Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jun 1.
Published in final edited form as: Alzheimers Dement. 2021 Sep 27;18(6):1248–1259. doi: 10.1002/alz.12465

Potential long-term effect of tumor necrosis factor inhibitors on dementia risk: A propensity score matched retrospective cohort study in US veterans

Chunlei Zheng 1,2,†, Nathanael R Fillmore 2,4,5,†, Jaime Ramos-Cejudo 2,7, Mary Brophy 2,6, Ricardo Osorio 8,9, Mark E Gurney 10, Wei Qiao Qiu 11,12,13, Rhoda Au 14,15, George Perry 16, Maureen Dubreuil 2,6, Shu G Chen 17, Xin Qi 18, Pamela B Davis 19, Nhan Do 2,6, Rong Xu 1,3
PMCID: PMC8957621  NIHMSID: NIHMS1733040  PMID: 34569707

Abstract

INTRODUCTION:

TNF inhibitors are widely used to treat rheumatoid arthritis (RA) and their potential to retard AD progression has been reported. However, their long-term effects on the dementia/AD risk remain unknown.

METHODS:

A propensity scored matched retrospective cohort study was conducted among 40,207 patients with RA within the US Veterans Affairs healthcare system from 2000 to 2020.

RESULTS:

A total of 2510 patients with RA prescribed TNF inhibitors were 1:2 matched to control patients. TNF inhibitor use was associated with reduced dementia risk (hazard ratio: 0.64, 95% confidence interval: 0.52–0.80), which was consistent as the study period increased from 5 to 20 years after RA diagnosis. TNF inhibitor use also showed a long-term effect in reducing the risk of Alzheimer’s disease (hazard ratio: 0.57, 95% confidence interval: 0.39–0.83) during the 20 years of follow-up.

CONCLUSION:

TNF inhibitor use is associated with lower long-term risk of dementia/AD among US veterans with RA.

Keywords: TNF inhibitor, Alzheimer’s disease, dementia, rheumatoid arthritis, US veterans, long-term effect, cohort study

1. Background

Dementia is a global health problem with an aging population in the world. No disease-modifying therapies are available.1 Alzheimer’s disease (AD) is the most common form of dementia, accounting for 50–80% of cases; other forms of dementia include vascular, Lewy-body, and frontotemporal dementia.2 A growing body of evidence indicates that cerebrovascular inflammation plays an important role in dementia, particularly in the pathological processes associated with amyloid-beta and tau accumulation in the brain of AD patients3. A sustained inflammatory response, mediated by over-activation of microglia and other immune cells, has been shown to exacerbate both amyloid and tau pathology.4,5 TNF is a key proinflammatory cytokine in the initiation and control of inflammatory processes.6 Multiple lines of evidence indicate that TNF may trigger or amplify aberrant microglia signaling in the brain and thereby contribute to AD pathogenesis.7–10 TNF produced systemically may enter the brain through receptor-mediated transcytosis.11 Elevated soluble levels of TNF have been previously reported in the cerebrospinal fluid of AD patients in association with disease progression.12

Rheumatoid arthritis (RA) is a common autoimmune systemic inflammatory disease affecting approximately 1% of the population worldwide.13 The overall prevalence of dementia in RA patients is 0.79% according to a recent study.14 RA may represent a good model to study the role of inflammation in the development of dementia since TNF plays an important role in both conditions,15 and because TNF inhibitor drugs are effective treatments for RA. Therefore, studying the effect of anti-TNF treatment on the dementia risk in RA patients may provide an opportunity in repurposing TNF inhibitors for the treatment of dementia and Alzheimer’s disease.

Inflammatory pathways have been targeted for dementia/AD treatment for many years. Numerous epidemiological studies of nonsteroidal anti-inflammatory drugs (NSAIDs) reported that NSAIDs are associated with reduced dementia/AD risk. However, most of these studies had small sample size16 and eventually, most NSAIDs-based randomized trials failed to show benefit for dementia/AD patients.17 Currently, conventional disease-modifying antirheumatic drugs (cDMARDs) are used as the first line drugs for RA treatment. Two studies have reported that an association of cDMARDs with dementia/AD risk but with opposite results.18, 19. A recent clinical trial examined the effect of hydroxychloroquine on progression of dementia in early AD in patients and showed no beneficial effects.20

Several observational studies have been conducted to investigate the association of TNF inhibitor use with dementia in RA patients. A retrospective case-control study using data from the Taiwan National Health Insurance Research Database from 2000 to 2011 showed no significant associations between TNF inhibitor use and dementia incidence (OR: 0.71, 95% CI: 0.46–1.10, 957 matched patients).18 A nested case-control study using health insurance claims data from 2000 to 2007 showed that TNF inhibitors as a class were associated with lower AD risk (adjusted OR 0.45; 95 % CI 0.23–0.90, 165 eligible AD patients). For individual TNF inhibitors, only etanercept showed decreased AD risk (adjusted OR 0.30; 95 % CI 0.08–0.89).14 However, these two studies were limited in sample size. Our recent study using nationwide electronic health record (EHR) database of 56 million adults including 514,440 RA patients also showed significant association of TNF inhibitor use and lower AD risk.21 All previous studies had limited study periods (7 and 11 years respectively) and the actual patient follow-up times were not clear, so the long-term effects of TNF inhibitor use on the risk of dementia and AD remain unknown.

In the present study, we utilized the nationwide Veterans Health Affairs (VHA) electronic medical record (EHR) from 2000 to 2020 to investigate the association of TNF inhibitor use with dementia risk in RA patients. The VA is the largest integrated health care systems in the United States, providing care to more than nine million patients annually. The VA serves a population of patients who are frequently under-represented in clinical trials, especially a large portion of older patients.22 In addition, the VA operates an EHR system for which records are aggregated in a centralized database regardless of VA site of care, enabling a longer follow-up study in a mobile real-world population. The large sample size, concentrated on older patients, and long follow-ups make the EHR data from the VA a unique resource to study the effects of TNF inhibitors on the risk of dementia and AD. In this retrospective cohort study, we evaluated the association of TNF inhibitors with dementia risk among patients with RA with follow-ups up to 20 years.

