Safety and efficacy associated with long-term low-dose glucocorticoids in rheumatoid arthritis: a systematic review and meta-analysis

Abstract

Objectives

The aim of this study was to assess the safety and efficacy of long-term low-dose glucocorticoids (GCs) in RA.

Methods

A protocolised systematic review and meta-analysis (PROSPERO No. CRD42021252528) of double-blind, placebo-controlled randomised trials (RCTs) comparing a low dose of GCs (≤ 7.5mg/day prednisone) to placebo over at least 2 years was performed. The primary outcome investigated was adverse events (AEs). We performed random-effects meta-analyses and used the Cochrane RoB tool and GRADE to assess risk of bias and quality of evidence (QoE).

Results

Six trials with 1078 participants were included. There was no evidence of an increased risk of AEs (incidence rate ratio 1.08; 95% CI 0.86, 1.34; P = 0.52); however, the QoE was low. The risks of death, serious AEs, withdrawals due to AEs, and AEs of special interest did not differ from placebo (very low to moderate QoE). Infections occurred more frequently with GCs (risk ratio 1.4; 1.19–1.65; moderate QoE). Concerning benefit, we found moderate to high quality evidence of improvement in disease activity (DAS28: −0.23; −0.43 to −0.03), function (HAQ −0.09; −0.18 to 0.00), and Larsen scores (–4.61; −7.52 to −1.69). In other efficacy outcomes, including Sharp van der Heijde scores, there was no evidence of benefits with GCs.

Conclusion

There is very low to moderate QoE for no harm with long-term low dose GCs in RA, except for an increased risk of infections in GC users. The benefit-risk ratio might be reasonable forusing low-dose long-term GCs considering the moderate to high quality evidence for disease-modifying properties.

RA, glucocorticoids, safety, efficacy, systematic review, meta-analysis, adverse events

Rheumatology key messages

• Long-term low-dose glucocorticoids are regularly used for rheumatoid arthritis, despite guidelines advising against this regimen.

• The present synthesis of long-term randomized trials found benefits in terms of reduced disease activity, disability, and radiographic progression.

• Overall adverse events were not increased, but there was a 40% increase in infections.

Introduction

Glucocorticoids (GCs) were first used in RA >70 years ago [1]. Today, the EULAR guidelines recommend GCs as a bridging therapy for RA [2, 3]. Several trials found that induction therapy with GCs plus conventional synthetic DMARDs (csDMARDs) results in a similar treatment response to biologic DMARDs (bDMARDs) plus csDMARDs, with similar toxicity [4–6].

Real-world data from North America [7] and Europe [8, 9] show that low-dose GCs are being regularly used in the long term, despite EULAR and ACR guidelines advising against the regimen [2, 10]. ACR guidelines even advise against using GCs as a bridging therapy [10]. Whether this treatment option has an acceptable benefit–risk ratio in RA when given in addition to any other treatment has been controversially discussed for years. GCs are cheap and easily available around the world. From a societal perspective, therapy with csDMARDs accompanied by low-dose GCs—if efficacious and safe—could result in considerable cost savings and gains in accessibility compared with treatment with bDMARDs or targeted synthetic DMARDs.

Many observational studies have investigated the risk associated with GC exposure in RA. However, their results must be evaluated cautiously as they are often biased by indication: patients with high disease activity are more frequently prescribed GCs, and both disease activity and medication are associated with adverse events (AEs) and serious AEs (sAEs). Randomized controlled trials (RCTs), though less affected by bias, are mainly powered to investigate efficacy rather than safety. However, research synthesis can increase power [11, 12]. As multiple trials have been conducted to evaluate several outcomes for long-term low-dose GCs in RA, we conducted a systematic review and meta-analysis—the first to summarize randomized, double-blind, placebo-controlled evidence for outcomes of low-dose GC treatment over 2 years or more. It provides a new perspective on the long-term management of RA.

Methods

This systematic literature review was conducted in compliance with the PRISMA 2020 statement [13]. A protocol was preregistered with PROSPERO (https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=252528). No patients were involved in the conduct of this study.

