Risk of severe infections after the introduction of biologic DMARDs in people with newly diagnosed rheumatoid arthritis: a population-based interrupted time-series analysis

Abstract

Objectives

To determine the impact of the introduction of biologic DMARDs (bDMARDs) on severe infections among people newly diagnosed with RA compared with non-RA individuals.

Methods

In this population-based retrospective cohort study using administrative data (from 1990–2015) for British Columbia, Canada, all incident RA patients diagnosed between 1995 and 2007 were identified. General population controls with no inflammatory arthritis were matched to RA patients based on age and gender, and were assigned the diagnosis date (i.e. index date) of the RA patients they were matched with. RA/controls were then divided into quarterly cohorts according to their index dates. The outcome of interest was all severe infections necessitating hospitalization or occurring during hospitalization after the index date. We calculated 8-year severe infection rates for each cohort and conducted interrupted time-series analyses to compare severe infection trends in RA/controls with index date during pre-bDMARDs (1995–2001) and post-bDMARDs (2003–2007) periods.

Results

A total of 60 226 and 588 499 incident RA/controls were identified. We identified 14 245 severe infections in RA, and 79 819 severe infections in controls. The 8-year severe infection rates decreased among RA/controls with increasing calendar year of index date in the pre-bDMARDs period, but increased over time only among RA, not controls, with index date in the post-bDMARDs period. The adjusted difference between the pre- and post-bDMARDs secular trends in 8-year severe infection rates was 1.85 (P = 0.001) in RA and 0.12 (P = 0.29) in non-RA.

Conclusion

RA onset after bDMARDs introduction was associated with an elevated severe infection risk in RA patients compared with matched non-RA individuals.

Rheumatology key messages.

• We observed a significant increase in severe infections trend among RA patients diagnosed after biologic DMARDs (bDMARDs) introduction.

• No change in trend was observed among age- and gender-matched general population controls.

• RA onset after bDMARDs introduction was associated with a significant increase in severe infection risk.

Introduction

Patients with RA are at a higher risk of severe infections than the general population, which contributes to an overall increased mortality [1]. Over the past several decades, there have been major advances in RA management including the introduction of biologic DMARDs (bDMARDs) and a shift in treatment approach targeting remission or low disease activity if remission is not obtained (i.e. treat to target) [2]. The paradigmatic change has resulted in significant improvements in the prognosis, clinical outcome and survival in RA [3–5].

Biologic agents, which prior to 2015 included TNF-α inhibitors, IL-1 receptor antagonist (anakinra), IL-6 inhibitors (e.g. tocilizumab), T cell co-stimulation modulator (abatacept) and anti-CD20 Ab (e.g. rituximab), are increasingly being used to treat patients with RA. Although very effective in suppressing inflammation and preventing joint damage, bDMARDs may be associated with increased risk of infections [6, 7]. Several systematic reviews and meta-analyses of randomized controlled trials (RCTs) have found an increased risk of severe infections requiring hospitalization among patients with RA treated with biologic agents [8–10], while other meta-analyses and individual RCTs did not have the same finding [11–14]. Evidence from observational studies is also contradictory, with some studies showing increased risk [15–22] whereas others report no significant increase [23–26]. Furthermore, there is a paucity of studies evaluating the long-term effect of bDMARDs on infections.

This study was undertaken to compare the 8-year risk of severe infections among newly diagnosed RA and non-RA individuals, before and after the introduction of biologics. We determined the secular trend in risk of severe infections in the population of incident RA patients in British Columbia (BC), Canada, over a 20-year period (1995–2015), and used a quasi-experimental study design to evaluate the long-term effect of the introduction of bDMARDs.

Methods

Study design

The interrupted time-series (ITS) design is a quasi-experimental method that is well-suited for assessing long-term effects of interventions (i.e. medication use) on disease outcomes [27, 28]. Using segmented regression, the ITS design allows us to determine how much an intervention changed an outcome by measuring the shift in levels and trends in the outcome before and after the intervention. In this study, we performed an ITS analysis on a population-based incident cohort of all RA patients in BC, Canada compared with non-RA individuals. We investigated the changes in 8-year severe infection rates in RA patients diagnosed in the pre- (1995–2001) and post-intervention (2003–2007) periods, compared with general population controls, with year of interruption set to 2002 to represent the introduction of bDMARDs.

