This study supports the findings of previous reports [1
] suggesting that use of NSAIDs for at least six months is associated with a reduced risk of AD. This association appears to be more robust among APOE-ε4 carriers than non-carriers, although the difference in associations between these two groups was not statistically significant.
The MIRAGE Study includes the largest number of well-characterized AD cases and family controls to date, and this large sample size permits adjustment for important potential confounders, as well as the power to examine effect modification by APOE genotype and ethnicity. The subjects without dementia were first-degree family members of AD cases, providing some degree of informal matching on age, socioeconomic status, and health-seeking behavior.
However, these results must be interpreted in light of some methodological limitations. Data on NSAID use was collected with a single retrospective question that did not distinguish between aspirin use and non-aspirin NSAID use. Moreover, while non-demented participants reported on themselves, a proxy historian reported on most of the demented individuals.
Differential reporting is a potential source of bias in a study that uses self-report on most of the non-demented subjects, yet relies on surrogate respondents for all of the subjects with AD. Asymmetric data collection is difficult to avoid when cases are cognitively impaired, but may be more accurate than expected in AD patients where the surrogate historian has a long association with the subject. We addressed this potential bias by performing an independent validation study to determine the accuracy of surrogate information on a number of questions, including the same questions used in this report about NSAID use. [15
] This study found substantial reliability on the NSAID item (kappa = 0.70). While a validation study comparing proxy historians for non-demented persons does not perfectly mirror the situation in which proxy historians report on demented individuals, our study revealed excellent concordance for surrogate responses from most categories of relatives.
This result is consistent with those of prior studies which found a similar association despite differences in study design (cross-sectional [1
] vs. case-control [3
] vs. prospective [7
]), sampling frame (family members [10
] vs. registry-based [11
] vs. general population [1
]), ascertainment of exposure, type of medication considered (aspirin [2
] vs. non-aspirin NSAIDs [1
] vs. 'any' NSAID [4
]), duration of exposure (current [1
] vs. any history of use, duration ranging from a week to at least six months [3
]), and degree of matching or adjustment (usually adjusted for age, sex, and education, less frequently APOE genotype [7
While many studies have examined the association between NSAID use and risk of AD, few have examined the impact of APOE genotype on this association. The Cache County Study [9
], the Canadian Study for Health and Aging (CSHA) [8
], and the Rotterdam Study [7
] adjusted for APOE and tested for effect modification and found none. But they had fewer AD cases, and, in the CSHA had a smaller proportion of genotyped subjects. The Rotterdam Study [7
] reported separate odds ratios for APOE-ε4 carriers and non-carriers, but this sample did not have any subjects who were both APOE-ε4 carriers and who reported long-term use of NSAIDs. They found no difference in risk between those with at least one ε4 allele compared to ε4 non-carriers among subjects who used NSAIDs between one month and two years.
Our data suggest an enhanced protective benefit of NSAID use among those with ε4. A smaller protective effect was also evident among those lacking ε4. In our sample there were relatively few AD cases who were not ε4 carriers, thus the appearance of different patterns of association between NSAID use and AD risk among APOE genotype subgroups may be spurious. This difference could also have arisen as a result of bias and confounding. The genotype-specific association could be explained by differential inclusion of subjects into the study on the basis of APOE-ε4 carrier status and NSAID use. This might occur if there were differential mortality, according to APOE genotype, among those with AD who had a history of NSAID use; or if for any reason among NSAID users APOE-ε4 carriers were less likely than non-carriers to be diagnosed with AD (or conversely, if among non-users of NSAIDs AD was more likely diagnosed in APOE-ε4 carriers compared to non-carriers). Differential recall could also give rise to this observation. However, these explanations are unlikely because subjects were not selected on the basis of APOE genotype. It is also possible that APOE-ε4 carrier status is a proxy for differentially distributed unmeasured confounders related to NSAID use such as inflammatory disease processes.
Alternatively, our results imply that NSAID use affects AD risk differently between APOE-ε4 carriers and non-carriers. For example, because ε4 carriers are inherently more vulnerable to AD, there is a greater opportunity for attributable risk reduction. This explanation does not imply biological interaction between NSAIDs and ε4. On the other hand, the ε4 isoform may have greater pro-inflammatory properties [21
] and ε4 individuals may be more responsive to the benefits of NSAID use than those lacking ε4.
Examination of this finding in prospective studies and clinical trials of sufficient power (such as the ADAPT Study [22
], a prospective trial of anti-inflammatory use in the prevention of AD) to detect effect modification by APOE-ε4 carrier status is needed. Such confirmation would provide critical insights into the mechanisms by which APOE isoforms modulate AD risk and into novel therapeutic strategies.