Tumor Necrosis Factor (TNF) Blocking Agents Reduce Risk for Alzheimer’s Disease in Patients with Rheumatoid Arthritis and Psoriasis

Authors: 8 Mengshi Zhou, Rong Xu, David C. Kaelber, and Mark E. Gurney 9 1 Department of Population and Quantitative Health Sciences, School of Medicine, Case Western Reserve 10 University, Cleveland, OH44106 11 Departments of Internal Medicine and Pediatrics and the Center for Clinical Informatics Research and Education, 12 The MetroHealth System, Cleveland, OH, USA 13 Tetra Therapeutics, 38 Fulton Street West, Grand Rapids, MI 49503 14 15

This large, retrospective case-control study of electronic health records from 56 million unique adult patients 23 examined whether or not treatment with a Tumor Necrosis Factor (TNF) blocking agent reduced risk for 24 Alzheimer's disease (AD) in patients with rheumatoid arthritis (RA), psoriasis, and other inflammatory diseases 25 which are mediated in part by TNF and for which a TNF blocker is an approved treatment. The analysis 26 compared the diagnosis of AD as an outcome measure in patients receiving at least one prescription for a TNF 27 blocking agent (etanercept, adalimumab, and infliximab) or for methotrexate. RA increased the risk for AD 28 with a prescription history for both a TNF blocker and methotrexate. Etanercept and adalimumab also reduced 36 risk for Alzheimer's disease and that this could be mitigated by a TNF blocking agent. A previous case-control 68 study of 8.5 million insured adults in the United States (US) reported increased risk for Alzheimer's disease in 69 patients with RA and that etanercept, a TNF blocking agent prescribed for the treatment of RA, was associated 70 with reduction of risk for AD (Adjusted Odds Ratio (AOR) = 0.30; 95% Confidence Interval (CI) = 0.08-0.87. P-71 value = 0.02) [13]. The study was based on 9,253 patients with Alzheimer's disease in the private insurance 72

database. 73
Alzheimer's disease is underrepresented in US private insurance databases as most patients with Alzheimer's 74 disease are older than 65, an age in the US when the majority of Americans transition from employer provided, 75 private insurance to Medicare, a national health insurance program. We therefore undertook a retrospective study 76 using the IBM Watson Healthcare Explorys Cohort Discovery platform which contains electronic health records 77 (EHR) from patients with private insurance, Medicare and Medicaid. The Explorys Cohort Discovery platform 78 contains the de-identified EHR of nearly 56 million unique patients with age ≥ 18 years from 26 healthcare 79 systems across all 50 states in the US from 1999 to 2018. We expanded our analysis to other inflammatory 80 diseases and to additional drugs in the class of TNF blocking agents to assess their potential effect on risk for 81 Alzheimer's disease. Recent studies have shown that with this unique EHR database and built-in informatics 82 tools, large hypothesis-driven case-control studies can be undertaken [14][15][16]. Sufficient patient EHR were 83 available for the evaluation of co-morbid AD in patients with RA or psoriasis that were treated with etanercept, 84 adalimumab, and infliximab. Etanercept (Enbrel®) is a fusion of soluble TNF receptor 2 with the Fc portion of 85 mouse immunoglobulin G1 [17]. Adalimumab (Humira®) is a fully human, anti-TNF monoclonal antibody [18], diagnosis of dementia (Table 1). SNOMED-CT disease names also were used to identify patients diagnosed with 116 one of eight inflammatory diseases: rheumatoid arthritis, ankylosing spondylitis, psoriasis, psoriatic arthritis, 117 inflammatory bowel disease, ulcerative colitis, and Crohn's disease. Inflammatory bowel disease is a SNOMED-118 CT diagnostic category that includes all patients with the subdiagnosis of ulcerative colitis but excludes patients 119 with Crohn's disease. Patients diagnosed with more than one inflammatory disease were excluded from the 120 analysis. For example, a patient diagnosed with rheumatoid arthritis who also was diagnosed with ankylosing 121 spondylitis, psoriasis, psoriasis with arthropathy, inflammatory bowel disease, ulcerative colitis, or Crohn's 122 disease was excluded from the analysis. The effects of each inflammatory disease on Alzheimer's disease or 123 dementia were compared to a non-inflammatory disease group that comprised patients who had no diagnosis of 124 any of the eight inflammatory diseases. 