Convert the format of a survfit object.

Description

Add the point for a starting time ("time 0") to a survfit object's elements. This is useful for plotting.

Usage

survfit0(x, start.time)

Arguments

Details

Survival curves are traditionally plotted forward from time 0, but since the true starting time is not known as a part of the data, the survfit routine does not include a time 0 value in the resulting object. Someone might look at cumulative mortgage defaults versus calendar year, for instance, with the ‘time’ value a Date object. The plotted curve probably should not start at 0 = 1970-01-01. Due to this uncertainty, it was decided not to include a "time 0" as part of a survfit object. Whether that (1989) decision was wise or foolish, it is now far too late to change it. (We tried it once as a trial, resulting in over 20 errors in the survival test suite. We extrapolate that it might break 1/3 of the other CRAN packages that depend on survival, if made a default.)

One problem with this choice is that some functions must choose a starting point, plots and computation of the restricted mean survival time are two primary examples. This utility function is used by plot.survfit and summary.survfit to fill in that gap.

The value used for this first time point is the first one below

1. the value of the start.time argument

2. a start.time argument used in the survfit call itself

3. for single state survival

   • min(0, time) for Surv(time, status) data

   • min(time1) for Surv(time1, time2, status) data

4. for multi state survival

   • min(0, time) for Surv(time, event) data, e.g., competing risks

   • min(time1) for Surv(time1, time2, event) data, if everyone starts in the same state

   • no addition: the timepoint used to estimate p0, the intial prevalence of states, will already be the first point of the curve. See survfit.formula

(Remember that negative times are allowed in Surv objects.)

This function will add a new time point at the front of each curve, but only if said time point is less than existing points in the curve. If there were a death on day 0, for instance, it will not add a (time=0, survival=1) point. (The question of whether the plotted curve in this case should or should not start with a vertical segment can be debated ad nauseum. It has no effect on the area under the curve (RMST), and the summary for time 0 should report the smaller value.) Likewise if the start.time argument is not prior to a curve's first time point.

The resulting object is not currently guarranteed to work with functions that further manipulate a survfit object such as subscripting, aggregation, pseudovalues, etc. (remember the 20 errors). Rather it is intended as a penultimate step, most often when creating a plot.

Value

a reformulated version of the object with an initial data point at start.time added. The time, surv, pstate, cumhaz, std.err, std.cumhaz and other components will all be aligned, so as to make plots and summaries easier to produce.