|
Abstract:
|
Increasing interest lies in personalized prediction of disease progression for soft tissue sarcoma patients. Currently, available prediction models are limited to predictions from time of surgery. The concept of dynamic prediction allows to include updated information as well as model time-varying covariate effects to make prediction of overall survival at different times during follow-up. To estimate a patient's probability of surviving an additional 5 years from a particular prediction time point a proportional landmark supermodel was used. Landmark models are able to make predictions from a particular time, by using all (updated) information of patients still alive and in follow-up at that time. For dynamic prediction, several time points are chosen and landmark models for each time point are combined to form a supermodel. The strong effect of time-dependent variables, such as a patient's status of local recurrence and distant metastases found, underlines the importance of making updated predictions. The validity of the prediction model was assessed using the heuristic shrinkage factor and dynamic cross-validated C-indices.
|