125 A patient was considered "taking a medication" if at least one outpatient prescription for the medication had been 126 written for the patient. Patients who took a drug treatment were identified by searching the generic drug names. 127 The Explorys Cohort Discovery platform has standardized all drug names to their generic names based on 128 RxNorm. There are five TNF blocking agents that have been approved by US Food and Drug Administration 129 (FDA): etanercept, adalimumab, infliximab, certolizumab pegol, and golimumab. Sufficient numbers of subjects 130 (at least 10 patients in each cell in a two by two table for at least one strata) were available to study the effects of 131 search tool. 141

Statistical Analysis 142
The crude Odds Ratio (OR), 95% Confidence Interval (CI), and P-values were calculated using the Fisher exact 143 test in univariate analysis. The Adjusted OR (AOR), 95% CI and P-values were calculated using the Cochran-144 Mantel-Haenszel (CMH) method by controlling for confounding factors such as age, gender, and race [25]. 145 Statistical tests were conducted with significance set at P-value < 0.05 (two-sided). 146 To compute the AORs of dementia for the drug use group versus the corresponding non-drug use group, or for the 147 inflammatory disease group versus the non-inflammatory disease group, we first calculated the OR for each of the 148 8 strata (tabulation of 2 age groups: 18-65, and >65 years; 2 genders: male and female; and 2 races: Caucasian and 149 Non-Caucasian) and then calculated the weighted average across all strata using the CMH method [25]. The AOR 150 comparing the drug use group versus the non-drug use group were calculated separately in patients with a 151 diagnosis code of rheumatoid arthritis and psoriasis. There were not enough patients (at least one cell in a two by 152 two table have counts lower than 10 for all the strata) to test the effect of TNF blocking agents in patients with the 153 other inflammatory diseases. To assess the effects of gender, age, and race on the risk associations between the 154 therapeutic drugs and dementia, we separately calculated the crude OR of dementia among the 2 age groups, 2 155 gender groups, and 2 race groups. To adjust for methotrexate prescription status in psoriasis patients, we 156 separately calculated the crude OR for a diagnosis of Alzheimer's disease or dementia among 16 strata (tabulation OR across all 4 strata (tabulation 2 genders: male and female; and 2 races: Caucasian and non-Caucasian) using 165 the CMH method [25]. 166 We used patient insurance types (e.g., Medicaid, Medicare, private insurance) as a proxy for their socioeconomic 167 status. We calculated the AOR of dementia by taking the insurance status into account. We performed this 168 procedure for patients with age between 18-65 and patients with age > 65 years old separately, since the majority 169 Alzheimer's disease is a clinical diagnosis made by a physician based on an interview and cognitive assessment. 189 Even with carefully selected patients enrolled in large, multi-center, Phase 3 clinical trials, Alzheimer's disease is 190 misdiagnosed based solely on clinical criteria in up to 25% of patients [26,27]. In a database such as the 191 Explorys Cohort Discovery platform that aggregates EHR from large, integrated health delivery systems and 192 community-based networks, the rate of misdiagnosis of Alzheimer's disease may be even higher. Therefore, we 193 structured our analysis to proceed in two stages. We first calculated the AOR for a SNOMED-CT diagnosis of Parkinson's disease. Of 338,400 patients with a diagnosis code for Alzheimer's disease, 172,750 (51%) received 210 at least one prescription for a cholinesterase inhibitor or memantine, while an additional 157,850 patients with a 211 diagnosis code for dementia were also treated with at least one drug approved for Alzheimer's disease (24%). A 212 All rights reserved. No reuse allowed without permission.
