Re: st: Risk tables for Cumulative Incidence Functions?

[Phil Clayton <philclayton@internode.on.net>]

Hi Richard,

As far as I know you can't easily add at-risk tables to graphs other than those drawn with -sts graph- (at least in version 11.2). But as a work-around you can draw an invisible K-M plot and add the cumulative incidence function(s) as an added plot.

This example, which draws the baseline CIF for two groups, is from the manual for -stcrreg postestimation-

use http://www.stata-press.com/data/r11/hiv_si, clear
stset time, failure(status == 2)
stcrreg if !ccr5, compete(status == 1) noshow
predict cif_si_0, basecif
label var cif_si_0 "ccr5 = 0"
stcrreg if ccr5, compete(status == 1) noshow
predict cif_si_1, basecif
label var cif_si_1 "ccr5 = 1"
twoway line cif_si* _t if _t<13, connect(J J) sort ///
	yscale(range(0 0.5)) title(SI) ytitle(Cumulative Incidence) ///
	xtitle(analysis time)


You can adapt the above line plot and put it inside an invisible K-M plot as follows:

sts graph, by(ccr5) risktable plotopts(lstyle(none)) /// hides the K-M plot
	addplot(line cif_si* _t if _t<13, sort c(J J) lstyle(p1 p2) /// lstyle to get first 2 colours/styles
		legend(order(3 "ccr5 = 0" 4 "ccr5 = 1"))) /// first 2 lines are hidden
	 title(SI)


This tends to squash the CIF plot a little, because the K-M plot y axis is forced to range from 0 to 1. You can use the -fail- option for -sts graph- but it will still squash it a bit because the K-M failure estimator overestimates the cumulative incidence in the presence of the competing risk. So I usually multiply the CIF by 100 and plot it as a %:

replace cif_si_0=100*cif_si_0
replace cif_si_1=100*cif_si_1
sts graph, by(ccr5) risktable plotopts(lstyle(none)) ///
	addplot(line cif_si* _t if _t<13, sort c(J J) lstyle(p1 p2) ///
		legend(order(3 "ccr5 = 0" 4 "ccr5 = 1"))) ///
	ylab(0(10)40, format(%1.0f)) /// override default K-M plot format of 1 decimal place
	yline(0(10)40, lstyle(grid)) ytitle("Cumulative incidence (%)") title(SI)


Of course you can adapt the above to draw at-risk tables with any plot, for example adjusted predictions after a regression.

There is also a program from 2004 called -stsatrisk- by Jean Marie Linhart and Jeff Pitblado which acts as a wrapper to -sts graph- - my guess is that this was a precursor to the current functionality of the -risktable- option. It may be possible to adapt it for your situation.

Hope this helps,
Phil

On 02/08/2011, at 4:59 AM, Richard Sylvester wrote:

> Hi,
> 
> Using sts graph, it is possible to show the number at risk at various time points for the Kaplan-Meier survivor function via risktable. However is it possible to show the number at risk when plotting the cumulative incidence function? stscurve doesn't seem to have this possibility.
> 
> Thanks,
> 
> Richard Sylvester
> 
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/