Trends in incidence of dementia among patients with rheumatoid arthritis: a population-based cohort study

Abstract

Objective:

We aimed to assess the incidence of dementia over time in patients with incident rheumatoid arthritis (RA) as compared to non-RA referents.

Methods:

This population-based, retrospective cohort study included Olmsted County, Minnesota residents with incident RA by ACR 1987 criteria, diagnosed between 1980 and 2009. We matched non-RA referents 1:1 on age, sex, and calendar year and followed all individuals until 12/31/2019. Incident dementia was defined as two codes for Alzheimer’s disease and related dementias (ADRD) at least 30 days apart. Cumulative incidence of ADRD was assessed, adjusting for the competing risk of death. Cox proportional hazards models calculated hazard ratios (HR) with 95% confidence intervals (CI) for incident ADRD by decade.

Results:

After excluding individuals with prior dementia, we included 897 persons with incident RA (mean age 56 years; 69% female) and 885 referents. The 10-year cumulative incidence of ADRD in individuals diagnosed with RA during the 1980s was 12.7% (95%CI:7.9–15.7%), 1990s was 7.2% (95%CI:3.7–9.4%), and 2000s was 6.2% (95%CI:3.6–7.8%). Individuals with RA diagnosed in 2000s had insignificantly lower cumulative incidence of ADRD than those in the 1980s (HR 0.66; 95%CI:0.38–1.16). The overall HR of ADRD in individuals with RA was 1.37 (vs. referents; 95%CI:1.04–1.81). When subdivided by decade, however, the risk of ADRD in individuals diagnosed with RA was higher than referents in the 1990s (HR 1.72, 95%CI:1.09–2.70) but not 2000s (HR 0.86, 95%CI:0.51–1.45).

Conclusions:

The risk of dementia in individuals with RA appears to be declining over time, including when compared to general population referents.

Introduction

Dementia is a leading cause of morbidity, mortality, and disability, especially among the elderly (1). In fact, it is the sixth overall leading cause of death in the United States, with costs to health care services of $236 billion per year (2). Although the prevalence of Alzheimer’s disease and related dementias (ADRD) was 1.6% of the United States population in 2014, it is estimated to double to 3.3% by 2060 (3). Currently, there are no disease-modifying therapies. Thus, prevention of ADRD and/or the development of novel effective therapies are essential to reduce this burden to both individuals and society.

Understanding the association between rheumatoid arthritis (RA) and dementia may help to clarify the pathogenesis of cognitive decline and provide insights for dementia prevention and treatment approaches. As early as the 1990s, several studies showed that anti-inflammatory medications including both non-steroidal anti-inflammatory drugs (NSAIDs) and glucocorticoids reduced the risk of ADRD (4). In the last decade, many studies have shown an association between RA and dementia (5–10). Together, these studies raised the possibility that inflammation may represent a key mediator of dementia. Indeed, numerous studies have begun to demonstrate evidence of an inflammatory response in the brains of individuals with AD/ADRD (11). However, more recent studies have shown no association between RA and dementia (12–14), or even a protective association (15).

One reason for these seemingly discrepant findings could be a difference in cardiovascular risk factors, as shown by a recent study (10). Two other explanations are also possible. First, all existing studies of dementia in RA used diagnosis codes for RA rather than confirmed cases. Although code-based algorithms for RA can have very high specificity (>97%), misclassification is still possible, as positive predictive values range up to 80–85% even with use of medications (16). Using code-based RA could also change the observed association over time as billing code practices have changed. Another possible reason for the recent discordant findings could be a change in the relationship between RA and dementia over time, for example, from a change in antirheumatic treatments.

To address these gaps, we leveraged a population-based cohort study of manually verified incident RA spanning 30 years. We aimed to assess the incidence of ADRD over time in patients with incident RA and to compare those trends to non-RA referents. We hypothesized that RA was associated with an increased risk of ADRD, but that the strength of the association has declined over the last four decades.

Materials and Methods

Study Design and Population

This retrospective, population-based cohort study utilized the Rochester Epidemiology Project (REP) medical-records linkage system in Olmsted County, Minnesota (17). Started in 1966, the REP captures medical data for over 500,000 unique residents, with capture rates over 97% (17, 18). Within this longitudinal cohort, we matched Olmsted County residents with incident RA to non-RA referents 1:1 based on age (±2 years), sex, and calendar year. We defined index date as the date of fulfillment of RA classification criteria, or a corresponding date for referents, and followed all individuals until death, migration, or 12/31/2019.

