0
$\begingroup$

I have to create a hypothetical study focusing on the relationship between sBCMA (soluble B-cell maturation antigen in blood) and the expression of BCMA on bone marrow cells in patients with multiple myeloma (MM). My main goal is to investigate whether there is a correlation between levels of sBCMA and the degree of BCMA expression on bone marrow cells in MM patients. I hypothesize that higher levels of sBCMA are associated with an increase in BCMA expression on these cells. To do this, I am using three groups of MM patients who either exhibit high sBCMA, medium sBCMA, or low sBCMA. Within each group, I then measure BCMA expression.

I am aware that a Pearson correlation analysis will be necessary for my experiment. However, I am wondering if additional tests are needed, such as the Kruskal-Wallis H-test...?

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Just to be clear, so there are basically two variables of interest: the 3-level sBCMA variable and the (continuous?) BCMA expression variable? And do you have one measurement of BCMA expression from each "patient" or several from each patient? $\endgroup$
    – Sointu
    Commented Apr 3 at 11:53
  • 1
    $\begingroup$ I've edited out claims of urgency and local details relating to your university. Being close to a deadline is what it is, but not our problem. Questions that are urgent to the OP don't get more attention here than any others. Details relating to your university could be embarrassing at best and are in any case irrelevant to your statistical question. $\endgroup$
    – Nick Cox
    Commented Apr 3 at 12:17
  • $\begingroup$ Yes Sointu, you're correct, there are indeed two main variables of interest. I only have one measurement of BCMA expression from each "patient". $\endgroup$
    – youknow 321
    Commented Apr 3 at 12:41

1 Answer 1

Reset to default
2
$\begingroup$

So, if you have 2 variables of interest, one of which (sBMCA) is basically an ordered factor, Pearson correlation is not the best choice (for that, both variables should be numeric continuous variables). I would start with linear regression with sBMCA as a categorical predictor of BCMA and check whether regression assumptions are met via regression diagnostics.

If your data meets the regression assumptions, you could proceed with the regression and use a linear contrast of sBMCA (meaning that the model tests whether BCMA expression increases when sBCMA level increases), or consecutive contrasts (low vs medium and medium vs high), or run all pairwise comparisons.

Also, not related to statistics but it doesn't sound like this is an experiment as it sounds like you are not manipulating anything, just measuring.

Improve this answer
$\endgroup$
2
  • $\begingroup$ I have categorized the groups for sBCMA based on existing literature, but I will also measure it at the beginning of the research, to confirm what I found on literature. At this point, I am not entirely certain whether sBCMA should be considered categorical or numeric continuous. What do you think? $\endgroup$
    – youknow 321
    Commented Apr 3 at 13:20
  • 1
    $\begingroup$ That's a substance knowledge question more than a statistics question. Very generally, you shouldn't categorize a continuous variable, but if this is the predominant way in the existing literature, I'd follow it especially in a thesis. Nothing stops you from doing both, though. $\endgroup$
    – Sointu
    Commented Apr 3 at 13:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.