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint high proportion of these patients likely represent cases of dementia due to suspected Alzheimer's disease. 213 Querying more broadly in the Explorys Cohort Discovery platform for any use of a cholinesterase inhibitor or 214 memantine, 145,860 patients were prescribed at least one of the four drugs without a SNOMED-CT diagnosis of 215 dementia or Alzheimer's disease. This included 9,540 patients with a diagnosis code for Parkinsonism, 216 3,900 with a diagnosis code for degenerative disease of the central nervous system, and 3,990 patients with a 217 diagnosis code for traumatic brain injury. Thus, Explorys captures both approved and off-label use of 218 prescription drugs. 219

Risk for Alzheimer's Disease is Increased by Systemic Inflammation 220
We hypothesized that systemic inflammatory diseases which in part are mediated by TNF increase the risk for 221 Alzheimer's disease. We therefore selected for study those inflammatory diseases for which a TNF blocking 222 agent is an FDA approved treatment [5]. Approved indications include rheumatoid arthritis, ankylosing 223 spondylitis, psoriasis, psoriatic arthritis, inflammatory bowel disease, ulcerative colitis, and Crohn's disease. 224 Five TNF blocking agents have been approved by FDA, of these, sufficient patients were prescribed etanercept 225 (44,210 patients), adalimumab (66,820 patients), and infliximab (40,290) to allow inclusion in the retrospective, 226 case-control study. In addition, 161,560 patients received at least one prescription for methotrexate but had not 227 been treated with a TNF blocking agent. Methotrexate is a small molecule inhibitor of immune cell activation 228 that suppresses systemic TNF [29]. In normal clinical practice, patients are started on methotrexate and those 229 who fail methotrexate may progress to a treatment with a TNF blocking agent. Of those who progress to the TNF 230 blocking agent, 60% will continue on methotrexate and 40% will drop the methotrexate [29, 30]. 231 The cumulative age distribution for patients with a diagnosis code for rheumatoid arthritis, dementia or 232 Alzheimer's disease is shown in Figure 2. Given that the Explorys Cohort Discovery platform includes at least 233 some information on patients making up over 20% of the US population, the cumulative age distribution should 234 reasonably represent the age-dependent prevalence of these diseases in the US. As expected, the age distribution 235 of patients with a diagnosis of rheumatoid arthritis skews younger, while patients with dementia or Alzheimer's 236 disease skew older. The average treatment retention time varies among the three TNF blocking agents selected 237 for study with etanercept having a better than 50% patient retention rate at 12 years of treatment, while the 50% 238 retention rates for adalimumab and infliximab are about 2 and 3 years, respectively [30,31]. To comply with 239 HIPAA compliant de-identification of EHR, the Explorys Cohort Discovery platform does not allow patient level 240 access to prescription data. Therefore, we compared the benefit of etanercept to adalimumab and infliximab as a 241 surrogate for duration of exposure as a modifier of treatment benefit. 242 Systemic inflammatory disease is a substantial risk factor for Alzheimer's disease (Figure 3). Greatest risk is 243 associated with inflammatory bowel disease and Crohn's disease followed by rheumatoid arthritis, ulcerative 244 colitis (as a subcategory of inflammatory bowel disease), ankylosing spondylitis and psoriasis. There is no 245 additional risk associated with psoriatic arthritis. Inflammatory bowel disease, Crohn's disease and rheumatoid 246 arthritis increase the risk for Alzheimer's disease over two-fold (AOR 2.06 (2.02-2.10) -AOR 2.46 (2.33-2.59)). 247 The increased risk contributed by systemic inflammation was replicated in patients with a diagnosis code for 248 dementia in which Alzheimer's disease is a subcategory (Figure 4). Risk for dementia was elevated two-fold in 249 patients with rheumatoid arthritis, inflammatory bowel disease, ankylosing spondylitis, ulcerative colitis and 250 Crohn's disease (AOR 2.00 (1.94-2.06) -AOR 2.69 (2.66-2.72)). Psoriasis also increased risk for Alzheimer's 251 disease and dementia although to a lesser extent (AOR 1.37 (1.31-1.42) and AOR 1.49 (1.46 -1.53), 252 respectively). Thus, systemic inflammation increases risk for Alzheimer's disease across multiple diseases 253 involving the joints, the gut and the skin. 254