Eligibility

We included double-blind RCTs in RA comparing the outcomes for long-term low-dose GC treatment (defined as ≤7.5 mg/d prednisone equivalent [14] over at least 2 years) with those for placebo, accompanied by any other treatment (e.g. standard of care). Further eligibility criteria are presented in Supplementary Table S1, available at Rheumatology online.

Data sources and search strings

MEDLINE (via PubMed), EMBASE (via Ovid), and the Cochrane Central Register of Controlled Trials (via Cochrane Library) were searched to satisfy the recommendations provided by the Cochrane Handbook [12]. A hand search was conducted to identify further articles. Additionally, we included data from the GLORIA trial [15], the results of which were unpublished when the search was conducted. PICOS and the search strings can be found in Supplementary Table S2 and Data S1, both available at Rheumatology online. The searches were conducted on 1 May 2021.

Study selection

Details of the study selection process, which was conducted by two authors independently, can be found in Supplementary Table S3, available at Rheumatology online.

Data management, items and collection

Notes on data extraction and management are provided in Supplementary Table S4 and Data S2, available at Rheumatology online. First authors of all included studies were contacted and asked for additional data: For four trials, the authors responded, and we received additional data for two of those four trials (Kirwan et al. [16] and Wassenberg et al.  17]), which was in part individual participant data (Kirwan et al. [16]).

Outcomes

Several benefit and risk outcomes were evaluated (all at the latest available time point). The incidence rate of AEs of any type (overall AEs) was defined as our primary outcome, and AE frequency (i.e. events per patient-year) was estimated. Selected secondary safety outcomes were assessed as well. We also evaluated a variety of benefit measures as secondary outcomes (including the WHO-ILAR RA core set [18]; Supplementary Table S5, available at Rheumatology online).

Risk of bias within and across trials

Risk of bias (RoB) in individual studies was assessed by two reviewers (Z.B. and A. Pankow) who were not involved in any of the included trials, using the Cochrane tool for assessing the RoB in randomized trials [19]. We planned to assess RoB across trials by inspection of funnel plots, but the number of included studies was too small.

Summary measures

Mean difference (MD) was used for continuous outcomes (e.g. DAS28) and risk ratio (RRs) for dichotomous outcomes (e.g. remission). A planned analysis using standardized mean difference (SMD) to pool radiographic outcomes measured on different scales had to be omitted: Some studies only reported final values, while others reported change-from-baseline values. The Cochrane Handbook advises against pooling with SMD in such cases [12]. For our primary outcome, we used incidence rate ratio (IRR), because in some studies there were more AEs than patients. To estimate patient-years of exposure, we assumed all randomized patients stayed in the trials for the full duration, which of course must be seen as an approximation. For GLORIA [15] and Kirwan et al. [16], we had access to exact patient-years of exposure.

Data synthesis

Restricted maximum likelihood random effects meta-analyses pooled results per outcome. We pooled end-of-study (post-intervention) and change-from-baseline data [20]. Further information can be found in Supplementary Table S6, available at Rheumatology online. No sensitivity analyses were performed, as the number of included trials was deemed too small to allow for meaningful conclusions. A significance test for overall effects was only conducted for our primary outcome (two-sided α = 0.05).

Confidence in body of evidence

We applied the GRADE approach (two reviewers: A. Palmowski and Z.B.) to make an overall judgement about the certainty of evidence [21]. RCTs start at high quality, but the level can be downgraded based on concerns across several domains. We did not assess publication bias, as the number of trials was too small.

Sensitivity analysis

In a sensitivity analysis suggested in peer review, we included the BARFOT trial [22]. This RCT meets all listed eligibility criteria except that it was not double-blind and placebo controlled. While the risk of both placebo and nocebo responses was higher in this trial, it increased our sample size further to improve statistical power.

Results

Our search yielded 922 articles (Fig. 1). After inclusion of the GLORIA trial [15], we arrived at six RCTs with a total of 1078 patients (Table 1). All but one trial [23] had a duration of 2 years, and all allowed concomitant anti-rheumatic treatment (e.g. with csDMARDs). Three trials [15, 16, 24], including the largest [15], were rated to have a low RoB in all domains (Supplementary Table S7, available at Rheumatology online). One [23] trial had a domain with a high RoB.

Figure 1.

Search flowchart. Inclusion of the GLORIA trial was already planned in the original protocol irrespective of its publication status. RCTs, randomized controlled trials

Table 1.