2. Methods

2.1. Data source

The patient data used in this study were obtained from the VA Corporate Data Warehouse (CDW), which is a centralized database of electronic health records (EHR) for patients seen at VA facilities nationwide.23 The CDW contains comprehensive patient information, including demographics, International Classification of Diseases, Ninth Revision and Tenth Revision (ICD-9 and ICD-10) diagnoses, procedures, prescription fills, and laboratory tests, which has been intensively used in observational studies.24 We used both ICD-9 (before October 2015) and ICD-10 (after October 2015) codes to obtain patient diagnosis. All ICD codes used in this study were included in Table S1 in Supporting information-1. The study period was from January 1, 2000 to August 24, 2020.

2.2. Study design

2.2.1. Study population

We used a cohort study to investigate the association of TNF inhibitors, as a whole or individually, with dementia risk in RA patients in the entire study period as well as different lengths of study periods after RA diagnosis (5, 10, 15, 20 years). Since patients with RA were often prescribed more than one TNF inhibitor in the clinical practice, we also compared the dementia risk between a combination of TNF inhibitors and single TNF inhibitor use in the entire study period.

Patients with RA were identified as patients with two or more RA diagnostic codes at least 6 months apart as described previously.25 The date of RA diagnosis was defined as the date of the first RA diagnostic code. The following exclusion criteria were applied: (1) We excluded patients with dementia diagnosis occurring before the first RA diagnosis. (2) We excluded patients with less than 6 months of VA utilization prior to RA diagnosis, in order to avoid capturing new VA patients with a prior diagnosis of RA outside the VA. (3) We excluded patients with less than one year of VA utilization after RA diagnosis. (4) We excluded patients younger than 18 years old. (5) We excluded patients diagnosed with dementia/AD within 120 days of RA diagnosis, in order to ensure sufficient temperal sequences in clinical practice, as previously described14. (6) We excluded patients diagnosed with dementia/AD within 3 months after TNF inhibitor use, for the same reason.

Individual TNF inhibitor use group was defined as the RA patients who were prescribed a single TNF inhibitor during different lengths of study periods respectively. A TNF inhibitor combination use group was defined as the patients with RA who were prescribed more than one type of TNF inhibitors during the entire study period. Patients with short-term TNF inhibitor use (less than 3 months) were excluded. Patients whose TNF inhibitor treatment took place after a dementia diagnosis were considered as a non-TNF use group.

A propensity-score based matching strategy was used to select control patients, i.e., non-TNF inhibitor use group, both in the entire study periods and different lengths of study periods. The propensity score, a predicted probability of each RA patient to be treated with TNF inhibitor, was computed using logistic regression on TNF inhibitor use (outcome) with covariates including all potential confounders mentioned below (Potential confounders section), and other anti-RA drugs (Drug exposure session). Each TNF inhibitor user was matched to two control patients with the nearest propensity score within a certain amount of standard deviation (caliper width of 0.25 on the logit of the propensity score).26,27, Choosing the optimal matching ratio is a trade-off of variance and bias,28, 29 we provied a sensitivity analysis using different matching ratio (from 1:1 to 1:4) in Figure S15 in supporting information-5. Matching was implemented using the R MatchIt package (Version 4.1.0).30 Covariate balance was checked after matching according to recent guidlines.31

2.2.2. Drug exposure

The drug exposure information was obtained from pharmacy data in the CDW, including the drug name, issue dates of prescription and order records. We considered the commonly used RA treatments according to the treatment guidelines.32 Five TNF inhibitors (adalimumab, certolizumab, etanercept, golimumab, infliximab) were included, and patients with RA were considered both by the individual TNF inhibitor, and as a class. Since conventional disease-modifying antirheumatic drugs (cDMARDs) are used as the first-line drugs for RA, we included four commonly used cDMARDs (methotrexate, sulfasalazine, hydroxychloroquine, leflunomide). In addition, we collected drug use information for prednisone and 18 commonly used nonsteroidal anti-inflammatory drug (NSAIDs) (indomethacin, sulindac, diclofenac, aceclofenac, piroxicam, ibuprofen, naproxen, ketoprofen, diflunisal, ketorolac, tiaprofenic acid, tenoxicam, meloxicam, celecoxib, rofecoxib, etoricoxib, nabumetone, mefenamic acid). Since many of these drugs were used transiently, we considered NSAIDs as a class to facilitate analysis. We did not consider non-TNF inhibitor biologics due to their low frequency of use (Table S2 in supporting information-2).

2.2.3. Potential confounders

Baseline variables included age, sex, race, marital status, year of diagnosis, geographic region, rurality, and known risk factors for dementia, including cardiovascular disease (CAD), peripheral vascular disease (PVD), hypertension, hyperlipidemia, diabetes, obesity, depression, axiety, alcohol use disorder, traumatic brain injury (TBI), and Charlson Comorbidity Index (CCI).33 Obesity was defined using body mass index at the time of RA diagnosis. Other baseline risk factors were defined using data within three years prior to the RA diagnosis as described previously.34

2.2.4. Outcomes

The outcome of interest was the diagnosis of dementia during the study period. Patients with dementia were identified as patients having two or more dementia diagnostic codes with 12 months, a criterion was previously shown with high accuracy in using EHR data35. The date of onset of dementia was defined as the date of patients’ first medical encounter for their diagnosis of dementia in the database. We examined the association of TNF inhibitors with dementia in general and Alzheimer’s disease (AD) in particular. Other subtypes of dementia including vascular dementia, Lewy-body, and frontotemporal dementia were not examined due to limited sample sizes (Table S2 in supporting information-2).

2.3. Statistical methods

Cox proportional hazards regression was used to investigate dementia risk over different lengths of study periods. The index date was the first RA diagnosis date. Time to first dementia/AD diagnosis was our outcome of interest, and patients were censored by the end of the study period, death, or last visit to VA, whichever came the first. TNF inhibitor use was coded as a time-dependent variable in the Cox regression. That is, only time after initiation of TNF inhibitor use was coded as being part of the TNF inhibitor exposure group, while time prior to initiation of TNF inhibitor use was coded as time in the control group. By using time-dependent variables, we avoid improperly ascribing immortal time to the treated group, since patients in the treated group cannot experience the event (dementia/AD) before TNF inhibitor use, while no such requirement exists in the control group. The proportional hazard hypthesis was examined and appeared valid.

