Overall Survival prognostic models: sensitivity analyses

Cited in :

New prognostic factors and calculators for outcome prediction in patients with recurrent glioblastoma: a pooled analysis of EORTC Brain Tumour Group phase I and II clinical trials. Thierry Gorlia, Roger Stupp, Alba. A. Brandes, Roy. R. Rampling, Pierre Fumoleau, Christian Dittrich, Mario M.Campone , Chris. C. Twelves, Eric.Raymond, Monika E. Hegi, Denis.Lacombe, Martin J. van den Bent, JCO 2011

Assessing the impact on the goodness of fit (measured by Schemper’s Percentage of Explained Variation, PEV) of removing WHO performance status or steroids use from the Overall Survival model in the whole pooled dataset.

The C-index was also computed for each submodel. The PEV was 15.7% and C-index was 0.68 when both WHO performance status and steroids use were included in the model. In the model without the WHO performance status, C-index was 0.64 and PEV was 14.1%. In the model without steroids use, C-index was 0.68 and PEV was 12.0%. Despite apparent violation of PH assumptions the model fit was (slightly) better when the two factors were included. The complete model was therefore used for the development of a prognostic calculator.

Assessing the impact on the goodness of fit (measured by Schemper’s Percentage of Explained Variation, PEV) of removing steroids use from the Overall Survival model in the dataset with patients pre-treated with TMZ chemoradiation.

In the model without steroids use, C-index was 0.68 and PEV was 16%. In the model including steroids, C-index was 0.68 and PEV was 19%. This model was considered for the development of a prognostic calculator.