Characteristics of included trials

First author and year	Trial name	Country	Type of GC	Dosagea	Industry funding	Disease duration (years)	Sample (randomized)	Demographics
								Age (years)	% women	Baseline DAS28	Main outcomes
Wassenberg 2005	Low-Dose Prednisolone Trial	Multiple	Prednisolone	5	Yes	<2	192	GC: 53 Placebo: 50	75 65	NA	Radiographic progression: benefit with GC
Choy 2005	Intramuscular Methylprednisolone Study	UK	Methyl-prednisolone (i.m.)b	5	No	2–10	91	GC: 59 Placebo: 56	75 81	5.52 5.39	No primary outcome defined
Capell 2004	WOSERACT	UK	Prednisolone	7	No	<3	167	GC: 55 Placebo: 56	66 64	NA	Radiographic progression: no benefit with GC
Kirwan 1995	Arthritis and Rheumatism Council Low-Dose Glucocorticoid Study	Multiple	Prednisolone	7.5	No	<2	128	GC: 48 Placebo: 50	66 62	NA	Radiographic progression: benefit with GC
Chamberlain 1976	–	UK	Prednisolone	4c	Unclear	3–25	49	GC: NA Placebo: NA	NA NA	NA	No primary outcome defined
Boers 2022	GLORIA	Multiple	Prednisolone	5	No	Any	451	GC: 73 Placebo: 73	71 69	4.43 4.60	Two co-primary outcomes: DAS28: benefit with GC Number of patients with AEs of special interestd: higher risk with GC

a

In mg/d prednisone equivalent.

b

120mg/month intramuscular methylprednisolone.

c

3mg and 5 mg groups were combined for analysis.

d

Adverse events of special interest defined by GLORIA: serious events, GC-specific events and those causing study discontinuation. AEs, adverse events; DAS28, DAS-28 joints; GC, glucocorticoid; GLORIA, Glucocorticoid Low-dose Outcome in Rheumatoid Arthritis Study; NA, not available; WOSERACT, West of Scotland Early Rheumatoid Arthritis Corticosteroid Therapy Study.

Adverse events

The incidence rate of overall AEs was not significantly elevated in GC groups compared with placebo (five trials; IRR 1.08; 95% CI 0.86, 1.34; P = 0.53; Fig. 2). In absolute numbers, there were 1281 and 1048 AEs in the GC and placebo groups, respectively, or 139.43 and 113.23 AEs per 100 patient-years, corresponding to a difference of 26 AEs per 100 patient-years of exposure. We judged this evidence to be of low quality, because (a) heterogeneity was substantial, with 71% of the variability in effect estimates between studies being due to heterogeneity (not chance) and (b) imprecision. The study from Chamberlain and Keenan [23] was not included in this analysis because it was not clear for all events in which group they occurred. Including the BARFOT trial did not change these results (Supplementary Table S8, available at Rheumatology online).

Figure 2.

Forest plot of the primary outcome: any AE. GCs, glucocorticoids; PYs: patient-years

Infections occurred more frequently in GC groups (RR 1.4; 1.19–1.65; Table 2). The risks of sAEs and death were similar in both groups, as were the risks for withdrawal due to AEs and the risk for all AEs of special interest typically associated with GC exposure apart from infections: osteoporosis, fracture, cardiovascular disease, hypertension, and diabetes or glucose intolerance. Inconsistency for all secondary safety outcomes was quasi-inexistent (sAE I²: 21%; all other I²: 0%). These results remained similar when including the BARFOT trial.

Table 2.

