Abstract
Objective
To evaluate seasonal patterns of early inflammatory arthritis (IA) onset and potential associations with IA symptom onset.
Methods
The Canadian Early Arthritis Cohort (CATCH) is an inception cohort study of adults with early (12 months or less) IA. We used patient reports of symptom onset as a proxy of IA onset and examined the seasonal distribution of IA onset over 10 years. Influenza time series was based on laboratory‐confirmed influenza A and B from the Canadian FluWatch surveillance from 2010‐2016. Bivariate analysis of influenza and IA was performed using cross‐correlations with different time lags and Poisson regression. IA and influenza were recorded as monthly total frequencies.
Results
Of 2519 IA patients, 88% had confirmed rheumatoid arthritis (RA). Significantly, more IA onsets occurred in winter compared with other seasons (P = 0.03); although IA onset was more frequent in January, the difference between months was not statistically significant. Compared to months with the lowest influenza rates, months with the highest influenza rates had a statistically significant, but trivial, increase of 0.003% in the incidence of IA (incidence rate ratio (95% confidence interval): 1.00003 (1.00005; 1.000053), P = 0.02).
Conclusion
Although IA symptom onset occurs more frequently in winter, we found that flu outbreaks were not associated with a meaningful increase in IA symptom onset in a large, well‐characterized cohort of Canadian adults over 6 years.
Keywords: Early inflammatory arthritis, rheumatoid arthritis, risk factors, influenza, time series analysis
Introduction
Early inflammatory arthritis (IA) is a recent‐onset arthropathy that may resolve spontaneously, develop into rheumatoid arthritis (RA), another definite arthropathy, or remain undifferentiated 1. RA is the most common IA worldwide, with 0.5%‐1% prevalence and annual incidence of 25‐55 cases per 100,000 population 2, 3, 4.
RA is an autoimmune disease perturbation 5, 6. The role of viral infections as potential environmental triggers of autoimmune diseases such as RA has been widely discussed 7, 8. Our recent meta‐analysis showed that parvovirus B19 and chronic hepatitis C infections are associated with the development of RA and that chikungunya virus is associated with the development of persistent IA 9.
Influenza is a common viral infection worldwide with usual annual epidemics during the late fall through early spring 10. Influenza has been linked to the development of autoimmune disorders 11, 12, 13, 14, but its effect on the risk of developing IA/RA is poorly understood. A significant association was found between rheumatoid factor (RF)‐positivity and the presence of anti‐influenza antibodies after an epidemic of influenza A in 1957 (odds radio [95% confidence interval (CI)] = 13.6 [1.4; 127.4], P = 0.02)15. Influenza was also accompanied by high levels of human leucocyte antigen D‐related expression on monocytes, which is an indicator for antigen‐presentation ability and T‐cell activation. Others have shown a significant increase in interleukin‐1, interleukin‐6, and tumor necrosis factor‐alpha levels in response to influenza stimuli 16. Thus, the similarities of the cytokines and cells involved in the pathogenesis of influenza and RA, coupled with the presence of RF+ in adults with RA years before its overt clinical manifestation 17, 18, 19, offer a potential mechanism through which exposure to the influenza virus may trigger development of RA.
Significance & Innovations.
RA has seasonal timing of onset.
This corresponds with the timing of influenza.
However, the association is implied but very weak (yet statistically significant enough) when adjusting for other known risk factors.
The aim of the study was to examine seasonal patterns of IA symptom onset and its relation to influenza outbreaks, including different lags (ie, induction periods) between exposure and outcome epochs.
Methods
Study methods and results are reported in accordance with Strengthening the Reporting of Observational Studies in Epidemiology Guidelines 20.
Study design
Data from a multisite inception cohort of patients with early IA suspected to be RA (Canadian Early Arthritis Cohort [CATCH]) were analyzed and combined with data from the Canadian national flu surveillance program (https://www.canada.ca/en/services/health/diseases-conditions.html) to evaluate exposure‐outcome relationships. The CATCH study has ethics approval obtained from each participating CATCH site, and patients signed informed consent.
Setting and Study Population
Inflammatory arthritis
Participants were adults with early IA recruited from 19 sites across Canada from January 2007 through January 2017. Within CATCH, RA is diagnosed using 1987 ACR or 2010 ACR/EULAR criteria, but patients are not required to meet these criteria at enrollment 3, 21. Diagnosis of IA is confirmed by a rheumatologist, with the majority (88%) meeting RA classification criteria. Patients with infectious, crystal‐induced, psoriatic arthritis or other connective tissue diseases at baseline or subsequent to enrollment 22 were excluded. Patients from CATCH receive treatment according to the discretion of the treating rheumatologist and in accordance with the Canadian guidelines for RA management 23, 24.
