Question: Problem for Running "multivariate Cox proportional hazards model"
0
gravatar for modarzi
7 days ago by
modarzi110
modarzi110 wrote:

Dear All,

For survival analysis, firstly I ran univariate Cox model between my 71 cases and 180 genes.based on that, 70 genes had p-value less than 0.05. Now I would like to fit a multivariate Cox proportional hazards model among 70 significant genes based on below function for Overal Survival with 70 genes expression values:

multiple_cox_green_genes <- coxph(Surv(coxSARC_green_data$OS.days, coxSARC_green_data$OS.status) ~ ENSG00000076662 + ENSG00000204475 + ENSG00000111796 + ENSG00000115607+ENSG00000093072 +ENSG00000172543 +ENSG00000255833 + ENSG00000121807 + ENSG00000227507 + ENSG00000122224 + ENSG00000172215 + ENSG00000073861+ENSG00000137078 + ENSG00000182866 + ENSG00000100351 + ENSG00000104814 + ENSG00000196684 + ENSG00000143185 +ENSG00000122188 + ENSG00000128284 + ENSG00000110448 + ENSG00000176083 + ENSG00000113088 + ENSG00000117091 +ENSG00000186810 + ENSG00000205045 + ENSG00000271503 + ENSG00000116824 + ENSG00000266094 + ENSG00000125910 +ENSG00000153563 + ENSG00000198821 + ENSG00000143184 + ENSG00000005844 + ENSG00000147168 + ENSG00000153283 +ENSG00000009790 + ENSG00000078589 + ENSG00000121380 + ENSG00000141293 + ENSG00000169442 + ENSG00000167984 +ENSG00000026751 + ENSG00000117090 + ENSG00000174946 + ENSG00000213886 + ENSG00000162739 + ENSG00000183918 +ENSG00000124203 + ENSG00000110848 + ENSG00000163519 + ENSG00000161405 + ENSG00000115085 + ENSG00000101082 +ENSG00000113263 + ENSG00000125735 + ENSG00000106560 + ENSG00000163508 + ENSG00000089012 + ENSG00000100385  +ENSG00000125347 + ENSG00000117560 + ENSG00000163564 + ENSG00000172673 + ENSG00000163600 + ENSG00000125637 +ENSG00000172116 + ENSG00000145649 + ENSG00000008516 + ENSG00000183813 + ENSG00000160185 + ENSG00000109684
                                  ,data =  coxSARC_green_data)

but I get below Error and warning:

Error in fitter(X, Y, strats, offset, init, control, weights = weights,  : 
  NA/NaN/Inf in foreign function call (arg 6)
In addition: Warning messages:
1: In fitter(X, Y, strats, offset, init, control, weights = weights,  :
  Ran out of iterations and did not converge
2: In fitter(X, Y, strats, offset, init, control, weights = weights,  :
  one or more coefficients may be infinite

I appreciate it if anybody shares his/her comment with me for solving this Error.

Best Regards

ADD COMMENTlink modified 6 days ago by Kevin Blighe56k • written 7 days ago by modarzi110
1
gravatar for Kevin Blighe
6 days ago by
Kevin Blighe56k
Kevin Blighe56k wrote:

Hi, the model has too many variables. You should, first, test each gene independently (that is, in separate models). At the end, you can create a 'final' model that contains just those genes that have a statistically significant p-value.

However, there are many ways of doing this.

Kevin

ADD COMMENTlink modified 6 days ago • written 6 days ago by Kevin Blighe56k

Thanks, Dr. Blighe. For 180 genes firstly I have done a univariate cox model. Now, I have 70 genes with a p-value<0.05. Now for final selection, I ran multivariate cox proportional hazards model which I got below error:

Error in fitter(X, Y, strats, offset, init, control, weights = weights,  : 
  NA/NaN/Inf in foreign function call (arg 6)

So, I appreciate it if you perform your solution for solving my problem. Best Regards,

ADD REPLYlink written 5 days ago by modarzi110

The model is still too large, I think. Try a lower p-value threshold.

However, it seems that there are also NA values somewhere in your data?

ADD REPLYlink written 5 days ago by Kevin Blighe56k

Ok, If I selected genes with p-value smaller than 0.03, Then 70 genes decrease to 40 genes. and my multivariable cox has no problem with 40 genes. but is this threshold acceptable for the scientific reviewers? could you pls guide me? Best Regards,

ADD REPLYlink modified 5 days ago • written 5 days ago by modarzi110

The lower p-value threshold, the better. Do the genes make sense?

Do you even need to produce this final model? The final message for your work could, instead, be just: 'we identified 40 genes as statistically significant independent predictors through a Cox proportional hazards survival analysis (p<0.03)'. After this, you could do, e.g., gene enrichment on these genes.

ADD REPLYlink written 5 days ago by Kevin Blighe56k

Ok. Dr. Blighe, In my problem, can I say that genes with lower P-values are more important than genes with higher values?  

ADD REPLYlink written 5 days ago by modarzi110

Yes, generally, but also check the Hazard Ratios. Can you access those?

ADD REPLYlink written 5 days ago by Kevin Blighe56k

yes. After univariate Cox analysis, for each gene I have below information:

"Beta" , "StandardError" , "Z" , "P" , "LRT" , "Wald" , "LogRank" , "HR", "HRlower" , "HRupper"

which items except P-Value can help me for the rank of genes?

ADD REPLYlink modified 5 days ago • written 5 days ago by modarzi110

You should be looking at these:

  • HR
  • HRlower
  • HRupper
  • LogRank
ADD REPLYlink written 5 days ago by Kevin Blighe56k

Ok, You mean that I should consider "LogRank", "HR", "HRlower" , and "HRupper" for ranking my genes. but how can I take into account them at the same time for ranking task?

Best Regards,

ADD REPLYlink written 5 days ago by modarzi110

The p-value that you use should be the LogRank p-value. We usually quote variables (genes) something like this:

ESR1, p=0.003 (1.5 [1.2; 1.9])

...for data:

  • LogRank = 0.003
  • HR = 1.5
  • HRlower = 1.2
  • HRupper = 1.9
ADD REPLYlink written 5 days ago by Kevin Blighe56k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1094 users visited in the last hour