I have to draw the Kaplan meier curve to show the proportion of clinical trials with and without results (on y-axis) and their time (which is the difference between trials completion and result reporting dates) (on x-axis). The event of interest is to see if the trial has either reported the result or not (if reported then 1 and 0 for not reported). For all those without results, we don’t have trial result reporting date to calculate difference and these trials are right censored because they did not report the results even till the date of our data collection.

Any idea to solve this issue because I want to show both trials- the trials with results and those without results (censored). How to deal with trials not reporting the results even till the time of our data collection. What I need to put in the result reporting date column for these trials. Any suggestion to deal with this issue will be appreciated.


Under your definition:

time ... is the difference between trials completion and result reporting dates

what you have is administrative censoring, just like with patients who haven't had an event by the end of data collection. The value of time to include for each censored case--one for which results haven't yet been reported--would be the difference between the trial-completion date and the date of your data collection.

One caution: you don't seem to be including trials that are registered but never are formally completed. That happens frequently in practice, so interpret your results carefully.

  • $\begingroup$ Thank you very much for answering. Yes, I already tried that-I used the date of data collection in result reporting column for all those trials with missing result reporting dates (censored). But I am confused because I think that by doing this they are not counted as right censored and it assumes they were reported at that day (data collection). Also, in the risk table under KMC, the number of trials at risk will not be accurate because we know that our all trials without results have not reported the result date even till the end of our data collection (and they all at risk) $\endgroup$ – bright dreamz May 20 at 18:53
  • $\begingroup$ @hania you need to have a separate event variable for each case, to keep track of whether time represents an event or censoring. In R, for example, censoring is usually represented by 0 and an "event" by 1. The survival software then expects both the time value and the associated event variable. (Other software might reverse the event/censored values, so be careful.) If you do this by hand, you calculate the Kaplan-Meier curve at each event time, using all the cases still at risk at that time. Censored cases thus contribute to the numbers at risk, at times while they are at risk. $\endgroup$ – EdM May 20 at 19:27
  • $\begingroup$ Once again thank you. Yes, I have that column in csv file and I am using python. I have one column for the difference, another named ‘events’ in which I have 0 value for censored trials (not reporting results) and 1 if they are reporting results, and one column for different types of trials. The thing which is making me confuse is as I used the date of data collection for those missing the date of reporting. If you kindly look at my risk table here (ibb.co/wSYSBCb) as it says: 5 trials at risk under type A and 0 at risk under each of type B and C. $\endgroup$ – bright dreamz May 21 at 19:11
  • $\begingroup$ But the fact is there are several thousand which did not report the result till the end of data collection (and those all should be at risk) so the number of trials at risk does not appear correct to me. Do you think all is good? This is the first time I am dealing with such kind of data $\endgroup$ – bright dreamz May 21 at 19:12
  • $\begingroup$ @hania presumably your time starts at 0 for each study when it is first completed. A study that was completed 5 years before your data collection would only be at risk through 60 months. if you are using the US clinical trials.gov database, they only started accepting results in 2008, so there shouldn't be many at risk at 160 months--those studies would have been completed 13 years before your data collection and still not reported! That's why you only show (5,0,0) at risk in types (A,B,C) at 160 months. $\endgroup$ – EdM May 21 at 21:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.