 Ethics approval was obtained from the University of British Columbia’s behavioural Research Ethics Board (H15-00887). All procedures were compliant with BC’s Freedom of Information and Privacy Protection Act.

Data sources

We used administrative health data (from 1990 to 2015) on all provincially funded health care services provided under Canada’s universal healthcare system. These include data on demographic information from consolidation files based on the Medical Services Plan (MSP) registration and premium billing [29]; all physician visits, with one diagnostic code per visit representing the reason for the visit, from the MSP database [30]; all hospitalizations from Discharge Abstract Database, which include up to 25 diagnostic codes per hospitalization representing the reason for admission, comorbid diseases or complications during hospitalization [31]; information on death from Vital Statistics Data [32]; and all medications dispensed to BC residents regardless of funding source, from the Pharmanet database [33].

Study cohort

Incident RA cohorts

We identified all incident RA patients over the age of 20 years, diagnosed between January 1995 and December 2007 using MSP physician billing data available from January 1990 onwards. RA cases were identified using previously published criteria [5] yielding a positive predictive value of 0.82 [34]: (i) at least two physician visits that are at least 2 months apart, within a 5-year period, with International Classification Disease (ICD) 9th version codes (ICD-9: 714.X.) and/or the 10th version (ICD‐10-CA M05.X–M06.X) for RA; (ii) individuals were excluded if over a 5-year period after their second RA visit, they had at least two subsequent visits with ICD codes for another form of inflammatory arthritis (SLE and other CTD, PsA, AS and other spondyloarthropathies); and (iii) cases were also excluded if a patient was diagnosed by a non-rheumatologist, but when the individual saw a rheumatologist the diagnosis of RA was not confirmed.

The date of the second code in the pair of ICD codes required for disease diagnosis was established as the diagnosis date (i.e. index date). To ensure we included only incident cases and did not mistakenly include prevalent cases who moved to BC, we implemented a 5-year run-in period, whereby cases needed to have at least 5 years of data available, with no RA diagnoses, prior to the first of two RA visits defining the RA diagnosis. Fig. 1 shows the cohort assembly and the reasons for exclusion.

Figure 1.

Incident RA cohort assembly

Non-RA cohorts

Non-RA individuals were randomly selected from the general population registered with the BC MSP with no physician visits for any type of inflammatory arthritis. These individuals were matched to incident RA patients (up to 10:1) based on birth year and gender, and were assigned the first RA visit date and the second RA visit date (i.e. index date) of the RA patients they were matched with. Non-RA individuals identified in earlier years could develop RA in later years during follow-up, and were censored at that time. Non-RA individuals were excluded if they had fewer than 5 years of available data prior to the assigned first RA visit date.

Outcome assessment

We defined severe infection as any infections requiring hospitalization or occurring during hospitalization [35], a case definition that yields a positive predicted value of 0.95 [36]. A total of 58 different types of infections selected a priori by a panel of experts were identified using ICD-9/10 diagnostic codes [35, 37], which are available in Supplementary Table S1, available at Rheumatology online. To account for the potential effect of a significant decrease in numbers of total joint arthroplasty (TJA) performed among RA patients diagnosed in recent years [38], we excluded any severe infections likely associated with TJA surgeries (i.e. any severe infections that occurred within 60 days before/after surgery).

Covariates assessment

Information on sociodemographic characteristics, comorbidities, and medication use known to be associated with RA and infection risk that were available in our administrative data were identified and included as covariates in the ITS models. Cohort characteristics included age, gender and neighbourhood income decile (a measure of neighbourhood socioeconomic status) at index date. Comorbidities and medication use included chronic obstructive pulmonary disease, Romano Charlson Comorbidity Index, diabetes, chronic kidney disease, alcoholism, cancer, prior hospitalization with infection, glucocorticoid use, all of which were measured at baseline (within 12 months before index date).

Follow-up

Follow-up for each patient began on the index date. Follow-up terminated on the date of death, out-of-province migration, an individual in non-RA cohort being diagnosed with RA or 8 years of follow-up, whichever happened first.

