Reduced prevalence of AD in users of NSAIDs and H2 receptor antagonists

Methods.

We planned the Cache County Study as a survey of the prevalence and incidence of AD and other dementias in relation to genotype at APOE or other genes, and to several environmental influences including medications. For the latter, we sought specifically to test hypotheses that medicinal use of cyclo-oxygenase inhibitors (e.g., aspirin, ibuprofen) or H2RAs would delay onset or reduce risk of AD. Because no association with AD had been suggested with several other drugs used for similar indications (non-aspirin non-NSAID analgesics, simple antacids, or stomach remedies other than H2RAs), we studied the latter agents as “control” exposures to discover whether effects with the target drugs, if any, would be specific. Such specificity could argue against the influence of underlying ailments (confounding by indication). It would also argue against artifact from generally impaired recall of medicines by demented individuals or their collateral informants (recall bias), or from a tendency of better educated individuals who use more medicines to develop less dementia (confounding with education or health habits).

To date, we have completed the ascertainment of individuals with dementia that was prevalent in 1995 through 1996, and have begun the detection of incident cases. Prospective analyses of associations between prior use of NSAIDs and H2RAs and incident AD will not be available for another year. Here, we report the comparison of medicinal drug use in prevalent AD cases and the cognitively intact remainder of the population.

Results.

Principal drugs included in the aspirin, NSAID, and H2RA categories, and their frequencies, are shown in table 1. Table 1 also lists drugs included in the control categories of non-aspirin, non-NSAID analgesics, antacids, and other stomach remedies. Table 2 shows adjusted prevalence odds ratios (aPORs) from the principal comparisons. Current use of aspirin alone, NSAIDs alone, and histamine H2RAs alone were all significantly associated with reduced prevalence of AD. The 95% CIs for each of the three control exposures included the null value of 1.0, with point estimates ranging between 0.63 and 1.46.

Current use of aspirin and non-aspirin NSAIDs (see table 2), or of two or more non-aspirin NSAIDs (aPOR = 0.17, 95% CI = 0.04 to 0.48, p = 0.0002), showed a stronger inverse association. We observed similar results with current use of two or more H2RAs (aPOR = 0.21, 95% CI = 0.01 to 1.08, p = 0.06). There was no evidence for additional strength of association among the subjects who used both aspirin and H2RAs (no NSAIDs), or both NSAIDs and H2RAs (no aspirin), but the strongest association was seen with use of all three target classes (see table 2). The numbers in all sets of doubly or trebly exposed subjects were small.

Addition to these models of terms for use of non-aspirin non-NSAID analgesics, simple antacids, and other stomach remedies yielded OR estimates with CIs that included the null value of 1.0 (see table 2). Although the CIs were predictably wider, we observed similar results throughout in the nested case-control analyses (table 3). For example, the case-control aPOR for NSAIDs alone was 0.47 (cf. 0.43 in the principal analyses); for aspirin alone it was 0.44 (cf. 0.50); for H2RAs, 0.50 (cf. 0.42).

Statistical adjustment for arthritic or other joint pain, peptic ulcer disease, or gastroesophageal reflux disease (indications that promote use of these treatments) produced no appreciable change in the related estimates from these models. Moreover, there was no real improvement in the models’ explanatory capacity when terms for these indications were added to the regression models (e.g., estimated OR for painful/inflammatory joint disease 1.29; 95% CI = 0.90 to 1.83; p = 0.16). Possible variation in AD occurrence across subgroups defined by joint disease and NSAID exposure was negligible (p = 0.477). Likewise, a subsidiary analysis of subgroups with varying numbers of APOE ε4 alleles suggested no appreciable variation (for all comparisons, p > 0.10).

Discussion.

We have described a series of prevalent case-cohort and case-control analyses in the Cache County Study, conducted to provide initial tests of a priori hypotheses that three classes of compounds might reduce the risk of AD. The analyses showed an inverse association between the use of each target medicine class and the odds of AD, whereas little or no such association was observed with other medicines used for similar indications. The strength of the inverse associations was increased among small numbers of subjects who were currently using two or more agents from the same class, or a combination of aspirin with non-aspirin NSAIDs.

