I bumped into this question by googling myself for the theory behind Hubaux&Voss limit of detection, LOD.
I can add to this topic that one can estimate such a LOD by using the Chromeleon platform's calibration feature. I doubt that my input has any value to the initial question after 3 years, but it might be an useful pointer to other (re)searchers.
In a nutshell, it is derived in terms of precision of the whole calibration at a certain preset confidence limit (e.g. set it to 95%).