For all analyses, a 2-sided p < .05 was regarded as statistically significant. All data analyses were performed through the Veterans Affairs (VA) Informatics and Computing Infrastructure (VINCI), where R (version 4.0.3) was used for statistical analysis.

3. Results

3.1. Characteristics of the study population

We identified 40,207 patients with RA in the VA system from January 1 to August 24, 2020. Among them, 20, 994 patients met our eligibility criteria (see details in methods) and 3,446 patients received TNF inhibitor treatment. The majority of patients were more than 65 years old, especially in the non-TNF inhibitor use group (80.1%). In addition, the majority of patients were male (90.5% and 85.9% in TNF non-treatment and treatment group, respectively). Before propensity score matching, baseline variables including demographics, comorbidities, and RA drug use, showed significant differences between the two groups (Table S2 in supporting information-2). After propensity score matching as described in Methods, these variables became more balanced (SMD < 0.1) Average age was 66.4 (SD:13.9) years old in non-TNF inhibitor use group and 65.6 (SD: 13.4) years old in TNF inhibitor use group. Median follow-up times were 13.7 (±5.2) years and 13.8 (±5.2) years in patients with and without TNF inhibitor treatments, respectively. (Table 1 and Figure S1–S7 in supporting information-3)

Table 1.

RA patient information between TNF inhibitor treatment and non-treatment group after propensity score matching - adult patients

VARIABLE NON-TNF INHIBITOR TNF INHIBITOR P VALUE SMD
N 5016 2510
AGE (SD) 66.43 (13.92) 65.59 (13.35) 0.012 0.062
GENDER= M (%) 4337 (86.5) 2147 (85.5) 0.289 0.027
RACE (%) 0.205 0.043
 White 3745 (74.7) 1854 (73.9)
 Black 790 (15.7) 383 (15.3)
 Other 481 (9.6) 273 (10.9)
REGION (%) 0.989 0.008
 Midwest 1341 (26.7) 665 (26.5)
 Northeast 470 (9.4) 232 (9.2)
 South 2186 (43.6) 1097 (43.7)
 West 1019 (20.3) 516 (20.6)
RURALITY = Urban (%) 3114 (62.1) 1580 (62.9) 0.48 0.018
MARRIGE (%) 0.675 0.043
 Married 3042 (60.6) 1486 (59.2)
 Divorced 1074 (21.4) 543 (21.6)
 Separated 156 (3.1) 83 (3.3)
 Single 452 (9.0) 253 (10.1)
 Widowed 259 (5.2) 126 (5.0)
 Unknown 33 (0.7) 19 (0.8)
COMORIDITIES
 Alcohol use = 1 (%) 147 (2.9) 70 (2.8) 0.784 0.009
 Anxiety = 1 (%) 311 (6.2) 145 (5.8) 0.5 0.018
 Depression = 1 (%) 375 (7.5) 168 (6.7) 0.234 0.031
 Obesity (%) 0.41 0.033
 Normal 1067 (21.3) 567 (22.6)
 Obesity 2214 (44.1) 1082 (43.1)
 Overweight 1735 (34.6) 861 (34.3)
 Diabetes (%) 599 (11.9) 293 (11.7) 0.763 0.008
 Hypertension (%) 1496 (29.8) 744 (29.6) 0.891 0.004
 Peripheral vascular disease (%) 129 (2.6) 69 (2.7) 0.707 0.011
 Cardiovascular disease (%) 574 (11.4) 289 (11.5) 0.958 0.002
 Hyperlipidemia (%) 1276 (25.4) 631 (25.1) 0.8 0.007
 Traumatic Brain Injury (%) 283 (5.6) 150 (6.0) 0.593 0.014
DRUG USE
 Methotrexate (%) 3766 (75.1) 1844 (73.5) 0.137 0.037
 Sulfasalazine (%) 1434 (28.6) 761 (30.3) 0.126 0.038
 Leflunomide (%) 1099 (21.9) 577 (23.0) 0.303 0.026
 Hydroxychloroquine (%) 2218 (44.2) 1116 (44.5) 0.86 0.005
 Prednisone (%) 3885 (77.5) 1907 (76.0) 0.16 0.035
 NSAID (%) 4208 (83.9) 2101 (83.7) 0.862 0.005
Follow-Up days (SD) 5049.70 (1880.52) 5005.89 (1881.94) 0.341 0.023
Charlson Index (SD) 0.57 (0.90) 0.57 (0.95) 0.768 0.007
Open in a new tab

SMD: standardized mean difference.

3.2. TNF inhibitor use is associated with lower risk of dementia in adult patients with RA

The incidence of dementia was 5.0% and 6.0% in US veterans with RA with and without TNF inhibitor treatment, respectively (Table S2 in supporting information-2). After adjusting for age, sex, race, region, rurality, comorbidities and non-TNF drug use, patients on TNF inhibitors had significantly lower risk for dementia compared with individuals not prescribed with TNF inhibitors (HR: 0.64, CI: 0.52–0.80), which is consistent with those reported in other populations with RA.14, 21 Two commonly used TNF inhibitors, adalimumab and etanercept, were associated with significantly lower risk for dementia with a hazard ratio of 0.57 (95% CI: 0.41–0.80) and 0.72 (95% CI: 0.52–0.98), respectively. Certolizumab and golimumab were not individually analyzed due to small sample sizes (Table S2 in supporting information-2). No significant difference in the dementia risk was found between patients on a combination of TNF inhibitors and those on a single agent (HR: 1.21, 95%CI: 0.79–1.83) (Figure 1).

Figure 1.

Figure 1

Open in a new tab

Forest plot shows the association of TNF inhibitor use with overall dementia risk in patients with RA using Cox proportional hazards regression. Note: For individual TNF inhibitor, control group was patients without TNF inhibitor use; For TNF inhibitor combination use group, control group was patients with one of TNF inhibitors.

3.3. Association of TNF inhibitor use with dementia risk at different lengths of study periods

We evaluated the association of TNF inhibitor use with dementia risk with different study periods (5, 10, 15, and 20 years after RA diagnosis). TNF inhibitor use as a class were associated with lower risk for dementia in patients with RA (HRs ranging from 0.67 to 0.72 over 20 years of study period, p < 0.05). For individual TNF inhibitors, adalimumab showed consistent association with lower dementia risk over 20 years of study period (HRs ranging from 0.35 to 0.57, p < 0.01). With longer study periods and more patients developing dementia, significant association of etanercept with reduced dementia risk emerged in 20 years of study period (HR: 0.73, 95%CI: 0.54–1.00). (Figure 2).

