LarryQC_SD said:Steady state is the thing most people miss.
This is where I part company with the consensus forming above. The trial means are being read too generously in this thread. STEP populations were selected, supported, and titrated by protocol, and the real-world curves are consistently a few points worse. That difference is not noise, it is what happens when you remove the study infrastructure.
If somebody has the primary source to hand I would rather cite it than paraphrase it.