Author: Bryan P. White
Original Publication: 09/28/2019
Recently, an association between non-steroidal anti-inflammatory drugs (NSAIDs) and decreased risk of Alzheimer’s disease (AD) or other dementias has been identified. Demonstrating that this association between NSAID use and AD is statistically significant would be an important first step in understanding the potential biological underpinnings of how inflammation might affect AD progression. In order to better understand that association, I selected two papers that had recently been included in a meta-analysis which determined there is most likely a statistically significant reduction of risk in the development of AD and compared the research methodologies of those papers. The research objective of both these papers was to determine the hazard ratios for the association of NSAID use and reduction of dementia and AD.
In the first study (Szekely et al., 2009), a prospective cohort was formed from an ongoing study, the Cardiovascular Vascular Health (CSV) study designed to understand factors leading to heart disease and stroke in adults 65 years or older. Beginning in 1989, participants were recruited from in four communities across the United States (Sacramento County, California; Washington County, Maryland; Forsyth County, North Carolina; and Allegheny County, Pennsylvania). In 1992, an additional cohort of African Americans was recruited from 3 of these communities. From the total population of participants, those that completed a Modified Mini Mental State Examination (3MSE) and brain MRI between 1992 and 1994 were enrolled in an additional CHS Cognition Study. Participants were followed for up to 10 years after initial evaluation.
Participants in the Cognition Study underwent additional behavior and mental screening during their visits, as well as recording prescription and over the counter (OTC) drug use. A total of 3,229 participants who did not have dementia at their initial screening were included in the analyses of NSAID effect on AD/dementia. Recorded exposures included OTC NSAIDs, aspirin, acetaminophen, and prescription NSAIDs. This study further classified NSAIDs into Aβ42-lowering and non-Aβ42-lowering, which is based on their ability to selectively lower Aβ42 protein deposits in previous animal studies. The measured outcomes of the study were incident all-cause dementia, Alzheimer's dementia (AD), and vascular dementia (VaD).
The statistical method used was Cox proportional hazards regression model to create crude and adjusted hazard ratios (HR) with 95% confidence intervals. Cox proportional regression is used because it can incorporate and rank a set of covariates along with a primary predictor variable. The covariates used in the analysis included factors that differed between NSAID and non-NSAID users, heart disease status, total number of prescription medications, alcohol consumption, and other general health ratings. Additionally, subgroup analysis separated out NSAID users by age at baseline, race (African American and White), presence of ε4 alleles at the APOE locus, and Aβ42-lowering vs. non-Aβ42-lowering NSAIDs.
The authors found that HRs were significantly reduced for participants less than 75 years of age, possessing any ε4 allele at the APOE locus, and having White ethnicity. This suggests that there is a strong genetic component to this association, as well as a possible socioeconomic factor with race/ethnicity, and an age factor. This study has a very strong research design and addressed some of key issues surrounding potential confounding factors. Since Alzheimer’s disease tends to have a genetic component, it would also seem plausible that any treatment (e.g. NSAID use) might also have a genetic component. Understanding this relationship would be critical for implementation of the treatment because genetic testing can be done on those with a potential AD risk to determine their APOE allele status. Those with the ε4 allele paired with increased AD risk might immediately be offered to begin an NSAID regimen, whereas those testing negative might not have as strong a case for beginning long-term NSAID use. Unfortunately because the race/ethnicities included in the study were limited, most likely be geographic area, it would be difficult to make generalizations to the broader population. However, within the populations studied here (White and African American), this study might be effective. In other words, because there is a strong genetic component to this association, and in many cases there are real genetic factors related to health that differ between race, this study probably cannot be used to make inferences about those with Asian, Latino, or potentially some other ancestries with shared genetic heritage related to health.
In the second study (Chang et al., 2016), a retrospective cohort study was conducted on a population of Taiwanese individuals having medical records in the National Health Insurance Research Database (NHIRD). This database was created in 1995 and contains records for 99% of the Taiwanese population. In this study, de-identified patient data was obtained from the Registry for Catastrophic Illness Patient Database (RCIPD), which contains a subset of records in the NHIRD, and was used to identify patients with Rheumatoid Arthritis (RA) diagnosed between 2000 and 2011. A non-RA cohort was created by frequency matching patients without RA from the RCIPD by age, sex, and year of RA diagnosis. A total of 33,229 patients with RA and 132,916 without RA were included in the study.
