WHAT IS ALREADY KNOWN ON THIS TOPICWHAT THIS STUDY ADDS

In this population-based real-world database, after controlling for confounding factors, the use of GC was associated with up to twofold increased risk of incident major adverse CV events on long-term follow-up.

No safe duration of use was found in patients receiving daily prednisolone ≥5 mg.

Very low dose prednisolone (<5 mg daily) did not appear to increase CV risk.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

GC should be discontinued or tapered to below prednisolone 5 mg daily as quickly as possible.

The effect of very low dose GC in symptom control and disease modification should be further evaluated.

Introduction

Rheumatoid arthritis (RA) is associated with accelerated atherosclerosis and increased risk of cardiovascular (CV) events.1 2 This excess CV risk in RA is mediated by the underlying chronic inflammation and increased prevalence of traditional CV risk factors.3 The role of various pharmacotherapeutic agents, glucocorticoid (GC) in particular, remained controversial.4 5 Among the extensive list of side effects, GC increases the risk of CV disease by adversely affecting blood pressure, glycaemic metabolism and lipid profile.6 It also leads to endothelial dysfunction.7 Despite its notorious toxicities, GC stands as a potent anti-inflammatory drug useful for prompt and reliable relief of synovitis symptoms. It is widely prescribed especially at the initial presentation of active disease while awaiting the effect of disease-modifying antirheumatic drugs (DMARDs).8 9 Real-world data reported that up to 60% of patients with RA use GC chronically.10 11

The latest RA management recommendations from the American College of Rheumatology (ACR, updated in 2021)12 and European Alliance of Associations for Rheumatology (EULAR, updated in 2022)13 are rather polarised in terms of GC use. Based on very low certainty of evidence, ACR conditionally recommended the initiation of conventional synthetic DMARD (csDMARD) without short-term GC (over with GC) for DMARD-naïve patients with moderate-to-high disease activity, meanwhile acknowledging that short-term GC is often needed for symptom alleviation. It also strongly recommended against using GC for longer than 3 months. On the other hand, EULAR recommended that short-term (defined as up to 3 months) GC should be considered when initiating csDMARD but should also be tapered and discontinued as fast as clinically feasible, given the robust evidence of efficacy of a combination of csMDARD with GC.14 Despite the distinctly different recommendation statements, both authorities clearly acknowledged the therapeutic effect as well as toxicity of GC, and emphasised on minimisation of the dose and duration of use.

Whether there exists a CV-safe dose or duration of GC in the treatment of RA is still debatable and of great clinical interest. While setting up a controlled trial to study the CV risk of GC is difficult due to the low event rate, real-world data showed that over preceding 6-month and 1-year intervals, initiating GC in GC-naïve patients with RA is associated with increased risk of CV events at daily prednisolone doses ≥5 mg.15 However, the safe duration of GC use was less well defined and the time-varying changes in C reactive protein (CRP), which is a known risk factor for CV events, was not adjusted for in that study. The longer-term effect of GC was not studied.15 Thus, this study aimed to determine the effect of systemic GC dose and duration on the risk of major adverse cardiovascular events (MACE) in patients with RA using a big data analytical approach controlling for systemic inflammation, traditional CV risk factors and other therapies.

MethodsData source and subjects

This was a population-based retrospective cohort study, using a citywide database (the Hospital Authority Data Collaboration Laboratory) in Hong Kong. This database captured all clinical data of patients managed in the public hospitals in Hong Kong.16 17

All patients diagnosed with RA, based on physicians’ diagnosis record according to International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9-CM) coding (ICD9-CM: 714), attending public hospital services from January 2006 to December 2015 were recruited with the following inclusion criteria: (1) age of 18 or above and (2) at least one follow-up visit after disease onset. Patients with MACE before the diagnosis date of RA were excluded.

Outcome and covariates

The outcome of interest was the first occurrence of MACE. MACE was defined in this study as a composite of myocardial infarction (MI), unstable angina, ischaemic or haemorrhagic cerebrovascular accident (CVA), transient ischaemic attack and CV death (ICD9-CM codes listed in online supplemental table 1). All patients were followed up till the end of December 2018. The exposure period was defined as days from study inclusion to first MACE or the end of follow-up.