Intervention

In Canada, the first bDMARDs were approved in the early 2000s. By the year 2002, Health Canada approved three types of anti-TNF-α including etanercept, infliximab and adalimumab for the treatment of RA and related inflammatory rheumatic conditions [39]. Therefore, we set the intervention time to a 1-year lag period from 1 January 2002 to 31 December 2002 to represent bDMARDs introduction. The lag period allows the effects of treatment transition to bDMARDs to manifest themselves.

Statistical analyses

RA patients and general population controls with index date between 1995 and 2007 were divided into quarterly incidence cohorts (four per year) according to their index date. They were subsequently followed for 8 years to capture all severe infection outcomes. We calculated 8-year all severe infection rates for each quarterly RA/non-RA cohort as the ratio of total number of severe infections occurred during the 8-year period following index date (numerator) to the total person-years (PY) accumulated during the follow-up (denominator).

We compared how the severe infection rates changed with increasing calendar time of index date (i.e. secular trends) between cohorts with index date before the intervention (pre-intervention cohorts) and cohorts with index date after the intervention (post-intervention cohorts). The ITS analysis assumes a linear relationship between time and the outcome within each period (pre- and post-period). The following segmented regression model was used:

$${SI}_i={\beta }_0+{\beta }_1\times {(T}_i-c)+{\beta }_2\times {\mathit{bDMARDs}}_i+{\beta }_3\times {(T}_i-c)\times {\mathit{bDMARDs}}_i+{\beta }_4\times X_i+{\mathit{error}}_i$$

where ${SI}_i$ is severe infection rates calculated for the i^th quarterly cohort; $T_i$ represents time for the i^th quarterly cohort diagnosed between 1995 and 2007 (i.e. 1995, 1995.25, 1995.5, 1995.75, 1996, …); $c$ represents the time point at which the intervention is introduced, and here c = 2002; ${\mathit{bDMARDs}}_i$ is an indicator variable for the intervention (${\mathit{bDMARDs}}_i=0$indicates pre-intervention cohorts where $T_i<c$; ${\mathit{bDMARDs}}_i=1$ indicates post-intervention cohorts where $T_i>c$); ${\beta }_1$ represents the secular trend in severe infection rates for pre-intervention cohorts; ${\beta }_1+{\beta }_3$ represents the secular trend for post-intervention cohorts; and${\beta }_2$ represents the level change (i.e. change in intercept). We obtained the ordinary least-squares model estimates of these model parameters and Newey–West robust standard errors that can account for autocorrelation and heteroskedasticity in the error terms of the segmented time-series regression model [40]. In the regression, we first accounted for changes in cohort composition by adjusting for age and gender at baseline and accounted for the potential effect of previous infections by adjusting for the presence of any prior hospitalization with infection diagnostic codes listed in Supplementary Table S1, available at Rheumatology online. We further adjusted for other covariates measured at baseline as mentioned in the ‘Covariates assessment’ section (listed in Table 1) using a stepwise selection procedure to identify the most parsimonious models (P-entry <0.05; P-exit ≥0.15). The variable selection approach employs Newey–West robust standard errors, which are based on generalized method of moments (GMM), appropriate to use when the basic assumptions (i.e. non-normality, autocorrelation or heteroskedasticity in the error terms) of least squares regression are violated. To estimate changes in severe infection rates attributable to bDMARDs introduction, we reported both the crude and adjusted changes in trends (i.e. ${\beta }_3$) of severe infection rates between the pre- and post-bDMARDs periods. Finally, we computed the differences between RA and controls in the pre-bDMARDs trends, post-bDMARDs trends, and the differences in pre- vs post-bDMARDs trends as well as their 95% CIs and P-values. All statistical analyses were performed using SAS (version 9.4).

Table 1.

