Risk of Staphylococcus aureus bacteraemia in patients with rheumatoid arthritis and the effect of orthopaedic implants on the risk: a nationwide observational cohort study

Objective It remains disputed how much the risk of Staphylococcus aureus bacteraemia (SAB) is increased in patients with rheumatoid arthritis (RA), and the extent to which orthopaedic implants explain the risk. We assessed SAB incidence rates (IRs) and incidence rate ratios (IRRs), comparing RA patients with a general population cohort (GPC) and individuals with versus without orthopaedic implants. Method Danish residents aged ≥ 18 years without prior RA or SAB (=GPC) were followed up for RA and microbiologically verified SAB events (1996–2017). IRRs were calculated by age- and sex-stratified Poisson regression adjusted for age, comorbidities, calendar year, and socioeconomic status. Results The GPC comprised 5 398 690 individuals. We identified 33 567 incident RA patients (=RA cohort) (median follow-up 7.3 years, IQR 3.6–12.3). We observed 25 023 SAB events (n = 224 in the RA cohort). IRs per 100 000 person-years were 81.0 (RA cohort) and 29.9 (GPC). IRs increased with age. Adjusted IRRs in 18–59-year-old RA patients were 2.6 (95% confidence interval 1.8–3.7) for women and 1.8 (1.1–3.1) for men, compared with same sex and age group GPC. IRRs declined with age. Compared with the GPC without implants, IRRs for RA patients with implants ranged from 1.9 (1.3–2.8) (women ≥ 70 years) to 5.3 (2.2–12.8) (18–59-year-old men). Conclusion In this nationwide registry-based cohort study RA was a risk factor for SAB, and orthopaedic implants further increased the risk. Clinicians should be aware of potential SAB in patients with RA and orthopaedic implants.

SAB in people with orthopaedic implants or evaluated the impact on the risk in patients with RA (18).
With access to linkage on the individual level of prospective, nationwide, and virtually complete Danish registries, we therefore aimed to investigate and compare the incidence of SAB (i) in patients with RA and in the general population, and (ii) in individuals with and without orthopaedic implants. We hypothesized that orthopaedic implants were a separate risk factor for SAB both in RA patients and in the general population.

Data sources
In this nationwide cohort study, we interlinked national registries by use of the unique personal identification number (CPR number) assigned to all residents at birth or upon immigration (19). In the Danish Civil Registration System (CRS), we identified birth date, vital status, immigration, and emigration (19). In the Danish National Patient Registry (DNPR), we identified RA patients, orthopaedic implants, selected comorbidities and procedures, and hospitalizations in relation to events of SAB (Supplementary Tables S1-S3) (20). In addition, we identified RA patients in the Danish Rheumatologic quality registry (DANBIO) (21). Microbiologically verified SAB events were identified in the Danish National S. aureus Bacteraemia Database (22,23). We used the Register of Medicinal Products Statistics (RMPS) for information on filled prescriptions to identify patients with diabetes mellitus, and the Population's Education Registry (PER) for highest attained educational level as a proxy for socioeconomic status (Supplementary  Table S4) (24,25).
The CRS, DNPR, RMPS, and PER are virtually complete (19,20,24,25). In DANBIO, 91% of incident cases of RA are registered (26). The Danish National S. aureus Bacteraemia Database (linked to CPR numbers since 1992) holds information on, for example, sample dates and antibiotic resistance patterns of 94-97% of SAB cases in the country (27). All registers have previously been described in detail (28).

Cohort identification and follow-up
We identified two main cohorts: patients with incident RA and a general population cohort. Individuals eligible for inclusion were ≥ 18 years old, alive and living in Denmark on 31 December 1996, or were included consecutively on their 18th birthday from this date until 31 December 2017 ( Figure 1). We excluded individuals who (i) had immigrated within 2 years before inclusion, (ii) had a history of RA (for details, see Exposure section, below), or (iii) had SAB before or on the date of inclusion.
Follow-up ended on the date of first time SAB, emigration, or death, or on 31 December 2017, whichever came first. All individuals contributed time at risk in the general population cohort. Patients with incident RA contributed time at risk in the general population cohort until their inclusion in the RA cohort ( Figure 1).

