Anti-inflammatory Drugs Protect Against Alzheimer Disease at Low Doses

Context  Anti-inflammatory medications have an inverse association with Alzheimer disease (AD).

Objectives  To examine at what doses this anti-inflammatory drug effect occurs and whether other medications and/or International Classification of Diseases, Ninth Revision, Clinical Modification diagnoses affect the association.

Design  Subjects 75 years and older from a random population sample were classified by consensus using International Classification of Diseases, Ninth Revision, Clinical Modification diagnoses. Drug associations with different types of dementia with and without the International Classification of Diseases, Ninth Revision, Clinical Modification diagnoses as well as dosage data were analyzed.

Setting  The Centre for Education and Research on Aging, Concord Hospital, Concord, Australia.

Patients  The Sydney Older Persons Study recruited 647 subjects (average age, 81 years). A total of 163 patients were given diagnoses placing them in different dementia categories and were compared with 373 control subjects. Of the patients with dementia, 78 had AD without vascular dementia, 45 had vascular dementia (permissive of other dementia diagnoses), and 40 had other dementia diagnoses (without AD or vascular dementia).

Main Outcome Measures  Fifty drugs or drug groups were subjected to a 2 (drug used vs drug not used) × 4 (dementia and control groups) χ² analysis. Drugs with inverse associations were identified and potential confounders (logistic regression) and dosage data (exact small sample 1-tailed tests) analyzed.

Results  As expected, there was an inverse association between nonsteroidal anti-inflammatory drugs and aspirin (and unexpectedly angiotensin-converting enzyme inhibitors) and AD. This association was not observed with vascular dementia or any other diagnoses. Analysis showed no evidence for a dosage effect, ie, responses were equivalent for low and high doses.

Conclusions  This study does not support a high-dose anti-inflammatory action of nonsteroidal anti-inflammatory drugs or aspirin in AD. Potential mechanisms for the beneficial effects of these medications are discussed.

ALZHEIMER DISEASE (AD) is the most common cause of progressive dementia in older people and is increasingly common with advancing age beyond the seventh decade of life. A small delay in age of onset of AD, with identification of modifiable risk factors, could improve the quality of life for older people and carry enormous public health and family carer benefits.¹ There is evidence that the use of anti-inflammatory medications is inversely associated with AD,² improves cognitive performance,^3,4 and slows decline in patients with AD.⁵ This has been highlighted in several,^6,7 but not all,⁸ longitudinal studies considering the protective effects of these drugs on AD. Because these drugs are classified as anti-inflammatory, it has largely been assumed that the mechanism underlying their effectiveness is against the brain inflammation associated with AD.⁹ This appeared particularly reasonable in the initial studies showing inverse associations between true inflammatory disorders (rheumatoid arthritis^10,11 and leprosy¹²) and AD. However, general populations have since been studied without consideration of drug dosages.

There is only limited data on whether the effect on AD is only found with high-dose anti-inflammatory drug use. At these doses many of the drugs have a high toxic effect in older persons,¹³ as highlighted in the double-blind, placebo-controlled studies performed to date.^3,4 Additionally, most studies have not assessed the interaction with other medications, and multiple drug usage is common in elderly persons. This study aims to determine whether the effect of anti-inflammatory drugs is restricted to usage at anti-inflammatory doses, whether the effect is confined to patients with AD, and whether there are any drug interactions that may underlie the effect. We report results on the use of 50 drugs in subjects with specific dementias (AD, vascular or multi-infarct dementia [VaD], and other dementias) in 536 community-living subjects aged 75 years or older. This age group was selected to analyze those populations at greatest risk of AD, and those in which apolipoprotein E genotype has limited, if any, effect on disease development.^14,15 Drugs with an association with AD diagnosis were further examined for potential confounders and the dose level for the effects to further understand the potential mechanism(s) of drug action.