TNF Blocking Agents Reduce Risk for Alzheimer's Disease in Patients with Systemic Inflammation 255
If the increased risk for Alzheimer's disease is associated with TNF, then TNF blocking agents should reduce risk 256 for Alzheimer's disease in patients with co-morbid inflammatory disease. Since TNF blocking agents frequently 257 are prescribed as an add-on to methotrexate, we compared subjects who received a TNF blocking agent without was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  Discovery platform does not permit access to patient level data, so a limitation of this analysis is that we do not 280 know if the TNF blocker was prescribed as an add-on to methotrexate or if the two drugs were prescribed 281 sequentially with methotrexate dropped once treatment with the TNF blocker was started. 282 We separately assessed the effects of gender, age, and race on the effect of TNF blocking agents on the risk of 283 dementia among patients with rheumatoid arthritis. Data are shown for etanercept in Figure 9. Female and male 284 patients had a similar benefit when treated with etanercept (P = 0.14), adalimumab (P = 0.14), or infliximab (P = 285 0.71). There was a significant difference in benefit when comparing Caucasian versus non-Caucasian patients 286 with a prescription history for etanercept (P = 0.022) but not for adalimumab (P = 0.34) or infliximab (P = 287 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (Figure 9). Greater benefit also was seen in younger patients prescribed adalimumab (P = 290 0.022) or infliximab (P = 0.00026) as compared to older patients. 291 We also explored insurance status as a potential confound for age (comparing younger patients with private 292 insurance as compared to patients over 65 years with Medicare) and socioeconomic status (patients with Medicaid 293 or private insurance). There was no difference between patients with private insurance as compared to Medicare, 294 nor does adjusting insurance status influence the effect of TNF blockers on the risk for dementia (Figure 10). 295 There were too few patients with Medicaid and Alzheimer's disease (21,030, 6.2%) or Medicaid and a diagnosis 296 of dementia (84,040, 8.4%) for analysis ( Table 1). 297 Finally, we sought to understand if treatment with a TNF blocking agent simply reduced the additional risk for 298 Alzheimer's disease contributed by a co-morbid inflammatory disease, or if the TNF blocking agent could reduce 299 risk for Alzheimer's disease below the risk in the general population in the absence of inflammatory disease as a 300 risk factor and excluding a prescription history for a TNF blocker or methotrexate. Indeed, the risk for 301 pathogenesis. Systemic TNF may promote neuroinflammation in the brain through receptor-mediated 317 transcytosis [12]. Biologic drugs such as etanercept and adalimumab distribute poorly to brain, so we presume 318 the TNF blockers act systemically to prevent TNF from reaching the brain and thereby prevent or delay the onset 319 of Alzheimer's disease. 320 Our study indicates that systemic inflammatory conditions involving TNF are a potentially treatable risk factor for 321 Alzheimer's disease. This is a small, but potentially treatable subgroup of people at risk for Alzheimer's disease. 322 The total number of patients with Alzheimer's disease in the US is estimated to be 5.3 million in 2018 [28]. 323 Based on our estimate of population attributable risk, 0.21 million cases (4.04%) of Alzheimer's disease might be 324 preventable by treating rheumatoid arthritis with a TNF blocking agent, 0.08 million cases (1.51%) of 325 Alzheimer's disease might be preventable by treating Crohn's disease with a TNF blocking agent, and 0.026 326 million cases (0.50%) of Alzheimer's disease might be preventable by treating psoriasis with a TNF blocker or 327 nearly 6% of cases of Alzheimer's disease. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. PDE4B allosteric inhibitor after acute brain injury reduces microglial activation, reduces production of TNF, and 357 provides neuroprotective benefit in rats [50]. Small molecule drugs targeting the TLR4-TNF pathway may 358 produce greater therapeutic benefit than large, biologic drugs for which brain penetration may be limited. 359 All rights reserved. No reuse allowed without permission.