Exposure
The main exposure of interest was incident RA identified either in DANBIO or in the DNPR, the latter to account for cases diagnosed before the implementation of DANBIO and to assure as high completeness of RA cases as possible.
In DANBIO, incident RA patients were defined as individuals with a first time diagnosis of RA by a rheumatologist (for more information on the codes and criteria used, see Supplementary Table S1). DAN-BIO holds a high proportion of true RA cases [positive predictive value (PPV) 96%] (26). In DNPR, incident RA was defined as at least two registrations of an International Classification of Diseases, 8th edition (ICD-8) or ICD-10 code of RA with ≤ 90 days' interval from departments of internal medicine or rheumatology (PPV ≈ 80%) (26). The date of the latter of these two diagnoses was assigned as the date of RA diagnosis. The 90 day interval was not applied when identifying and excluding prevalent RA.

Outcome
The outcome of interest was a first time microbiologically verified SAB registered in the Danish National S. aureus Bacteraemia Database. For sensitivity analyses, we defined SAB as hospital acquired if the positive blood culture was obtained > 2 days after hospitalization and otherwise as non-hospital acquired. Information on bacteraemia by methicillin-resistant Staphylococcus aureus (MRSA) was collected for descriptive purposes.

Orthopaedic implants and other covariates
Orthopaedic implants were identified in the DNPR as a hospital contact with a relevant surgical procedure code either before or during follow-up, and included prosthetic joints, osteosyntheses, and internal fixations (see Supplementary Tables S2-S4 for more information on covariates). Other covariates included sex, age (1 year time spans), calendar year (3 year time spans), socioeconomic status (highest attained educational level), and the following comorbidities: diabetes mellitus, chronic heart failure, chronic liver disease, chronic obstructive pulmonary disease, human immunodeficiency virus (HIV) infection, cancer (except for nonmelanoma skin cancer), solid organ transplantation, vascular devices/pacemakers, and chronic dialysis treatment. All comorbidities were identified as far back as registrations allowed, except for cancer, for which the look-back period was 5 years before inclusion. A recent surgical procedure (30 day look-back period) was defined as a hospital contact with a registered surgical procedure code.

Statistical analysis
We performed data management and analysis with SAS software, version 9.4 (SAS Institute, Cary, NC, USA). All covariates (except for sex) were allowed to vary with time, and follow-up time was split according to exposure and covariate status by a lexis macro (29). Individuals were considered exposed to a condition or an implant from the date of diagnosis or procedure. Removal of implants was not considered.
We explored the effect of RA on the risk of SAB by calculating IRs in the RA and general population cohorts overall, and in strata of sex (men/women) and age (18-59/60-69/≥ 70 years). Adjusted IRRs were assessed by stratified (sex/age) multivariate Poisson regression adjusted for age (1 year time spans) and calendar year, and further adjusted for socioeconomic status and selected comorbidities (excluding conditions not occurring in RA patients with SAB).
Similarly, the impact of orthopaedic implants on SAB risk was assessed by calculating (i) IRs in individuals with orthopaedic implants overall (RA patients and the general population) and subgrouped in strata according to sex, age, and orthopaedic implant status (with versus without implants, 12 strata per cohort); (ii) IRRs comparing RA and general population individuals with/ without orthopaedic implants (general population individuals in the same sex/age stratum without implants as reference); and (iii) IRRs comparing RA patients with and without orthopaedic implants within same sex/age stratum. All IRRs were assessed by multivariate Poisson regression.
To account for possible overrepresentation of hospital-acquired SAB in the RA cohort (hospitalizations expected to be more frequent among RA patients), we performed a sensitivity analysis restricting the outcome to non-hospital-acquired events (30). Furthermore, we performed model control by testing for relevant interactions and by adding calendar year in 1 year time spans. DANBIO RA patients are well characterized, and we performed a sensitivity analysis restricting the RA cohort to incident patients from DANBIO to compare these IRRs (age and sex stratified) with our main results from the full cohort.

Ethics and data security
Ethical approval is not required by Danish legislation for registry-based research. Data were handled in accordance with the General Data Protection Regulation (EU) 2016/679, and data processing agreements were approved by the Capital Region of Denmark (RH-2015-209). Statistics Denmark hosted and linked the data. Personal information was encrypted.

