The only thing you can do is to set the value of the non-detectables to something below the LLOQ. If the LLOQ is 20pg/mL, I usually set it value to something like 19 or so. You won't be able to calculate a true mean or median. On the graph you can clearly see that it's below the detection limit. And this will also allow you to run stats.

The only other option is to remeasure the samples using a ultrasensitive kit...

Dear you can double the required volume of the sample and later divide the concentration by 2.

Shen-An Hwang is right. For graphing purposes, you could just write LOQ, if possible. This may require using more concentrated or higher volumes of samples, improving the sensitivity of your ELISAs or using a technique that is more sensitive.

First a clarification: limit of detection is the concentration (LOD) of any compound that can be detected. Conventionally, it is 3x s (standard deviation) of the concentration of the read-out. Lower limit of quantification (LLOQ) is the lowest concentration that can be measured quantitatively. Conventionally, it is 10x s. This implies that the lowest concentration is not per se the LLOQ. As many standard curves in ELISAs are not linear, extrapolations of values from calculated standard curves beyond the standard range concentrations are at risk of giving erroneous results. Moreover, at lower concentrations, there is a risk of cross-reacting compounds giving falsely elevated or lowered results.

What can you do? I presume you did not check LLOQ or it is not given by the kit provider

- re-analyse with a more sensitive kit (a good lesson: first select a kit by testing it first. Also carefully read whether the kit is suitable for your sample. Try to convince the provider to give you a test kit or oblige him to re-embourse in case of no result)

- Dilute the lowest standard concentration (check if it is possible with higher standards too)

- Use more sample (check if it is possible with higher samples too). even in combination with above. BUT: always check

- Use non-parametric or categorial statistics when many samples are under the lowest standard. When you have many samples under the lowest standard you do not have a gaussian or other parametric distribution when you set them al at LOQ of 1/2LOQ (as some do). I cannot advise you as I am not a statistician.

I'm having the same problem. I was planning to just not compare groups where there were significant numbers of samples below the LOQ but I have mused that an alternative would be to perform categorical statistics in this case as Robert suggests. Ie. 60% of serum samples from participants with exacerbation contained measureable levels of IL17 vs 20% measureable in subjects who were stable p