Baseline characteristics and outcomes of RA/non-RA cohorts for pre- and post-bDMARDs periods

		RA cohort				Non-RA cohort
		Total	1995–2001	2003–2007	P-valuea	Total	1995–2001	2003–2007	P-valuea
Demographics
	N	60 226b	32 459	23 369		588 499	319 168	224 331
	Age at diagnosis (years), mean (s.d.)	59.1 (15.9)	59.1 (16.2)	59 (15.6)	0.6834	58.9 (15.9)	59.0 (16.2)	58.8 (15.6)	<0.0001
	Female, %	66.8	66.8	66.9	0.0298	66.9	67.0	67.0	0.8012
	Neighbourhood income decile, mean (s.d.)	5.2 (2.9)	5.2 (2.9)	5.3 (2.9)	<0.0001	5.5 (2.9)	5.5 (2.9)	5.5 (2.9)	<0.0001
Comorbiditiesc
	CCI, mean (s.d.)	0.4 (1.0)	0.4 (1.0)	0.5 (1.1)	<0.0001	0.2 (0.7)	0.2 (0.7)	0.2 (0.7)	0.0221
	Diabetes, %	2.7	2.5	3.1	<0.0001	1.1	1.1	1.1	0.0022
	COPD, %	11.5	11.8	11.2	0.0276	4.2	4.5	3.9	<0.0001
	CKD, %	0.5	0.3	0.8	<0.0001	0.2	0.2	0.4	<0.0001
	Alcoholism, %	1.0	1.2	0.8	<0.0001	0.4	0.4	0.4	0.6207
	Cancer, %	9.5	10.1	8.6	<0.0001	5.7	5.7	5.7	0.3099
	Previous infection, %	2.5	2.9	2.0	<0.0001	1.1	1.3	0.9	<0.0001
Medication used
	Glucocorticoidsc, %	23.7	21.0	27.1	<0.0001	1.9	1.9	1.9	0.0224
	tDMARDse, %	8.2	4.9	12.5	<0.0001
Outcome
	Follow-up (PY)	384 710	205 586	150 848		3 982 833	2 162 252	1 529 632
	Total number of severe infections	14 245	8041	5178		79 819	45 491	28 600
	8-year severe infection rates (per 1000 PY), mean (s.d.)
	severe infections	37.0 (4.5)	39.1 (4.5)	34.3 (3.5)		20.0 (1.7)	21.0 (1.6)	18.7 (0.8)

^a P-values obtained from two-sample t-test (pre-period vs post-period).

^bA total of 60 227 incident RA cases were diagnosed between 1995 and 2007. Excluding one patient for whom we were unable to find a suitable match, we identified a total of 60 226 incident RA cases.

^cComorbidities and glucocorticoids use were assessed over 12 months prior to index date.

^dMedication use data accessed from PharmaNet is available from 1996. Thus, glucocorticoids and csDMARDs use in pre-period does not include 1995.

^ecsDMARDs use is defined as the proportion of cohort using at least two traditional DMARDs (MTX, SSZ, chloroquine, HCQ, LEF or ciclosporin) within the first year of diagnosis. This measure is used as a proxy for more aggressive use of csDMARDs and to capture the change in the management strategy of RA.

CCI: Charlson Comorbidity Index; COPD: chronic obstructive pulmonary disease; CKD: chronic kidney disease; csDMARDs: conventional synthetic DMARDs; tDMARDs: traditional DMARDs; PY: person-years.

Sensitivity analyses

We conducted four sensitivity analyses in this study. First, we considered the potential effect of a concomitant change to more aggressive use of conventional synthetic DMARDs (csDMARDs) such as HCQ, SSZ, MTX and LEF during the study period. We adjusted for csDMARDs use, calculated as the proportion of patients who used two or more types of csDMARDs within 1 year after diagnosis, in our regression model. Second, we set the intervention time to year 2001 instead of 2002 as in our primary analysis and repeated the analyses. Third, we analysed all severe infection outcomes without excluding those that likely associated with TJA surgeries. Fourth, we checked the robustness of the study findings to the use of stepwise selection by fitting a full model with all the listed covariates.

Results

We identified a total of 60 226 RA patients [67% women; mean (s.d.) age at diagnosis 59.1 (15.9) years] who fulfilled our inclusion criteria for incident cases and 588 499 matched non-RA individuals [67% women; mean (s.d.) age at index date 58.9 (15.9) years] during the study period (1995–2015). Table 1 displays baseline characteristics of patients and outcomes according to time period of RA incidence/index date. In total, we identified 14 245 severe infections in RA and 79 819 severe infections in non-RA, with a mean (s.d.) follow-up time of 7.21 (0.17) and 7.05 (0.25) years for RA and non-RA, respectively.