Incidence of SAB in patients with RA and in the general population cohort
A total of 224 individuals experienced a first time SAB after their RA diagnosis: women comprised 59% (n = 133), the median age was 72 years (IQR 62-79 years), median RA disease duration was 15.3 years (IQR 10.9-18.7 years, range 0.6-20.9 years), and 40% (n = 89) had an orthopaedic implant.
The overall IR of SAB was 81.0/100 000 person-years in the RA cohort, versus 29.9/100 000 person-years (24 799 events) in the general population cohort, corresponding to a crude IRR of 2.7. When we stratified by sex and age, the IRs appeared to be higher in men than in women in both cohorts and higher in RA patients compared with the general population cohort ( Table 2). IRs increased with age in both cohorts, although the agerelated increase appeared more pronounced in the general population. In the adjusted analyses, the incidence remained higher among RA patients compared with the general population (IRRs > 1.0), except for men ≥ 70 years of age. The highest IRRs were observed in women younger than 70 years (Table 2 and Supplementary Figure  S1). In a post-hoc analysis, we estimated IRRs associated with RA disease duration (≤ 2/3-9/≥ 10 years) stratified by sex and age and found the IRRs in each subgroup compared with the general population to be similar to the IRRs obtained when not taking disease duration into account ( Table 2 and Supplementary Table S5).
Overall, 1.2% of SAB events were MRSA; for the RA cohort this figure was 1.3% (n = 3).

Incidence of SAB among individuals with orthopaedic implants
We identified 8841 RA patients and 704 992 individuals from the general population cohort who had orthopaedic implants either before or during follow-up. Of these, 89 and 5376, respectively, acquired SAB, corresponding to IRs of 178.8 and 99.2 per 100 000 person-years. Recent surgery preceded bacteraemia in 27% (n = 24) of RA patients and in 36% (n = 1939) of individuals from the general population.
Orthopaedic implants appeared to be associated with higher IRs of SAB in both women and men in both cohorts, and IRs increased with age ( Figure 2). For women with RA (with general population women without implants in the same age stratum as the reference), adjusted IRRs ranged from 1.9 (≥ 70 years) to 5.2 (60-69 years). Similarly, for men with RA, the adjusted IRRs ranged from 2.1 (≥ 70 years) to 5.3 (18-59 years). In the general population cohort, IRRs for women with implants ranged from 1.7 (≥ 70 years) to 2.9 (18-59 years), and for men from 1.5 (> 70 years) to 2.3 (18-59 years). The IRRs of  RA patients without implants were increased compared with general population individuals without implants (except for men ≥ 70 years).
Within the RA cohort, orthopaedic implants were associated with increased adjusted IRRs compared with RA patients without implants, although the confidence intervals (CIs) were wide (Figure 3).

Sensitivity analysis and model control
Overall, 25% of SAB cases (n = 55) in the RA cohort were hospital acquired. Restricting the outcome to nonhospital-acquired SAB, comparing RA patients with the general population, we observed similar stratified IRRs (Supplementary Figure S2).
Statistical interactions were observed between RA and age, and RA and sex, which had already been addressed in the stratified analyses. Adding calendar year in 1 year intervals to the regression models changed the estimates only marginally (Supplementary Table S6).
When we restricted the RA cohort to include only individuals identified in DANBIO as incident RA patients, we saw near-identical IRRs in the age-and sex-stratified analysis of risk of SAB in patients with RA compared with the general population (Supplementary Table S7).

