I am injecting prostate cancer cell lines PC3 and LNCaP (1 million and 2 million respectively) into the flanks of nude/J mice. In some of the mice (from both control and treatment group) the tumors do not grow and the read is "zero". Is there justification to use these as missing values in statistical analysis? Or do I have to keep them in as "zero" values, which skews the data a lot and I am not reaching significance, when it is clear that in the tumors that did grow there is a large difference. Any help is appreciated!
Dear Jean,
I would certainly not include non growing tumors as zero values in statistical analyses.
You have to provide a maximum of details in your "M&M", explaining that:
1) you grafted x mice / experimental group, and that there were y groups (deatailing each group),
2) after xx days, yy mice developed tumors with "measurable sizes" of a - b (ranges of min - max values; median; mean +/- SEM) in groups I, II, III, etc...
3) "thus", you discarded from any subsequent analyses those tumors which size was not measurable at day xx, meaning that group I included x tumor-bearing mice, group II y, group III z, etc...
4) the treatment began on day x post-tumor grafting procedure.
In that case, you offer a crystal-clear story to your potential readers.
Best regards
Robert
Thanks for the input guys. I think I also agree with you both, the zero data was skewing the outcome a lot and my error bars were huge!
i have another question, what about if the "treatment" group was hypothesized to decrease tumor growth, in this case would you also exclude any zero growth tumors? Thanks!
Lets come to the first question zero growth in stats flank assay using Nude mice? What Robert has described is a typical M&M and I agree that the story should clear but you don't want to include the zero in your stats because it will give huge error bars etc. Consider this way you have to satisfy the larger no of your reader, again I understand the concern but I will report the zero and will explain in my results and again in discussion why there is a huge error bars completely my opinion and as a matter of fact I am doing in one of my publication in preparation.
Second question yes this time it is more important as the value of zero has increased in your second study due to the nature of the study.
Dear Jean,
You can certainly exclude from your statistical analyses any mouse whose tumor growth = "0" BEFORE starting treatment (which was my understanding of your question in my first response).
I you experience "zero growth" during treatment, you have to notice these events as such in your statistical analyses.
As I do not know how many tumors / mice you have per group, I also do not know which statistical test(s) you intend to do.
I would perform a statistical test relying on median, not on mean (+/- SEM) values. In that case, you can satisfy any reader's sensitivity to biological + statistical data ... as suggested by Syed (if and only if my understanding is right ...).
Best regards
Robert
OK thanks everyone.
I am injecting one group of mice with a cell line that has an empty vector (control), the "treatment" group is injecting a cell line with a vector that contains a gene of interest, where the gene may inhibit tumor growth, or suppress it. I use a t-test assuming unequal variance, two tailed. Do you think I would still exclude the zero in this case, or include them?
I am thinking that the cell line itself may be inappropriate to inject (LNCaP prostate cell lines) I have spoken with others who have had inconsistent tumor growth with this cell line, they always include the zero in their analysis (hence why I am asking here on RG as I thought excluding them would be the best option).
In fact, Jean, the control should be mice with injected LNCaP cells without any construction.
The possibility remains that even the empty vector modifies the behavior of LNCaP cells when injecting them to mice.
My opinion is that if you have also "zero" values in a group with actually "naive" LNCaP cells, then I think that you can remove the zero values from your other two groups. In that case, you should use a statistical test analyzing the median, not the mean values. You should thus demonstrated that your empry vector did not modify the growth as compared to the actual control group.
This is the situation I described in my previous response.
If you have no zero values in the naive / actual control group, but you have them in the other two groups, you should (must?) keep the zero values in the two vector-modified LNCaP group (empty and real) and see what happens in the three groups. Once more I woould look on the median variations with an appropriate test, not on the mean onces.
It is also important to know at which timing are you "looking" at the growth. In other words, after how long after having grafted the cells?
Best regards
Robert
Thanks Robert,
I will have a chat with the lab and see what they think about your suggestions, you make very valid points.
I am measuring tumor growth on a weekly basis after injection, for approximately 4 weeks.
Thanks,
Jean
We are measuring tumor growth 3 times a week (Monday - Wednesday - Friday) for rapidly growing tumors and 2 times a week (Monday - Thursday) for "less rapidly" growing tumors.
When are you starting the treatment post tumor grafting?
On which basis?
All the best
Robert
- Badges
- Science topic