How to predict overall survival (Cox) based on a population?

MrDo

New Member
#1
Hi, there! First of all, I am quite excited about having found this forum, and look very much forward to learning from here. :)

Say I want to devise a score index for predicting patient overall survival (OS) based on a handful of co-variates, what would be the best way to do it?

Briefly, I have performed a Cox regression analysis (e.g. by using coxph in R) on a number of variables (~15 variables), and found 4 to be significantly associated with OS. Noteworthy, these predictors are protein level measurements (continuous), and from the Cox regression results I see that high levels (anything above the sample median) of 3 of these proteins (A, B, C) and low level of 1 (D) of them is associated with a shorter survival.
So, if I artificially classify the individuals based on their protein levels in 2 groups risk classes: (1) "high risk" (A, B, and C levels above sample median AND D below sample median); (2) "low risk" (vice-versa), the classification works, showing a very distinct difference in OS between these groups (plot below).

Rplot.png

However, again, this classification is artificial and its application is quite limited. If there was a way to use some sort of a cumulative hazard within time and use it as an index score that would be great.
What I am thinking is: how to calculate such index in a way that patients are still classified in 2 groups, only this time the criterion would be that the "high risk" should be the group where the estimated survival is less than 60% in 1-year period, and "low risk" more or equal to 60% in one year?
Any help is greatly appreciated!

Thanks a lot in advance!


PS: if there is a function in R that handles it, that would be great. If not, it is also fine.