Influenza
In September 2010, the Public Health Agency of Canada launched a national surveillance program (FluWatch) to monitor influenza and influenza‐like illnesses across Canada. FluWatch identifies influenza cases by reviewing hospital, laboratory, and admission databases and infection control logs for patients hospitalized during influenza season with a documented positive influenza test (https://www.canada.ca/en/services/health/diseases-conditions.html).
Measures
Primary Outcome
The main outcome was the total monthly cases of IA symptom onset based on patient reports from CATCH across all participating sites. IA onset is often insidious, and there are often several months of delay before the patient is diagnosed by a physician 25. At enrollment, patients were asked, “When did you have your first episode of joint pain and swelling lasting for at least 6 weeks?” Physician‐reported IA symptom onset was recorded by rheumatologists based on patient reports of the first episode of joint pain and swelling lasting longer than 6 weeks. All patients enrolled in CATCH could have had IA symptoms for up to 12 months prior to enrollment. IA onset was recorded as month/year.
Exposure variable
Regional reports of influenza diagnoses were the exposure of interest. We used laboratory‐confirmed new influenza A and B cases from September 1 2010 to December 30 2016 in each province from the FluWatch database. Influenza time series was recorded as monthly total frequencies through the same time period. It is important to note that this was not a study of confirmed influenza in people developing early RA/suspected RA but was a correlation of the timing of RA onset and the background rate of influenza cases over each month, with data of 6 years superimposed.
Sociodemographic variables
Patient age, sex, ethnicity, education, and smoking status were obtained from a sociodemographic questionnaire.
Confounders
Because the main unit of analysis was the month of disease onset (and not individuals), common risk factors (eg, age, sex, education) that do not change from month to month were not included.
Statistical analysis
Baseline descriptive statistics were calculated and characteristics between patients with confirmed RA vs. undifferentiated IA were compared using independent t tests and χ2. The number of IA and influenza onset cases was recorded as the monthly cumulative values. All results were reported at a two‐sided level of significance of 0.05.
Median monthly IA onset was aggregated over a 10‐year period from January 2006 to January 2016 and presented graphically using box plots. One‐way analysis of variance was used to compare differences in symptom onset across the 12 months of a year. Potential seasonal variations in aggregate IA onset across winter (December‐February) vs. all other seasons were compared using two‐sample t test.
The association between monthly influenza and IA onset from September 2010 to December 2016 inclusively was assessed using cross‐correlations 26. The cross‐correlation method is based on identifying lags of the input series x at time t (influenza) that can be predictive of the response series y at time t (IA). A maximum number of 12 lags for influenza (1 lag = 1 month based on our data) that might be predictive of IA were included, with zero lag representing cross‐correlation of IA with influenza onset during the same month. As we were interested in whether influenza leads to chronic IA (suspected or proven RA), we reported correlation coefficients only for negative lags. We investigated whether short‐term variation in the IA onset can be partially explained by changes in the influenza onset over time using Poisson regression for count data. Time series Poisson regression analysis 27 (with adjusted standard errors to account for overdispersion) was used to examine incidence rate ratio (IRR) after adjustment for seasonality and long‐term trends that might obscure the true exposure‐outcome relationship. By definition, IRR is a ratio of two incidence rates in exposed and unexposed individuals. In the context of our study that uses a month as the main unit of analysis, months with the highest influenza activity were considered as exposed months, and months with the lowest influenza activity were considered as unexposed months. Statistical analysis was performed using SAS 9.4 software 28 and Stata SE 13 software 29.
Physician and patient symptom onset dates were compared using intraclass correlation coefficients (ICCs) 30 and were categorized into three levels based on agreement 31. We prespecified the minimum ICC with at least moderate reliability (ICC = 0.60‐0.74) to use patient‐reported date of symptom onset to measure IA onset over time.
