Which statistical approach should be used to compare healthy and sick patients?

#1
Hello everyone,

I'm stuck trying to figure out what the best strategy is for analyzing and distinguishing between healthy and sick patients.

I have a dataset that was analysed with 20 healthy and 20 sick patients (each of them is different person).

I collect data using one of the spectroscopic techniques (Raman). Regarding this technique, you analyse the data in the matter of the interactions between the incident light and sample. My goal here is to use statistics to distinguish between two groups in their spectral bands. (Do they differ or not? Is the difference significant if they are? or How much they differ?, etc.)

Is there any suggestion? or anything that you can transfer me to learn more?

Thanks for any feedback.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Is the spectroscopy used to define them as sick? Is the 50% prevalence the natural prevalence in the population for sick or is it artificial. How is the sprectral band data formatted - what does it look like (provide an example).
 
#3
Is the spectroscopy used to define them as sick? Is the 50% prevalence the natural prevalence in the population for sick or is it artificial. How is the sprectral band data formatted - what does it look like (provide an example).
In my case, they have been defined as whether sick or not in the hospital. I already knew what the samples were about, but they had never been subjected to the technique until I started.)

I've attached an image from a 2019 study, just to clarify what I mean. (thanks for the contribution; Nargis, H.F. et al. Raman spectroscopy of blood plasma samples from breast cancer patients at different stages. DOI: 10.1016/j.saa.2019.117210)

The image shows wavenumbers on the x-axis. Different intensities, mean, and standard deviation can be seen in the y-axis in relation to those numbers.

My interest begins here. Some bands (for example, the very sharp one near 1000cm-1) can be statistically analyzed to differentiate between control and sick groups. What confuses me is which method is correct or should be used. (perhaps a basic student t-test?)

nargis2019.jpg
 

Karabiner

TS Contributor
#4
A robust method could be the "nonparametric" Mann-Whitney U test.
It does not compare means, as the t-test does, but ranks. It is not affected
by problems such as outliers or peculiar distributions.

Did you determine beforehand which areas and/or parameters you want to
compare between groups? Otherwise, eyballing complex patterns and then
applying statistsical tests of significance to interesting-looking part of the
pattern will probably lead to false results (see Texas sharpshooter fallacy).

With kind regards

Karabiner