Arthritis and anti-inflammatory agents as possible protective factors for Alzheimer's disease

Case-control studies where arthritis was considered as a risk factor.

Seven case-control studies were conducted in which arthritis was included as a risk factor (see Table 1). They involved cohorts from the United States, ^[5,6,11-13] Australia, ^[7] China, ^[9] and Canada. ^[10] In the initial study of Heyman et al., ^[5] a nested cohort of 40 AD cases, who had previously participated in a comprehensive, genetic, and epidemiologic study at Duke University Medical Center, was compared with a matched community control group selected by random-digit telephone dialing. Exposure to various factors was determined by personal interview. This was followed by the study of French et al., ^[6] who compared 78 clinically diagnosed male AD cases from the Veterans Administration Medical Center in Minneapolis with two different control groups, a hospital control group selected from records at the same medical center and nonhospital controls from surveys of the surrounding areas. Exposure was determined by interview with random reconfirmation in 15% of the interviews. Odds ratios were given only for the hospital control group. Broe et al. ^[7] compared a group of 178 clinically diagnosed AD cases obtained from a Sydney, Australia, catchment area and 178 controls selected from names suggested by the general practice physicians referring the AD cases to the clinic. Exposure to various risk factors was determined by personal interview. Breteler et al. ^[8] published a reanalysis of these early data, but this report was not included in the overall analysis of Table 1 because of the redundancy of the data presented. Li et al. ^[9] compared a series of 70 clinically diagnosed AD patients from several cities in China with a control group drawn from the registration offices of the neighborhoods of the patients. Relatives of both the patients and the controls were used as informants for risk factors, which were analyzed by direct interview. In the Canadian Health Study, ^[10] AD subjects were recruited from both communities and institutions across Canada through initial cognitive impairment screening, followed by clinical examination. They were matched against cognitively normal people identified by the same screening procedure. Risk factor questionnaires were completed by proxy respondents, usually a close relative, for both cases and controls.

The two studies of Breitner et al. ^[11,12] were designed to eliminate, or at least minimize, genetic factors. In the initial study, Breitner ^[11] identified 50 elderly twin pairs from across the United States, 26 of whom were homozygotes, where one twin developed AD 3 or more years before the other. Subjects suspected of having AD were examined at home by standard neuropsychological test procedures. Similar procedures were followed for unaffected twins. Exposure information was obtained by self-reporting or, where that was not possible, from surrogates. In the second study, ^[12] sibships were selected from 45 pedigrees that had been studied at Duke University. They were not connected to the twin studies. For inclusion sibships met one of two criteria: presence of two or more cases of AD with onsets that differed in age by 3 or more years; or at least one affected individual and one or more unaffected sibs who had survived 3 or more years beyond the onset age of the index case. Telephone interviews were then conducted with subjects or collateral informants to determine exposures.

In all but the final study of Breitner et al., ^[12] arthritis was included as one of a multitude of medical conditions and stood out as the only one consistently identified as a protective or negative risk factor. Details regarding the type of arthritis were not specified in any of these studies. Age was carefully controlled in all the studies but genetic factors were addressed only in the two studies of Breitner et al. ^[11,12] In any event, genetic factors would only confound the results if arthritis was distributed differently in sub-populations genetically susceptible to AD from in the general population, and if these subpopulations constituted a significant proportion of the total AD cohort included. The odds ratios for the two studies by Breitner et al. (0.64 and 0.45, see Table 1) were in the same range as in the other studies, diminishing the possibility that genetic makeup was a confounding factor in any of the studies.

The results of the seven studies were tested for homogeneity (see Table 1). The hypothesis of homogeneity was not rejected (p(H) = 0.194). Therefore a combined odds ratio was calculated, with a value of 0.556 (95% confidence interval 0.442 to 0.700), which was highly significant (p < 0.0001).

Case-control studies where rheumatoid arthritis was considered as a risk factor.

Rheumatoid arthritis (RA) is much less common than osteoarthritis. It is typically of earlier onset and is more severe than osteoarthritis, requiring aggressive anti-inflammatory treatment. Only two case-control studies considered rheumatoid arthritis as a risk factor. ^[13,14] Graves et al. ^[13] selected 130 clinically diagnosed AD cases from two Seattle, Washington clinics. They were matched with a control group made up of friends, surrogates, or nonblood relatives of the patients. Exposure to a variety of factors, including rheumatoid arthritis, was determined through telephone interviews with patient and control surrogates. The study of Jenkinson et al. ^[14] was restricted to AD and rheumatoid arthritis. They surveyed 192 consecutive inpatients of a London geriatric unit, four of whom were excluded from the study due to schizophrenia. The remainder included 96 AD and 92 non-AD cases. The presence or absence of RA was determined in each case by clinical examination using criteria established by the American Rheumatism Association.