Results
Sample characteristics at study entry
At the time of data extraction (12 October 2017), 2519 patients constituted the final sample of patients with less than 13 months of symptoms at enrollment. Among them, 88% (n = 2209) fulfilled RA classification criteria. Patients’ characteristics at study entry are provided in Table 1. The majority of IA patients were female (71%), white (82%), had more than a high school education (58%), and were current/past smokers (56%), with a mean (SD) age of 55 (15) years and mean (SD) symptom duration of 5 (3). Baseline disease activity based on Disease Activity Score 28 (DAS28)–C‐reactive protein (CRP) three‐variable and DAS28–Erythrocyte Sedimentation Rate (ESR) three‐variable was 4.6 ± 1.3 and 4.9 ± 1.5, respectively. Patients with IA had mean ± SD ESR of 27 ± 23 mm/h, CRP of 14 ± 18 mg/l, Health Assessment Questionnaire Disease activity score of 1.0 ± 0.7, and were RF‐ and Anti‐citrullinated protein antibody (ACPA)‐positive in 44% and 53% of cases, respectively. However, there were n = 12 (0.5%) missing data for RF status and n = 702 (28%) for ACPA status.
Table 1.
Sample characteristics at study entry
| Characteristics | All IA Cohort (n = 2519) |
|---|---|
| Sociodemographic characteristics | |
| Age, mean (SD), years | 55 (15) |
| Female, n (%) | 1791 (71) |
| Caucasian ethnicity, n (%) | 2053 (82) |
| Education ≤high school, n (%) | 1051 (42) |
| Never smoked, n (%) | 1109 (44) |
| Disease characteristics at baseline | |
| Patient‐reported symptom duration mean (SD), months (from baseline visit) | 5 (3) |
| Physician‐reported symptom duration mean (SD), months (from baseline visit) | 5 (3) |
| Tender 28 joint count, mean (SD) | 8 (7) |
| Swollen 28 joint count, mean (SD) | 7 (6) |
| DAS28 (CRP) three‐variable, mean (SD) | 4.6 (1.3) |
| DAS28 (ESR) three‐variable, mean (SD) | 4.9 (1.5) |
| ESR, mean (SD), mm/h | 27 (23) |
| CRP, mean (SD), mg/l | 14 (18) |
| HAQ DI, mean (SD) | 1.0 (0.7) |
| RF positive,* n (%) | 1103 (44) |
| ACPA positive,* n (%) | 958 (53) |
Abbreviation: RA, rheumatoid arthritis; UIA, undifferentiated inflammatory arthritis; DAS, disease activity score; CRP, C‐reactive protein; ESR, erythrocyte sedimentation rate; HAQ DI, Health Assessment Questionnaire Disease activity; RF, rheumatoid factor; ACPA, antibodies against citrullinated proteins.
Missing data affected the number of subjects for these variables.
Univariate analysis of inflammatory arthritis time series
We visually assessed the number of IA onsets from the CATCH cohort by month and season (Figure 1) and quantitatively assessed for the entire study period. The number of IA symptom onset for 10‐year aggregated data varied; the highest being in January, but the overall differences between months were not statistically significant (Figure 1, top panel; P = 0.6). There were significantly more IA onsets in winter than in other seasons (mean 22 (95% CI (18; 25)) and 18 (95% CI (17; 20)), respectively, P = 0.03) (Figure 1, bottom panel).
Figure 1.
Inflammatory arthritis onset aggregated by month (top panel) and aggregated by season (bottom panel), 2006‐2016.
Bivariate time series analysis of inflammatory arthritis and influenza
Figure 2 shows plots of the IA and influenza time series over six years. IA symptom onset tended to have noticeable peaks in January and troughs in summer, with, however, the exception of August 2014. The downward trend at the end of the study period reflects an artifact of decreased enrollment in CATCH, not in incidence. Cross‐correlations between new influenza and IA symptom onset are presented in Figure 3. As you can see from the figure, all the correlation coefficients on the y axis are clustered around the null value (ie, zero).
Figure 2.
Time series of inflammatory arthritis and influenza onset over time, 2010‐2016.
Figure 3.
Cross‐correlations of new inflammatory arthritis and influenza onsets, 2010‐2016.
As there was moderate agreement between patient‐ and physician‐reported symptom month of onset (ICC (95% CI) = 0.70 (0.67; 0.79), we used patient reports of onset for the time series calculations.
