1
$\begingroup$

I have done a survey on " dosgnosis delay for breast cancer".My response variable of interest is the " time taken to seek medical care" which is the interval between a patient notice a symptom and first consultancy with a doctor. As there is no censoring is involved is it right to fit a survival model or any suggestions of a better model?The response is non normal as well.

$\endgroup$
3
  • $\begingroup$ How are you defining "patient notices symptoms"? Retrospective interview? Medical chart text-mining? $\endgroup$
    – AdamO
    Oct 20 '17 at 18:10
  • $\begingroup$ Are you coding screen-detected cancers as 0 delay? Or are you using the interval between screening and diagnostic mammogram? $\endgroup$
    – AdamO
    Oct 20 '17 at 18:14
  • $\begingroup$ No only the symptotic patients were oncluded.that is who have discovered a symptom by themselves. $\endgroup$
    – peshala
    Oct 21 '17 at 1:35
3
$\begingroup$

To answer your question: yes, you can use survival models without censoring.

The bigger problem you face is selection. I suspect your sample consists of people who have sought help, which is why you have no censoring. But this way you miss people who do not seek help.

$\endgroup$
3
  • $\begingroup$ If by "seeking help" you mean they schedule diagnostic mammogram, then if this is the US, then our medical record system is junk: many providers code such mammograms as screening so they can be reimbursed by medicare/insurance. Putting that aside, I agree there is no "study base"; the study needs to include "symptomatic" participants who did not schedule some type of follow-up mammogram; survival analysis theory states they are still waiting for breast cancer... even if they are dead (unless you include competing risks). $\endgroup$
    – AdamO
    Oct 20 '17 at 18:13
  • $\begingroup$ Thank you for the response.I'm considering the cancer patients who have already diagnosed and there's no way to find people who haven't seek medical care. Most of the similar studies have done using logistic regression by making the response dichotomus.I wanted try a different model since the information is lost by categorizing the response.do you have any suggestions other than survival? $\endgroup$
    – peshala
    Oct 21 '17 at 1:52
  • $\begingroup$ Selection is a general problem, so it will affect all methods. My laymen's intuition is that eventually all cancer patients will end up at the doctor (some too late) so that would mitigate the problem. This is how I would approach this problem: think about it, understand it, and only that start thinking about potential solutions. Don't rush into new models. $\endgroup$ Oct 21 '17 at 8:35

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.