Hello,
I'm trying to get an event plot for a survival analysis with left and right censored data, plus failed items.
I uploaded data table and event plot. My problem is that I can not insert "x" to the right end of failed items (line 3 and 5).
If you look at line 4 and 5, they have different censor codes but they have same plotting. I was expecting to see an "x" on the right side of line 5.
What am I missing?
thank
[*Edit by staffer sseligman: Including another relevant part of poster's question here.]
I attached a part of my real table:
event: 1= failed 0=no event or censored
months to event: time from surgery to event
from surgery to today: from the date of surgery to today (disregarding the presence or absence of an event, drop out, etc...)
months from oldest: I inserted this column in order to insert left censoring (maybe it is not the right term). Given 0 the time of surgery of the first patient, months from oldest are the delay, in months, of the surgery of all the other patients
difference: it is the time to event, calculated for each patient from a 0, that is the time of the first surgery of the first patient.
For example, patient #5 was operated 9 months after the first one and failed after 102,9 months. In a time line where 0 is the surgery of the first patient, his failure should be placed at 112 months. Tha's why I'm expecting an "x" at 112 months and a triangle at 9 months.
Accepted Solutions
Thank you for the example and the explanation. I understand better what you want to plot. Unfortunately, that is not how the Event Plot works. It's behavior is fixed and cannot be altered.
Your description leads me to believe that the proper analysis (correct estimates) involves using the difference column for the time to event role and the event column for the censor role, where event = 0 is censored. With that set up for launch, I get this Event Plot:
All of the subjects entered at difference = 0, their individual start time to the event (or last observed non-event). There were 7 events in this sample. The rest were censored. Here is the associated analysis, using the options to show the survival curve and not show the mid-points.:
The 7 events appear in the empirical survival curve and the remainder appear in the scatter plot at the top.
I realize that the Event Plot is not the graphic that you want. Perhaps we can make it in the Graph Builder. I will think about it, now that I understand what you want.
I assume that censor = 1 means that the life time is censored. Given that interpretation, you are not using the start and end values properly. If you use interval censoring, Life Distribution ignores the censor variable. Entering an interval indicates a failure event in that interval. If the observation is right censored, then life goes in the starting column and missing goes in the end column. A left censored life value goes in the end column and missing goes in the start column. An exact failure time is entered in both start and end columns.
I think that this way should work:
That leads to this event plot, which seems correct to me if I interpreted your data correctly:
Thanks for the reply,
this is what I get reversing the order...
You are not using interval censoring properly. If you enter 0 and 40 for observation 3, you are telling JMP that the failure occurred in this interval of time. If it is uncensored, then enter 40 for both start and end so that JMP knows it is an exact life.
It doesn't matter when the subject life began. It is assumed that all begin at life = 0. The question is how do you indicate the life value?
Most cases involve only right censoring, so Life Distribution makes this special case easy. You enter the life value and, if present, the censor value. All other cases (left censoring, interval censoring) use two life values. See this example in which life is observed every 10 time units for the different kinds of censoring when using two life data columns:
This data results in the following Event Plot in Life Distribution:
Thank you for the example and the explanation. I understand better what you want to plot. Unfortunately, that is not how the Event Plot works. It's behavior is fixed and cannot be altered.
Your description leads me to believe that the proper analysis (correct estimates) involves using the difference column for the time to event role and the event column for the censor role, where event = 0 is censored. With that set up for launch, I get this Event Plot:
All of the subjects entered at difference = 0, their individual start time to the event (or last observed non-event). There were 7 events in this sample. The rest were censored. Here is the associated analysis, using the options to show the survival curve and not show the mid-points.:
The 7 events appear in the empirical survival curve and the remainder appear in the scatter plot at the top.
I realize that the Event Plot is not the graphic that you want. Perhaps we can make it in the Graph Builder. I will think about it, now that I understand what you want.