The Sydney Older Persons Study, its subjects, and medical assessments have been described elsewhere.^16,17 In brief, 647 community-living subjects 75 years of age and older were recruited using local area probability sampling schemes with a response rate of 78% yielding approximately equal numbers of men and women (average age, 81.0 years; age range, 75.0-97.8 years). Five hundred thirty-seven subjects participated in medical and neurological assessment by several of us experienced in geriatric medicine (G.A.B., H.M.C., L.M.W., and W.S.B.). This included a standardized, 2½-hour medical history taking examining both past and current health with a neuropsychological test battery, a disability assessment, and detailed medical and neurological examination, resulting in systemic, neurological, and psychiatric diagnoses. A history of current drug usage, with dosage, was obtained including inspection of: drug prescriptions; all prescribed and over-the-counter drugs, and drug containers; and the medicine cabinet in the home.

Case notes with field diagnoses from all 537 clinical interviews were reviewed by a consensus group (neurologist [H.M.C.], geriatrician/interviewer [L.M.W.], and neuropsychologist [H.P.B.]), who applied International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) categories¹⁸ for up to 40 possible systemic and neurological diagnoses. This yielded 163 cases derived from 8 ICD-9-CM dementia or organic brain syndrome categories. Subjects in the AD category fulfilled National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association (NINCDS–ADRDA) criteria for AD¹⁹; the other 7 categories defined by the consensus group using Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition and ICD-9-CM criteria were VaD, alcohol-related dementia, head injury–related dementia, other dementias, frontal lobe syndrome, parietal lobe deficits, and organic brain syndrome from systemic insults. The remaining 373 subjects were considered to be control subjects, free of any dementia or organic brain syndrome diagnosis within these categories. The following 4 mutually exclusive groups were identified: controls; subjects with AD, non-VaD; subjects with VaD ± AD; and subjects with other dementias, non-AD, non-VaD. (One control group subject was lost to analysis because of lack of drug data.)

Controls (n = 373) were those with no diagnosis of dementia or organic brain syndrome.

Subjects in this group were those with a diagnosis of AD alone (n = 78) by NINCDS–ADRDA probable AD criteria (n = 66) and AD by NINCDS–ADRDA possible criteria mixed with any dementia or organic brain syndrome diagnosis—excluding VaD (n = 12).

Subjects in this group were those with a diagnosis of VaD ± AD (n = 45) by Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria either alone or mixed with any dementia or organic brain syndrome diagnosis—including AD.

Subjects in this group were those with a diagnosis of other dementias, non-AD, non-VaD (n = 40), ie, any other dementia or organic brain syndrome diagnosis not mixed with AD or VaD.

Drug usage, for the 50 drugs reported, was analyzed among these 4 groups with specific attention to the 3 anti-inflammatory drugs—aspirin, nonsteroidal anti-inflammatory drugs (NSAIDs), and corticosteroids. Statistical analyses of drug prevalences in controls against drug prevalences in other groups were all based on the Fisher exact (randomization) test to address the following questions.

1. Is the inverse association between dementia and anti-inflammatory drug use specific to AD or also seen with other dementias? To examine this question the anti-inflammatory drugs were assessed across all groups.

2. Are the inverse associations due to subjects with dementia forgetting, or underreporting, or having a lower number of prescriptions, or a reduced need for drug use? To examine this we included group 4 in the analysis, a group with dementia excluding AD and VaD but open to these sources of potential bias. We also analyzed acetaminophen (Paracetamol), as a commonly used (but not anti-inflammatory) analgesic, predicting its use would not vary across the 4 groups.

3. If vascular risk factors lead to high aspirin pre scription and use (in low dose), could failure to isolate subjects with clear vascular risk factors (ie, those diagnosed with VaD) lead to a genuine AD-protective effect of (low-dose) aspirin therapy going unnoticed? To examine this, we excluded all subjects with VaD (pure or mixed) from group 2 and analyzed VaD ± AD separately in group 3.