Figure 2.

Figure 2

Open in a new tab

Forest plots of dementia risk in TNF inhibitor-treated patients with RA with different lengths of study periods, compared to those without TNF inhibitor treatment. (A) Dementia risk over five years of study period. (B) Dementia risk over 10 years of study period. (C) Dementia risk over 15 years of study period. (D) Dementia risk over 20 years of study period. Note: Multivariate-adjusted Cox proportional hazards regression was used in the analysis.

3.4. Association of TNF inhibitor use with dementia risk in elderly RA patients

Since age at onset of RA determines severity and choice of treatment,36 and is therefore a critical confounder by indication, we next examined the association of TNF inhibitor use with dementia risk in elderly patients with RA. A total of 817 elderly patients (> 65 years old) with RA taking TNF inhibitor were matched to control patients. The average age was 71.4 (SD:5.3) years old in non-TNF inhibitor use group and 71.3 (SD: 5.1) years old in TNF inhibitor use group. Median follow-up times were 14.2 (SD: 5.0) years and 14.1 (SD: 5.1) years in patients with and without TNF inhibitor treatment, respectively (Table 2 and Figure S8–S14 in supporting information-4). The elderly patients treated with TNF inhibitors had lower risk for dementia compared with matched elderly patients not treated with TNF inhibitors (HR: 0.67, 95% CI: 0.48–0.95) (Figure 3A). This significantly reduced dementia risk in elderly patients on TNF inhibitors was consistent in longer study periods (HRs were 0.67 and 0.69 for 15 and 20 years of study periods respectively, p < 0.05) (Figure 3B–D). Elderly patients on individual TNF inhibitors adalimumab and etanercept had decreased risk for dementia, with overall hazard ratios of 0.50 (CI: 0.28–0.90) and 0.60 (0.36–0.99) respectively (Figure 3A).

Table 2.

RA patient information between TNF inhibitor treatment and non-treatment group after propensity score matching - elderly patients (> 65 years old)

VARIABLE NON-TNF INHIBITOR TNF INHIBITOR P VALUE SMD
N 1616 817
AGE (SD) 71.39 (5.27) 71.33 (5.09) 0.799 0.011
GENDER= M (%) 1557 (96.3) 784 (96.0) 0.718 0.02
RACE (%) 0.972 0.01
 White 1267 (78.4) 644 (78.8)
 Black 103 (6.4) 51 (6.2)
 Other 246 (15.2) 122 (14.9)
REGION (%) 0.911 0.031
 Midwest 447 (27.7) 230 (28.2)
 Northeast 194 (12.0) 102 (12.5)
 South 621 (38.4) 316 (38.7)
 West 354 (21.9) 169 (20.7)
RURALITY = Urban (%) 932 (57.7) 475 (58.1) 0.86 0.009
MARRIGE (%) 0.989 0.033
 Married 1160 (71.8) 585 (71.6)
 Divorced 216 (13.4) 109 (13.3)
 Separated 32 (2.0) 17 (2.1)
 Single 60 (3.7) 30 (3.7)
 Widowed 134 (8.3) 71 (8.7)
 Unknown 14 (0.9) 5 (0.6)
COMORIDITIES
 Alcohol use = 1 (%) 44 (2.7) 24 (2.9) 0.862 0.013
 Anxiety = 1 (%) 46 (2.8) 25 (3.1) 0.867 0.013
 Depression = 1 (%) 55 (3.4) 24 (2.9) 0.623 0.027
 Obesity (%) 0.841 0.025
 Normal 345 (21.3) 181 (22.2)
 Obesity 649 (40.2) 319 (39.0)
 Overweight 622 (38.5) 317 (38.8)
 Diabetes (%) 250 (15.5) 128 (15.7) 0.946 0.005
 Hypertension (%) 508 (31.4) 268 (32.8) 0.524 0.029
 Peripheral vascular disease (%) 53 (3.3) 25 (3.1) 0.866 0.013
 Cardiovascular disease (%) 304 (18.8) 163 (20.0) 0.536 0.029
 Hyperlipidemia (%) 452 (28.0) 228 (27.9) 1 0.001
 Traumatic Brain Injury (%) 59 (3.7) 34 (4.2) 0.611 0.026
DRUG USE
 Methotrexate (%) 1164 (72.0) 579 (70.9) 0.581 0.026
 Sulfasalazine (%) 390 (24.1) 213 (26.1) 0.319 0.045
 Leflunomide (%) 362 (22.4) 189 (23.1) 0.722 0.017
 Hydroxychloroquine (%) 653 (40.4) 331 (40.5) 0.995 0.002
 Prednisone (%) 1162 (71.9) 598 (73.2) 0.533 0.029
 NSAID (%) 1168 (72.3) 590 (72.2) 1 0.001
Follow-Up days (SD) 5133.54 (1860.10) 5166.65 (1838.35) 0.583 0.018
Charlson Index (SD) 0.70 (0.95) 0.71 (1.02) 0.868 0.007
Open in a new tab

Figure 3.

Figure 3

Open in a new tab

Forest plots of dementia risk in TNF inhibitor-treated patients with RA older than 65 years old, compared to those without TNF inhibitor treatment. (A) Overall dementia risk during the entire study period. (B) Dementia risk over five years of study period. (C) Dementia risk over ten years of study period. (D) Dementia risk over 15 years of study period. Note: Multivariate-adjusted Cox proportional hazards regression was used in the analysis.

3.5. Association of TNF inhibitor use with AD development in RA patients

We then examined the association of TNF inhibitor use and AD development. Patients with RA treated with TNF inhibitors as a class had significantly lower risk for AD development compared with matched control population (HR: 0.57, 95% CI: 0.39–0.83) (Figure 4A), which was also observed in elderly patients with RA (age > 65 years) (HR: 0.58, 95% CI: 0.33–1.00) (Figure 4B). Individual TNF inhibitor drugs didn’t show associations with AD development in elderly patients with RA.

Figure 4.

