# Predict method for CoxBoost fits Obtains predictions at specified boosting steps from a CoxBoost object fitted by `CoxBoost`. ```r ## S3 method for class 'CoxBoost' predict( object, newdata = NULL, newtime = NULL, newstatus = NULL, subset = NULL, weights = NULL, stratum = NULL, at.step = NULL, times = NULL, type = c("lp", "logplik", "risk", "CIF"), ... ) ``` ## Arguments - `object`: fitted CoxBoost object from a `CoxBoost` call. - `newdata`: `n.new * p` matrix with new covariate values. If just prediction for the training data is wanted, it can be omitted. - `newtime, newstatus`: vectors with observed time and censoring indicator (0 for censoring, 1 for no censoring, and any other values for competing events in a competing risks setting) for new observations, where prediction is wanted. Only required if predicted partial log-likelihood is wanted, i.e., if `type="logplik"`. This can also be omitted when prediction is only wanted for the training data, i.e., `newdata=NULL`. - `subset`: an optional vector specifying a subset of observations to be used for evaluation. - `weights`: weights for each time. - `stratum`: vector specifying different groups of individuals for a stratified Cox regression. In `CoxBoost` fit each group gets its own baseline hazard. - `at.step`: scalar or vector of boosting step(s) at which prediction is wanted. If `type="risk"` is used, only one step is admissible. If no step is given, the final boosting step is used. - `times`: vector with `T` time points where prediction is wanted. Only needed for `type="risk"` - `type`: type of prediction to be returned: `"lp"` gives the linear predictor, `"logplik"` the partial log-likelihood, `"risk"` the predicted probability of not yet having had the event at the time points given in `times`, and `"CIF"` the predicted cumulative incidence function, i.e., the predicted probability of having had the event of interest. - ``...``: miscellaneous arguments, none of which is used at the moment. ## Returns For `type="lp"` and `type="logplik"` a vector of length `n.new` (`at.step` being a scalar) or a `n.new * length(at.step)` matrix (`at.step` being a vector) with predictions is returned. For `type="risk"` or `type="CIF"` a `n.new * T` matrix with predicted probabilities at the specific time points is returned. ## Examples ```r # Generate some survival data with 10 informative covariates n <- 200; p <- 100 beta <- c(rep(1,10),rep(0,p-10)) x <- matrix(rnorm(n*p),n,p) real.time <- -(log(runif(n)))/(10*exp(drop(x %*% beta))) cens.time <- rexp(n,rate=1/10) status <- ifelse(real.time <= cens.time,1,0) obs.time <- ifelse(real.time <= cens.time,real.time,cens.time) # define training and test set train.index <- 1:100 test.index <- 101:200 # Fit CoxBoost to the training data cbfit <- CoxBoost(time=obs.time[train.index],status=status[train.index], x=x[train.index,],stepno=300,penalty=100) # mean partial log-likelihood for test set in every boosting step step.logplik <- predict(cbfit,newdata=x[test.index,], newtime=obs.time[test.index], newstatus=status[test.index], at.step=0:300,type="logplik") plot(step.logplik) # names of covariates with non-zero coefficients at boosting step # with maximal test set partial log-likelihood print(cbfit$xnames[cbfit$coefficients[which.max(step.logplik),] != 0]) ``` ## Author(s) Harald Binder binderh@uni-mainz.de

predict.CoxBoost function