Many of the 50 drugs or drug groups were used by few subjects. Aspirin (n = 142), NSAIDs (n = 112), and corticosteroids (n = 75) (nonsystemic corticosteroids—topical and aerosol [n = 52] and systemic [(n = 15]) were selected for more detailed analysis. Of the other 47 drugs, those used by 50 or more subjects were subjected to 2 (drug used vs drug not used) by 4 groups χ² analysis, examining whether or not drug usage was independent of dementia subtype. Those with P values less than .10 were also selected for further analysis; this yielded 3 further drugs–angiotension-converting enzyme (ACE) inhibitors (n = 68; P = .09), other antirheumatics (n = 52; P = .01), and diuretics (n = 169; P = .02). Dosage data were examined for NSAIDs and aspirin in only controls and subjects with AD who were taking these drugs. Nonsteroidal anti-inflammatory drug dosage levels (for 17 NSAIDs) were evaluated by a rheumatologist who applied a naproxen (Naprosyn) dose equivalent as follows: low dose, less than 500 mg/d; medium dose, 500 to 1000 mg/d; and high dose, more than 1000 mg/d. Aspirin therapy was classified similarly as a low dose if less than 175 mg/d was taken or medium if more than 175 mg/d was taken. Only 1 control subject of 111 was taking in excess of 650 mg/d (1200 mg/d). The Fisher exact (small sample) 1-tailed test was used to determine if there was a significant difference in the proportion of persons in each group taking low vs medium and high drug doses.

Table 1 lists the 50 drugs on which simple usage data (used vs not in use) were available. Thirteen drugs or drug groups, selected for analysis because they had 50 or more users, showed a pattern of reported drug usage that was not significantly different between controls and cases in any of the 3 dementia groups. These drugs were aerosols (atrovent or ventolin), antacids with aluminium, any other prescription drugs, benzodiazepines, β-blockers, calcium channel blockers, digoxin, histamine₂ (H₂) antagonists, laxatives (bulkers or osmotics), nitrates, other topical preparations, acetaminophen, and vitamins, iron, and minerals. That is, overall reported drug usage was not higher in controls. This suggests that problems with underreporting and/or taking medications can generally be discounted in this data set.

Table 2 gives the prevalence by group of the drugs of interest: NSAIDs, aspirin, corticosteroids, other antirheumatics, acetaminophen, ACE inhibitors, H₂ antagonists, and diuretics. In terms of the specific questions to be addressed (see "Subjects, Materials, and Methods" section): (1) Both NSAID and aspirin usage was significantly lower in the AD, non-VaD group (group 2), the effect being stronger with NSAIDs; (2) usage of aspirin and NSAIDS was the same in the other dementias (group 4) as in the controls (group 1); usage of corticosteroids and acetaminophen (and the other high-use drugs) was the same across all groups. These latter findings argue against the likelihood that forgetting reporting or prescribing biases could explain the (inverse) associations of NSAIDS and aspirin with the group with AD, non-VaD; (3) the exclusion of VaD comorbidity from this group with AD may have been justified, as group 3 (VaD ± AD) shows higher aspirin use than controls (albeit not significantly so). Unexpectedly, ACE inhibitors and other antirheumatics showed the same inverse correlation with the group with AD, non-VaD, and lack of correlation with the groups with VaD ± AD and other dementias (groups 3 and 4), as did NSAIDs and aspirin. Diuretics were used more frequently by those with VaD ± AD than by controls and will not be addressed further.

For the AD-specific assocations in Table 2, we examined potential confounders in the following manner: each of the 4 drugs was first fitted as a predictor of AD, non-VaD vs controls in a logistic regression; then each of 22 chosen possible confounders was singly added as a second predictor of AD, non-VaD vs controls to see if the AD-drug odds ratio was reduced in size and/or statistical significance. The confounders fell into 3 groups: the other 3 drugs with an AD association (aspirin, NSAIDs, other antirheumatics, and ACE inhibitors), 10 systemic diagnoses chosen because they may relate to usage of the 4 drugs (osteoarthritis, any arthritis, hypertension, atrial fibrillation, congestive cardiac failure, angina, coronary artery bypass graft, any heart complaint, stroke or transient ischemic attacks, or migraine), and 9 risk factors known or thought to be possibly related to AD (age at interview, sex, educational level, occupation, cigarettes smoked, weekly alcohol consumption, before age 65 years physical activity, before age 65 years walking, and current physical activity). The original AD-drug odds ratios remained unaffected in magnitude in all these 88 ancillary analyses, while statistical significance remained less than .05 in 84 analyses, increasing to a maximum of only .08 in the remaining 4 analyses.

