Hi, anyone out there can help with finding an answer to how we calculate the sample size to validate the Protein biomarker(s) in the field with ELISA? Initially, we identified some proteins with Quantitative proteomics analysis, and now we want to test those proteins with ELISA kits in the field samples.
Any help will be appreciated.
Cheers,
There are several things to consider. If you are analyzing each of the biomarkers independently, then you will need an estimate of the mean and standard deviation of the abundances between the two cohorts (e.g., disease and control). The higher the overlap, the more samples you will need to establish statistical difference. I encourage you to look at the wiki page on receiver operator characteristic curve. If you are interested in cross-products between different biomarkers (i.e., a multi-marker panel), then you will need a whole lot more.
A rule of thumb is to expect 80 independent samples for each biomarker in each cohort, if you are looking at the biomarkers independently. If you aren't looking at cross-products, you could get by with 80 disease and 80 controls. If you are looking at cross-products, you have to add 80 additional independent disease samples for each biomarker evaluated. Note, 80 is a rule of thumb value based on where the p-scores asymptote. Of course, this assumes that the biomarker abundances in each cohort are normally-distributed. Sometimes they aren't, so you have to test that using a Chi-squared test to a normal distribution. If biomarker abundances aren't normally distributed, then you are into non-parametric statistics.
The lower the standard deviation overlap between the test cohorts, the fewer samples that you can get away with. Here is where a pilot study can be helpful. You can screen a smaller number of samples in each cohort to get an estimate of the fraction of individuals that will overlap in biomarker abundance. Armed with that estimate of the means and standard deviations for each cohort, you can predict the n that will reach your desired degree of confidence.
Only if you are looking at multiplexing the biomarkers. If you just want to evaluate these 5 independently, then you can get by with n in cohort 1 and n controls. You can use the same samples for all the biomarkers IF you aren't going to try to use the biomarkers in combination with each other. What is n is the question? At n = 80 you get an F value that approaches a constant (i.e., it doesn't pay much to go for higher n). You can see this if you plot the F values corresponding to p=0.05 as a function of n. If the means of the biomarkers are so far apart in each cohort that the 95% confidence limits about those means don't overlap, then you can get by with lower n. The issue is the variance around the mean. The issue is the number of overlapping abundances (i.e., false positives and false negatives). You should have some idea of the variances for these biomarkers from your scouting experiment. This estimate can be used to predict what n will give you 95% confidence in the result, assuming the mean and variance remain the same in future samples. Note, the above discussion assumes that the abundances of the biomarkers in each cohort are normally distributed. You have to test for this first using a Chi-squared test.
Thanks a lot, Luke V Schneider for such a helpful response. As we will be testing these candidates with commercially available ELISA kits, I don't think we can multiplex them, We will test individual biomarkers. I will get in touch if need further help regarding calculations.
Thanks again.
Multiplexing is a multimarker assay which is usually an algorithm that includes the quantities of each marker from different ELISA assays. Genomic Health does this with qrtPCR of various mRNA markers for the OncotypeDx Breast Cancer treatment guidance assay.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size