On 2013 Jul 01, lu tian commented:

It is a very interesting idea of using observable short-term event information such as PFS status up to a landmark time point to predict the long term survival status. Predicting the residual life based on most updated current information (including short-term event status) is THE real-life problem. The "baseline" used in traditional survival analysis is somewhat artificial and pertains to the study from which the data is collected.

On the other hand, it will be interesting to study the proposed prediction performance measure such as AUC as a function of time t0, i.e., the landmark time. This can provide useful information on the choice of landmark time to evaluate the short-term event status such as tumor growth to maximize the performance boost in the predicting the long-term event outcome.

On 2013 Jul 01, lu tian commented:

It is a very interesting idea of using observable short-term event information such as PFS status up to a landmark time point to predict the long term survival status. Predicting the residual life based on most updated current information (including short-term event status) is THE real-life problem. The "baseline" used in traditional survival analysis is somewhat artificial and pertains to the study from which the data is collected.

On the other hand, it will be interesting to study the proposed prediction performance measure such as AUC as a function of time t0, i.e., the landmark time. This can provide useful information on the choice of landmark time to evaluate the short-term event status such as tumor growth to maximize the performance boost in the predicting the long-term event outcome.