The results are shown in Table 1. The numbers of cases were very small, and the results differed. Graves et al. ^[13] found 6.2% of their AD patients to be exposed compared with 3.8% of their controls. The age-adjusted odds ratio of 1.18 was not considered significant. By contrast, Jenkinson et al. ^[14] found only 2% of their AD cases suffering from rheumatoid arthritis compared with 13% of their controls, for a highly significant odds ratio of 0.17. The test for homogeneity of the two studies gave a p(H) value of 0.0393, i.e., <0.05, indicating that homogeneity was not acceptable. Further case-control studies are necessary for reliable odds ratios to be calculated.

Population-based studies in which rheumatoid arthritis was considered as a risk factor.

Three population-based studies considered rheumatoid arthritis as a risk factor (see Table 2). McGeer et al. ^[4] examined the records of 973 rheumatoid arthritics aged 65 and older who had been followed for long periods of time in established clinics in Saskatchewan and Arizona. Only four cases of dementia (0.41%) were identified. They also gathered statistics from the discharge records of hospitals in three Canadian provinces and Arizona. Of 7,490 patients aged 65 or older with the diagnosis of rheumatoid arthritis, a concomitant diagnosis of AD was found in only 0.39% of these cases. The prevalence of AD in the 65 and older general population was conservatively estimated to be about 2.7% based on studies using comparable survey methods. This is considerably less than the 5.1% reported later by the Canadian Study of Health and Aging Working Group from their 1991 to 1992 survey. ^[27] However, that survey used detailed diagnostic methods that detected early any mild AD cases that are often missed in more general surveys.

Beard et al. ^[15] examined data from the Rochester, Minnesota AD incidence cohort covering the years 1950 to 1975. They identified 521 cases of rheumatoid arthritis, of which 23 (4.4%) subsequently developed AD. They concluded that the prevalence of AD among rheumatoid arthritics reported by McGeer et al. ^[4] was too low. Since the age specificity of their survey was not included, it cannot be determined whether this incidence was higher or lower than would be expected in an age-matched general population. Nevertheless, it is unlikely that age distribution alone could account for such a large difference. One possibility might be differing methods of treatment for rheumatoid arthritis used in the periods covered by the two surveys. The series of Beard et al. was derived from records of an era when salicylates were widely used for rheumatoid arthritis. All but one of their AD patients had received such treatment. By the time of the survey of McGeer et al., salicylates had been largely replaced by the more powerful NSAIDs. Breitner et al. ^[12] did an orthogonal comparison of the effect of aspirin and other NSAIDs in reducing the risk of AD. Their data suggested that aspirin alone produced a similar but weaker effect compared with other NSAIDs.

Myllykangas-Luosujarvi and Isomak, ^[16] using official statistics, investigated the causes of death during the year 1989 for the whole Finnish population over age 55. Only two patients with rheumatoid arthritis died of AD, which was 0.12% of the rheumatoid arthritic subpopulation. In the general population, 227 individuals died of AD, which was 0.54% of the cohort. The odds ratio was 0.23 with a 95% confidence interval of 0.065 to 0.826. They concluded their data supported the suggested negative association between AD and rheumatoid arthritis.

For overall statistical analysis, the rheumatoid arthritic clinic data of McGeer et al. ^[4] and Myllykangas-Luosujarvi and Isomaki ^[16] were combined, and the hypothesis of homogeneity was not rejected (p(H) = 0.735, see Table 3). The overall odds ratio was 0.194 (95% confidence interval 0.092 to 0.410, p <0.0001, see Table 3). The study of Beard et al. could not be included due to the lack of age specificity.

The most logical explanation for the negative associations between arthritis generally, and rheumatoid arthritis specifically, is the long-term use of anti-inflammatory drugs. NSAIDs are by far the most commonly used agents for these conditions. Steroids are probably the next most commonly used class although their use is restricted to the more severely affected patients, and the administration is usually intermittent. Both classes of agents have been considered as possible protective factors in epidemiologic studies of AD.

Case-control studies in which steroids were considered as a risk factor.

(Table 3) summarizes the four previously described case-control studies in which steroids were included as a risk factor. ^[10-13]

Each of the four studies showed an odds ratio below 1, although only in the twin study of Breitner et al. ^[11] was the odds ratio (3/12 or 0.25) considered to be statistically significant (p = 0.04). Interestingly, in their subsequent larger study of sibships at equal risk for AD, Breitner et al. ^[12] found the odds ratio to be of considerably lower significance than in the initial study, although the numbers exposed in the AD and non-AD groups combined were still small (9/151, odds ratio 0.57). The case numbers in the studies by Graves et al. ^[13] and the Canadian Health Study ^[10] were considerably larger than in the Breitner studies, with the odds ratios being very close (0.73 and 0.75, respectively). The hypothesis of homogeneity was not rejected (p(H) = 0.538, see Table 4), and the overall odds ratio was estimated. The value was 0.656 (95% confidence interval 0.431 to 0.999, p = 0.0492).