Figure 4

Open in a new tab

Forest plots of AD risk in TNF inhibitor-treated patients with RA, compared to those without TNF inhibitor treatment. (A) Overall AD risk in adult RA patients during the entire study period. (B) Overall AD risk in RA patients older than 65 years during the entire study period. Note: Multivariate-adjusted Cox proportional hazards regression was used in the analysis.

4. Discussion

In the present study, we used nationwide database of patient EHR from the US Veterans Affairs healthcare system to investigate the association of TNF inhibitor use with dementia/AD risk in patients diagnosed with RA. We not only showed that TNF inhibitor use is associated with reduced dementia/AD risk in US veterans, but further provided evidence that TNF inhibitor use may have a long-term effect on decreased risk of dementia and AD in patients with RA.

Our study has two strengths. One is the long patient follow-up in the VA system. The median follow-up was 13.7 (±5.2) years and 13.8 (±5.2) years in TNF inhibitor-treated and untreated patients with RA, respectively, which allowed us to study the long-term effect of TNF inhibitor use on dementia risk. We observed that the associations of TNF inhibitor use with reduced dementia risk were consistent over study periods from 5 to 20 years, while the median time of patients with RA on TNF inhibitors was only 4.1 years, suggesting that TNF inhibitor use has long-term effect on dementia. Another strength is that we used propensity score matching for patient cohort selection, which allowed us to better control confounders than traditional covariate adjustment methods.37 We noticed differences in age, race, drug use, and prevalence of comorbidities between TNF use and non-TNF use groups before propensity score matching. RA patients receiving TNF inhibitors tended to be younger, have a lower prevalence of comorbidities, and a higher percentage of cDMARDs and NSAIDs use (Table S2 in supporting information-2). All these observations in our study were consistent with current clinical usage since cDMARDs are considered as the first-line treatment for RA. To address this confounding by indication, we used a propensity score to match each patient receiving TNF inhibitor to non-TNF use patients on a wide range of confounders, including age, sex, race, marriage status, region and rurality, comorbidities, and in particular, drug use, such as commonly used cDMARDs and NSAIDs. After matching, two groups of patients were balanced in all these variables.

The overall hazard ratios of dementia risk both in adult and elderly patients with RA receiving TNF inhibitors were similar (0.64 and 0.67 respectively), which was little higher than a cohort study (HR: 0.48, 95%CI: 0.39–0.58).38 However, that study examined the association of all DMARD use on dementia risk probably due to limited study period (2000–2005). The overall hazard ratio of AD risk was 0.57 (95%CI: 0.39–0.83), which is comparable with a nested case-control study (OR: 0.45, 95% CI: 0.23–0.90) using patient claim data from 2000 to 2007.14 The larger sample size and longer follow up enabled us to examine the effect of use of the individual TNF inhibitors on dementia risk. We showed that both adalimumab and etanercept were associated with lower risk to develop dementia over longer study periods, while they were not significant in shorter study periods (Figure 1 and 2).

Drug combinations are commonly used in RA patient treatment. According to current RA treatment guideline32, combination of traditional DMARDs or TNF inhibitor/MTX combination are recommended for patients with moderate or high disease activity. While this drug combination is usually achieved optimal outcome for RA39,40, an interesting question is that how this drug treatment scheme affects dementia/AD risk. We did not observe that TNF inhibitor combinations further lower dementia risk beyond the TNF inhibitor itself. Since these TNF inhibitors have the similar mechanism of action (targeting TNF-mediated inflammatory pathways), they may not provide additional effect on further reducing dementia risk. For other non-TNF inhibitor anti-RA drugs, including traditional DMARDs and NSAIDs, epidemiological studies and randomized clinical trials don’t show significant associations with dementia/AD risk, although some contradictory evidence exists.16–20 Hence, we don’t expect that order/timing of TNF inhibitors and other non-TNF inhibitor use has a significant impact on dementia/AD risk. However, we hypothesize that initial timing of TNF inhibitor is important for preventing dementia/AD risk given that inflammation is a major contribution to dementia/AD development. This hypothesis predicts that earlier treatment of RA patients would achieve a better outcome in terms of delaying the onset of dementia/AD, and needs to be further investigated.

The potential long-term effects of TNF inhibitors on dementia/AD could be related to the role of neuroinflammation in the development of dementia/AD. A growing body of evidence suggests that inflammation can occur early in patients with dementia and AD, independent of amyloid deposition.41–43 Therefore, early reduction of neuroinflammation may have a profound impact on the pathogenesis and progression of dementia/AD. Our findings support suggestions that TNF inhibitors could be tested as a preventive drug to delay the onset of dementia/AD in patients with RA.45,46,47 Encouraging results have recently been reported for etanercept, showing that the drug is well-tolerated44 and improved cognitive function.45, 46 An ongoing interventional study is examining the effect of a TNF inhibitor on improving memory and cognitive abilities in patients with mild cognitive impairment without any rheumatic disease.47

Confounding by indication is common in observational studies. To control potential confounding effects by RA itself, we used propensity score to match the exposure group and non-exposure group by drug use, age and other co-variates. we tested several different RA drug classes, which may partially reflect different degree of RA severity. Based on a recent study showing that age is a main determinant for RA severity and types36, we matched the patients based on age at RA diagnosis. All these patient matching strategies are expected to mitigate this confounding by indication. However, we can’t exclude the possibility that RA in patients on TNF inhibitors may be still different from RA in patients not on TNF inhibitors, both in disease severity and underlying pathophysiology of the disease, especially given that RA is a highly complex and heterogeneous disease. Recently, a self-controlled cohort design, originally developed for the study of the association of acute outcomes with transient exposures48,49, has been used to study the association of drug treatment with chronic diseases.50,51 The advantage of this method is that all fixed confounders are implicitly controlled, which provided an alternative approach to handle this confounding. However, this method has strong assumptions and not easy to control the time varying co-variates such as age, a major risk factor for dementia/AD52,53. In addition, the non-exposure period is difficult to establish for the study of the long-term effect of the drug due to limited follow-up time in the electronic health record. Despite these limitations, ability to self-control fixed co-variates is particularly attractive and is an interesting future direction when sufficient data are available.