The 8-year severe infection rates among RA patients diagnosed in the pre-bDMARDs period decreased with increasing calendar year of RA onset with a slope of –1.12 (95% CI –1.74, –0.50), but changes observed among those diagnosed in the post-period were not statistically significant, with a slope of 0.72 (95% CI –0.08, 1.52) (Fig. 2). This means severe infection rates decreased by an average rate of 1.12 severe infections per 1000 PY for every increasing calendar year of RA incidence, for RA diagnosed between 1995 and 2001, while severe infection rates increased by an average rate of 0.72 severe infections per 1000 PY among those diagnosed between 2003 and 2007. The crude and adjusted difference between the pre- and post-bDMARDs secular trends in 8-year severe infection rates was 1.84 (95% CI 0.83, 2.84) and 1.85 (95% CI 0.81, 2.89), respectively (Table 2). Overall, the results indicate a significant increase in severe infection rates over time among RA patients diagnosed after bDMARDs introduction.

Figure 2.

8-Year severe infection rate. Unadjusted 8-year severe infection rates (per 1000 PY). GP: general population; bDMARDs: biological DMARDs; ITS: interrupted time-series; PY: person-years

Table 2.

ITS analysis results of 8-year severe infection rates adjusting for age, gender and other selected baseline covariates

Cohort	Pre-intervention trend (95% CI)	Post-intervention trend (95% CI)	Unadj. diff (95% CI)	Adj. diff (95% CI)
	P-value	P-value	P-value	P-value
RA	−1.12 (–1.74, –0.50) 0.0010	0.72 (–0.08, 1.52) 0.0863	1.84 (0.83, 2.84) 0.0009	1.85 (0.81, 2.89) 0.0011
Non-RA	−0.63 (–0.81, –0.45) <0.0001	−0.34 (–0.52, –0.16) 0.0005	0.28 (0.04, 0.53) 0.0305	0.12 (–0.10, 0.34) 0.2877
Difference between RA/non-RA	−0.49 (–1.13, 0.15) 0.1365	1.06 (0.24, 1.88) 0.0114	1.56 (0.50, 2.62) 0.0040	1.73 (0.67, 2.79) 0.0014

Estimations and P-values were derived from GMM. 95% CI were obtained using critical value = 1.96.

Stepwise regressions were used for variable selection in the covariates-adjusted model. RA model adjusted covariates include age, gender and previous infection. Non-RA model adjusted covariates include age, gender, previous infection, and chronic obstructive pulmonary disease. ITS: interrupted-time series; unadj: unadjusted; adj: adjusted; diff: difference.

The 8-year severe infection rates among general population controls with index date in the pre-bDMARDs period decreased over time with a slope of –0.63 (95% CI –0.81, –0.45), and continued to decrease at a slightly slower rate among those with index date in the post-period with a slope of –0.34 (95% CI –0.52, –0.16) (Fig. 2). This means severe infection rates decreased by an average rate of 0.63 severe infections per 1000 PY for every increasing calendar year of index date for controls with index date between 1995 and 2001 and decreased by an average rate of 0.34 severe infections per 1000 PY among those between 2003 and 2007. The crude and adjusted difference between the pre- and post-bDMARDs secular trends in 8-year severe infection rates was 0.28 (95% CI 0.04, 0.53) and 0.12 (95% CI –0.10, 0.34), respectively (Table 2). This indicates that there is no significant change in severe infection rates over time among general population controls after bDMARDs introduction.

The between-group comparison suggests that there was no significant difference in the pre-bDMARDs period trends between RA and non-RA, with the trend difference estimated to be –0.49 (95% CI –1.13, 0.15) (Table 2). The difference between the post-bDMARDs trends was 1.06 (95% CI 0.24, 1.88), suggesting that the post-period secular trends in severe infection rates among RA individuals were significantly higher than among controls. The estimated differences between RA individuals and controls in the unadjusted and adjusted difference in the pre- and post-bDMARDs secular trends were estimated to be 1.56 (95% CI 0.50, 2.62) and 1.73 (95% CI 0.67, 2.79), respectively (Table 2).

