Multi-Arm Randomized Controlled Trial for Diagnostic Imaging
After highlighting the weaknesses of cohort data, Norton provides a concrete description of what a proper RCT would look like. He outlines a 3-group design: one group gets real SPECT imaging and receives customized treatment based on the results; a second gets real imaging but only standard care advice; a third undergoes a sham scan—sitting in the machine, hearing the noises, but no actual imaging—and receives standard care. This design separates the effects of (a) the imaging ritual and expectation, (b) the image itself, and (c) the personalized advice. He explains that without such a design, any observed improvement cannot be attributed to the scan because baseline group differences and placebo are unaccounted for. The sham group is critical because it tests the hypothesis that being told you had a "super duper special imaging" is what drives improvement, not the scan or the tailored advice.
Randomization ensures groups are statistically equivalent at baseline so that outcome differences reflect the intervention. Blinding (sham) prevents expectation bias, which is especially potent for subjective mental health outcomes and can mimic drug-level effects. Comparing tailored advice vs standard care within the real-imaging arm tests whether the scan adds useful information beyond standard practice.
What you would typically do, this could be where you have two groups or even four groups. ... basically, you're putting them in the apparatus and everything, but you're not actually doing any imaging. And then you are giving them standard of care. So, I guess it'd be three groups. And that is going to reduce bias.