This analysis received ethics committee approval (20-008066), followed the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines for observational studies (19), and complied with the Declaration of Helsinki.

Measures

All individuals with RA met 1987 ACR criteria for RA between 1/1/1980 and 12/31/2009 (20). We defined incident dementia as presence of two ICD-9/10 codes for ADRD including “dementia” (290.x, 294.1, F03.x) AD (F00.x, G30.x), “vascular dementia” (F01.x), “dementia in other diseases” (F02.x), or “senile degeneration brain” (331.2, G31.1), all at least 30 days apart. We defined the onset of ADRD as the date of the first dementia code. Anyone with dementia prior to RA incidence/index date was excluded from this analysis. Covariates including age, sex, race, education level, body mass index (BMI), smoking status, hypertension, diabetes mellitus, dyslipidemia, myocardial infarction, stroke (ischemic or hemorrhagic), erythrocyte sedimentation rate (ESR), and disease-modifying anti-rheumatic drug (DMARD) use, including both conventional synthetic DMARDS (csDMARDs) and biologic DMARDs (bDMARDs), were manually abstracted from medical records by trained nurse abstractors. For this study, we defined “other csDMARDs” as leflunomide, sulfasalazine, azathioprine, gold, d-penicillamine, cyclosporine, and cyclophosphamide. We defined “seropositivity” as positivity for either rheumatoid factor (RF) and/or anti-cyclic citrullinated peptide antibody (anti-CCP) at any time.

Statistical Analysis

Descriptive statistics (means, percentages, etc.) were used to summarize the data. Chi-square and Wilcoxon rank-sum tests were used to compare characteristics between the cohorts. Cox proportional hazards models were used to calculate hazard ratios (HR) with 95% confidence intervals (CI) for incident ADRD by decade, adjusting for age (continuous) and sex, as these were our pre-specified confounders in our study protocol. We assessed the cumulative incidence of ADRD adjusting for the competing risk of death.

To ensure the validity of our results, we performed three sensitivity analyses. First, we performed analyses using age as the time scale for the Cox models instead of time since index date. Second, we truncated follow-up of each decade to make the length of follow-up comparable (e.g., the 1980–89 cohort was truncated at 12/31/1999). Third, we included additional adjustors in the model: educational status, and time-dependent covariates for myocardial infarction, stroke, hypertension, diabetes mellitus and dyslipidemia prior to RA incidence/index or at any time during follow-up. There were no missing data. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC) and R 3.6.2 (R Foundation for Statistical Computing).

Results

Demographics

Within the study period, we identified 904 individuals with incident RA and 904 referents. Of these, 897 RA and 885 referents had no dementia prior to index date of RA onset and were thus included in the study. Mean age of individuals with RA was 56 (standard deviation 16) years, and 69% were female (Table 1). Follow-up past age 65 occurred in 70% of individuals with RA and 71% of non-RA referents. Of the 894 with available serological testing, 603 (67%) were seropositive. The 885 non-RA referents had similar characteristics aside from lower education level, lower prevalence of current or former smoking, higher prevalence of hypertension and dyslipidemia, and lower prevalence of stroke (Table 1). Median length of follow-up for individuals with RA was 15 years (interquartile range 10–21) and for non-RA referents was 16 (interquartile range 11–23). Loss to follow-up occurred for only 50 (6%) of RA and 65 (7%) of non-RA patients, with mean age 53 years.

Table 1.