The sensitivity analyses showed robust results. After adjusting for aggressive csDMARDs use in RA patients in the sensitivity analysis, the difference in trends remained significant and only slightly changed to 1.69 (95% CI 0.51, 2.86) (see Supplementary Table S2, available at Rheumatology online). The sensitivity analyses with intervention year set at 2001 provided similar estimates as the primary results (see Supplementary Table S3, available at Rheumatology online). The crude and adjusted difference between the pre- and post-bDMARDs secular trends in 8-year severe infection rates in RA was 1.73 (95% CI 0.82, 2.64) and 1.79 (95% CI 0.85, 2.73), respectively. The crude and adjusted difference in trends in general population controls was 0.26 (95% CI 0.02, 0.50) and 0.09 (95% CI –0.12, 0.30), respectively. The results also remained similar after including the TJA-related severe infections in the outcomes (see Supplementary Table S4, available at Rheumatology online). The crude and adjusted difference in trends in 8-year severe infection rates in RA was 1.86 (95% CI 0.81, 2.90) and 1.86 (95% CI 0.86, 2.87), respectively. The crude and adjusted difference in trends in controls was 0.27 (95% CI 0.03, 0.52) and 0.12 (95% CI –0.10, 0.33), respectively. The adjusted results obtained from the full model remained similar to the reduced model yielded by stepwise selection (see Supplementary Table S5, available at Rheumatology online). The adjusted difference in trends in 8-year severe infection rates was 1.67 (95% CI 0.40, 2.94) in RA and 0.14 (95% CI –0.16, 0.43) in general population controls. The estimated differences between RA and controls in the adjusted difference in the pre- and post-bDMARDs secular trends were estimated to be 1.54 (95% CI 0.23, 2.84).

Discussion

In this study, we examined the longitudinal impact of the introduction of bDMARDs on severe infections among population-based cohorts of incident RA patients diagnosed between 1995 and 2007 in British Columbia. We observed a significant increase in the trend in 8-year severe infection rates among RA patients diagnosed after bDMARDs introduction, whereas no change in trend was observed among non-RA population controls.

Previous studies evaluating changes in severe infection rates have yielded mixed results [15–26], due to differences in study design and population, and limited follow-up duration. A number of studies utilized a sizable RA cohort of >10 000 [19–21, 23–25]. Among these, three US studies with sample sizes n = 15 597 [19], n = 10 484 [24] and n = 14 586 [25], with a mean follow-up time of 1.29, 1 and 0.26 years, respectively, evaluated the short-term effect of TNF-α inhibitors initiation on infection rates. They found that TNF-α inhibitors initiation was not associated with higher risk of severe infection compared with non-biologic medications. In contrast, three studies with n = 24 530 [20], n = 11 798 [21] and n = 20 814 [22], with a slightly longer mean follow-up time of 2.2, 3.9 and 2.7 years, respectively, showed that bDMARDs use was associated with a higher risk of infection than treatment with other csDMARDs.

Our study differs from previous studies in a number of ways. First, to our knowledge, our study uses the largest population-based RA cohort (n = 60 226), making the results highly generalizable. Our use of population-based incident cohort of RA patients protects against potential selection bias and survival bias that could result from cohorts of selective patients, such as registries, and from a prevalent cohort, respectively [41]. Second, our study evaluated the long-term impact of bDMARDs introduction as an intervention at the population level, whereby not all individuals received the intervention. We established a uniform follow-up interval of 8 years (substantially longer than past RCTs and population-based observational studies), allowing for accumulation of any severe infection cases occurred within 8 years after RA incidence. We adopted an ITS analysis design which is considered one of the strongest quasi-experimental designs [27, 28] to determine the long-term impact of an intervention. It makes full use of the longitudinal nature of data, accounts for pre-intervention trends and effectively mitigates confounding effects of unobserved time-varying covariates, so only those that change suddenly, concomitantly with the intervention, can bias the results [27, 28]. Applying a simpler before-and-after comparison to our data would have demonstrated an opposite effect (i.e. a reduction in risk of severe infection from 39.1 in the pre-bDMARDs period to 34.3 in the post-bDMARDs period) (Table 1) and would have led to an erroneous conclusion, by failing to capture the change in trends over time discovered by the ITS analysis. Also, our analysis accounts for the potential effect of infections prior to RA diagnosis, an important predictor of future severe infections. By considering medication use, such as glucocorticoids and csDMARDs, the study takes into account the potential confounding effect of changes in RA treatment regimens that happened during the study period.