Time series regression analyses, which are unadjusted for seasonal and long‐term patterns, may risk spurious conclusions. Therefore, we applied the flexible spline function of time fitted to the IA time series to remove these patterns from our data. As we can see in Figure 4, the model with splines does not reveal any seasonal pattern over time. Adjusted for seasonality and long‐term trend regression of IA on influenza, time series estimated that, compared with months that had the lowest influenza rates, months with the highest influenza rates had a statistically significant but trivial increase of 0.003% in the incidence of IA (IRR [95% CI] 1.00003 [1.000005; 1.000053], P = 0.02). No effect of influenza on IA onset was evident when IA onset was lagged by 1 and 3 months (0.99 [95% CI (0.99; 1.00), P = 0.84], and 0.99 [95% CI (0.99; 1.00), P = 0.33], respectively).
Figure 4.
Trends in the inflammatory arthritis over time, 2010‐2016 (flexible spline function).
Discussion
There are few studies that evaluate seasonal variability of IA/RA. Our time series analysis is, to our knowledge, the first to assess the monthly and seasonal distribution of early‐incident IA onsets over 10 years of observation. We found that in a large incident cohort of Canadians with IA, significantly more reported onset in winter compared with other seasons.
Jacoby, Jayson, and Cosh also reported that in Bath, United Kingdom, more RA onset occurred in winter, with December being the highest onset month 32. Söderlin et al observed that in Swedish RA patients observed in 1996‐2004, symptom onset was most likely to occur from December through March as compared with other months 33. Silman et al found inconsistent peaks of incident IA in the United Kingdom in April and October over 2 years of observations 34.
Similarly, when we examined IA symptom onset and influenza by month over 6 years, we observed similar peaks of onset in January. Results from cross‐correlation analysis showed random scattering of cross‐correlations with no evidence of association between IA and influenza. Furthermore, after adjusting for seasonal and long‐term trends, the risk of IA that is due to influenza was trivial, only explaining 0.3% of IA per 100 cases of influenza.
Our study had several strengths. Our research is based on a large sample of patients with well‐characterized IA inception cohort with patients drawn from 19 centers across Canada. Patient demographics in CATCH are similar to the population‐based studies 4, 35. Including patients with recent symptom‐onset IA (12 months or less) helped to minimize the risk of recall bias. We minimized the risk of disease misclassification and reporting bias by using only laboratory‐confirmed influenza A and B cases retrieved from the influenza surveillance program. We applied the time series regression analysis at the population level, at which traditional risk factors, such as age, sex, and health status, that do not significantly change from month to month are not necessary to include as confounders. Even though participation in CATCH is voluntary and not every incident IA case was captured in this cohort, the estimated IRR with months determining the exposed and unexposed groups was unlikely affected by study participation as the rates of influenza were not determined by recall but by regional surveillance. Additionally, we did not prove that the patients in the CATCH cohort had influenza at similar rates as those in the regional provincial surveillance data. Our study also has limitations. First, there is a risk of ascertainment bias and underreporting of influenza cases. There might be additional confounding factors that were not measured. For example, low socioeconomic‐status individuals may be predisposed to influenza and may also experience greater difficulty accessing care. Use of ecological data on influenza exposure preclude making causal inferences about influenza and IA onset in individuals. Because essentially, there was no association with timing of influenza, we did not subset patients with RA meeting criteria vs. those not meeting criteria, as all patients enrolled in the CATCH cohort are thought to have RA, and since the association was statistically significant but weak, the results would not change if we eliminated the 12% who did not meet ACR criteria for RA. Additionally, even if an association had been found, this would only be a study of correlations between onset of IA and onset of influenza from surveillance programs and would not have necessarily been a cause‐and‐effect relationship as influenza serology was not collected in this cohort.
In conclusion, we found that although early IA onset tends to cluster in winter months, we found no meaningful increase in IA onset that was due to influenza.
Author Contributions
All authors were involved in the study design, data analysis, initial outlines, drafting, and final approval of the manuscript.
Acknowledgments
All authors gratefully acknowledge the Canadian Early Arthritis Cohort (CATCH) investigators: Murray Baron, MD, Louis Bessette, MD, Gilles Boire, MD, Vivian Bykerk, MD, Ines Colmegna, MD, Sabrina Fallavollita, MD, Derek Haaland, MD, Paul Haraoui, MD, Glen Hazlewood, MD, Carol Hitchon, MD, Shahin Jamal, MD, Raman Joshi, MD, Ed Keystone, MD, Bindu Nair, MD, Peter Panopoulos, MD, Janet Pope, MD, Laurence Rubin, MD, Carter Thorne, MD, Edith Villeneuve, MD, Michel Zummer, MD.
