In initial analyses, we compared the rate of cognitive decline in those who reported taking anti-inflammatory agents during the previous 2 weeks at their baseline interview, to those with no recent use (table ​(table2),2), after adjusting for the primary potential confounding factors (age, sex, education, race). We did not find an association either between aspirin and cognitive decline, or between other NSAID use and cognitive decline.

In separate analyses, we also examined lifetime duration of use of anti-inflammatory agents (table ​(table3).3). After adjustment for age, education, sex and race, we did not find a difference in the yearly rate of cognitive decline in those with short or moderate duration of aspirin use, compared to no use. There was a suggestion that 20 or more years of use may be associated with worse rates of cognitive decline (mean difference = −0.008, p = 0.06). When we conducted analyses further adjusting for major vascular factors, findings were largely similar (for 20+ years: mean difference = −0.008, p = 0.08). In analyses stratified by vascular factors, there were no clear interactions with aspirin use. In addition, because lifetime duration of use could include sporadic use (e.g. for occasional headaches), we separately examined relations after excluding subjects who reported taking an average of less than four tablets per week. However, findings were generally similar in these analyses.

In contrast, for other NSAID use (table ​(table3),3), we found statistically significant associations with lifetime use. For those who reported taking other NSAIDs for less than 5 years, the yearly rate of decline was 0.008 units (p = 0.03) slower than for those with no history of other NSAID use. To help interpret these results, we contrasted the effect estimate for NSAIDs to the effect estimate for 1 year of age in this population. We found that each year of age was associated with a decline of 0.004 standard units; thus, our finding that cognitive decline was slower by 0.008 units among those with short-term NSAID use, might suggest that NSAID use delays cognitive aging by 2 years. The relation appeared stronger with longer duration (5+ years: mean difference = 0.012, p = 0.07). In addition, there was a statistically significant trend of less cognitive decline with increasing duration of other NSAID use (p trend = 0.03). Findings were similar after adjustment for major vascular factors and within strata of vascular factors, and were also consistent when we excluded subjects who reported taking less than four tablets per week.

Although we did not request information on the dose of anti-inflammatory medications, we found a high prevalence of cardiovascular disease and vascular factors among those reporting aspirin use (table ​(table1),1), suggesting that aspirin may have primarily been of low doses for inhibiting platelet function; moreover, long-term use of regular-dose aspirin is generally accompanied by gastrointestinal side effects and is rarely advised or tolerated. Thus, our findings regarding aspirin appear consistent with a recent clinical trial [4], in which an average 10 years of treatment with low-dose aspirin had no effect on cognitive decline, as well as with several observational studies [11,12,13].

Few large-scale, prospective studies have examined the association of other anti-inflammatory agents to cognitive decline, and many existing studies have important limitations. For example, one study found that among 3,809 older persons, there was little relation between either aspirin only or use of any NSAID and risk of cognitive decline over 6 years (RR = 0.94, 95% CI 0.79–1.12, and RR = 0.89, 95% CI 0.69–1.16, respectively) [12], but there was no information on the duration of use; thus, this study likely had limited ability to detect associations. In the largest investigation among 16,128 older women, Kang and Grodstein [13] reported findings consistent with those observed here: there was a borderline significant decreased risk of cognitive decline among long-term users of nonaspirin NSAIDs (primarily ibuprofen: RR = 0.77, 95% CI 0.57–1.05). Interestingly, a recent clinical trial including two nonaspirin NSAIDs (naproxen and celecoxib) reported that their use increased the risk of AD [2]. With very short follow-up in the trial, the authors suggested that NSAIDs might be harmful at late stages of AD development, and only provide benefits in older persons with intact cognitive health. Consistent with this hypothesis, we found slightly stronger inverse relations between ibuprofen and cognitive decline after excluding subjects with the worst cognitive function at baseline (defined as the bottom 10% of the distribution). Clearly, this issue merits additional research.

There are several potential mechanisms by which anti-inflammatory agents might prevent early cognitive decline. Inflammation appears to be directly associated with neuronal dysfunction and death [14], and NSAIDs may reduce neurotoxic inflammatory responses in the healthy brain [15]. Specifically, NSAIDs appear to mitigate neurotoxicity of microglia, which are believed to regulate inflammatory reactions in the central nervous system [16]. Moreover, in transgenic mice, inflammatory proteins induce amyloid-β accumulation and deposition [14]. Noninflammatory mechanisms could be involved as well. For example, ibuprofen lowers levels of amyloid-β in the brain independent of cyclooxygenase activity, while aspirin does not have this effect [17]. Finally, the antiplatelet effects of anti-inflammatory drugs, especially aspirin, are known to improve cerebral blood flow and reduce vascular disease [18], and thus might also prevent cognitive decline via vascular mechanisms.

Interestingly, our epidemiologic study may provide clues as to the most likely mechanism of action. The lack of apparent relation between aspirin and cognition in this study and in clinical trials suggests that anti-platelet or vascular effects may not be important. However, the aspirin results likely provide little information on the anti-inflammatory effects of NSAIDs in the brain. Although we did not ask subjects their reason for taking anti-inflammatory drugs, regular ibuprofen use may have been for arthritis, as indicated by the high prevalence of joint pain in those reporting nonaspirin NSAID use (table ​(table1),1), and thus of sufficient dose to limit inflammation. Additionally, reduced amyloid-β levels with ibuprofen [17] may also explain those findings.

Our study has some limitations: the baseline question on current use of anti-inflammatory agents only inquired about use in the last 2 weeks. Because it is somewhat unlikely that short-term use would impact a chronic condition such as cognitive decline, the analyses of recent anti-inflammatory drug use may underestimate any benefits of regular, long-term use. However, the additional data from the cycle two interview on lifetime duration of anti-inflammatory drug use along with the number of tablets taken per week should substantially reduce this problem. Nonetheless, there may be misclassification of reported lifetime history, as well as timing of use (e.g. at middle age vs. at older ages), which would lead us to underestimate relations. To assess the possible extent of such misclassification, we compared subjects’ reports of their history of use, provided at the cycle 2 interview, to their reports 3 years earlier (at the baseline interview) of recent anti-inflammatory use. Responses provided were highly consistent; for example, 95% of those who reported no lifetime history of aspirin use had also reported no recent use 3 years earlier. Confounding is an additional limitation, and is an important issue in observational studies. While we cannot eliminate confounding, we adjusted all results for the major risk factors for cognitive decline, including vascular factors. Moreover, especially for nonaspirin NSAID use, there were very few differences in major known demographic or health factors between users and nonusers, reducing the possibility that unmeasured confounding factors might have substantially influenced findings. Finally, in studies of older persons, high death rates and the possibility of competing risks can be a problem. In the analyses presented here, all participants had at least one follow-up cognitive assessment, and the large majority (approximately 80%) were still alive for additional follow-up; thus, it is unlikely that competing risks could have had a substantial influence on our findings.