Of 142 subjects taking aspirin, 139 had dosage data available; and of 112 subjects taking NSAIDs, 111 had dosage data available. For the subjects analyzed, there was no overlap between NSAID and other antirheumatics users with approximately 25% of the NSAID-using subjects also taking aspirin. We reasoned that if the AD-drug association was related to higher doses of the drug in question, then among the subjects with AD, non-VaD low-dose usage should be overrepresented for the controls. The distribution of cases in each group according to dose of drug are given in Table 3. While there is no trend toward higher doses of either drug among controls, relative to subjects with AD, non-VaD, the power of the analysis is small because of the number of subjects with AD, non-VaD who were taking such medications. If we examine controls vs subjects with AD, non-VaD and low vs medium to high doses, we obtain Fisher exact (small sample) 1-tailed P values of .57 for NSAIDs and .64 for aspirin. Thus, there is no evidence for a dosage effect, with most of the elderly subjects in our sample taking low doses of these drugs.

In this community-living random sample of older persons, a strong inverse association was shown between the usage of anti-inflammatory medications and the presence of AD (but not VaD or other dementias). Analysis of many potential confounders suggests that the effect occurs because of drug usage. However, the major finding of this study, in which drug doses have been carefully assessed, is that these effects seem unrelated to drug dosage. The use of NSAIDs and aspirin at antiplatelet levels seemed to provide equivalent protection compared with their usage at anti-inflammatory levels. In our sample, most subjects were taking these medications at the lower antiplatelet doses. This suggests that effective protection for elderly persons may be gained at these low doses in addition to the higher anti-inflammatory doses. The low-dose effect of anti-inflammatory medications is likely to occur via a noninflammatory mechanism.

Our results have broad significance as we studied a community-living population of subjects aged 75 years and older. These are the ages at which AD and other neurodegenerative disorders are most likely to occur,¹ and also ages at which adverse drug effects are more common for physiological reasons.¹³ By examining older persons still living in the community, we have knowingly biased the sample toward survivors as these are most representative of the elderly population at risk of AD. However, the use of such subjects is likely to reduce variability in a number of clinical factors that may be important for disease pathogenesis. In particular, we selected this age group to decrease the effect of apolipoprotein E genotype on the disease development.^14,15 Other clinical effects may be similarly reduced in this population. This would be countered by the increased prevalence of other pathologic brain conditions in elderly persons.¹ Because of such brain aging, it was important to characterize all types of dementia in this population. In particular, the exclusion of subjects with VaD, including mixed cases with both AD and VaD, from the analysis proved to be important. Using this strategy our results suggest that protection against AD may occur at these older ages with a range of anti-inflammatory medications.