Discussion
In this nationwide study of approximately 25 000 events of SAB, we found that the risk was increased, with rates up to five-fold higher among RA patients with orthopaedic implants and doubled in RA patients without implants. Orthopaedic implants were also associated with an increased incidence in the general population (approximately doubled rates). This evidence of increased risk in patients with RA, after adjusting for possible confounders, as well as the additional risk associated with orthopaedic implants both in patients with RA and in the population in general, is new. In RA, the highest absolute risk of SAB was observed in men ≥ 70 years old; however, the highest relative risk was observed among women and younger individuals. This indicated that several other factors besides RA were likely to affect the risk among elderly individuals, both with and without RA. The risk increase in RA was not explained by more hospitalizations, since restriction to non-hospital-acquired cases yielded similar results (30). In addition, the risk associated with orthopaedic implants was not limited to the postoperative period, since two out of three cases of SAB were not preceded by a recent surgical procedure (17).
The magnitude of the unadjusted risk increase in SAB in patients with RA remains disputed (13)(14)(15). The highest estimate was from a Swedish study, but this included only six events (invasive S. aureus infections, including bacteraemia) (15). Two Canadian studies reported relative risks of 2.5 (95% CI 1.7-3.6) and 2.6 (95% CI 1.8-3.7), indicating an occurrence comparable to the IRRs of 2.7 in our study (13,31). One study showed that more than half of individuals with SAB had an implanted device, but lacked a reference background prevalence of implanted devices to compare with, and no studies have investigated the long-term incidence of SAB after insertion of orthopaedic implants (18).
Our study has several strengths, especially the nationwide design and the ability to link data on an individual level from virtually complete registries, which allowed us to investigate a relatively rare outcome in a large cohort of RA patients. With nationwide registration of both vital status and residence, loss to follow-up was negligible. Restricting the RA cohort to incident cases had the advantage of avoiding selection bias. Thus, including all living RA patients at a certain date would have resulted in an underrepresentation of frailer RA patients, since only individuals who survived until that date would be included. In addition, the validity of the RA diagnosis in this study was high because validated algorithms for identifying incident cases of RA were used (26). However, this approach may potentially have resulted in misclassification of RA patients with fewer than two registrations of RA in the DNPR or for whom the registrations were > 90 days apart (i.e. being included in the general population cohort). Such bias would have reduced the IRRs in this study. Yet, we anticipated little or no error because of the expected small size of this group compared with the size of the general population cohort, and this was supported by our observation of almost identical estimates in the sensitivity analysis restricting the RA cohort to the DANBIO RA patients only. Although, in our study, SAB occurred at all timepoints after RA diagnosis, a potential limitation of our design was that fewer patients with long-standing RA contributed follow-up time, which may have limited the number of events and underestimated the true incidence; however, we did not see any clear evidence of longer disease duration resulting in higher IRRs in our post-hoc analysis of the impact of disease duration on the age-and sex-stratified IRRs. The study also had other limitations. We included individuals with orthopaedic implants inserted prior to the start of follow-up, who were thus past the postoperative period where the incidence of infections is high, potentially leading us to underestimate of the true incidence of SAB in individuals with orthopaedic implants (32). Our study lacked information on, for example, S. aureus nasal carriage, alcohol consumption, and illicit injection drug use; however, we consider these to be minor limitations regarding the risk of bacteraemia among RA patients (14,33,34). Thus, for RA patients compared with controls, S. aureus nasal carriage is similar or only slightly elevated, and alcohol consumption is lower (35)(36)(37)(38)(39). Furthermore, in our study, we observed few HIVpositive patients at RA diagnosis, indicating that illicit injection drug use was rare (40). Individuals who contracted SAB before 1992 (i.e. before data on SAB were linked to CPR numbers) may have been erroneously included in the study. We consider this to be a minor issue owing to the high mortality of SAB and the advanced age of most infected individuals (41). Similarly, people who had immigrated may have had a history of SAB or RA or have risk factors that were not available in the Danish registries. To reduce this source of error, we excluded individuals with less than 2 years of residence in Denmark. As expected, bacteraemia caused by MRSA was infrequent (23). Studies have found conflicting evidence on whether RA is associated with a higher or lower risk of MRSA bacteraemia compared with methicillinsusceptible Staphylococcus aureus (MSSA), and our findings may not be generalizable to populations where MRSA is more frequent (13,14). Finally, since this is an observational study, residual confounding cannot be excluded.
For patients with RA, both the disease itself (disease activity and disability) and the anti-rheumatic treatments increase the infection risk (2,42,43). RA disease severity and treatment could potentially mediate some of the excess risk of SAB associated with having both RA and orthopaedic implants; however, it was beyond the scope of this study to explore the impact of RA-specific factors such as disease activity and immunosuppressive treatments on the risk. Information on injectable glucocorticoids, other antirheumatic treatments, and disease activity was not available in this study. The effect of oral glucocorticoids without corresponding information about disease activity would have been heavily subjected to confounding by indication and thus the true effect would still be unclear. In addition to the risk from orthopaedic implants, future studies should investigate whether subgroups of RA patients are more susceptible to this serious infection.

Conclusion
This nationwide registry-based cohort study explored the impact of RA and of orthopaedic implants on the risk of SAB in incident patients with RA and in the general population. Overall, RA patients had a doubled risk of SAB compared with the general population, and orthopaedic implants further increased the risk. Thus, RA and orthopaedic implants both increased the risk, with the highest risk in RA patients with implants. Physicians should be aware of potential SAB in this patient group. clinical trial from Eli-Lilly; CTP received research grants from Bayer and Novo Nordisk Foundation; MLH received research grants from AbbVie, Biogen, BMS, Celltrion, Eli-Lilly, Janssen Biologics B.V, Lundbeck Foundation, MSD, Pfizer, Roche, Samsung Biopies, Sandoz and Novartis and chairs the steering committee of the Danish Rheumatology Quality Registry (DANBIO), which receives public funding from the hospital owners and funding from pharmaceutical companies and co-chairs EuroSpA, which generates real-world evidence of treatment of psoriatic arthritis and axial spondyloarthritis based on secondary data and is partly funded by Novartis.