These findings may be noteworthy for several reasons. The studies were population-based, and should thus be relatively invulnerable to selection biases. They were undertaken specifically to test a priori hypotheses and, accordingly, included salient control exposures as well as specific questions to control confounding by indication and other sources of error. The findings with aspirin are interesting because 65% of subjects who used this drug stated they did so for cardiovascular prophylaxis (hence, presumably, in low dosage). A significant inverse association with AD thus appeared with sustained doses of aspirin that would not be thought of as anti-inflammatory in any conventional sense. When the explanatory models were revised to allow for confounding by indication (e.g., painful/inflammatory joint disease), there were no appreciable changes in the study estimates for medicine exposures. These estimates should have moved appreciably toward the null value (1.0) if the NSAID indicators were standing in for what truly is an effect of joint disease on occurrence of AD.

The study’s main limitation is probably its classification of exposure status. There is some chance that physicians were not prescribing H2RAs or other medicines for AD patients because of concerns about worsening their cognitive deficits, or about decreased tolerance of these medicines leading to confusion or delirium. Ideally, one should investigate occurrence of AD in relation to medication use before onset of dementia. Unfortunately, after the clinical assessments of the nondemented participants in the 19% subsample, a comparison of self-reported lifetime history of drug use versus collateral informant reports for the same subjects showed chance-corrected degree of agreement to be substandard (kappa values between 0.25 and 0.40), with substantial under-reporting by collateral historians. This finding was unexpected because we had seen better agreement between self- and collateral-reported information in prior studies of highly motivated research volunteers.10 Upon discovering it, we opted to rely on observed current drug use, as others have done,21 and correspondingly to specify current age (rather than age at onset of AD) in the regression analyses.

In these circumstances, we speculate that the observed relationship (stronger inverse association with use of two cyclo-oxygenase inhibitors or two H2RAs) might reflect a tendency toward more sustained use or higher doses. An alternate explanation is that the number of cyclo-oxygenase inhibitors and H2RAs is an index for a still-undiscovered and therefore uncontrolled confounding variable. In the latter case, until this confounding variable can be specified and measured or otherwise eliminated, it will remain a source of distortion in studies of AD protection by these agents.

The apparent lack of an additive effect with NSAIDs and H2RAs contradicts an earlier finding.10 We have no explanation for this discrepancy other than the small numbers of doubly exposed AD cases in both studies.

An obvious trivial explanation for the association with H2RAs—that these medicines were being taken to treat NSAID-induced gastropathy—is not supported here. Rather, these analyses appear to confirm earlier observations of an inverse association of AD with H2RA use.9 The actions of cyclo-oxygenase inhibitors and H2RAs might converge biologically if both drug classes decreased efficiency of synaptic response or postsynaptic signaling in cells with NMDA-type glutamate receptors.3 Such cells are vulnerable to excitotoxic death under conditions of excessive stimulation, a phenomenon that has recently received attention in relation to the pathogenesis of AD.22-24

One recent investigation from the Netherlands failed to confirm the H2RA–AD association.25 We have speculated26 that the apparent contradiction may result from the Dutch study’s exclusion of OTC formulations and the limited duration (typically, 8 weeks) of prescription H2RA use that is typical in the Dutch health service. Nonetheless, it is possible that different H2RAs have different effects. Our sample is not large enough to yield reliable estimates for the individual H2 blocking agents. Further resolution of this conflict may come from incidence cohort analyses in the Cache County Study next year, or may require even larger sample sizes. Such prospective studies avoid the so-called prevalent case bias (association of exposures with factors that affect prevalence by altering duration of illness). They also avoid the need to rely on collateral drug history data from demented subjects, because those who later develop AD will have previously given an adequate self-reported exposure history. However, even prospective studies remain vulnerable to unsuspected sources of confounding. Definitive demonstration of the efficacy of NSAIDs, aspirin, or H2RAs in preventing AD can come only from randomized controlled prevention trials.

Appendix

Other Cache County Study investigators involved in the project: James Burke, MD; Tony Calvert, RN; Barbara Gau, MSW; Michael Helms, MS; Ara Khachaturian, BS; Carole Leslie, MA; Tiffany Newman, MA; Brenda Plassman, PhD; David C. Steffens, MD; Martin Steinberg, MD; JoAnn Tschanz, PhD; Kathleen Welsh-Bohmer, PhD; Nancy West, MS; and Bonita Wyse, PhD.