As mentioned in my comment, I don't think it is possible to get KM-estimates for individual patients. The KM-estimator gives the observed probability of survival at a certain timepoint on a population level. The observed survival probability for an individual, however, is either 0 (death) or 1 (alive), nothing in between.
Instead of observed survival probabilities you will have to use some sort of model (e.g. Cox PH, accelerated failure time model, neural network etc.) to get predicted survival probabilities. These probabilities inform you about the risk of an individual with that particular variable combination to be alive at a particular timepoint.
UPDATE: with example code based on code the OP provided here
library(pec) ; library(rms)
# Simulate data
set.seed(1)
examp.data <- SimSurv(3000)
# fit a Cox model with predictors X1+X2
coxmodel <- cph(Surv(time,status)~X1+X2, data=examp.data, surv=TRUE)
# predicted survival probabilities can be extracted at selected time-points:
ttt <- quantile(examp.data$time)
ttt
# 0% 25% 50% 75% 100%
#6.959458e-03 9.505409e+00 3.077284e+01 7.384565e+01 7.100556e+02
# Get predicted survival probabilities at selected time-points:
preds <- predictSurvProb(coxmodel, newdata=examp.data, times=ttt)
# Store in original data
examp.data$predict.surv.prob.Q1 <- preds[,1] # pred. surv. prob. at 0.006959458
examp.data$predict.surv.prob.Q2 <- preds[,2] # pred. surv. prob. at 9.505409
examp.data$predict.surv.prob.Q3 <- preds[,3] # pred. surv. prob. at 30.77284
examp.data$predict.surv.prob.Q4 <- preds[,4] # pred. surv. prob. at 73.84565
examp.data$predict.surv.prob.Q5 <- preds[,5] # pred. surv. prob. at 710.0556
Now you have predictions of the survival probabilities at those 5 timepoints for each individual in your data. Of course, you do need to evaluate the predictive performance of your model in terms of discrimination (e.g. with the function cindex
in pec-package) and calibration (with calibration plot, see rms-package).