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. frequently associated with osteoarthritis (46-48% of cases and controls). The risk reduction associated with use of 363 an NSAID was less than the risk reduction of Alzheimer's disease due to use of methotrexate in patients with lack of efficacy seen with NSAIDs in Alzheimer's disease may be associated with the lack of a strong suppressive 372 effect on TNF production. Low dose prednisone (10 mg/day) is used to treat rheumatoid arthritis, presumably by 373 suppressing immune activation and TNF production, but long term use is associated with increasingly 374 unacceptable side effects. Treatment with low dose prednisone for 1 year failed to show benefit in patients with 375 mild-to-moderate Alzheimer's disease [56]. 376

Strengths and Limitations 377
The strength of the current study is that the massive-scale of the EHR database allowed us to access real-world 378 patient populations. With 64 million unique patient EHR, the IBM Watson Health Explorys Cohort Discovery 379 platform has aggregated the medical records from approximately 20% of the US population. Such a massive 380 database covers the full spectrum of disease diagnoses and the use of prescription drugs for both approved and 381 off-label indications. Access to large data sets is particularly useful in the case of Alzheimer's disease where a 382 substantial fraction of patients in a clinical setting are either misdiagnosed as having the disease when they do not 383 or for whom the physician is unwilling to make a specific diagnosis even though Alzheimer's disease is suspected 384 and an approved medication for Alzheimer's disease is prescribed. This allowed us to test for the association of 385 co-morbid Alzheimer's disease with systemic inflammatory disease and to test the benefit of TNF blocking agents 386 by first selecting patient cohorts with a specific diagnosis of Alzheimer's disease and then by replicating that 387 finding in a larger cohort of patients with dementia which better reflects the uncertainty of the diagnosis of 388 Alzheimer's disease in a community health setting. 389 The scale of the Watson Health Explorys Cohort Discovery platform allowed us to maintain adequate statistical 390 power while we studied disease association and the benefit of TNF blockers across multiple inflammatory 391 diseases. This often is difficult in traditional observational studies which frequently are limited in sample size. 392 Compared to the private insurance database used in the previous study [13], the Explorys Cohort Discovery 393 platform contains a study population that includes patients with private insurance and Medicare. This is 394 particularly critical for the study of diseases that primarily affect an elderly population as the majority of patients 395 older than 65 years in the US transition from private insurance to Medicare. 396 The current study has limitations. First, the quality or incompleteness of the EHR for individual patients may 397 affect the sensitivity and specificity of the analysis and misdiagnosis will contribute to false positives and false 398 negatives within the patient cohorts selected for study Second, the retrospective case-control analysis suggests a 399 potential therapeutic benefit of TNF blocking agents in Alzheimer's disease, however, the analysis does not 400 distinguish an effect on treatment of prevention, nor does our analysis address how the severity of disease may 401 influence benefit. A limitation of the Explorys Cohort Discovery platform is the inability to access patient level 402 data for reasons of confidentiality. We are unable to extract age at diagnosis, duration of treatment, or years of 403 follow up which are potential confounding factors to this analysis, nor are we able to build an exposure-response 404 relationship for the reduction in risk due to use of a TNF blocking agent. Due the fact that we used population-405 level data, we relied on the odds ratio for the statistical analysis and cannot calculate an incidence rate ratio [36]. 406 Additionally, the Explorys Cohort Discovery platform considers patients to be "taking a medication" if at least on 407 outpatient prescription has been written for the medication, but does not record if medications were prescribed 408 sequentially after the first drug failed, or whether two or more medications were prescribed concurrently. These 409 All rights reserved. No reuse allowed without permission.
