In their letter, Solfrizzi and colleagues raise concerns similar to those they have previously expressed in response to our work on this topic. They suggest the possible use of the alternate cut point method applied by Mitrou et al1 (ie, rather than using the 50th percentile, assigning points to intakes ≥25th percentile or <75th percentile for “beneficial” or “detrimental” components of the MeDi calculation) for a sensitivity analysis. Unfortunately, using this method is not practical in our study owing to power limitations. More specifically, in applying this method in the cognitively normal to incident MCI analyses (N = 1393), a MeDi score of 6 would be assigned to 23 subjects (1.7%), a score of 7 to 9 subjects (0.6%), a score of 8 to 1 subject (0.1%), and a score of 9 to 0 subjects. Overall, only 33 of 1393 subjects (2.4%) would obtain a MeDi score greater than 5. In adjusted analyses, the risk of incident MCI for those 33 subjects was 0.38 (95% confidence interval, 0.12-1.19; P = .09). In applying this alternate cutoff method in the MCI to incident AD analyses (N = 482), a MeDi score of 6 would be assigned to 6 subjects (1.2%), a score of 7 to 1 subject (0.2%), and a score of 8 or 9 to 0 subjects. Overall, only 7 of 482 subjects with MCI (1.5%) would obtain a MeDi score greater than 5. In adjusted analyses, the risk of incident AD for those 7 subjects was 0.23 (95% confidence interval, 0.03-1.71; P = .15). Summarizing, in both the incident MCI and incident AD analyses, the effect sizes (as judged by the coefficient magnitudes) are clearly quite strong, but owing to obvious power limitations the confidence intervals are wide, precluding statistical significance. To put things in perspective, Mitrou and colleagues were able to apply the alternate MeDi score calculation method because of the luxury of including in the study 214 284 men and 166 012 women from the American Association of Retired Persons database. Overall, our study was conducted in a multiethnic urban cohort from New York, New York, and these individuals are unlikely to strictly consume the foods typical of Mediterranean countries. A “true” MeDi may have even more beneficial effects in cognition, but this does not invalidate our MeDi score calculations and certainly does not negate the benefit of a diet closer to the Mediterranean type in our cohort.
Scarmeas N, Stern Y, Manly JJ, Schupf N, Luchsinger JA, Mayeux R. Mediterranean Dietary Pattern, Mild Cognitive Impairment, and Progression to Dementia—Reply. Arch Neurol. 2009;66(7):910-916. doi:10.1001/archneurol.2009.129