The CATCH study was designed and implemented by the investigators and financially supported through unrestricted research grants from Amgen and Pfizer Canada, founding sponsors since January 2007; UCB Canada, AbbVie Corporation and Bristol‐Myers Squibb Canada since 2011; Medexus Inc. since 2013; Eli Lilly Canada since 2016; Merck Canada since 2017, and Sandoz Canada Pharmaceuticals since 2018. Previously funded by Hoffmann‐LaRoche and Janssen Biotech from 2011 to 2016, and Sanofi Genzyme from 2016 to 2017.
No potential conflicts of interest relevant to this article were reported.
References
- 1. Van der Helm‐van Mil AH, le Cessie S, van Dongen H, Breedveld FC, Toes RE, Huizinga TW. A prediction rule for disease outcome in patients with recent‐onset undifferentiated arthritis: how to guide individual treatment decisions. Arthiritis Rheum 2007;56:433–40. [DOI] [PubMed] [Google Scholar]
- 2. Alamanos Y, Voulgari PV, Drosos AA. Incidence and prevalence of rheumatoid arthritis, based on the 1987 American College of Rheumatology criteria: a systematic review. Semin Arthritis Rheum 2006;36:182–8. [DOI] [PubMed] [Google Scholar]
- 3. Aletaha D, Neogi T, Silman AJ, Funovits J, Bingham CO III, Birnbaum NS, et al. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum 2010;62:2569–81. [DOI] [PubMed] [Google Scholar]
- 4. Widdifield J, Paterson JM, Bernatsky S, Tu K, Tomlinson G, Kuriya B, et al. The epidemiology of rheumatoid arthritis in Ontario, Canada. Arthritis Rheumatol 2014;66:786–93. [DOI] [PubMed] [Google Scholar]
- 5. Shoenfeld Y, Isenberg DA. The mosaic of autoimmunity. Immunol Today 1989;10:123–6. [DOI] [PubMed] [Google Scholar]
- 6. Smolen J, Aletaha D, McInnes IB. Rheumatoid arthritis. Lancet 2016;388:2023–38. [DOI] [PubMed] [Google Scholar]
- 7. Shoenfeld Y, Levin‐Agmon N, Rose NR, editors. Infection and Autoimmunity. 1st ed Amsterdam: Elsevier Science; 2004. [Google Scholar]
- 8. Bach JF. The effect of infections on susceptibility to autoimmune and allergic diseases. N Engl J Med 2002;347:911–20. [DOI] [PubMed] [Google Scholar]
- 9. Kudaeva FM, Speechley MR, Pope JE. A systematic review of viral exposures as a risk for rheumatoid arthritis. Semin Arthritis Rheum 2019;48:587–96. [DOI] [PubMed] [Google Scholar]
- 10. Fiore AE, Shay DK, Broder K, Iskander JK, Uyeki TM, Mootrey G, et al. Prevention and control of influenza: recommendations of the Advisory Committee on Immunization Practices (ACIP), 2008. MMWR Recomm Rep 2008;57:1–60. [PubMed] [Google Scholar]
- 11. Horner FA. Neurologic disorders after Asian influenza. N Engl J Med 1958;258:983–5. [DOI] [PubMed] [Google Scholar]
- 12. Tam CC, O'Brien SJ, Petersen I, Islam A, Hayward A, Rodrigues LC. Guillain‐Barré syndrome and preceding infection with Campylobacter, influenza and Epstein‐Barr virus in the General Practice Research Database. PLoS One 2007;2:e344. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Toplak N, Avčin T. Influenza and autoimmunity. Ann NY Acad Sci 2009;1173:619–26. [DOI] [PubMed] [Google Scholar]
- 14. Peltola V, Ziegler T, Ruuskanen O. Influenza A and B virus infections in children. Clin Infect Dis 2003;36:299–305. [DOI] [PubMed] [Google Scholar]
- 15. Svec KH, Dingle JH. The occurence of rheumatoid factor in association with antibody response to influenza A2(Asian) virus. Arthritis Rheum 1965;8:524–9. [DOI] [PubMed] [Google Scholar]
- 16. Giamarellos‐Bourboulis EJ, Raftogiannis M, Antonopoulou A, Baziaka F, Koutoukas P, Savva A, et al. Effect of the novel influenza A (H1N1) virus in the human immune system. PLoS One 2009;4:e8393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Bos WH, Wolbink GJ, Boers M, Tihuis GJ, de Vries N, van der Horst‐Bruinsma IE, et al. Arthritis development in patients with arthralgia is strongly associated with anti‐citrullinated protein antibody status: a prospective cohort study. Ann Rheum Dis 2010;69:490–4. [DOI] [PubMed] [Google Scholar]