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint potential confounds may lead our findings to underrepresent the "true findings." Without access to patient level 410 data, we were only able to control for known confounding factors using the Cochran-Mantel-Haenszel (CMH) 411 method [25] instead of systematically identifying and controlling for potential confounding factors using a more 412 rigorous multivariable regression method. Even with these limitations, the results of our study are consistent with 413 a previous study using patient-level data [13] and with a unpublished study reported in the popular press. As 414 shown in Table 2, the therapeutic benefit of etanercept on risk for co-morbid Alzheimer's disease that we report 415 in patients with rheumatoid arthritis is consistent with a previous case-control analysis of 8.5 million privately 416 insured patients [13] and with an unpublished study that reportedly analyzed claims data from 254,000 patients. 417 Thus, our study conducted with a massive EHR database is consistent with smaller studies conducted with 418 patient-level data collected using private insurance claims databases. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint Figure 9. Effect of age, gender and race on risk for a diagnosis of dementia in rheumatoid arthritis patients treated 518 with etanercept. Young patients (18-65 years) showed greater benefit than did older patients (over 65 years). The 519 effect of gender and race was not significant. 520 521 522 523 All rights reserved. No reuse allowed without permission.
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint Figure 11. Comparison of the risk for a diagnosis of Alzheimer's disease or dementia in patients with a diagnosis 530 of rheumatoid arthritis or psoriasis with a prescription history for etanercept, adalimumab, or infliximab who were 531 not prescribed methotrexate against the general population. 532 533 534 535 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint Supporting information 536 S1 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease 537 associated with a diagnosis for an inflammatory disease versus the non-inflammatory disease group. 538 S2 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of dementia associated with 539 a diagnosis for an inflammatory disease versus the non-inflammatory disease group. 540 S3 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease or 541 dementia in patients with a diagnosis of rheumatoid arthritis with a prescription history of a TNF blocker or 542 methotrexate versus the no-drug group. The analysis excluded patients with a prescription history of a TNF 543 blocker and methotrexate. 544 S4 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease or 545 dementia in patients with a diagnosis of rheumatoid arthritis with a prescription history of a TNF blocker and 546 methotrexate versus the no-TNF blocker but with methotrexate patient group. 547 S5 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease or 548 dementia in patients with a diagnosis of psoriasis with a prescription history of a TNF blocker or methotrexate 549 versus the no-drug group. The analysis excluded patients with a prescription history of a TNF blocker and 550 methotrexate. 551 S6 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease or 552 dementia in patients with a diagnosis of psoriasis with a prescription history of a TNF blocker versus the no-drug 553 group adjusting for methotrexate prescription status. 554 S7 Table. Two by two tables investigating the effects of age, gender and race on the association between the 555 prescription of a TNF blocker and dementia (or Alzheimer's disease) in patients with a diagnosis code for 556 rheumatoid arthritis. 557 All rights reserved. No reuse allowed without permission.
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which this version posted September 27, 2019. ; https://doi.org/10.1101/19007666 doi: medRxiv preprint S8 Table. Two by two tables calculating the insurance status-adjusted and crude ORs of dementia for each TNF 558 blocker use group compared to the non-drug use group in patients with a diagnosis of rheumatoid arthritis. The 559 analysis excluded patients with a prescription history of a TNF blocker and methotrexate. 560 S9 Table. Two by two tables for each of the strata comparing the odds for a diagnosis of Alzheimer's disease or 561 dementia in rheumatoid arthritis or psoriasis patients who have prescribed a TNF blocker but no methotrexate 562 prescription history against the general population. 563 564 565 566 All rights reserved. No reuse allowed without permission. was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.