Dr. Louria asked whether nut consumption confounded the inverse association between coffee consumption and type 2 diabetes mellitus in our study. We reran the analyses controlling for nut intake; the results remained virtually unchanged, suggesting that this confounding is unlikely to explain our findings.

Dr. Schaefer also raised the issue of confounding by other dietary and lifestyle factors. However, coffee consumption tends to be associated with smoking, unhealthy diet, and less exercise. Thus, such confounding would have biased the relationship between coffee consumption and type 2 diabetes mellitus toward a positive association. In an ecological analysis, Dr. Schaefer did not find an association between consumption and diabetes prevalence in 17 countries. However, such analyses can be highly misleading because they are potentially confounded by economic developments, local coffee production, and important dietary and lifestyle factors.

In a cross-sectional analysis, Soriguer and colleagues found a significant inverse association between coffee consumption and risk for diabetes or impaired glucose tolerance. They found significantly lower glucose and insulin levels 120 minutes after an oral glucose tolerance test among coffee drinkers, but it is unclear whether this effect is due to caffeine or other ingredients in coffee.

The Glasers suspected that adding milk or sugar to coffee might explain the inverse association between coffee consumption and diabetes. Such an explanation is unlikely because the amount of sugar and milk added to coffee is presumably much smaller than that consumed in other foods or beverages. Although we did not specifically assess the amount of milk added to coffee, further adjustment for total milk consumption did not appreciably alter our results.

Finally, Dr. Gerber suggested that exclusion of diabetic patients at baseline might have resulted in an artificial inverse association between coffee consumption and incidence of diabetes because coffee-induced early diabetes would have been selectively excluded from the cohort analysis. Such a scenario is highly unlikely because of the long-term follow-up (12 to 18 years) of the participants and a large number of patients with incident cases. Moreover, even at baseline, diabetic participants tended to consume less coffee than nondiabetic participants. Because such a cross-sectional correlation may be confounded by changes in coffee-drinking behavior after a diagnosis of diabetes, it is critical to conduct prospective analyses of coffee consumption and diabetes risk.