Reduced incidence of AD with NSAID but not H₂ receptor antagonists

Methods.

Results.

Table 1 compares the characteristics of participants categorized by their reported use of the six different classes of medicines. The few participants (only 1 to 2%) who were unable to provide data on medicine use tended to be older and less educated. Users of each medication class were more likely than nonusers to be women, except for aspirin compounds. Users of aspirin compounds and other non-NSAID pain relievers tended to be slightly older than nonusers, but there were no other notable differences between users and nonusers of these medicines in age or level of education.

Table 2 shows the crude and covariate-adjusted estimated hazard ratios of developing AD among users versus nonusers of each class of medicine. Any lifetime use of NSAID was associated with a marginally reduced risk of AD. No evidence suggestive of a protective effect was apparent for aspirin compounds, H₂RA, or any of the “control” medicines.

Table 3 presents the results of additional analyses specifically with nonaspirin NSAID as a target exposure group. We first tested interaction terms of NSAID with age (Model 1) or with the presence of one or more APOE ε4 alleles (Model 2). The age interaction term may suggest that the inverse association of NSAID and AD decreases with age. By contrast, the interaction with APOE genotype was weak.

We then examined the apparent influence of duration and recency of NSAID use. Longer duration was associated with greater reduction in risk, so that a significant inverse association appeared between AD and use of NSAID for >2 years (Model 3). Separation of current and former NSAID users revealed a reduced risk only for the latter (Model 4). Among both current and former users, there was a suggestion of greater reduction in risk with increasing duration of use (Model 5). While there was suggestive evidence of an effect among current users who reported substantial prior exposure, the nearly null results among current users were not appreciably altered for those with longer periods of exposure extending forward into the Wave I to II interval (data collected at Wave II) but before the estimated year of onset for dementia (data not shown).

A similar pattern of association with duration and recency was observed with aspirin compounds. Individuals who used aspirin compounds for >2 years had an estimated hazard ratio of 0.57 (95% CI 0.31 to 0.99) for development of AD. Former users with >2 years of reported aspirin exposure showed a still lower estimated hazard ratio of 0.24, but with relatively small numbers under analysis, the 95% CI (0.01 to 1.11) was relatively broad and included 1.0.

Discussion.

These analyses replicate and extend earlier findings from prevalence data in Cache County suggesting a protective effect of NSAID, but with an important clarification. Specifically, in this prospective study, we found increasing duration of nonaspirin NSAID use was associated with stronger reduction in AD risk, whereas current use was not associated with reduced risk unless its duration had extended >2 years prior to the Wave I interview.

Our findings are consistent with those from the Baltimore Longitudinal Study on Aging23 and the Rotterdam Study,25 and they appear to reinforce the “lag period” for effect noted in those two studies. Considered together, the three studies suggest that sustained use of NSAID may delay or prevent AD only if they are taken during a critical window in the latent stages of the disease, before damage to the integrity of the brain is sufficient to provoke demonstrable cognitive symptoms.40 This critical window may end several years prior to the onset of dementia.1,41⇓ If correct, this interpretation could help explain some conflicting findings in the literature and might also explain the disappointing negative results from recent treatment trials with anti-inflammatory agents.28-30⇓⇓

We also found that sustained use of aspirin compounds for >2 years may be similarly associated with reduced risk of AD. This observation reinforces the possible neuroprotective effect with aspirin noted in our previous analyses of prevalence data,21 although results from other studies on aspirin as a separate category have been mixed.22,23,25⇓⇓ As we noted previously, more than two-thirds of Cache County respondents who reported regular aspirin use stated that such use was in low dosage (≤325 mg/day) for purposes of cardiovascular prophylaxis. Aspirin, like other NSAID, inhibits COX, which catalyze the synthesis of prostaglandins. Notably, however, aspirin’s inhibition of COX results from its irreversible covalent bonding (hence, steric inhibition) of the enzyme’s active site.42

