0
$\begingroup$

I have survival data (n ~ 1000) with ~40% censoring and ~10% missing data. I want to partition the subjects into 4 prognostic groups based on survival. Misganow B and Vidyasagar M (Exploiting Ordinal Class Structure in Multiclass Classification: Application to Ovarian Cancer. IEEE Life Science Letters, 2015) have previously done this by simply partitioning the survival times into quartiles while completely ignoring the censoring (i.e., event indicator variable). I have the following questions:

  1. Is their method statistically sound? (i.e., can I do what they did?)
  2. Are there concerns with the application of this method due to censoring? (which is my concern)
  3. Are there other methods? (e.g., died between T0 - T1, died between T1 - T2, died between T2 - T3, died sometime after T3. The last group would include all data, censored and uncensored, with T > T3. Note that this method would ignore censored data between T0 - T3.)
  4. Can you provide references relevant to this issue?

I have done extensive searches with search terms "partition survival data" and "stratify survival data" and "divide survival times into groups" etc. The results lead me to methods that separate subjects based on predictor variables, which is not what I'm looking for. My objective is to separate into groups based just on the target variables (survival time, survival status)

$\endgroup$
0

1 Answer 1

2
$\begingroup$

This is completely unsound. You'll find prognostic variability within the so-called strata. Risk is a continuous notion and it is so easy to get predicted risk with equations, line graphs, nomograms, and point tables. See Steyerberg's book Clinical Prediction Models for example, and RMS.

One way to show that those authors' approach is wrong is to see that a patient at the edge of one prognostic stratum is more similar to a patient in a different stratum than she is to patients in her own stratum.

$\endgroup$

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.