Demographic data and traditional CV risk factors including diabetes mellitus (DM), hypertension (HT) and hyperlipidaemia (HL) were collected. The definition of DM, HT and HL at baseline or ever was based on the ICD9-CM; and/or by the dispensed prescription records of treatment for these conditions at baseline or during the follow-up (online supplemental table 1). Medications were tracked according to the British National Formulary codes (online supplemental table 2). RA-specific data including serial erythrocyte sedimentation rate (ESR) levels, CRP levels and medication exposure (including GC, methotrexate (MTX), sulfasalazine (SLZ), leflunomide (LEF), hydroxychloroquine (HCQ), biological DMARD (bDMARD) and non-steroidal anti-inflammatory drug (NSAID) online supplemental table 2 were collected at each visit. The daily dose of GC was derived from drug strength (eg, 5 mg), drug frequency (eg, once a day), dosage (eg, 2 tablets) and dispensed quantity (eg, 14 tablets). The duration of GC use was calculated by dividing the dispensed quantity by the prescribed daily dose. Given the differences in the relative anti-inflammatory effects of different types of GC, the dose of each GC was converted into prednisolone-equivalent dose.

Statistical analysis

Descriptive statistics were presented as a number with percentage or mean±SD. Time-dependent Cox proportional hazard regression models were used to analyse the association between covariates (including age, sex, duration of RA, HT, HL, DM, ESR, CRP, use of NSAID, DMARD and GC) and MACE. ESR and CRP were recorded as time-varying covariates with an annual update. The ESR and CRP levels of each year were expressed as the mean of all ESR and CRP measured in that year, respectively. Medications dispensed during the follow-up were updated annually and analysed as a time-varying exposure. Hence, ESR/CRP levels and drug exposure were expressed as yearly intervals in the regression models. Covariates mechanistically related to MACE development with p value less than 0.05 from univariate analysis were subjected to multivariable analysis. The users of DMARDs, NSAIDs and GC were compared against the respective non-users. A forest plot was used to summarise the HR of the multivariable model. Kaplan-Meier analysis was used to compare MACE-free survival between subgroups of different GC doses. If there were missing data, the intervals would be discarded. All analyses were performed in R V.4.0 (https://www.r-project.org/) using ‘base’, ‘survival’ and ‘survminer’ packages. As an alternative method for confounder control to validate the effect of GCs on MACE, time-varying inverse probability treatment weighting (IPTW) was conducted as a sensitivity analysis. To perform IPTW analysis, stabilised weighting factors using the predicted probabilities from the propensity score model were calculated. The use of stabilised weights is a widely accepted method to adjust for confounders in observational studies.18 19 IPTW analysis was applied to both the ESR and CRP adjusted models, comparing ‘maximum daily prednisolone 0 mg vs 1 to <5 mg’' as well as ‘maximum daily prednisolone 1 to <5 mg vs ≥5 mg’.

Results

A total of 12 233 patients with RA were included in this study for analysis, with 105 825 person-years of follow-up. The mean and median follow-up time of the cohort were 8.7 (± 3.2) and 10 (IQR=2–13) years, respectively. Their baseline characteristics are summarised in table 1. At baseline, 34.5%, 7.9% and 3.4% of them had HT, HL and DM, respectively. GC was prescribed at baseline in 37.7% of patients. MTX was the most commonly prescribed csDMARD (55.6%), and only a very small proportion of patients were on bDMARDs (2.8%) initially.