Our study has several limitations. First, information on drug usage duration, dosage and patient adherence is limited in the EHR database. We assumed that patients on TNF inhibitor started at the time of the first prescription and stopped at the last prescription. However, in clinical practice, patients may continue to use TNF inhibitors after the last prescription, or prescriptions may be obtained from a non-VA provider. In addition, we could not assess compliance with the prescribed drugs. Due to these limitations, we could not evaluate how the duration, dosage and compliance of drug use affect the observed inverse association of TNF inhibitors with dementia risk. Second, patients in the VA EHR database are predominately men (90.4%), and 25.5% of population lacked information on race, which prevent generalizability of these results to women and diverse ethnic groups. Third, we were unable to control some known risk factors for dementia/AD, such as education and APOE4 status, due to lack of this information in the database. In addition, data on smoking, a potential risk factor for dementia/AD, was often missing (50.7%). Fourth, dementia and AD diagnosis were solely based on ICD codes, which may lead to under-, mis- or over-diagnoses, especially for prodromal early stages of the disease.54 An additional limitation may come from propensity score matching. Though patient matching can make distributions of covariates more balanced to avoid many confounders, we still could miss some unobserved and unmeasured confounders due to the observational nature of the study design itself, which could lead to increase the data imbalance and bias instead.55 Fifth, while our study used standard statistical analyses for cohort studies, it will benefit from additional sensitivity analysis as well as negative control analysis. Finally, this study is associational, not causal. Validation of our findings in other population is needed and mechanistic understanding of this potential long-term effect may identify new drug targets for dementia/AD.

5. Conclusions

In summary, our study provides further evidence that TNF inhibitor use is associated with reduced dementia/AD risk in patients with RA. More importantly, our findings indicated that TNF inhibitor use may have a long-term protective effect on dementia/AD development. Further investigation of the mechanism and potential therapeutic use of TNF inhibitors in delay of dementia/AD is warranted.

Supplementary Material

tS1
tS2
fS1-7
f8-14
f15-17

Research in Context.

Systematic review:

We reviewed the literature using PubMed up to 2021 and identified several studies examining the association between TNF inhibitor use and incidence of dementia. These studies showed overall benefit of TNF inhibitor use on dementia, but were limited in short follow-up time. Long-term effects of TNF inhibitor use on dementia/AD risk remains unknown.

Interpretation:

We leveraged the long follow-up available for patients in the US Veterans’ Health Affairs. In this population, TNF inhibitor use is associated with reduced dementia/AD risk during up to 20 years of follow-up, suggesting that TNF inhibitors have long-term effect and supporting the rationale for studying their therapeutic potential in reducing dementia/AD risk in patients with RA.

Future directions:

Validation of our findings in other population is needed. Understanding the mechanism of this long-term effect may identify new drug targets for dementia/AD.

Funding/Support

We acknowledge support from NIH National Institute of Aging R01 AG057557, R01 AG061388, R56 AG062272, The Clinical and Translational Science Collaborative (CTSC) of Cleveland 1UL1TR002548-01.

Role of the funding source

The funder of the study had no role in study design, data collection, data analysis, data interpretation, and writing of the report. The corresponding author (R.X) had final responsibility for the decision to submit for publication.

Funding:

Xu (AG057557, AG061388, AG062272, 1UL1TR002548-01), Fillmore (VA Cooperative Studies Program, American Heart Association Strategically Focused Research Network Aging Initiative Pilot Award), Ramos-Cejudo (VA Cooperative Studies Program), Brophy (VA Cooperative Studies Program), Au (NIH/NIA, Alzheimer’s Drug Discovery Foundation, American Heart Association), Osorio (AG067549, AG056031, AG056531), Chen (AG061388, U01NS112010, R01NS118760), Davis (1UL1TR002548-01), Perry (NIH and Kleberg foundation), Dubreuil (NIH K23 and R03 grants). Zheng, Qiu, Qi, Perry and Do don’t have related funding to declare.

Disclosure of outside interests:

Mary Brophy servers a member of the Board of Trustees for Boston VA Research Institute (BVARI). Mark E Gurney was a Chairman, CEO and Director of Tetra Therapeutics, Inc; held stock and options at the time of the sale of the company to Shionogi Co Ltd in 2020; have a U.S patient application titled “SALTS AND POLYMORPHS OF A PDE4 INHIBITOR”. Rhoda Au has served as a consultant for Biogen and Signant Health; received honoraria from NIH/NIA and University of Southern California. Boston University School of Medicine has received restricted gifts to be used for research from Robert Thomas, M.D. and Gates Ventures. George Perry has served as a consultant for Synaptogenix and Nervgen; participated on advisory boards for Synaptogenix and Nervgen; held stock and options from Synaptogenix, InvestiCure and Neurotez; received payment from John Hopkins University (Alaska) and University of Rhode Island. Maureen Dubreuil has received honoraria from Spondyloarthritis Research and Treatment Network and Rockpointe Education; received meeting and travel support from NIH and Spondyloarthritis Research and Treatment Network; participated on UBC Inc (Bimekizumab) DSMB/advisory board; played a leadership role on Spondyloarthritis Research and Treatment Network and Spondylitlis Association of American. Shu G Chen has served as a consultant for AMRIF BV, Wageningen, the Netherlands. Pamela B Davis has served as a consultant for Trinity College, Dublin, Ireland; participated on University of Chicago External Advisory Board, Clinical Research Forum-administrative board and Judson Services Board. Zheng, Fillmore, Cejudo, Osorio, Qiu, Qi, Do and Xu have no outside interests to declare.

Footnotes

Conflict of Interest Disclosures

None of the authors have financial interests to disclose.

Human subject consent: Informed human subject consent was not necessary for this study.

