FormalPara Key Summary Points

A decrease in hospitalization rates for tuberculosis and pneumocystosis occurred over three decades for Australian patients with rheumatoid arthritis. However, TBC rates remain above population level.

In contrast, over that time period, rates for non-tuberculosis mycobacterial, mycotic infections, and Cytomegalovirus (CMV) infections requiring hospitalization increased, while Zoster infection rates remained stable for patients with rheumatoid arthritis.

For patients with psoriatic arthritis and ankylosing spondylitis, overall TBC incidence rates remained low, but both Candida and non-candida opportunistic infections rates increased over time.

Candidiasis was the most prevalent opportunistic infection followed by Zoster infections across all three disease categories.

Hospitalization for any opportunistic infections was associated with a 6.5% in-hospital mortality rate.

Introduction

The management of patients with immune-mediated inflammatory joint diseases (IJD) has changed dramatically over the last 30 years during which methotrexate became the established anchor drug for peripheral arthritis followed by an ever-increasing arsenal of cytokine and kinase-inhibiting drugs [1,2,3]. Methotrexate use in rheumatoid arthritis (RA) patients has doubled in the last three decades, while subsidized biological therapy with etanercept became available for RA patients in Australia in 2003, followed by subsequent approval for use in other IJD and for other TNF inhibitors (2004, 2010, 2014), rituximab (2007), abatacept (2008) and tocilizumab (2010) [4]. This more intense treatment approach provides valuable symptomatic relief and protection against joint complications, but also predisposes the patient to side effects, including infections [5]. Multiple studies have analyzed rates of severe bacterial infections in IJD [6,7,8,9], but there are only limited data on opportunistic infections (OI), which mainly stem from clinical trials. Exposure time during clinical trials is usually limited, whereas IJD patients typically require longer-term immune-modulating therapy that places patients at risk of OI over a protracted period of time. Registry-based studies from Europe and the USA reported OI rates between 152 and 3000 per 100,000 for various IJD [10, 11] with the wide range due to heterogeneity in patient selection, differences in defining OI (e.g., in- or exclusion of herpes zoster) and regional influences as tofacitinib use in Asia led to double the H. zoster rate seen in Europe [12]. Population-based studies, which are less prone to selection bias and allow long-term follow-up, can improve our understanding of the scarcity of data on OI for specific IJD [8, 13,14,15]. Based on the above, we performed a study to investigate the rate of overall and IJD-specific hospital admission rates for OI among patients with rheumatoid arthritis (RA), psoriatic arthritis (PsA) and ankylosing spondylitis (AS) across Western Australia (WA) over the period 1985–2014 when early more intensive immunomodulation for IJD gradually became embedded in clinical practice.

Methods

This population-based retrospective observational study used prospectively collected, state-wide longitudinal administrative health data for IJD patients as recorded in the Western Australian Rheumatic Disease Epidemiological Registry (WARDER). WARDER data were extracted and linked through the Western Australian Data Linkage System (WADLS), which covers all public and private hospitals in Western Australia (WA) (pop 2.5 million) and applies probabilistic matching to provide individual longitudinal health data across the Hospital Morbidity Data Collection (HMDC), WA Cancer Registry, WA Mortality Registry or the Emergency Department Data Collection (EDDC) for registered IJD patients.

We selected patients with prevalent IJD defined as having at least two International Classification of Diseases (ICD-9-CM or ICD-10-AM) diagnostic codes separated by at least 30 days for one of the three defined IJD (rheumatoid arthritis (RA), psoriatic arthritis (PsA), ankylosing spondylitis (AS) (suppl Table 1) in the HMDC between January 1, 1985 and December 31, 2014. This algorithm has demonstrated accuracy in identifying patients with various inflammatory rheumatic diseases [16,17,18].

Opportunistic infections (OI) were defined as a microbiologically confirmed infection with the pathogen of interest (suppl Table 1) registered in HDMC. Included infections were tuberculosis (TBC) and nontuberculous mycobacterial infections (NTMB), fungal infections (candidiasis, aspergillosis, cryptococcosis, histoplasmosis), pneumocystosis, and viral infections (varicella-zoster virus (VZV), Cytomegalovirus (CMV). This selection of OI follows the EULAR recommendations for reporting OI during surveillance studies and is consistent with data presented in an earlier meta-analysis [11, 18].