Table 1

Baseline characteristics of rheumatoid arthritis (RA) patients in the entire cohort reported as mean±SD or n (%)

A total of 860 (7.0%) patients developed their first MACE. The crude incidence rate was 8.13 per 1000 person-years. As depicted in online supplemental table 3, patients who developed MACE had a higher prevalence of traditional CV risk factors, more severe RA (higher CRP and ESR, and longer disease duration), but were less likely to use csDMARD, bDMARD and NSAID at baseline. The results of the time-varying univariate analysis for unadjusted HR of risk factors associated with MACE are presented in table 2. All conventional CV risk factors (including age, male gender, DM, HT and HL) and inflammatory burden (disease duration, ESR and CRP) were significantly associated with MACE. When comparing with no GC, prednisolone ≥5 mg daily was associated with higher risk of incident MACE (HR 2.88, 95% CI 2.51 to 3.31, p<0.05) in the univariate analysis; while prednisolone <5 mg daily did not increase the risk of MACE (HR 1.26, 95% CI 0.94 to 1.70, p=0.12). Most of the other antirheumatic medications (NSAID, MTX, HCQ, bDMARDs) were associated with lower risk of MACE except SLZ, while no association was noted between LEF and MACE.

Table 2

Univariate Cox proportional hazards regressions using the demographic variables as time fixed and the other features as time-dependent predictors (being updated at each visit) in the entire cohort

In the multivariable analysis, compared with no GC use, the adjusted HRs of prednisolone ≥5 mg daily were 2.02 (95% CI 1.87 to 2.37, p<0.001) and 1.87 (95% CI 1.60 to 2.18, p<0.001) in the time dependent ESR and CRP Cox regression models, respectively, after adjusting for traditional CV risk factors and other medication use (table 3). The use of prednisolone <5 mg daily was not associated with incident MACE (ESR model: HR 0.83, 95% CI 0.60 to 1.14; CRP model: HR 0.84, 95% CI 0.62 to 1.15). The forest plot summarising the adjusted HRs of all candidate covariates of the CRP model is shown in figure 1 (ESR model shown in online supplemental figure 1). Other independent predictors of MACE included older age, male gender and elevated ESR/CRP, while NSAIDs and MTX exposure were associated with lower risk of MACE.

Table 3

Multivariable Cox proportional hazards regressions using the demographic variables as time fixed and the other features as time-dependent predictors (being updated at each visit) in the rheumatoid arthritis cohort

Figure 1

Forest plot illustrating the multivariable Cox regression analysis with time-varying covariates (CRP model) for the predictors of MACE in the RA cohort. CRP, C reactive protein; RA, rheumatoid arthritis. MACE: major adverse cardiovascular event, RA: rheumatoid arthritis, CRP: C reactive protein, TNF: tumor necrosis factor, MTX: methotrexate, SLZ: sulphalazine, HCQ: hydroxychloroquine, COX2: cyclooxygenase 2.

The MACE-free Kaplan-Meier survival curves of the groups receiving different maximum GC doses are shown in figure 2. While the curves of those with maximum prednisolone dose of <5 mg daily overlapped with the group receiving no GC, patients exposed to prednisolone ≥5 mg daily had significantly reduced MACE-free survival, with an apparent separation from the above two groups after the third year.

Figure 2

Major adverse cardiovascular event-free survival Kaplan-Meier curves of rheumatoid arthritis patients on different daily doses of glucocorticoids.

Among patients receiving daily prednisolone ≥5 mg, the risk of incident MACE was increased by 7% per month in the multivariable models (table 4). Both intermediate-term (up to 180 days) and long-term (longer than 180 days) use of prednisolone ≥5 mg daily were significantly associated with increased risk of MACE (online supplemental table 4).

Table 4

Multivariable Cox proportional hazards regressions using the demographic variables as time fixed and the other features as time-dependent predictors (being updated at each visit) in the rheumatoid arthritis cohort with ever 5 mg daily prednisolone (equivalent) taken.

After applying IPTW, the findings were consistent with the main results from the regression models. Compared with prednisolone 1 to <5 mg, the adjusted HRs of prednisolone ≥5 mg daily were 2.04 (95% CI 1.74 to 2.41) and 2.30 (95% CI 1.87 to 2.82) in the ESR and CRP models, respectively, while daily prednisolone dose of 1 to <5 mg was not associated with excess MACE risk compared with no GC (ESR model: HR 0.85, 95% CI 0.61 to 1.18; CRP model: HR 0.82, 95% CI 0.60 to 1.13) (online supplemental table 5).