Characteristics of 897 individuals with incident RA 1980–2009 and referents at index date

Characteristic	RA (N=897)	Non-RA (N=885)	p-value
Age, years, mean (SD)	56 (16)	55 (15)	0.68*
Female sex, N (%)	616 (69)	606 (68)	0.93*
Decade, N (%)			0.99*
1980s	198 (22)	196 (22)
1990s	298 (33)	291 (33)
2000s	401 (45)	398 (45)
White race, N (%)	832 (93)	831 (94)	0.33
Education level			<0.001
Less than high school	90 (10%)	83 (9 %)
High school	293 (33%)	256 (29%)
Technical school/college	431 (48%)	396 (45%)
Graduate school	52 (6%)	121 (14%)
No documentation	31 (3%)	29 (3%)
		27
Body mass index, kg/m2, median (IQR)	27 (24,32)	(24,31)	0.86
Smoking status, N (%)			0.001
Never	409 (46)	481 (54)
Former	301 (34)	246 (28)
Current	187 (21)	158 (18)
Hypertension	365 (41%)	315 (36%)	0.027
Diabetes mellitus	96 (11%)	77 (9%)	0.15
Dyslipidemia	507 (57%)	446 (50%)	0.01
Myocardial infarction**	22 (2%)	26 (3%)	0.53
Stroke (ischemic or hemorrhagic)**	10 (1%)	21 (2%)	0.042

IQR = interquartile range, RA = rheumatoid arthritis, SD = standard deviation

^*matched factor

^**Prior to or on RA incidence/index date

RA Treatment Characteristics

Among the individuals with RA, ESR at index and maximum ESR in the first year declined with each decade of RA onset (Table 2). This decline coincided with a strong increase in methotrexate and bDMARD use, slight increase in hydroxychloroquine use, and decline in other csDMARD use in the first year (Table 2).

Table 2.

Treatment characteristics of the 897 incident RA by decade of index date

Characteristic	1980s	1990s	2000s
ESR at index, mm/hr, median (IQR)	25 (31,40)	20 (8,34)	18 (9,32)
Max ESR first year, mm/hr, median (IQR)	36 (20,50)	27 (12,42)	23 (12,40)
Methotrexate use in first year, N (%)	4 (2)	83 (28)	233 (58)
Hydroxychloroquine use in first year, N (%)	50 (25)	133 (45)	225 (56)
Other csDMARD use in first year, N (%)	56 (28)	44 (15)	37 (9)
Biologic DMARD use, N (%)	0 (0)	1 (0)	39 (9)

cs = conventional synthetic, DMARD = disease-modifying anti-rheumatic drug, ESR = erythrocyte sedimentation rate, IQR = interquartile range, RA = rheumatoid arthritis

Dementia Incidence over Time

The 10-year cumulative incidence of ADRD for incident RA in the 1980s was 12.7% (95% CI 7.9–15.7%), but with a decline to 7.2% (95% CI 3.7–9.4%) in the 1990s and 6.2% (95% CI 3.6–7.8%) in the 2000s. Compared to incident RA in the 1980s, the adjusted hazard ratio for ADRD among incident RA in the 1990s was 0.86 (95% CI 0.54–1.36) and in the 2000s was 0.66 (95% CI 0.38–1.16) (Table 3). Of note, the risk of ADRD in seropositive cases was similar to that of seronegative cases (HR 1.26 for seropositive compared to seronegative RA, 95% CI 0.84–1.89).

Table 3.

Incident dementia by decade for the 897 incident RA and 885 non-RA referents

	Total (Dementia Events)		Hazard Ratio (95% CI) *
Decade	RA	Non-RA	RA	Non-RA
1980–1989	198 (34)	196 (36)	Ref	Ref
1990–1999	298 (44)	291 (34)	0.86 (0.54,1.36)	0.75 (0.45,1.23)
2000–2009	401 (26)	398 (30)	0.66 (0.38,1.16)	0.97 (0.56,1.68)

CI = confidence interval, RA = rheumatoid arthritis, ref = reference

^*From cox models adjusting for age and sex

Comparison to General Population

The 10-year cumulative incidence of ADRD among the non-RA referents in the 1980s was 9.3% (95% CI 4.6–11.9%), in the 1990s was 5.0% (95% CI 2.2–6.3%) and in the 2000s was 7.1% (95% CI 4.3–8.9%). Overall, the risk of ADRD among individuals with RA was significantly higher than the non-RA referents (HR 1.37; 95% CI 1.04–1.81). When subdivided by decade, the risk of ADRD in patients diagnosed with RA was higher than referents in the 1980s and 1990s but not 2000s (Figure 1).

Figure 1.

Cumulative incidence of dementia (based on ICD 9/10 codes) versus age-and sex-matched non-RA referents by decade of RA incidence/index.