Study entry was at the first-ever hospital contact for each patient, while index admission was the first IJD-related admission. Primary outcomes were OI incidence rates per 1000 person-years (IR) for all OI across IJD categories, temporal trends over 10-year periods for specific OI within each IJD category and in-hospital (30 days) as well as all-cause mortality per 1000 person-years (MR) based on date of death in WA Death Registry during subsequent follow up. Comorbidity was calculated using a modified Charlson comorbidity index (mCCI), that excluded the scores for rheumatic disease (weighted score 1) to minimize confounding with mCCI at index hospitalization based on accumulated health data from prior hospital separations. The Human Research Ethics Committee at the WA Department of Health considered this project to be of low risk and waived the need for individual consent given the use of deidentified data and approved this project (WADOH HREC # 2016.24).

Statistical analyses descriptive statistics include median and interquartile range (IQR) for continuous variables compared by non-parametric methods (Kruskal–Wallis), categorical data described with a frequency and proportion and group comparisons tested with odds ratios (OR) and Fisher’s exact test. OI incidence rates were calculated per 1000 person-years with 95% CI derived from Poisson distribution and compared across disease categories by incidence rate ratios using conditional maximum likelihood estimates [19]. Changes in OI rates over time were assessed by (a) linear least squares regression analysis using the coefficient of determination (R-squared, R2) as the goodness-of-fit measure, where higher coefficients indicate a better fit for increasing or decreasing incidence rates and (b) analysis of variance (ANOVA) to compare mean IR values across three decades. Mortality rates (MR) per 1000 person-years were compared across categories by mortality rate ratios with 95% confidence intervals. All analyses were performed using SPSS software v23.0 (IBM, USA) and OpenEpi software with two-sided p values (p) < 0.05 considered to be statistically significant.

Results

Descriptive characteristics of patients with RA (n = 7330), PsA (n = 529), and AS (n = 1126) (Table 1) showed age and gender distribution within the expected range for each IJD category with a relatively underrepresentation of patients identifying as Indigenous (population prevalence at 3.5%) and a higher frequency of baseline comorbidity observed in PsA patients. There was no clear trend in any of six comorbid conditions including diabetes mellitus in each IJD category over time (Suppl. Figure 1).

Table 1 Patient characteristics at index hospitalization across the three categories of inflammatory joint disease. Figures indicate frequency (%) or median (interquartile range) unless otherwise stated
Full size table

During almost 20 years of follow-up across all IJD categories, OI occurred more often in RA (17.9%) than PsA (12.6%) or AS patients (13.3%) (p = 0.002) with a significantly higher overall OI incidence rate in RA patients than in PsA and AS patients (both p < 0.01) while the rate difference between PsA and AS was borderline significant (p = 0.05) (Table 2). Over the three study decades, the rate of overall OI increased in all three disease categories (Fig. 1A).

Table 2 Overall frequency, incidence rate per 1000 person-years (95% confidence intervals), and outcome data for patients hospitalized with opportunistic infection (OI) across three categories of inflammatory joint diseases
Full size table
Fig. 1
figure 1

Incidence rate per 1000 person-years for all opportunistic infections (OI) (A), Candida (B) and non-Candida infections (C) by joint disease category over time. D The increasing usage of conventional (cDMARD) and biological (bDMARD) over two decades in RA patients in Western Australia expressed as defined daily dosage per 1000 individuals (adapted from Ref. 4). RA rheumatoid arthritis, PsA psoriatic arthritis, AS ankylosing spondylitis. Stippled lines give trend and coefficient of determination (R2) by least squares regression analysis. p values derived from ANOVA

