Coffee and diabetes incidence: decaf matters to the interpretation
A large prospective-study synthesis found inverse associations for both caffeinated and decaffeinated coffee, without proving prevention.

Illustrative food image, not a photograph of the study intervention.
Large numbers, observational comparisons
The coffee review pooled 28 prospective studies involving 1,109,272 participants and 45,335 cases of type 2 diabetes. Follow-up ranged from 10 months to 20 years. Investigators searched PubMed and Embase for cohort or nested case-control studies published from 1966 through February 2013, then modelled the relationship between coffee intake and diabetes incidence. These are prospective observational comparisons, not trials assigning people to drink coffee. The large participant count makes this a substantial body of follow-up evidence, but does not change that design. Its central finding should be expressed as an association between reported consumption and later diabetes, rather than as proof of prevention.
Ding M, Bhupathiraju SN, Chen M, van Dam RM et al.. Diabetes care, 2014. PMID 24459154 · DOI
Reading the dose-response without prescribing it
Compared with no or rare coffee consumption, the reported relative risks of diabetes were 0.92 for one cup per day, 0.85 for two, 0.79 for three, 0.75 for four, 0.71 for five and 0.67 for six. For the one-cup category, the 95% confidence interval was 0.90–0.94; for the six-cup category, it was 0.61–0.74. These figures describe the modelled observational gradient across intake categories. They are not trial results showing what happens when an individual starts drinking that amount. In particular, the lowest reported relative risk should not be converted into a recommendation to consume six cups each day.
Ding M, Bhupathiraju SN, Chen M, van Dam RM et al.. Diabetes care, 2014. PMID 24459154 · DOI
Decaffeinated coffee was associated too
For each additional cup per day of caffeinated coffee, the pooled relative risk was 0.91, with a 95% confidence interval of 0.89–0.94. The corresponding estimate for decaffeinated coffee was 0.94, with an interval of 0.91–0.98. The reported P value for the difference between those associations was 0.17. Thus the abstract supports inverse associations for both types, not a demonstrated statistically significant difference between them. The decaffeinated finding is also a reason not to recast this review as a caffeine trial. The investigators compared coffee-consumption patterns in prospective studies; they did not isolate caffeine by randomly assigning it to participants.
Ding M, Bhupathiraju SN, Chen M, van Dam RM et al.. Diabetes care, 2014. PMID 24459154 · DOI
Incidence is not treatment response
The outcome in this synthesis was the risk of developing type 2 diabetes, with 45,335 cases recorded across the included studies. That differs from asking whether coffee improves HbA1c in people who already have diabetes. The review's observational design also leaves causal prevention unestablished, even though both caffeinated and decaffeinated coffee showed inverse associations. A careful summary therefore keeps three labels together: coffee consumption, subsequent diabetes incidence and prospective observational evidence. None of the reported intake gradients constitutes a personal treatment instruction. The study is informative about patterns seen during follow-up, while leaving the effect of deliberately changing an individual's coffee intake unresolved.
Ding M, Bhupathiraju SN, Chen M, van Dam RM et al.. Diabetes care, 2014. PMID 24459154 · DOI