Magnesium and diabetes: separate intake associations from supplement trials
One review combined two evidence streams that answer different questions about magnesium and glucose metabolism.

Illustrative food image, not a photograph of the study intervention.
One review with two research questions
This review examined both magnesium intake in prospective cohorts and magnesium supplementation in randomized trials. Its searches covered studies available through May 15, 2019. For the intake question, 26 publications contributed 35 cohorts. For the supplementation question, 26 randomized trials contributed 1,168 participants. Those numbers describe separate bodies of evidence, not a single experiment. The cohorts addressed the incidence of type 2 diabetes, while the trials assessed glucose metabolism and other markers. Treating them as one continuous proof would blur the crucial distinction between an association with future disease and a measured response to a supplement intervention.
Zhao B, Deng H, Li B, Chen L et al.. Diabetes Metab Res Rev, 2020. PMID 31758631 · DOI
The dose-response estimate was observational
The highest magnesium-intake group had 22% lower type 2 diabetes risk than the lowest-intake group in the pooled cohort analysis. Each additional 100 milligrams of daily intake was associated with a 6% lower risk. These are associations across prospective studies, not evidence that adding 100 milligrams to an individual's diet will reduce that person's risk by 6%. The abstract does not supply an absolute risk reduction for a particular reader. It also does not establish that a specific food or a strictly vegan, zero-added-fat dietary pattern was responsible for the association. The intake estimate should retain its observational label.
Zhao B, Deng H, Li B, Chen L et al.. Diabetes Metab Res Rev, 2020. PMID 31758631 · DOI
Trial effects need their measurement labels
In the randomized trials, supplementation reduced fasting plasma glucose with a standardized mean difference of minus 0.32, with a 95% confidence interval from minus 0.59 to minus 0.05. The corresponding estimate for the two-hour oral glucose tolerance test was minus 0.30, with an interval from minus 0.58 to minus 0.02. These are standardized effects, not changes of 0.32 or 0.30 in a familiar glucose concentration unit. They summarize measured glucose responses across studies. Reporting them without the standardized label would make the result appear more directly usable for an individual glucose reading than the abstract supports.
Zhao B, Deng H, Li B, Chen L et al.. Diabetes Metab Res Rev, 2020. PMID 31758631 · DOI
A marker benefit is not a prevention trial
Supplementation also reduced fasting insulin and the HOMA-IR insulin-resistance score, with standardized mean differences of minus 0.17 and minus 0.41 respectively. The review reported reductions in triglycerides and systolic and diastolic blood pressure as well. Its trial sequential analysis supported stability of many glucose-metabolism findings. Even so, the abstract's supplement trials concern markers rather than an assigned intervention shown to prevent new diabetes cases. It supplies no universal dose or formulation for readers to adopt. The cautious reading keeps the encouraging randomized marker findings separate from the cohort evidence about incidence and does not convert either into personal supplement advice.
Zhao B, Deng H, Li B, Chen L et al.. Diabetes Metab Res Rev, 2020. PMID 31758631 · DOI