Case-control studies in which NSAIDs were considered as a risk factor.

Three of the four case-control studies in which steroids were considered as a risk factor also included NSAIDs as a risk factor ^[10-12] (see Table 3). Here the total numbers were larger, no doubt due to the more common use of NSAIDs. The Canadian Health Study ^[10] had an odds ratio of 0.55, while that of Breitner et al., ^[12] considering NSAID users not taking histamine H2 blockers, had an odds (hazard) ratio of 0.19.

The odds ratios for the three NSAID case-control studies did not support rejection of the hypothesis of homogeneity (p(H) = 0.264, see Table 3), and so the combined odds ratio was estimated. This was 0.496 (95% confidence interval 0.343 to 0.716, p = 0.0002). Thus, even though fewer studies were done than for steroids, the odds ratio was smaller and the level of significance greater.

When all the steroid and NSAID studies were considered together, as a single class of anti-inflammatory agents, a combined odds ratio estimate of 0.559 was obtained (95% confidence interval 0.424 to 0.738), which was very highly significant (p < 0.0001).

Population-based studies in which NSAIDs were considered as a risk factor.

Three population-based studies considered NSAIDs as a risk factor (see Table 4). ^[17-19] The methods in each case were completely different, so the results did not lend themselves to combined analysis. Lucca et al., ^[17] in their Italian study, approached the problem by considering whether AD patients used NSAIDs to a greater or lesser extent than elderly controls. They reviewed the data from two clinical trials involving 195 AD patients. They found NSAID users to be only 0.8% and 0% of the total cohorts in these trials designed to test the efficacy of other agents. They compared these NSAID use figures with data obtained from medical records in three general surveys of elderly patients: a sequential sample of the elderly from general practice and two randomized surveys of very old patients drawn from general practitioner records. They found NSAID users were 22.8%, 20.3%, and 18.5% of these control cohorts. They compared osteoarthritic pathology in the AD and control groups as a criterion to determine if they differed in their indication for NSAID use. They found that osteoarthritic pathology was present in 22% of the AD patients compared with 26% of the controls. They concluded that, because of the comparable osteoarthritic pathology, NSAID use was far lower than expected in the AD group. They considered that NSAID use may have reduced vulnerability to AD in the control group.

Andersen et al. ^[18] surveyed all eligible community-dwelling and institutionalized individuals in the Ommoord district of Rotterdam aged 55 years and over. They found only 1.4% of 365 NSAID users to be suffering from AD, compared with 2.5% in the total cohort of 5,893. The odds ratio adjusted for gender, age, education, and benzodiazepine use was 0.38. As a control subgroup of drug users, they chose a cohort of 365 topical medication users. The odds ratio using this control group increased to 0.54. In either case, they concluded their findings were compatible with a possible protective effect of NSAIDs on the risk of developing AD.

Rich et al. ^[19] reviewed the records of 210 patients in the Johns Hopkins Alzheimer Disease Research Center to evaluate the role of NSAIDs. Thirty-two AD NSAID users were compared with 177 AD non-NSAID users on clinical, cognitive, and psychiatric measures. They found that the NSAID group had a significantly shorter duration of illness at study entry, and, over a 1-year longitudinal study in the clinic, a slower progression of AD. They concluded that NSAIDs may serve a protective role in AD.

These three studies, involving very different approaches in three different countries, all reached the same general conclusion that NSAIDs have a protective function against the development of AD.

Antileprosy agents.

Physicians in Japan have long noticed that elderly leprosy patients, who are carefully supervised by leprosy hospitals, seem to be relatively spared from dementia. McGeer et al. ^[20] surveyed all living leprosy patients in Japan age 65 or over for the use of antileprosy drugs and the prevalence of dementia. They found the overall prevalence of dementia to be 2.9% in 1,410 leprosy patients 65 years or over who had been continuously treated with dapsone or closely related agents. This compared with 6.25% of 1,761 leprosy patients who had been untreated for at least 5 years. The age-adjusted odds ratio for the protective effect of dapsone was 0.63, with a 95% confidence interval of 0.43 to 0.92. Dapsone has significant anti-inflammatory activity and has been used to treat a number of allergic and autoimmune disorders. ^[20]

(Table 5) summarizes the statistical meta-analytical results for the various studies detailed in Table 1, Table 2, Table 3, Table 4. The type of study, the risk factor analyzed, number of studies in the group, overall odds ratio, 95% confidence interval, and level of statistical significance are given. Also shown are the calculated numbers of null studies required to overturn the significance of the results in each case.