Methotrexate and relative risk of dementia amongst patients with rheumatoid arthritis: a multi-national multi-database case-control study

Dementia is one of the largest unmet medical needs, and with growing numbers of older people in a majority of countries worldwide, it is set to become an increasing burden on health services and economies [1]. Despite considerable progress in the understanding of the pathological lesions associated with the diseases causing the commonest forms of dementia—Alzheimer’s disease (AD), vascular dementia, dementia with Lewy bodies and fronto-temporal dementia—this has not resulted in therapeutic progress, with multiple trials of compounds designed for disease modification failing, often at a great cost [2]. Whilst part of the failure of such trials might be that they have largely been conducted in patients with established disease, possibly too late for therapeutic benefit, there is an increasing realisation that it is likely forms of dementia are part of a complex set of pathological processes [3]. Understanding these causal pathways, and in particular, understanding their timing in relation to the onset of dementia, has become a pressing task in order to progress a novel approach to therapeutic development.

One of the processes that has come under intensive investigation in recent years has been that of inflammation [4‐6]. In the analysis of large-scale genome-wide association studies, inflammatory pathways such as complement signalling were consistently identified as altering susceptibility to disease [7, 8]. Post-mortem studies reflect this association with inflammation, as microglial and astrocyte numbers are increased and associated with pathological lesions in disease [9]. However, such genetic and pathological studies are unable to distinguish between cause and effect, and it might be that an inflammatory reaction is part of the defence mechanism or is contributing to disease progression [10].

An alternative approach to understanding the direction and timing of the effect of inflammation in relation to dementia is to utilise existing observational clinical data where individuals have received anti-inflammatory drugs. It was such an approach that first highlighted the role of inflammation in neurodegeneration when it became apparent that non-steroidal anti-inflammatory compounds were associated with reduced dementia risk [11‐14].

Many subsequent studies have reproduced this finding, although trials of such compounds have not been successful to date, perhaps because the association is only seen in those taking medication for extended periods of time before the onset of dementia [15]. In line with this, we recently found non-steroidal anti-inflammatory drugs (NSAIDs) were associated with higher cognitive function in participants in the UK Biobank cohort study of largely healthy individuals [16]. Rather than being cognitive enhancers, a potential explanation of this finding is that NSAIDs are reducing the decline in cognitive function in pre-clinical dementia states. Together, these real-world observational clinical data studies, therefore, have made a significant contribution in the following: first, adding support to the association with inflammation in disease; second, in adding support suggesting a direction of that effect; and third, in providing evidence for timing in relation to disease and therefore informing the design of clinical studies.

In recent years, more specific anti-inflammatory drugs have become part of the pharmacotherapy of disorders such as rheumatoid arthritis, providing an opportunity to begin to explore the effects of different mechanisms of inflammation in relation to neurodegeneration [17, 18]. Recognising this, we recently used Clinical Practice Research Datalink (CPRD), a large dataset derived from primary care in the UK, to explore the relationship between different anti-inflammatory compounds and showed that disease-modifying antirheumatic drugs (DMARDs), and especially methotrexate, are associated with a reduction in risk of incident AD [19]. However, although a relatively large study, the data was from a single source and in a single health care context and thus vulnerable to unanticipated confounding. Here, we set out to replicate and extend this study across multiple real-world observational datasets across Europe.

In this uniquely large and geographically diverse sample using four European harmonised electronic health record datasets using the EMIF platform, we found evidence that in adults older than 50 years of age with rheumatoid arthritis, taking methotrexate was associated with a lower risk of a subsequent dementia diagnosis, specifically recorded methotrexate use for more than 4 years. We also investigated sulfasalazine use but found no such association.

However, the lack of association between sulfasalazine and dementia could also be attributed to the low frequency of sulfasalazine usage in these databases, and this could provide an account for the lack of statistically significant reduction in dementia risk.

The data used in this study does not permit any conclusive causal information regarding the biological mechanisms of the anti-inflammatory properties of methotrexate relative to sulfasalazine in relation to dementia risk. However, by identifying the potential mechanisms and how they differ between the two drugs will be helpful in pointing to future directions for understanding the underlying mechanisms in basic research of inflammation and dementia, which would need to be understood before any such compounds would be tested in intervention trials. The exact mechanism of action of methotrexate or sulfasalazine is not well understood; however, several overlapping and distinct mechanisms have been proposed. For methotrexate, these mechanisms include folate-dependent processes, adenosine signalling, inhibition of methyl-donor production, generation of reactive oxygen species, alteration of cytokine profiles and downregulation of adhesion molecule expression, eicosanoids and matrix metalloproteinases [34]. For sulfasalazine, it is still not clear whether the parent drug and/or its main metabolites (sulfapyridine and mesalazine) are responsible for the beneficial effects in rheumatoid arthritis [35, 36]. Based on in vitro and in vivo studies, sulfasalazine appears to exert effects on inflammatory cell function, cytokine and antibody production, inhibition of folate-dependent enzymes, inhibition of synovial neovascularization and free radical scavenging activity [37]. Both methotrexate and sulfasalazine appear to be involved in adenosine signalling [38‐40] and affect T cell activity via CD8 and inhibit B cells by interrupting Ig synthesis whereas only methotrexate interferes with IL-2, and only sulfasalazine causes prostaglandin inhibition. There is evidence in the literature that methotrexate is involved in the inhibition of inflammatory cell proliferation, interference with T cell activity and cytokine secretion, and augmented release of adenosine, macrophage and polymorphonuclear leukocyte inhibition where this has not been shown with sulfasalazine [37, 38]. These drugs are involved in overlapping and distinct mechanisms, and this could potentially explain why we did not see an association between sulfasalazine and dementia risk. However, the lack of cumulative measures reflecting inflammatory disease burden precludes definitive conclusions regarding the causal relationship between methotrexate use, reduced inflammation and lowered risk of dementia.