Discussion

From this population-based retrospective cohort study with relatively short disease duration, the CV risk of systemic GC use was again demonstrated after controlling for confounding factors, consistent with previous understanding. Yet more importantly, this study revealed the adverse CV effect in RA was dependent of the dose and duration of GC use. While prednisolone dose of less than 5 mg daily might not increase the risk of MACE.

The prevalence of GC use at baseline was 37.7% in this study involving patients recruited between 2006 and 2015. This appears to be lower than what is expected in real-world cohorts. From another study in Hong Kong, it was noted that the prevalence of baseline GC use in historical patients with RA (2002–2006) and contemporary patients with RA (2012–2016) was 55% and 30%, respectively.20 This is consistent with the declining trend of GC use in patients with RA observed in an Australian study from 2001–2005 (55%) to 2012–2015 (39%).21 The low prevalence of GC use might reflect the regional practice of increasingly restrictive GC prescription.

The multidirectional interplay of metabolic and anti-inflammatory effects have made the interpretation of any association between GC and CV events in RA difficult. Different results have been reported regarding the CV effect of GC. A small prospective cohort study (n=353) reported that the increased CV risk associated with GC use at baseline was confounded by the higher RA disease activity and disability among GC users.22 Nonetheless, the study was not powered to analyse the effects of GC use during the follow-up period, with adjustment for cumulative 28-joint Disease Activity Score (DAS28) and the Health Assessment Questionnaire (HAQ) on the risk of CV events.22 Conversely, results from the Quantitative Patient Questionnaires in Standard Monitoring of Patients with Rheumatoid Arthritis (QUEST-RA) study showed that GC was associated with a small reduction of CV disease with an HR of 0.95 (95% CI 0.92 to 0.98).23

Knowing that the key inflammatory cytokines in RA, for example, TNF-alpha and IL-6 also promote atherosclerosis,3 GC may be able to exert some CV protective effect via suppression of these cytokines at transcription level.24 25 In addition, nuclear factor-kappa B (NF-κB) has been revealed to be a pivotal upstream source for these proinflammatory signals involved in the pathogenesis of both RA and atherosclerosis,26 27 and GC is a known potent inhibitor of NF-κB expression.28 29 More precise understanding of the molecular mechanism of GC’s anti-inflammatory effect could facilitate the development of target-based therapeutics, so as to minimise the untoward side effects.

Studies showed that the risk of all-cause and CV mortality increased in a dose-dependent manner when the daily dose of prednisolone was above 7.5–8 mg.30 31 However, the threshold dose beyond which GC increases CV risk was not well determined. The salient findings of our study are that prednisolone <5 mg daily was not associated with MACE after adjusting for conventional CV risk factors, inflammatory markers and other medications, while prednisolone ≥5 mg daily doubled the risk of incident MACE, when compared with non-GC-users. Our findings are in line with the recent study using the CorEvitas registry by Ocon et al, which reported a dose-dependent increased short-term risk of CV events with prednisolone ≥5 mg daily, while the dose <5 mg daily was found to be CV safe.15 On the other hand, while a safe GC duration was not identified in our patients on daily prednisolone ≥5 mg, no increased risk with duration of GC use ≤100 days in the preceding 1 year was found in their study. However, there appeared to be ‘noise’ in the duration-of-use data as dose was not accounted for. In the Glucocorticoid LOw-dose in RheumatoId Arthritis (GLORIA) trial, patients with RA aged above 65 on standard of care with DAS28≥2.6 were randomised to receiving add-on prednisolone 5 mg daily for 2 years vs placebo.32 Among 224 patients receiving add-on prednisolone, 10 developed CV adverse events (composite of MI, CVA and peripheral arterial vascular event); while 6 out of 225 patients in the placebo arm developed events. There was no statistically significant difference between the two groups.32 However, the follow-up duration of 2 years might not be adequate to reveal the long-term detrimental effect as the CV event-free survival curves of our study showed the separation of the prednisolone ≥5 mg daily group being more obvious after 3 years (figure 2).