The primary outcome variable for the study was the development of any dementia, which included Alzheimer’s Disease (AD). Patients were monitored in the study until either the occurrence of dementia, death, or December 31, 2011. Crude and adjusted hazard ratios (HR) with 95% confidence intervals were constructed using Cox proportional regression to adjust for covariates. The primary cofactor analyzed for NSAID use in this study was duration of NSAID use. These authors found that while NSAID use reduced HRs in the ranges of 731-1460 days and 1461-2190 days, only durations greater than 2191 days had a significant effect. This suggests that duration of NSAID exposure is a key factor in the association between NSAID use and AD/dementia risk reduction. Unrelated to NSAID exposure, this study also assessed HRs due to comorbidities (alongside RA) and reported that the presence of any other type of comorbidity also increased risk of AD/dementia. This suggests that while having one chronic condition that triggers the use of NSAIDs (e.g., RA) might decrease AD/dementia risk, having other chronic conditions in addition to that eliminates most of the benefits.
I think this study has a strong design and took advantage of the fact that it is a retrospective study by including a very large sample size (> 30,000 in the exposure group, > 130,000 in the control group). However, as we learned in the previous study, genetics plays a strong factor in both AD/dementia risk and the potential for NSAIDs to have an effect. Unfortunately, this study does not have APOE status recorded, so the genetic factor cannot be ranked in the multivariate analysis. However, on the plus side, this study had additional information about the duration of the exposure of NSAID use, which ended up being a significant factor.
If we combine the information between both studies, we can learn that genetics, race/ethnicity, age, and duration of NSAID use are all strong factors linking together the association of NSAIDs on AD/dementia risk. Overall, although both studies had different primary results, these results are not in conflict with each other, and instead paint a bigger picture about the hypothesized association. I agree with the results of both papers and think they were both conducted with sound statistical methods and correctly interpret the results of their data. Between the two studies, I think the Szekely, et al. (2009) study was the stronger study because it contributed a more unique assessment related to the biological component of the underlying association. Given this information, specific laboratory experiments can be designed using cellular models of human tissue to better understand the specific physiologic mechanisms surrounding neuroinflammation and other interactions between neurons and microglia, such as done in Lee et al. (2015).
To better understand this association, I would design a prospective cohort study based on individuals that had a strong genetic predisposition to developing AD/dementia. I would separate the patients into groups based on APOE status and NSAID type and recruit broadly across the United States. Controls would be identified as having AD/dementia risk but monitored for NSAID usage. Those individuals selected as the treatment group would begin taking NSAIDs regardless of comorbidity for the duration of the study. If controls developed a comorbidity that required the use of NSAIDs, they would have to be removed as controls and transferred into a 2nd group (secondary treatment). This study would have to be conducted for at least 5 years (~2100 days) since that is what the expected dosage duration to produce a significant effect is according to Chang et al. (2016). Initially a greater number of controls than treatment group individuals should be recruited because as individuals age it would be expected that more controls will have to begin taking NSAIDs for other comorbidities.
Chang, K. H., Hsu, Y. C., Hsu, C. C., Lin, C. L., Hsu, C. Y., Lee, C. Y., … & Kao, C. H. (2016). Prolong exposure of NSAID in patients with RA will decrease the risk of dementia: a nationwide population-based cohort study. Medicine, 95(10).
Lee, M., McGeer, E., & McGeer, P. L. (2015). Activated human microglia stimulate neuroblastoma cells to upregulate production of beta amyloid protein and tau: implications for Alzheimer's disease pathogenesis. Neurobiology of aging, 36(1), 42-52.
Szekely, C. A., Breitner, J. C., Fitzpatrick, A. L., Rea, T. D., Psaty, B. M., Kuller, L. H., & Zandi, P. P. (2008). NSAID use and dementia risk in the Cardiovascular Health Study*: Role of APOE and NSAID type. Neurology, 70(1), 17-24.