Sensitivity Analyses

The first sensitivity analysis utilized an age scale instead of time since index date. The main difference in results using this approach was that the point estimate for risk of ADRD among individuals with RA in the 1990s was the same as the 1980s, though it did still decline in the 2000s (Table 4). Just like the main analyses, however, the risk of ADRD among incident RA was higher than non-RA referents in the 1980s and 1990s but not 2000s (Table 4). The overall HR for ADRD among individuals with RA, compared to those without, was nearly identical to the primary analysis, at 1.39 (95% CI 1.06–1.84).

Table 4.

Sensitivity analysis of incident dementia by decade using age as the time scale

	Total (Dementia Events)		Hazard Ratio (95% Confidence Interval) *
Decade	RA	Non-RA	RA (N=897)	Non-RA (N=885)	RA vs. Non-RA
1980–1989	198 (34)	196 (36)	Ref	Ref	1.59 (0.98,2.57)
1990–1999	298 (44)	291 (34)	1.00 (0.64,1.57)	0.89 (0.56,1.43)	1.69 (1.08,2.65)
2000–2009	401 (26)	398 (30)	0.75 (0.45,1.26)	1.31 (0.80,2.14)	0.90 (0.53,1.52)

CI = confidence interval, RA = rheumatoid arthritis, ref = reference

^*Adjusting for sex

A second sensitivity analysis truncated the follow-up of each decade’s participants for comparability. Using this approach, the point estimates for risk of ADRD among RA patients decreased from the 1980s to the 1990s to the 2000s in an almost identical fashion to the primary analysis (data not shown). Similarly, the risk of ADRD for RA patients versus referents was increased in the 1990s (HR 1.97, 95% CI 1.02–3.80) but not the 2000s (HR 0.86, 95% CI 0.51–1.45), for an overall HR of 1.19 (95% CI 0.84–1.69).

The third sensitivity analysis additionally adjusting for hypertension, diabetes mellitus, dyslipidemia, myocardial infarction, stroke, and education level showed nearly identical results to the main analysis (Table 5). In particular, compared to incident RA in the 1980s, the adjusted hazard ratio for ADRD among incident RA in the 1990s was 0.89 (95% CI 0.56–1.44) and in the 2000s was 0.68 (95% CI 0.38–1.21) (Table 5). Just like the main analysis, when compared to the general population, the risk of ADRD in patients diagnosed with RA was higher than referents in the 1980s (HR 1.57, 95% CI 0.94–2.62) and 1990s (HR 1.79, 95% CI 1.13–2.85) but not 2000s (HR 0.84, 95% CI 0.49–1.46).

Table 5.

Sensitivity analysis of incident dementia by decade adjusting for prior cardiovascular disease events (myocardial infarction or stroke), cardiovascular risk factors (hypertension, diabetes mellitus, dyslipidemia) and education level

	Hazard Ratio (95% Confidence Interval) *
Decade	RA (N=897)	Non-RA (N=885)	RA vs. Non-RA
1980–1989	Ref	Ref	1.57 (0.94,2.62)
1990–1999	0.89 (0.56,1.44)	0.79 (0.47,1.31)	1.79 (1.13,2.85)
2000–2009	0.68 (0.38,1.21)	1.10 (0.62,1.96)	0.84 (0.49,1.46)

CI = confidence interval, RA = rheumatoid arthritis, ref = reference

^*Adjusting for age, sex, education level, and time-dependent adjustment for myocardial infarction, stroke, hypertension, diabetes mellitus and dyslipidemia prior to RA incidence/index or at any time during follow-up

Discussion

This population-based cohort study of individuals with validated RA across 40 years demonstrated that the incidence of dementia among incident RA patients has declined over time. Further, the incidence of dementia was higher among incident RA patients compared to age-, sex- and calendar-year-matched referents. However, this gap has narrowed over time. These time trends may explain some of the discrepancy in existing literature and provide insights regarding the role of inflammation in dementia pathogenesis.