Age seems to be a more important factor than commonly recognized in evaluating disease mechanisms and appropriate treatment strategies. The Baltimore Longitudinal Study of Aging⁶ recruited 1686 volunteers that were young (<65 years) with educational levels higher than a college degree. Both younger age and higher educational level decrease the risk of AD.²⁰ This study found a reduced risk of AD among those volunteers taking NSAIDs or aspirin, and that there was an increased protective effect the longer NSAIDs were used. The Medical Research Council Treatment Trial of Hypertension in Older Adults⁷ recruited 2651 subjects mean (±SD) age 70 ± 3 years with moderate hypertension and high premorbid intelligence. The results of this study showed the following: (1) a small, but significant, protection against cognitive decline in those taking NSAIDs and (2) that this effect was absent in persons older than 74 years. The latter finding may explain the negative results obtained by Henderson et al⁸ in a study of 1045 Australians aged 70 years or older (mean age, 80 years). They found that neither NSAIDs nor aspirin therapy provided protection against cognitive decline or incidence of dementia over a 3- to 4-year period, although drug usage data were only collected at one time point. These longitudinal studies suggest that anti-inflammatory effects may be most potent in persons younger than 75 years. However, the reason for cognitive decline was not evaluated in these elderly populations^7,8 and, therefore, the loss of the effect may be due to confounding neurodegenerative conditions in elderly persons. Our data would support this contention.

The analysis of dosage data for both NSAIDs and aspirin indicates that the AD association may occur at very low doses, although further research on larger samples is required. Of 139 aspirin users no one was receiving a high dose and more than 80% were receiving 175 mg/d or less. For NSAIDs, only 12% of 111 subjects were receiving high-dose naproxen (dose equivalent of >1000 mg/d) and more subjects were on a low (<500 mg/d) than a medium (500-1000 mg/d) dosage. Nonsteroidal anti-inflammatory drug suppression of a primary inflammatory process in the brain affected by AD is unlikely at these low doses. Few studies have investigated the effect of drug dosage. In those studies of the elderly persons examining dosage effects, subjects taking high-dose NSAIDs have poorer memory and decline at a faster rate than those taking low doses.¹³ The protective effect of high-dose nonaspirin NSAIDs in elderly persons seems to be reversed as the risk for memory decline significantly increases in those taking high doses.²¹ Up to 40% of NSAID perscriptions are written for persons older than 70 years, and at least 10% to 15% of the persons at these ages take these medications.¹³ Although the duration of dose was not formally assessed in this study, most of the elderly subjects assessed were taking prescription medications, as described in the literature. Thus, previous studies, along with the results of this study, do not support an anti-inflammatory dose effect and suggest that alternate low-dose mechanisms may underlie the protective effect of these drugs against cognitive decline and AD.

The major known effects of low-dose aspirin and other NSAID therapy are on blood vessels and platelets with inhibition of cyclooxygenase.^22,23 Cyclooxygenase is a major tissue oxidant converting the oxygen in hemoglobin to superoxide radicals. In platelets inhibition of cyclooxygenase reduces platelet aggregation and the release of platelet factors.²³ As platelets are the primary source of β-amyloid in the blood,^24,25 NSAIDs may directly reduce the amount of circulating β-amyloid derived from platelets. Cylcooxygenase also produces superoxide radicals within vascular endothelial cells,²⁶ and decreasing superoxide radicals abolish β-amyloid–mediated vasoactivity and damage.^27,28 Thus, NSAIDs may also reduce endothelial damage through inhibiting cyclooxygenase. Our results suggest that the anti-inflammatory drug hypothesis of AD prevention should be reviewed and alternate mechanisms of low-dose NSAID and/or aspirin drug action considered. Platelet and endothelium dysfunction in particular can be ameliorated at such low doses.

Accepted for publication May 23, 2000.

The work was funded by the Public Health Research and Development Council of Australia project grant support (Drs Broe and Creasey); the Research and Development Grant Application Committee project grant support (Dr Broe); the Sir Zelman Cowen Universities Fund, Sydney, Australia (Drs Broe, Creasey, and Halliday and Mr Brooks); and the Australian and New Zealand Army Corps Health and Medical Research Foundation, Syndey (Drs Broe and Halliday).

We thank Bain Shenstone, MD, for providing the nonsteroidal anti-inflammatory drug dosage rating and Dorothy Edelbrock, BA(Hons), and Enid Sawley, BA, for study management and assessments.

Corresponding author: G. Anthony Broe, FRACP, Prince of Wales Medical Research Institute, High Street, Randwick 2031, Australia (e-mail: BroeT@sesahs.nsw.gov.au).