It is widely conjectured that, to the extent NSAID protect against AD, they do so by attenuating inflammatory responses in the brain.43 We note, however, that low-dose aspirin is not believed to have potent anti-inflammatory effects. Unfortunately, a lack of precise exposure data limited our abilities to investigate dose effects with NSAID. However, new results from the Rotterdam Study25 and a recent report from Australia22 both note that the apparent neuroprotective effect of NSAID is at least as great with low (analgesic) as with higher (anti-inflam-matory) doses. These findings could therefore suggest other possible mechanisms for NSAID including antiplatelet effects,22 reduction of glutamate-related excitotoxicity (thought to involve COX-2 in postsynaptic signal processing),44 or inhibition of free radical production by intraneuronal COX-2.45 In addition to these effects, recent studies have shown that NSAID may reduce amyloid β-protein load and plaque deposition in the brains of transgenic mouse models.46,47⇓

As was true in our previous analysis of prevalence data from Cache County,21 we found no evidence that the inverse association of NSAID and AD varied with APOE genotype. To our knowledge, in fact, no study has suggested a strong influence of APOE genotype on this relationship. We did, however, find suggestive evidence that the association between NSAID and AD varies with age. One previous study reported a stronger apparent effect of NSAID in those younger than 85,24 and another study reported a similar association between age and a possible AD-protective effect of statin use.48 The current results are no more than suggestive, however, and more investigation is needed.

In what we believe to be the first prospective study to examine another question, we found no evidence that current use of H₂RA is associated with incidence of AD. These results are at odds with those from earlier retrospective and cross-sectional investigations, including our own.8,21,31⇓⇓ We have no direct explanation for the discrepancy but wonder about the declining use of H₂RA (and, especially, of potent prescription formulations) in recent years, as these agents have been largely supplanted by proton pump inhibitors. Such declining use was presumably responsible for our observation that very small numbers of Cache County subjects had used H₂RA for periods extending several years prior to the Wave I interview. These small numbers precluded meaningful analyses of the relation between extended former use and incident AD—exactly the sort of analyses that were needed here to clarify the apparent effect with NSAID. If we are correct, the opportunity for new observational studies relating H₂RA use and incident AD may be behind us.

Among its strengths, this new investigation in the Cache County Study attempted to follow an entire population (not a sample). Response rates were high, thus reducing the threat of selection bias. The large size of the study enabled us to test several a priori hypotheses with reasonable statistical power while controlling for important potential confounders. Prospective studies such as this one can avoid two important limitations of cross-sectional designs. First, they classify exposures from data collected at a time when participants, having been effectively screened for dementia, should be able to provide more reliable information about their current and former use of NSAID, including use of OTC formulations not typically available from other sources. Second, analysis of incidence data mitigates the so-called prevalence case bias in which exposures that are associated with altered survival after onset of illness may be mistaken for those that alter incidence.

Our study has several limitations, however. We ascertained exposures to the six classes of medicines by self-report at the initial contact, and the period of follow-up was only 3 years. Individuals who developed AD during this relatively brief interval may have already begun experiencing cognitive impairment at the time of their initial contact, thus promoting their selective failure to recall prior medication use, especially former use dating back >2 years previously. We suggest that this is not a likely explanation of our findings with NSAID, however, because similar inverse associations were not observed with any of the “control” medications when we examined their lifetime use or former versus current use (data not shown). It seems difficult to explain why participants should selectively under-report the former use of NSAID but not these other medications. Nonaspirin, non-NSAID analgesics, in particular, are typically used for many of the same indications as OTC NSAID and therefore should be vulnerable to the same reporting biases. The relevant difference, we suggest, is that these other analgesics do not inhibit COX.

We also note that this study was conducted in a relatively homogeneous population with unusual sociocultural attributes. Over 90% of the population in Cache County are members of the Church of Jesus Chris of Latter-Day Saints and therefore tend not to use tobacco or alcohol. They are also relatively highly educated and have access to quality health care. Probably as a result, they have unusually low rates of chronic disease and long life expectancy.32 Although we have no reason to believe that such characteristics influence the relation of NSAID use and AD, it is certainly possible that our findings would not generalize to other populations.

In the end, the only sure way to demonstrate whether NSAID can delay or prevent the onset of AD is to conduct randomized prevention trials. Our findings suggest that several years of observation may be required before significant differences would likely emerge in such trials between NSAID-treated and placebo-treated groups. As we have suggested recently in relation to similar findings with postmenopausal hormone replacement therapy (Zandi et al., submitted for publication),⁴⁹ caution may be in order when interpreting early results from such trials, and patience may be rewarded only in the long run.