Meeting Presentation: No

References

  • 1.Gale SA, Acar D, Daffner KR. Dementia. Am J Med. 2018; 131:1161–1169. [DOI] [PubMed] [Google Scholar]
  • 2.Barker WW, Luis CA, Kashuba A, et al. Relative frequencies of Alzheimer’s disease, Lewy body, vascular and frontotemporal dementia, and hippocampal sclerosis in the State of Florida Brain Bank. Alzheimer Dis Assoc Disord. 2002; 16:203–12. [DOI] [PubMed] [Google Scholar]
  • 3.Mason A, Holmes C, Edwards CJ. Inflammation and dementia: Using rheumatoid arthritis as a model to develop treatments? Autoimmun Rev. 2018; 17:919–925. [DOI] [PubMed] [Google Scholar]
  • 4.Bolós M, Perea JR, Avila J. Alzheimer’s disease as an inflammatory disease. Biomol Concepts. 2017; 8:37–43. [DOI] [PubMed] [Google Scholar]
  • 5.Kinney JW, Bemiller SM, Murtishaw AS, Leisgang AM, Salazar AM, Lamb BT. Inflammation as a central mechanism in Alzheimer’s disease. Alzheimers Dement (NY). 2018; 4:575–590. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Akiyama H, Barger S, Barnum S, et al. Inflammation and Alzheimer’s disease. Neurobiol Aging. 2000; 21:383–421. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Cheng X, Shen Y, Li R. Targeting TNF: a therapeutic strategy for Alzheimer’s disease. Drug Discov Today. 2014. [DOI] [PubMed] [Google Scholar]
  • 8.Long H, Zhong G, Wang C, et al. TREM2 Attenuates Abeta1–42-Mediated Neuroinflammation in BV-2 Cells by Downregulating TLR Signaling. Neurochem Res. 2019; 44:1830–9. [DOI] [PubMed] [Google Scholar]
  • 9.McAlpine FE, Tansey MG. Neuroinflammation and tumor necrosis factor signaling in the pathophysiology of Alzheimer’s disease. J Inflamm Res. 2008;1:29–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Liddelow SA, Guttenplan KA, Clarke LE, et al. Neurotoxic reactive astrocytes are induced by activated microglia. Nature. 2017; 541:481–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Pan W, Kastin AJ. TNFalpha transport across the blood-brain barrier is abolished in receptor knockout mice. Exp Neurol. 2002; 174:193–200. [DOI] [PubMed] [Google Scholar]
  • 12.Tarkowski E, Andreasen N, Tarkowski A, Blennow K. Intrathecal inflammation precedes development of Alzheimer’s disease. J Neurol Neurosurg Psychiatry. 2003; 74:1200–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Gibofsky A Overview of epidemiology, pathophysiology, and diagnosis of rheumatoid arthritis. Am J Manag Care. 2012;18:S295–302. [PubMed] [Google Scholar]
  • 14.Chou RC, Kane M, Ghimire S, Gautam S, Gui J. Treatment for Rheumatoid Arthritis and Risk of Alzheimer’s Disease: A Nested Case-Control Analysis. CNSDrugs. 2016; 30:1111–1120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Farrugia M, Baron B. The role of TNF-α in rheumatoid arthritis: a focus on regulatory T cells. J Clin Transl Res. 2016; 2:84–90. [PMC free article] [PubMed] [Google Scholar]
  • 16.McGeer PL, Schulzer M, McGeer EG. Arthritis and anti-inflammatory agents as possible protective factors for Alzheimer’s disease: a review of 17 epidemiologic studies. Neurology 1996; 47 : 425 – 32. [DOI] [PubMed] [Google Scholar]
  • 17.Jaturapatporn D, Isaac MGEKNG, McCleery J, Tabet N. Aspirin, steroidal and non-steroidal anti-inflammatory drugs for the treatment of Alzheimer’s disease. Cochrane Database Syst Rev 2012; 2 : CD006378. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Chou MH, Wang JY, Lin CL, Chung WS. DMARD use is associated with a higher risk of dementia in patients with rheumatoid arthritis: A propensity score- matched case-control study. Toxicol Appl Pharmacol. 2017;334:217–222. [DOI] [PubMed] [Google Scholar]
  • 19.Judge A, Garriga C, Arden NK, Lovestone S, Prieto-Alhambra D, Cooper C, Edwards CJ. Protective effect of antirheumatic drugs on dementia in rheumatoid arthritis patients. Alzheimers Dement (N Y). 2017. Nov 9;3(4):612–621. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Van Gool WA, Weinstein HC, Scheltens P, Walstra GJ. Effect of hydroxychloroquine on progression of dementia in early Alzheimer’s disease: an 18-month randomised, double-blind, placebo-controlled study. Lancet Aug 11 2001;358(9280):455–60. [DOI] [PubMed] [Google Scholar]
  • 21.Zhou M, Xu R, Kaelber DC, Gurney ME. Tumor Necrosis Factor (TNF) blocking agents are associated with lower risk for Alzheimer’s disease in patients with rheumatoid arthritis and psoriasis. PLoS One. 2020; 15:e0229819. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.US Department of Veterans Affairs: About VHA: Veterans Health Administration. https://www.va.gov/health/aboutVHA.asp. Accessed 15 March 2021
  • 23.Turano A, Eibling D. Using Voogle to Search Within Patient Records in the VA Corporate Data Warehouse. Fed Pract. 2019; 36:518–523. [PMC free article] [PubMed] [Google Scholar]
  • 24.Price LE, Shea K, Gephart S. The Veterans Affairs’s Corporate Data Warehouse: Uses and Implications for Nursing Research and Practice. Nurs Adm Q. 2015; 39:311–8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Singh JA, Holmgren AR, Noorbaloochi S. Accuracy of Veterans Administration databases for a diagnosis of rheumatoid arthritis. Arthritis Rheum. 2004; 51:952–7. [DOI] [PubMed] [Google Scholar]
  • 26.Rosenbaum PR and Rubin DB. Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score. The American Statistician. 1985; 39:33–38 [Google Scholar]
  • 27.Austin PC. Some methods of propensity-score matching had superior performance to others: results of an empirical investigation and Monte Carlo simulations. Biom J. 2009; 51:171–84. [DOI] [PubMed] [Google Scholar]
  • 28.Austin PC. Statistical criteria for selecting the optimal number of untreated subjects matched to each treated subject when using many-to-one matching on the propensity score. American Journal of Epidemiology 2010; 172: 1092–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Rassen JA, Shelat AA, Myers J, Glynn RJ, Rothman KJ, Schneeweiss S. One-to-many propensity score matching in cohort studies. Pharmacoepidemiol Drug Saf. 2012. May;21 Suppl 2:69–80. [DOI] [PubMed] [Google Scholar]
  • 30.Ho DE, Imai K, King G, Stuart EA. “MatchIt: Nonparametric Preprocessing for Parametric Causal Inference.” Journal of Statistical Software. 2011; 42:1–28. [Google Scholar]