The first key and novel finding from this study was that the incidence of dementia in RA patients has decreased over time. Indeed, we observed declines in both the cumulative incidence of ADRD over time and the hazard ratios for ADRD over subsequent decades. The latter was not statistically significant, likely due to small sample size. Interestingly, this decrease in dementia incidence over time coincides with both a decline in dementia over time in the United States and Europe (21), as well as better RA disease control over the same time frame, both in our study as well as others (22, 23). Use of bDMARDs may also contribute to this decline, perhaps by lowering cardiovascular risk factors (10). Between 1999 and 2006, utilization of bDMARDS increased from 3% to 26% in a sample from the United States (24), with a similar increase in our population. Indeed, the risk of dementia on a biologic tumor necrosis factor inhibitor compared to csDMARDs was lower in two recent studies (25, 26), though another study showed no association between bDMARD use and dementia (27). Regardless of the etiology for this decline, future studies should adjust for calendar year, as it appears to be a key factor in the relationship between RA and dementia.

A second main finding from this study was that RA increases the risk of dementia. Although a handful of previous studies have also observed this association (5–9), these studies used diagnostic codes to identify patients with RA rather than validated RA cases. Furthermore, only one previous study used a population-based design as this study did (5). This study also verified the results using an age scale, which is important in dementia research since age is the most important risk factor for dementia (28). Thus, the high-quality design of this study represents a significant contribution to existing literature on this topic.

A third novel finding from this study was that the difference in risk of dementia between RA patients and non-RA referents also declined over time. As with the incidence of dementia in incident RA patients alone, previous studies have not compared trends in dementia incidence to non-RA referents over decades. The increased risk for ADRD among individuals with RA in the 1980s and 1990s, but not 2000s, may explain the discrepancy in existing literature on this topic. Older studies show a positive association between RA and dementia (5–9), whereas newer studies show no association (12–14), or even a reverse association (15). Together, these findings of decreased dementia risk over time, and in comparison to individuals without RA, suggest that chronic, systemic inflammation may be important in dementia pathogenesis.

The question of an association between inflammation and dementia arose in the 1980s when immune-related proteins were identified histologically in association with beta-amyloid plaques (29). This notion has further biologic plausibility, as other autoimmune diseases including systemic lupus erythematosus (9, 30), psoriatic arthritis (31), and Sjögren syndrome (9, 31, 32) are also associated with increased risk of dementia. Some studies have shown that anti-inflammatory drugs including DMARDs are associated with decreased risk of dementia (4, 6, 15, 33, 34), while others have not (35, 36). Relatively recently, however, a mutation in a part of the immune system termed Triggering Receptor Expressed on Myeloid Cells2 (TREM2) was identified, which results in a substantial increased risk of Alzheimer’s disease (37, 38). This finding supports the notion that inflammation may play a role in dementia pathogenesis after all. An exciting corollary of this conclusion is that anti-inflammatory drugs could be considered as drug repurposing candidates in efforts such as the ongoing Drug Repurposing for Effective Alzheimer’s Medicines (DREAM) study (39). Further work in this area is essential for this devasting disease.

This study has several important limitations. First, results from this single population-based cohort may not generalize to more diverse populations or other geographic regions (28). Furthermore, results from this cohort where the mean age at index date of 56 years may not generalize to older RA populations, though the majority of this cohort had follow-up past age 65 years. Second, the modest sample size was not large enough to evaluate whether the trend in dementia incidence was statistically significant. Third, health care utilization bias might have artificially increased the hazard ratio point estimates, since individuals with RA might be more likely to have routine medical attention and have their dementia detected compared to those without RA. On the other hand, a prior study showed no association between ambulatory clinical encounters and dementia risk (10). Fourth, because of the timeframe studied, serostatus calculations largely reflect RF positivity and not anti-CCP positivity. Studying the association with CCP in particular may be important, as citrullination is associated with the pathogenesis of both RA and Alzheimer’s dementia, and both show an epidemiological association with periodontitis (40). Fifth, residual confounding for factors such as socioeconomic status, NSAIDs, DMARDs, and unknown confounders is possible. Finally, this study used a code-based definition of dementia, which may have resulted in misclassification. Future studies should work to develop standardized algorithms for identification of dementia cases from medical records (for example, using natural language processing), which can be applied to similar studies.

Conclusion

Our findings show an increased risk of dementia among patients with RA in the 1980s and 1990s, but a declining trend in risk of dementia in patients with RA over time, including when compared to the general population. Further studies should investigate this association using validated algorithms for dementia case identification, and should also elucidate the role of inflammation, autoimmunity, and anti-rheumatic treatments in risk of dementia.