Investigations of anti-inflammatory DMARDs on dementia risk have reported conflicting results. Judge et al. [19] showed a decreased risk of dementia with methotrexate use. Using the UK Clinical Practice Research Datalink, Judge et al. found that DMARD users were at reduced risk of dementia (hazard ratio 0.60; 95% confidence interval 0.42–0.85), and the effect was strongest in methotrexate users (hazard ratio 0.52; 95% confidence interval 0.34–0.82). Our study extends these findings by utilising datasets across multiple countries as well as highlighting the stronger associations of longer methotrexate use with reduced dementia risk.

Judge et al. contradict the study carried out by Chou et al. [41]. In this study, using the Taiwan National Health Insurance Research Database, people taking non-biologic DMARDs had a higher risk of any dementia type (AD, vascular) compared with people who did not take DMARDs or used biologic DMARDs. This study used a cut-off age of people equal or older than 20 years; however, the majority of people used in the analysis were older than 65 (> 80%). Furthermore, this study also showed that the numbers of days of use of either methotrexate or sulfasalazine increased dementia risk. The contradictory results from this study could be for a number of reasons, firstly differences in geographic locations (European versus Asian). Further studies are still required to clarify the relationship between prior methotrexate use and duration as well as dementia risk with the use of other DMARDs, particularly biological therapies. Future experimental studies are warranted to determine the specific mechanisms for the protective association between prior methotrexate treatment and dementia.

Our study has several strengths. First, the use of multiple routine EHR databases under the EMIF platform has allowed for aggregation and harmonisation of huge volumes of suitable data to investigate the association between dementia and inflammatory exposures. The access to large amounts of data has meant that we have been able to select people with a relatively rare condition such as RA and have sufficient numbers of incident cases of dementia. By only including people who had RA in this study, we have increased the likelihood of cases and controls being comparable rather than just comparing medication use or not within the total samples and therefore could consider the effect of medications and dementia risk. In this study, we used a matched population, which minimises bias due to confounding by indication but also limits the generalizability of the findings. In particular, we only utilised those people with RA also limiting our generalizability. In this study, four EHR databases, primary care and hospital-based, from four European countries, were utilised, giving general applicability across multiple European countries. Second, we followed an EMIF harmonisation process where a single protocol was utilised for the study, single format of data extracts, semantic harmonisation process ensuring comparability of clinical code lists and a statistical analysis plan with the same statistical analysis steps applied on the extracted data ensuring consistency in results.

Our study also had limitations. We only investigated the effect of two anti-inflammatory drugs, methotrexate and sulfasalazine, without considering the impact of other drug treatments potentially administered such as biologics and concomitant use of biologics and other non-biologic anti-inflammatory medications. Therefore, confounding by indication could exist as the severity of RA may impact the choice of medication where methotrexate and sulfasalazine are usually the first-line treatments for RA. Therefore, the characteristics of patients receiving methotrexate versus sulfasalazine could be different due to the underlying disease severity (as measured by the Disease Activity Score 28 (DAS28), presence/absence of joint erosions and other inflammatory biomarkers) and degree of inflammation and therefore could contribute to confounding by indication. However, Judge et al. [19] used propensity score matching to account for confounding by indication and showed similar results to ours regarding a reduced risk of dementia with methotrexate use. Furthermore, drug adherence could affect the results with drug side effects which could affect the compliance of taking the medications utilised in this study. Additionally, discontinuation of drug treatment could impact the degree of undefined inflammation leading up the onset of dementia in cases. Based on the study design, it is feasible that some patients had continuous treatment and others did not. Additionally, cases and controls who used both methotrexate and sulfasalazine were included in the analysis, adding another potential confounder that weakens the overall statistical association between methotrexate use and lowered risk of dementia. Owing to the observational nature of the study, there remains the potential for residual confounding that could attenuate or explain the observed associations, due to unmeasured variables such as arthritis disease severity, health, concomitant therapy and lifestyle effects such as physical activity and alcohol consumption. However, in spite of this, the data used for this study does reflect the wider rheumatoid arthritis patient population in terms of gender distribution [42].

Regarding the clinical diagnosis of dementia, there is the possibility of non-uniform case assignment. However, it is reasonable that all countries (the UK, Spain, Denmark and Holland) have dementia assessment services and the potential for specialist diagnoses (which are more likely to vary in coverage within rather than between sites) and that this will be more likely to have obscured our findings through case heterogeneity and measurement error rather than give rise to false-positive findings. Furthermore, the diagnosis of dementia is usually assessed by a general practitioner and not by a standardised research diagnosis, which may differ depending on the type of dementia, so this may attenuate the associations. We also defined dementia to encompass any type of dementia including Alzheimer’s disease and vascular dementia; therefore, it might be possible that a specific subtype of dementia is driving the association.

Further work, from this study, would be to establish the following: firstly, if other anti-inflammatory medications such as hydroxychloroquine and leflunomide affect dementia risk; secondly, whether biological anti-inflammatories reduce dementia risk; thirdly, if these associations are effected by NSAID and prednisone usage; and finally, whether methotrexate treatment affects the risk in the different dementia subtypes.

In summary, this case-control study across multiple EHR databases found that prior methotrexate use, but not sulfasalazine use, was associated with dementia diagnosis even after adjusting for patients’ demographics and comorbidities. Furthermore, we showed that there is a gradual decrease against dementia risk the longer the use of methotrexate before the index date. These findings are relevant from a public health perspective since they highlight the potential of methotrexate to protect against dementia; however, the mechanism behind this remains to be elucidated.