  • 31.Zhang Z, Kim HJ, Lonjon G, Zhu Y; written on behalf of AME Big-Data Clinical Trial Collaborative Group. Balance diagnostics after propensity score matching. Ann Transl Med. 2019; 7:16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Singh JA, Saag KG, Bridges SL Jr, et al. 2015 American College of Rheumatology Guideline for the Treatment of Rheumatoid Arthritis. Arthritis Care Res (Hoboken). 2016; 68:1–25. [DOI] [PubMed] [Google Scholar]
  • 33.Silva MVF, Loures CMG, Alves LCV, de Souza LC, Borges KBG, Carvalho MDG. Alzheimer’s disease: risk factors and potentially protective measures. J Biomed Sci. 2019; 26:33. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Orkaby AR, Nussbaum L, Ho YL, et al. The Burden of Frailty Among U.S. Veterans and Its Association With Mortality, 2002–2012. J Gerontol A Biol Sci Med Sci. 2019; 74:1257–1264. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Harding BN, Floyd JS, Scherrer JF, Salas J, Morley JE, Farr SA, Dublin S. Methods to identify dementia in the electronic health record: Comparing cognitive test scores with dementia algorithms. Healthc (Amst). 2020;8(2):100430. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Innala L, Berglin E, Möller B, et al. Age at onset determines severity and choice of treatment in early rheumatoid arthritis: a prospective study. Arthritis Res Ther. 2014. Apr 14;16(2): R94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Elze MC, Gregson J, Baber U, et al. Comparison of Propensity Score Methods and Covariate Adjustment: Evaluation in 4 Cardiovascular Studies. J Am Coll Cardiol. 2017; 69:345–357. [DOI] [PubMed] [Google Scholar]
  • 38.Huang LC, Chang YH, Yang YH. Can Disease-Modifying Anti-Rheumatic Drugs Reduce the Risk of Developing Dementia in Patients with Rheumatoid Arthritis? Neurotherapeutics. 2019; (3):703–709. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Breedveld FC, Weisman MH, Kavanaugh AF, et al. The PREMIER study: a multicenter, randomized, double-blind clinical trial of combination therapy with adalimumab plus methotrexate versus methotrexate alone or adalimumab alone in patients with early, aggressive rheumatoid arthritis who had not had previous methotrexate treatment. Arthr Rheum. 2006; 54:26–37 [DOI] [PubMed] [Google Scholar]
  • 40.Witte T Methotrexate as combination partner of TNF inhibitors and tocilizumab. What is reasonable from an immunological viewpoint? Clin Rheumatol. 2015. Apr;34(4):629–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Heneka MT, Carson MJ, El Khoury J, et al. Neuroinflammation in Alzheimer’s disease. Lancet Neurol. 2015. Apr;14(4):388–405 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Eikelenboom P, Hoozemans JJ, Veerhuis R, van Exel E, Rozemuller AJ, van Gool WA. Whether, when and how chronic inflammation increases the risk of developing late-onset Alzheimer’s disease. Alzheimers Res Ther. 2012; 4:15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Fan Z, Okello AA, Brooks DJ, Edison P. Longitudinal influence of microglial activation and amyloid on neuronal function in Alzheimer’s disease. Brain. 2015; 138:3685–98. [DOI] [PubMed] [Google Scholar]
  • 44.Butchart J, Brook L, Hopkins V, et al. Etanercept in Alzheimer disease: A randomized, placebo-controlled, double-blind, phase 2 trial. Neurology. 2015; 84:2161–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Tobinick EL, Gross H. Rapid improvement in verbal fluency and aphasia following perispinal etanercept in Alzheimer’s disease. BMC Neurol. 2008; 8:27. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Chen YM, Chen HH, Lan JL, Chen DY. Improvement of cognition, a potential benefit of anti-TNF therapy in elderly patients with rheumatoid arthritis. Jt Bone Spine. 2010; 77:366–7. [DOI] [PubMed] [Google Scholar]
  • 47.McGuinness B Understanding whether drugs for rheumatoid arthritis can reduce the risk of Alzheimer’s disease. https://www.alzheimers.org.uk/research/our-research/research-projects/understanding-whether-drugs-rheumatoid-arthritis-can-reduce-risk-alzheimersdisease; Accessed 15 Nov 2020
  • 48.Farrington CP. Relative incidence estimation from case series for vaccine safety evaluation. Biometrics. 1995; 51:228–235. [PubMed] [Google Scholar]
  • 49.Farrington CP, Nash J and Miller E. Case series analysis of adverse reactions to vaccines: a comparative evaluation. American Journal of Epidemiology 1996; 143:1165–1173. Erratum 1998; 147:93. [DOI] [PubMed] [Google Scholar]
  • 50.Kern DM, Cepeda MS, Lovestone S, Seabrook GR. Aiding the discovery of new treatments for dementia by uncovering unknown benefits of existing medications. Alzheimers Dement (N Y). 2019; 5:862–870. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Cepeda MS, Kern DM, Seabrook GR, Lovestone S. Comprehensive Real-World Assessment of Marketed Medications to Guide Parkinson’s Drug Discovery. Clin Drug Investig. 2019;39(11):1067–1075. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Whitaker HJ, Farrington CP, Spiessens B, Musonda P. Tutorial in biostatistics: the self-controlled case series method. Stat Med. 2006;25(10):1768–97 [DOI] [PubMed] [Google Scholar]
  • 53.Petersen I, Douglas I, Whitaker H. Self controlled case series methods: an alternative to standard epidemiological study designs. BMJ. 2016; 354: i4515. [DOI] [PubMed] [Google Scholar]
  • 54.Butler D, Kowall NW, Lawler E, Michael Gaziano J, Driver JA. Underuse of diagnostic codes for specific dementias in the veterans affairs New England healthcare system. J Am Geriatr Soc. 2012;60(5):910–915. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.King G and Nielsen R. Why Propensity Scores Should Not Be Used for Matching. Political Analysis. 2019;27(4):435–454 [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

tS1
NIHMS1733040-supplement-tS1.docx (14.7KB, docx)
tS2
NIHMS1733040-supplement-tS2.docx (21KB, docx)
fS1-7
NIHMS1733040-supplement-fS1-7.docx (461.1KB, docx)
f8-14
NIHMS1733040-supplement-f8-14.docx (513.5KB, docx)
f15-17
NIHMS1733040-supplement-f15-17.docx (578.8KB, docx)

RESOURCES