Full size image

Analysis of specific OI demonstrated that candidiasis was the most frequent OI by number and with significantly increasing rates over time across all three IJD categories (Table 3, Fig. 1B). Candidiasis was more frequent in female patients with RA (14.9 vs. 12.1%, p < 0.01), PsA (13.5 vs. 6.3%, p < 0.01) and AS (14.5 vs. 7.1%, p < 0.01) but age at Candida diagnosis did not differ by gender in each IJD (all p > 0.1). Rates for non-Candida OI also rose over time across all categories (Fig. 1C). TBC hospitalization was rare in PsA, while in RA and AS hospitalization for TBC declined but rates for NTBM increased (Suppl Fig. 1). The TBC incidence rate in the latest decade was still higher than the population background rate of TBC in WA [20]. Overall, non-Candida mycotic infection rates increased in all categories especially since 2005 (Suppl Fig. 2). Pneumocystosis occurred in RA patients only pre-2005, while Cryptococcus incidence rates increased significantly over time in RA patients (IR from 0.02 to 0.63, R2 0.75, p = 0.13); aspergillosis incidence also increased in RA (IR from 0.22 to 0.50, R2 0.99, p = 0.004) and nearly doubled in AS (IR from 1.05 to 1.74, 0.72.p = 0.15). Rates for other mycotic infections than above mentioned also rose across all three categories over time (Suppl Fig. 3). VZV infection rates rose over the last decade in AS patients with no significant change seen in RA or PsA patients (Suppl Fig. 4) and with CMV testing not available before 1995, a rate increase was seen in RA and AS patients after 2005.

Table 3 Incidence rates per 1000 person-years for specific opportunistic infections (OI) across three decades in each disease category
Full size table

Patients with OI spent an average between 7 and 11 days in the hospital with longer stays observed in RA patients (Table 2). The need for ICU admission (4.9% overall) and the frequency of in-hospital death (6.4%) did not differ significantly between groups, while long-term mortality rates ratios were increased for patients with OI compared to patients in the same disease category not experiencing OI.

Discussion

This observational study included three decades during which management options and guidelines for IJD patients have changed significantly following the introduction of MTX as anchor drug in the 1990s and targeted biologic drug therapy available in Australia since 2005 [4, 21]. The results of this study indicate that despite significant reductions in incidence rates for TBC and pneumocystosis in RA patients, the hospitalization rates for all other OI have increased in RA, PsA, and AS patients with the steepest increases seen since 2005.

RA has been the most widely studied IJD in terms of serious and opportunistic infections since the start of the biological therapy era [22]. The overall non-Candida OI rate in RA patients in this study fits well with the 5.6/1000 incidence of serious OI requiring admission reported from Spain [23]. In a meta-analysis of clinical trials, the recalculated overall OI rate was 3.3/1000 person-years (96 OI among 32,504 patients over 24 weeks) in the period 1998–2012, with a higher rate observed in patients on biological drugs compared to placebo or sDMARD [24]. A recently reported study from Korea, which used less stringent OI definitions and included outpatient visits, reported an OI rate of 37.6/1000 in RA patients with minimal change over time [25], while national insurance in- and outpatient data from Taiwan found OI incidence to be 25/1000 person-years in RA patients [26]. Although restricted to inpatient data, our data support this lack of improvement in overall OI rates in RA patients despite the fact that admissions for TBC infection rates decreased over time. This is likely due to prophylactic measures taken in RA patients initiating biologic therapy but other explanations (e.g., increased outpatient treatment for TBC) cannot be ruled out. TBC infections requiring admission in RA patients however still occurred at a rate of 0.38/1000 person-years in the last study decade, which remains significantly higher than the stably low TBC incidence rate of 5–6 cases per 100,000 Australian population [20]. Furthermore, other mycobacterial infection rates increased in that time frame. This increased susceptibility to mycobacterial infections in RA patients has been found across multiple studies and seems to be due to a poorly defined combination of patient and disease characteristics possibly anti-TNFi therapy [22, 27,28,29,30]. While the rate of pneumocystosis admissions pleasingly decreased to zero in the last decade, most likely as a result of increased prophylactic measures [31], the IR of both Candida infections and non-Candida fungal infections increased over time. While our data predated the introduction of anti-IL17 therapy, candidiasis was also the most frequent fungal OI observed in the Taiwanese national insurance study [26]. Candida infections were twice as frequent in female IJD patients and while we were unable analyze this predisposition further, it would suggest increased surveillance for and potential use of Candida prophylaxis in female IJD patients. The IR for cryptococcal infections in this study was similar to the 0.22 IR seen in a single center retrospective study from Taiwan where higher risk was observed for patients on biologics [32]. The most frequent viral OI observed in RA patients was VZV (IR 2.8/1000 person-years) in line with IR in other RA studies between 1.6 and 10.9 /100 person-years [27]. Despite the well-known association between VZV and immunosuppressive therapy and the availability of prophylaxis, VZV rates unfortunately did not decline over time. This suggests that clinicians faced significant difficulties in defining important characteristics in estimating and preventing the risk of zoster infections, already before the VZV increase seen following the introduction of Jak inhibitor therapy in 2015 [33, 34].

