KM curves with 1mil observation, curves difficult to tell difference

[Martin Imelda Borg]

Hi there, I am currently using a large database with over 1 million observation.

I am trying to plot a KM curve for different surgeries. I can not give you the exact data as the data is stored on a remote platform with no internet ! <sigh> So in order to understand what's going on I created a smaller dataset to see what's going on with data and ensure I'm using the right code. I am sure some people will complain about this, but there is nothing I can do except complain to the government who won't give me internet on our remote system - so sorry about this !


Here's the sample data

* Example generated by -dataex-. For more info, type help dataex
clear
input float(type revision yearofsurgery yearofrevision timetorevision_years) byte(_st _d) double _t byte _t0
1 1 14610 16438  5.008219 1 0                 3 0
1 0 15310     .         . 0 .                 . .
0 0 16468     .         . 0 .                 . .
1 1 17867 18263 1.0849315 1 1 1.084931492805481 0
1 0 17932     .         . 0 .                 . .
1 1 18298 19422  3.079452 1 0                 3 0
0 1 19029 20794  4.835617 1 0                 3 0
0 0 16109     .         . 0 .                 . .
0 1 15745 16111 1.0027397 1 1 1.002739667892456 0
0 1 18303 20498  6.013699 1 0                 3 0
end
format %td yearofsurgery
format %td yearofrevision
label values type surgery
label def surgery 0 "Sling", modify
label def surgery 1 "Pessary", modify

Now I've attempted to create two KM curves using this code
yearofsurgery = date when surgery took place
yearofrevision = date revision of surgery took place
type = type of surgery performed

//Generate time variable from operation to revision 
gen timetorevision_years = (yearofrevision - yearofsurgery)/365

stset timetorevision_years, failure(revision=1) exit(time 3)
sts graph, by(type)  //Creates two KM curves for the different type of procedures 
stdescribe

Now my sample data create creates a nice KM curve which looks just right as seen below



Now when I replicated the code on my remotedata set I got these curves (the lines are really close together) .

Question 1: How can I address the problem to separate out the KM line graphs ?


In fact when I then plot a cum hazard plot , the curves separate out

Question 2:
In my current sample data provided above when I use

stsum, by(type)

I get missing values for 50% and 75% . Why does this happen?

[Tiago Pereira]

You can try to adjust the vertical axis:

sts graph, ylabel(0.8 1)

Sometimes graphs are not useful. You can also present a table with the survival function per group.

[George Ford]

change the yscale might help. ylabel(0.9(0.1)1.1)

you may have to do it in the graph editor.

[Scott Merryman]

On your timetorevision variable you are not accounting for leap years. The function -datediff_frac()- will take care of this.

. gen diff = datediff_frac(yearofsurgery, yearofrevision, "year")
(4 missing values generated)

. list year* time diff in 1, ab(20)

     +-----------------------------------------------------------------+
     | yearofsurgery   yearofrevision   timetorevision_years      diff |
     |-----------------------------------------------------------------|
  1. |     01jan2000        02jan2005               5.008219   5.00274 |
     +-----------------------------------------------------------------+

. disp 1/365
.00273973

. disp 3/365
.00821918

The difference is due to the years 2000 and 2004 being leap years.

[Martin Imelda Borg]

Thanks for all this, just realised I had a coding error and was taking the wrong variable.

With regards to my question2
does anyone know why I am getting missing values

 
 stsum, by(type)

I get missing values for 50% and 75% . Why does this happen?