Inquiry Regarding Cutoff for Survival Analysis

Question

[ali]

Dear Friends,

I am reaching out to seek guidance on the cutoff values for survival analysis in the context of my COAD (Colon Adenocarcinoma) data. Specifically, I have normalized the data using limma voom and am now preparing for COX regression analysis.

I have two main queries:

Normalization to Z-Scores:

Is it appropriate to convert my data to Z-scores?

Cutoff for High and Low Expression:

In determining high and low expression, I am contemplating the use of Z-scores. Could you kindly advise on the choice of cutoff? Specifically, I am unsure whether to consider Z > 0 as indicative of high expression or to adopt a more stringent threshold such as Z > 1.96. Your expertise on this matter would be invaluable.

[ali]

Thanks a lot for your response

Answer 1

[ATpoint]

Typically you define groups by something like quantiles of gene expression. For example top 25% vs bottom 25%. For this it doesn't matter whether you use Z-score or logcounts or anything else as long as it is normalized. Percentages are the same.

[ali]

Thank you. but I want more to know that which cut off is suitable if i convert my data to z-score? actually, usually we use median for high exp and low exp and its a common.

[ATpoint]

As I said, use quantiles, not fixed cutoffs. Different genes have wildly different Z-scores, but it's a relative measure that depends on standard deviation. Use top-25 vs bottom-25%, or anything else that you find reasonable.

[dsull]

If you're using cox, why do you even need cutoffs/groups? Gene expression is a continuous variable and can simply serve as the independent variable / a covariate in your regression model.