.BG .FN avg.surv .TL This function calculates adjusted survival curves following the "corrected group prognosis" (average of individual survival curve) method. Based on a fitted Cox proportional hazards model, for each specified value of a given variable, the function returns the mean of the family of fitted survival curves relative to the specified variable value and a set of "reference" values for the remaining covariates. .DN Model-based adjusted survival curves. .CS avg.surv(cfit, var.name, var.values, data, weights) .RA .AG cfit An object of class "coxph", typically produced by the application of the coxph function. .AG var.name A variable from the model represented in cfit. .AG var.values Values of the variable defining strata for which mean survival curves are to be calculated. .AG data A dataframe with variables corresponding to those represented in cfit. The values for the covariates (i.e. model variables setting aside the response and the variable specified in var.name) define the reference population. If data is not specified, the function looks first to see if cfit has a model component (from selecting model=T in the original application of coxph) to use as the data, and failing that, tries to construct a dataframe using model.frame and the parent frame. .AG weights An optional vector of weights of length equal to the number of rows of data, used to weight the mean survival curve over the set of reference covariate values. .OA .RT An object of class "survfit", suitable for plotting (see example). .DT The function organizes a sequence of calls to survfit.coxph which does the real work. The function returns curves which are the averages of covariate specific survival curves, NOT a fitted survival curve at the mean of the covariate values. .SH REFERENCES Nieto, F.J., Coresh, J. (1996), Adjusting survival curves for confounders: a review and a new method, \fIAmerican Journal of Epidemiology\fP, \fB143:10\fP, 1059-1068. .SA survfit.coxph .EX # Variables for fit should be defined prior to application of coxph # so that model formula contains only already defined variables, i.e. no # "transformations" of variables, such as Surv or log. stime <- Surv(fu.time, status) cfit <- coxph(stime ~ sex + age + diabmell + hyperten + creat + chf) # select diabetes variable, a factor with levels "Y" and "N". afits <- avg.surv(cfit, "diabmell", c("Y","N")) # plot unadjusted Kaplan-Meier curves plot(survfit(stime ~ diabmell)) # add adjusted curves for comparison matlines(afits$time,afits$fits) .KW survival data .WR