As prednisolone dose below 5 mg daily appeared to be CV risk neutral, it would be important to know whether this very low dose is in fact clinically efficacious in RA. The GLORIA trial showed that add-on prednisolone 5 mg daily led to significantly lower DAS28 and significantly less radiographic progression, compared with placebo.32 The Circadian Administration of Prednisone in Rheumatoid Arthritis (CAPRA-2) study also showed that add-on modified-release prednisolone 5 mg daily resulted in doubling of ACR50 response.33 The Steroid EliMination In Rheumatoid Arthritis (SEMIRA) trial compared stable patients with RA on ‘tocilizumab with prednisolone 5 mg/day’ versus ‘tocilizumab with prednisolone tapering 1 mg every 4 weeks to 0 mg’.34 Although it was concluded that continuing prednisolone 5 mg/day was better for disease control (relative risk 0.83; 95% CI 0·71 to 0·97) at the end of the study (24 weeks), the flare-free proportion at 8 weeks was 0.96 and 0.97 in the tapered prednisolone and continued prednisolone groups respectively. This provided indirect evidence that prednisolone at 3 or 4 mg daily could still be effective. In another double-blind GC withdrawal trial, patients with RA originally on prednisolone 5 mg daily were randomised to gradual tapering with 1–4 mg prednisolone daily or placebo (ie, direct discontinuation of GC).35 It was found that patients in the tapering group were more likely to withdraw GC without flare, indicating that very low dose GC could help RA disease control. The effect of prednisolone <5 mg daily in controlling symptoms and preventing structural progression should be formally evaluated in clinical trials.

Our study involved a large real-world sample of patients from a territory wide database with long follow-up periods, enabling good statistical power and representativeness. We also tried to homogenise the baseline CV event risk as far as possible by excluding all subjects with MACE prior to the diagnosis of RA. Most importantly, unlike some other similar studies,15 30 we were able to adjust for multiple potential confounding factors including time-varying changes in CRP and NSAID use, which are known predictors of MACE.36 37

There are some limitations of this study. Data on some traditional CV risk factors, for example, smoking status, obesity, exercise habit, family history of CV disease and genetic background were not analysed as they were not available from the database. For the same reason, we were unable to include validated DAS into the analysis due to the lack of information pertaining to some clinical parameters. However, we used inflammatory markers as surrogate for disease activity. Moreover, as with all observational studies, there is a risk of channelling bias and causality cannot be established. The dynamic and time-varying interplay among inflammation, treatment (including GC use) and conventional CV risk factors poses a challenge to traditional statistical analyses, especially for retrospective observational studies. GC can affect the MACE risk directly or through mediation of inflammation and conventional CV risk factors, potentially leading to treatment confounder feedback. Our main results were supported by the IPTW sensitivity analysis which might be better suited for causal inference.

There appears a trend in the research field to revisit the effect and safety of low dose GC in rheumatic diseases, as we observed more studies on low dose GC have been conducted. Back in 2016, the EULAR task force re-evaluated the harm and safety of low dose GC use. It was suggested that previous evidence regarding the harm of low dose GC was often derived from irrelevant, inconsistent or biased information, and proposed that the risk of long-term prednisolone ≤5 mg daily was acceptably low.38 Evidence in favour of the safety of low dose GC in patients with rheumatic diseases has been emerging, also in the field of osteoporosis, where prednisolone ≤5 mg daily did not seem to be associated with reduced bone-mineral density in the recent Glucocorticoid-induced Osteoporosis in Patients With Chronic Inflammatory Rheumatic Diseases or Psoriasis (Rh-GIOP) study.39

Conclusion

To conclude, in patients with RA, the use of systemic GC was associated with a dose and duration dependent risk of incident MACE. This association appeared to be negated for patients on prednisolone dose below 5 mg daily. We advocate to use systemic GC judiciously in RA, balancing the risks and benefits, and to discontinue or taper to prednisolone <5 mg daily as soon as possible.

Data availability statement

Data are available on reasonable request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study was approved by the Joint Chinese University of Hong Kong—New Territories East Cluster Clinical Research Ethics Committee (no. 2021.024). Written informed consent was waived.

Acknowledgments

We would like to express our gratitude to all the staff of the Hospital Authority Data Collaboration Laboratory (HADCL) for their assistance.