Several limitations exist in the study and should be acknowledged. Although the ITS study design reduces potential selection bias or unmeasured confounders, the ITS design is still subject to confounding due to co-intervention or concomitant change that occurred at the same time as the intervention [27, 28]. In our study population, while the time to first glucocorticoid use after RA diagnosis remained unchanged, the time to first bDMARDs/csDMARDs use significantly decreased throughout the study period (though changes were less pronounced but still significant in csDMARDs; see Supplementary Fig. S1, available at Rheumatology online), reflecting more aggressive use of bDMARDs and csDMARDs during the study period. In our sensitivity analyses, we controlled for increased aggressive use of csDMARDs (i.e. used two or more types of csDMARDs within 1 year after diagnosis) and a reduction in TJA surgeries which occurred around the same time as the intervention. Results remained robust, still indicating a significant increase in infection risk over time in the post-intervention period. However, it is still possible that there was other unmeasured confounding which changed concurrently with bDMARDs introduction. In our study, we included matched non-RA individuals randomly selected form the general population as a comparator group. This controlled ITS study design allows both within- and between-group comparisons and allows to control for time-varying confounders (including contemporaneous events) that affect both non-RA and RA cohorts [42, 43], thus strengthening the results. For example, using non-RA controls limits the confounding effect of any changes in the Canadian healthcare system and policies (e.g. around health services utilization). Nevertheless, this does not eliminate the effect of any changes that happened only in the RA population, such as the concomitant change in RA treatment approach towards more aggressive control of inflammation aiming at remission. Thus, the increase in severe infections among newly diagnosed RA patients we observed in this study may not solely be due to bDMARDs introduction. One assumption for our ITS model is the linear relationship between time and the outcome within each period. We considered more complex models by adding a quadratic term of time within each period into the model but found them to be statistically insignificant. Hence, we dropped these nonsignificant quadratic terms of time for model parsimony.

We considered each infection-related hospitalization record as a distinct infection case. It is possible that a patient was readmitted to the hospital within a short time frame. While such readmissions are very rare, there is still a possibility that they could be linked to the previous hospitalization due to one period of infection. Supplementary Fig. S2 and Table S6 (both available at Rheumatology online) shows the results of identifying severe infection outcomes with a minimum 2-week time window between two consecutive hospital admissions. The findings remained unchanged. Another limitation is that not all patients in the post-intervention period used bDMARDs, and some patients diagnosed prior to 2002 may have been exposed to bDMARDs during follow-up. Although the goal of the ITS analysis is to inform the change after the introduction of biologics in the long-term infection risk after first RA diagnosis, without such information on treatment actually used, the observed change from the ITS analysis likely underestimates the impact of biologic usage on infection risk. Lack of clinical, laboratory and radiographic information (markers of disease severity) in administrative data also limits our ability to accurately determine the underlying mechanisms of bDMARDs’ influence on severe infection. bDMARDs may increase severe infection via their immunosuppressive effect. It is also possible that bDMARDs significantly improved disease outcome and, in turn, improved survival, which allowed for the survival of cases who are more prone to severe infections. Future studies using detailed data (e.g. clinical, laboratory and radiographic information) on causes of severe infections are warranted to assess the extent to which the increased risk of infections is attributable to bDMARDs use.

In this longitudinal study using a large population-based incident RA cohort, we found a significant increase in the secular trend in severe infection rates among RA patients diagnosed after the introduction of bDMARDs, but not in general population controls. The adjusted difference between the pre- and post-bDMARDs secular trends in 8-year severe infection rates was 1.85 (P = 0.001) in RA and 0.12 (P = 0.29) in non-RA. The results suggest that RA onset after bDMARDs introduction is associated with a significant increase in risk of severe infections.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

All the data are made available via Population Data BC (https://www.popdata.bc.ca/).

Funding

This work was supported by the Canadian Institute of Health Research [CIHR grant: THC-135235 and MOP-130480]; and Natural Sciences and Engineering Research Council [NSERC discovery grant: RGPIN-2018–04313].

Disclosure statement: The authors have declared no conflicts of interest.