Cannot get ipdfc to work

[Geoff Hammond]

I happened to come across a user written program which might be the solution to what I am attempting to do (compare a published KM curve to one we are working on locally).

I am using ipdfc which was written to reconstruct individual participant data from a published Kaplan-Meier curve (findit ipdfc).

I have digitised the KM curve from the following paper and wanted to use the ipdfc command to reconstruct the individual participant data.

https://www.ncbi.nlm.nih.gov/pmc/art...v038p00769.pdf

based on this, i have extracted the following data and points.

femalesagexdermott femalesproportionydermott
2.250259 1.0009734
6.7189907 .99996194
11.187561 1.000384
15.656271 .99956804
20.125786 .99158437
24.48281 .98801388
29.064506 .97837551
33.345984 .97288928
38.048789 .93842834
42.476369 .92490446
47.000088 .9079631
51.273208 .87864681
54.617883 .81446671
58.031029 .73895572
62.397216 .70367233
65.917298 .62334907
67.837164 .58112485
68.801202 .52348231
70.192767 .44853756
72.75203 .3972092
75.418717 .33673176
77.021841 .27274933
79.189122 .21373451
81.609278 .16217862
83.74858 .06082447

You may note that the source publication does not include a risk table, so i am clean out of luck there. However, i have read from the help file associated with ipdfc, an approach is to enter 0 for the trisk option and the total number of participants (60) for the nrisk option.

Armed with all this, i have put together the following syntax

ipdfc, surv(femalesproportionydermott) tstart(femalesagexdermott) trisk(0) nrisk(60) generate(FEMALEDERMOTT1 FEMALEDERMOTT2) saving(TEST2) proportion isotonic

This returns:

0 invalid name
(error in option trisk())

if i stipulate a dummy variable for both

gen DUMMYZERO=0
gen POPULATION=60

Re-running the command:

ipdfc, surv(femalesproportionydermott) tstart(femalesagexdermott) trisk(DUMMYZERO) nrisk(POPULATION) generate(FEMALEDERMOTT1 FEMALEDERMOTT2) saving(TEST2) proportion isotonic

this returns:

'0' invalid observation number

It is of note that the command is at least doubling the number of lines in the file at some stage in its execution, but provides the error.


I am using Stata 15.1 IC. I have also attempted to specify previous versions 14 and 13 using the version command to get this working. A colleague has tried on Stata 16 also, but to no avail.

I have also tried to contact the original authors of the program, but their Stata profiles did not come up when i looked them up by name.

A Google search and a search on the Stata forum has provided little further insight.

Does anyone out there have experience using this ado in situations where the life table is not available?

Thanks in advance,

Geoff

[Marcos Almeida]

I haven't used - ipdfc - but, by looking at the help files, I see the option - probability - but I fail to see the - proportion - option.

By the way, Stata is also telling you there is an "error in option trisk". The help files recommend to use "varname" instead of zero for trisk, unless "the total number of patients in the sample is knwon".

Hopefully that helps.

[Michael Yan]

Hi Geoff,

I'm also using this program and am stuck in the same position where no risk table is available. I'm wondering if you ever figured this out?

Thanks,

[Kyle Monahan]

Hi all,

I am also using this program, and ran into the same issue.

Taking what Marco said, I created a new variable, trisk_na and nrisk_na where I set trisk to zero and nrisk to the number of people within the study. Note that in the original paper, the authors state that we need to provide the number of patients in the sample. This solved the initial, error, which was:

 
 '0' invalid observation number

However, this gave me a new error:

observation number out of range
    Observation number must be between 154 and 2,147,483,619.  (Observation
    numbers are typed without commas.)

I have reached out to the author of the program, but I would appreciate any suggestions. Reproducible code and data are available below.

Best,
Kyle

// Conversion of extracted KM curve data to time-to-event data
// Kyle Monahan
// Tufts Technology Services
// 5/4/2020

// Turn on debugging
 set trace on

// Load the package if you haven't already
net install st0498.pkg

// Clear previous data
clear

// Note to future users - change directory or place SampleData_AllData.xlsx in the same directory
// Download sample data here: https://tufts.box.com/shared/static/7znmk3azhh9a5v3q9787e7mtngtt154a.xlsx

// Load sample Data
import excel "SampleData_AllData.xlsx", sheet("Sample") firstrow


// Run model
ipdfc, surv(s0) tstart(ts0) trisk(trisk) nrisk(nrisk0) isotonic ///
 generate(t_ipd event_ipd) saving(output)

// This gives an error: observation number out of range
    // Observation number must be between 154 and 2,147,483,619.  (Observation
    // numbers are typed without commas.)
    
// Try to run the model without trisk. From Wei et al.:
// "Set trisk() as 0 only if the total number of patients in the sample is known. trisk() is required."
// "If no risk table is available, specify nrisk() as the number of patients in the sample, and specify trisk() as 0."

// Run model without risk table
ipdfc, surv(s0) tstart(ts0) trisk(trisk_na) nrisk(nrisk_na) isotonic ///
 generate(t_ipd2 event_ipd2) saving(output2)

[Abdullah Shah]

Hi all,

Has anyone been able to contact the authors or seem to get the code to work?

I am also working on extracting the number of events from a KM curve but am running into trouble when the number at risk is greater than 1000. In particular when I perform stset on the resulting data the last observed exit is the second time point (out of 92 time points) taken from the digitized plot.

Best,
Abdullah

[Courtney Balentine]

Not sure if this helps anyone, but I was able to replicate the scenario using the data provided with the ipdfc command and got it to work. Maybe the problem is that it was meant to be run separately for the control and treatment groups and then merged into one file for analysis. You also can't specify trisk(0) in the code, there needs to be a variable in the data file that provides the 0

import delimited using "head_and_neck_arm0.txt"

*I manually went in and deleted all values for trisk and nrisk other than the initial 0 and 211 (where 211 is the sample size)

ipdfc, surv(s) tstart(ts) trisk(trisk) nrisk(nrisk) isotonic generate(t_ipd event_ipd) saving(temp0a)

import delimited using "head_and_neck_arm1.txt", clear
*did the same thing here, erasing the values for trisk and nrisk aside from 0 and 213 (n = 213)

ipdfc, surv(s) tstart(ts) trisk(trisk) nrisk(nrisk) isotonic generate(t_ipd event_ipd) saving(temp1a)


use temp0a, clear
gen byte arm = 0
append using temp1a
replace arm = 1 if missing(arm)

label define ARM 0 "Radiotherapy" 1 "Radiotherapy plus cetuximab"
label values arm ARM

stset t_ipd, failure(event_ipd)

sts graph, by(arm) title("") xlabel(0(10)70) ylabel(0(0.2)1) ///
    risktable(0(10)50, order(2 "Radiotherapy" 1 "Radiotherapy plus")) ///
    xtitle("Months") l2title("Locoregional control") ///
    scheme(sj) graphregion(fcolor(white)) ///
    plot1opts(lpattern(solid) lcolor(gs12)) ///
    plot2opts(lpattern(solid) lcolor(black)) ///
    text(-0.38 -9.4 "cetuximab") ///
    legend(off) ///
    text (0.52 53 "Radiotherapy plus cetuximab") text(0.20 60 "Radiotherapy")