That is a very good question - did you read the paper (
Enhanced Wellbeing of Adults with Type 2 Diabetes following Multi-Vitamin and Mineral Supplementation for Three Months in a Randomised, Double-blind, Cross-over Pilot Study? Curiously, it's not nearly as positive as the Diabetone website seems to suggest:
In our pilot study, a multinutrient supplement failed to significantly improve the glycemic and lipemic profile of patients with type 2 diabetes
Our results must, however, be treated with caution as not only was this a pilot study carried out without prior estimation of study volunteer numbers required to reach statistical signifance using a power calculation, but the measurement of wellbeing was a secondary outcome.
Furthermore, and on account of these uncertainty factors, no corrections to the level of significance were applied for errors incurred in multiple testing.
Simply put, since they are looking at so many things they can't be sure that any of the findings are true (and not due to chance):
If you test a hypothesis at p=0.05 each then there's a chance of 5% that you will get be positive result by pure chance. If you do that 20, say, times then you should expect at least one positive result by pure chance. If you want your overall conclusion to be correct with 95% chance (p=0.05) then each hypothesis has to be tested with p= 1 - (0.95) ^ (1/20) = 0.0025
XKCD provides
an illustration
[Here, there are at least 6 hypothesis being tested (3 wellbeing scores, HbA1c, triglyceride, cholesterol), so each hypothesis needs to be tested with p=0.00639; none of the tests are significant at that level]
Furthermore, regardless of
statistical significance there's the question of
practical significance - e.g. anxiety is down by 1.042 from 3.5ish to 2.5ish out of 18. Is that good enough value for money?