- 18. Nielsen SF, Bojesen SE, Schnohr P, Nordestgaard BG. Elevated rheumatoid factor and long term risk of rheumatoid arthritis: a prospective cohort study. BMJ 2012;345:e5244. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Rantapää‐Dahlqvist S, de Jong BA, Berglin E, Hallmans G, Wadell G, Stenlund H, et al. Antibodies against cyclic citrullinated peptide and IgA rheumatoid factor predict the development of rheumatoid arthritis. Arthritis Rheum 2003;48:2741–9. [DOI] [PubMed] [Google Scholar]
- 20. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007;370:1453–7. [DOI] [PubMed] [Google Scholar]
- 21. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Asscociation 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24. [DOI] [PubMed] [Google Scholar]
- 22. Bykerk VP, Jamal S, Boire G, Hitchon CA, Haraoui B, Pope JE, et al. The Canadian Early Arthritis Cohort (CATCH): patients with new‐onset synovitis meeting the 2010 ACR/EULAR classification criteria but not the 1987 ACR classification criteria present with less severe disease activity. J Rheumatol 2012;39:2071–80. [DOI] [PubMed] [Google Scholar]
- 23. Bombardier C, Hazlewood GS, Akhavan P, Schieir O, Dooley A, Haraoui B, et al. Canadian Rheumatology Association recommendations for the pharmacological management of rheumatoid rrthritis with traditional and biologic disease‐modifying antirheumatic drugs: part II safety. J Rheumatol 2012;39:1583–1602. [DOI] [PubMed] [Google Scholar]
- 24. Bykerk VP, Akhavan P, Hazlewood GS, Schieir O, Dooley A, Haraoui, et al. Canadian Rheumatology Association recommendations for pharmacological management of rheumatoid arthritis with traditional and biologic disease‐modifying antirheumatic drugs. J Rheumatol 2012;39:1559–82. [DOI] [PubMed] [Google Scholar]
- 25. Silman A, Harrison B, Barrett E, Symmons D. The existence of geographical clusters of cases of inflammatory polyarthritis in a primary care based register. Ann Rheum Dis 2000;59:152–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Becketti S. Introduction to Time Series Using Stata. College Station (TX): Stara Press; 2013. [Google Scholar]
- 27. Bhaskaran K, Gasparrini A, Hajat S, Smeeth L, Armstrong B. Time series regression studies in environmental epidemiology. Int J Epidemiol 2013;42:1187–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. SAS/STAT software: release 9.4. Cary (NC): SAS Institute; 1994. [Google Scholar]
- 29. Stata statistical software: release 13. College Station (TX): StataCorp; 2013. [Google Scholar]
- 30. Bartko JJ. The intraclass correlation coefficient as a measure of reliability. Psychol Rep 1966;19:3–11. [DOI] [PubMed] [Google Scholar]
- 31. Portney LG WM. Foundations of clinical research: applications to practice/Leslie G. Portney, Mary P. Watkins. 2nd ed. Upper Saddle River (NJ): Prentice Hall; 2000. [Google Scholar]
- 32. Jacoby RK, Jayson MI, Cosh JA. Onset, early stages, and prognosis of rheumatoid arthritis: a clinical study of 100 patients with 11‐year follow‐up. Br Med J 1973;2:96–100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Söderlin MK, Bergsten U, Svensson B, BARFOT Study Group . Patient‐reported events preceding the onset of rheumatoid arthritis: possible clues to aetiology. Musculoskeletal Care 2011;9:25–31. [DOI] [PubMed] [Google Scholar]
- 34. Silman A, Bankhead C, Rowlingson B, Brennan P, Symmons D, Gatrell A. Do new cases of rheumatoid arthritis cluster in time or in space? [research support]. Int J Epidemiol 1997;26:628–34. [DOI] [PubMed] [Google Scholar]
- 35. Humphreys JH, Verstappen SM, Hyrich KL, Chipping JR, Marshall T, Symmons DP. The incidence of rheumatoid arthritis in the UK: comparisons using the 2010 ACR/EULAR classification criteria and the 1987 ACR classification criteria. Results from the Norfolk Arthritis Register. Ann Rheum Dis 2013;72:1315–20. [DOI] [PMC free article] [PubMed] [Google Scholar]