For those who are inclined and interested as to how synthetic EPO testing is performed in the laboratory setting, this is a really illustrative manuscript from a group in the Netherlands looking at the performance of sarcosyl-PAGE and isoelectric focusing on detecting recombinant human EPO.
I haven't done an SDS-PAGE in over 8 years (my research stays strictly in silico now), but other people in my lab routinely do similar techniques for research purposes. Just from discussions that occur in lab meetings when the students and staff are showing their gel blots to the group, I can see how any diagnostic or classification problem that relies upon looking at a PAGE blot is going to be subject to some diagnostic variability from inter-observer, inter-specimen, and inter-test sources. It's hard sometimes to decide what you're looking at.
Methodologic highlights from this study: the investigators randomly assigned 48 trained cyclists to either 8 weeks of rHuEPO or placebo and were able to study the detection performance of both the biochemical methods as well as an approximation of the athlete biological passport system -- which I didn't have a great appreciation for how ABP worked until reading this article. Interesting findings (to me) were (1) false-positive rates on the screening tests were in the ~4-6% range; (2) sensitivity on the screening tests were disappointingly low in the ~60% range, so around 40% of the drugged samples were inappropriately called as clean; and (3) sensitivity of ABP seemed much better than PAGE or IEF without any major loss of specificity.