With limited data on OI rates in PsA and AS patients outside short-term clinical trial results [22, 35], we also analyzed serious OI for these patient groups. Candidiasis was again the most frequent OI observed in both categories and its IR rose significantly over time, especially in PsA patients. With few data on fungal infections available for PsA and AS [36], our data (although from a selected group of PsA and AS patients requiring admission) indicate rising IR for a number of fungal infections in these patients (Table 3) and illustrate that immunomodulating therapy over longer periods of time than observed in trials in these patients groups does increase the risk of OI [37]. No PCP infections were observed in PsA and AS patients, while mycobacterial infection rates were significantly lower than in RA. A recent pooled analysis of PsA and AS patients treated with secukinumab reported a latent TBC IR over 2 years of 0.02 and 0.08/1000 person-years, respectively, with no cases of active TBC observed [38]. There are few real-life data on zoster in PsA [36], but a recent metanalysis confirmed a lower risk of zoster in PsA than RA and found only a slightly elevated risk of zoster in PsA patients requiring multiple agents for severe disease [39]. A recent systemic review of nine clinical trials demonstrated that VZV risk was higher across multiple biological therapies than in non-biological therapy (odds ratios 1.48 CI: 1.18–1.86 [40]. The rate of VZV in a combined psoriasis/PsA/AS cohort treated with TNF inhibitors was 4.4/1000 person-years [41], like the combined rate (4.8/1000 person-years) reported here in the same study period. While our results support an overall lower risk of OI in PsA/AS patients, they also raise the question of whether safety data from RA studies can be meaningfully extrapolated to PsA and AS patients [42]. Finally, while our data confirm that OI are relatively infrequent complications in the lifelong disease facing IJD patients, they require considerable health care resource (25.696 days spent in the hospital at an average (2020 price) cost of AU$ 2332/day) and more importantly associate with a high case fatality rate.

Limitations for this study include the strict criteria for IJD inclusion, which provided a high specificity [17] but may have reduced the sensitivity. Comparative data from the general population could have clarified whether the observed changes were unique to IJD patients. Also, the use of hospital admission data for OI detection may have led to a selection bias towards more severe OI as OI patients treated on an outpatient basis were not included. The low proportion of Indigenous patients with IJD has been reported before and the lack of shared epitope in the restricted MHC polymorphisms of Australian Aborigines considered as explanation for their apparent resistance to IJD [43]. Increasing awareness of diseases and more advanced laboratory technology may also have confounded our data. Although we found no clear changes in the frequency of comorbid conditions over time, they remain potential risk factors for OI in patients receiving immunomodulating therapy. Finally, whereas the use of ICD-based administrative health data reflects the physician-based diagnoses, more detailed data such as for inflammatory markers, details of specific antirheumatic drug therapy, or anti-infectious treatment were not available. The strengths of this study lie in the large number and population-wide representation of patients with a rare complication, a median follow-up of nearly 20 years to allow trend analysis of OI over time, the comparison between different IJD categories, and the inclusion of infectious disease codes with an unequivocal microbial test result (culture, PCR or other) to ensure a high specificity of the presented OI data.

Conclusions

Despite improvements in hospitalization rates for TBC and pneumocystosis, we observed a rising incidence rate over three decades for a wide range of other OI requiring hospitalization across all three categories of inflammatory joint disease. This suggests that the more aggressive treatment approach in IJD to achieve disease remission and reduce comorbidity comes at an increasing expense of OI complications. These results indicate the urgent need for wider implementation of preventative measures for OI in IJD.