
The talk will address three issues:
-
How to obtain valid prognostic model based on
high-dimensional genomic data.
-
How to assess the added value of the genomic
information.
-
How to obtain robust dynamic predictions (predictions
available later on in the follow-up)
|
