1
$\begingroup$

I have data of 5794 subjects which were followed for 22 years, for a total of 18729 events (e.g., stroke, myocardial infarction, coronary heart disease death, etc). Additionally some of those subjects had ultrasound examinations at 5 and 11 years.

Let's say I want to study fatal strokes and limit the study to 5 years so that I can use ultrasound data at 5 years. Only 61 patients experienced fatal stroke. How should I treat all the other 5794 - 61 = 5733 subjects that didn't die from stroke or had other unrelated events ? Should I censor them or discard them ? How about the events ? I was thinking of filtering the 18729 events to take only whose with time_to_event <= 365.25 * 5 (or 11) but it seems kind of odd.

For example, subject XYZ might have had two non-fatal events and one fatal event (which is not stroke) before the fifth year. Another subject might have had only non-fatal events and still be alive at the fifth year, but again, no stroke.

I'm kind of lost since this isn't my area of expertise. Thank you for the help.


Edit: There are two columns that are used to indicate time of censoring (CENSTIME) and time to event (TTOEVENT) from study entry (this is time = 0). Additionally, there is another column, FATAL that indicates whether an event is fatal (i.e., led to a death). If FATAL == 1, CENSTIME is equal to TTOEVENT and is time of death; for each subject, in all events with FATAL == 0, TTTOEVENT indicates time to event and CENSTIME indicates time of censoring. Finally, if a subject did not experience any event during the study, TTOEVENT is NaN and CENSTIME indicates time of censoring.

There is also a column that indicates event type EVTYPE and has 12 different values:

  • 0 - "No event"
  • 1 - "MI"
  • 2 - "Angina",
  • 3 - "Stroke",
  • 4 - "CHF",
  • 5 - "Claudication",
  • 6 - TIA",
  • 7 - Angioplasty",
  • 8 - Coronary Artery Bypass Surgery",
  • 9 - Other Deaths (non-CHD)",
  • 10 - ECG MI (silent)",
  • 11 - other CHD Deaths"

Of these, 1, 3, 9 and 11 are potentially fatal (can lead to death).

$\endgroup$
10
  • $\begingroup$ How is time = 0 defined in your study? Is it the date of a diagnosis of some kind, or the date that an individual joined the study (perhaps at random), or just some calendar date that is common to all? Please provide that information by editing the question, as comments are easy to overlook and can be deleted. $\endgroup$
    – EdM
    Sep 30, 2022 at 15:28
  • $\begingroup$ @EdM just edited adding details about time to event, time to censoring and event types. $\endgroup$ Sep 30, 2022 at 20:47
  • 1
    $\begingroup$ My main questions are (1) how were you planning to incorporate the ultrasound reports into the survival analysis (or should it be a separate secondary post hoc endpoint)? (2) were there two cohorts in this study? One on the study drug and another on placebo? $\endgroup$
    – Alex
    Sep 30, 2022 at 21:06
  • $\begingroup$ @Alex (1) I was thinking naively to transform the events database to a format where I have one row for each subject in the study, and then add relevant ultrasound variables acquired at 5 years. In general, each subject may have had 1+ events. Thus, referring to my example of fatal strokes and considering death from other events as a competing event, I though of taking the event that came first (death from stroke = 1, death from other events = 2, censoring = 0). Basically I don't care of the other non fatal events. (2) Only one cohort of 65+ years old, no study drug or specific treatments. $\endgroup$ Oct 1, 2022 at 10:31
  • $\begingroup$ Thanks. But if you have only one cohort, what is the hypothesis? $\endgroup$
    – Alex
    Oct 1, 2022 at 10:37

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.