Secondary safety outcomes

Outcome	k	Risk ratio	I²	Quality of evidence		Downgrading due to
Serious adverse event	6	1.07 (0.83–1.37)	21.0%	⊕⊕	Low	Inconsistency, Imprecision
Death	6	1.14 (0.37–3.50)	0.0%	⊕⊕	Low	Imprecision (×2)
Osteoporosis	3	1.11 (0.58–2.11)	0.0%	⊕⊕	Low	Imprecision (×2)
Fracture	6	1.20 (0.78–1.83)	0.0%	⊕⊕	Low	Imprecision (×2)
Cardiovascular disease	6	1.02 (0.72–1.44)	0.0%	⊕⊕⊕	Moderate	Imprecision
Diabetes or glucose intolerance	6	1.18 (0.31–4.43)	0.0%	⊕	Very low	Indirectness, Imprecision (×2)
Hypertension	6	1.04 (0.71–1.53)	0.0%	⊕⊕⊕	Moderate	Imprecision
Infections	6	1.40 (1.19–1.65)	0.0%	⊕⊕⊕	Moderate	Indirectness
Withdrawal due to adverse event	6	0.97 (0.69–1.37)	0.0%	⊕⊕	Low	Indirectness, imprecision

k, number of trials. ×2 means downgraded by two levels.

Efficacy

Over 2 years, our efficacy findings indicate disease-modifying properties. We found a benefit of GCs in the overarching disease activity assessment (DAS28: MD −0.23; −0.43 to −0.03; Table 3), and in radiographic progression measured by Larsen scores (MD –4.61; –7.52 to –1.69). Other positive effects of GCs were less disability measured with HAQ (−0.09; −0.18 to 0.00) and lower ESR values (MD −2.53 mm/h; –4.07 to −0.99). There was zero to very little heterogeneity across the selected studies in the above-mentioned outcomes [I²: 0% in all except ESR (10%)]. Interestingly, with BARFOT data, both DAS28 (MD −0.32; −0.53 to −0.10) and HAQ (MD −0.12; −0.20 to −0.05) meta-analyses moved further into the direction of benefit with GCs.

Table 3.

Secondary efficacy outcomes

Outcome	k	Pooled effect (95% CI)	I² (%)	Quality of evidence		Downgrading due to
DAS–28 joints	2	–0.23 (–0.43, –0.03)	0.0	⊕⊕⊕⊕	High	–
HAQ	4	–0.09 (–0.18, 0.00)	0.0	⊕⊕⊕	Moderate	Imprecision
Larsen Score	2	–4.61 (–7.52, –1.69)	0.0	⊕⊕⊕	Moderate	Imprecision
Modified Sharp van der Heijde Score	3	–3.83 (–8.21, 0.54)	95.4	⊕	Very low	Inconsistency (×2), imprecision
Paina	5	–0.18 (–0.43, 0.07)	24.2	⊕⊕⊕	Moderate	Inconsistency
Patient global assessment of disease activitya	4	0.32 (–0.41, 1.06)	95.5	⊕⊕	Low	Inconsistency (×2)
Physician global assessment of disease activitya	3	0.50 (–0.18, 1.17)	79.3	⊕⊕	Low	Inconsistency (×2)
Tender joint count	2	0.32 (–0.75, 1.38)	0.0	⊕⊕⊕⊕	High	–
Swollen joint count	2	–0.24 (–2.34, 1.85)	69.6	⊕⊕	Low	Inconsistency (×2)
Duration of stiffness, min	2	–5.96 (–18.32, 6.39)	61.9	⊕	Very low	Inconsistency (×2), Imprecision
ESR, mm/h	5	–2.53 (–4.07, –0.99)	10.0	⊕⊕⊕⊕	High	–
CRP, mg/l	3	2.20 (–1.61, 6.02)	7.3	⊕⊕⊕⊕	High	–
Fatigueb	1	–0.10 (–0.69, 0.49)	–b	–b	–b	–
Remissionc  (risk ratio)	2	1.40 (0.88, 2.22)	0.0	⊕⊕	Low	Imprecision (×2)

Effect size is mean difference unless otherwise noted. k, number of trials; MD, mean difference. ×2 means downgraded by two levels.

a

Numerical Rating or Visual Analogue Scale 0–10.

b

Only one trial with respective data.

c

Boers 2022: Boolean definition according to ACR/EULAR criteria; Wassenberg 2005: ACR criteria.

In another measure of radiographic progression [modified Sharp van der Heijde scores (mSHS)], we observed a trend in favour of GCs, but CIs projected above 0. Of note, heterogeneity was considerable, with an I² of 95%, and the quality of evidence indicating no difference between groups was very low. After including the BARFOT trial, we found benefit with GCs in mSHS: MD –3.88; –6.70 to –1.05.

We also tried to differentiate the effects of GCs on disease activity by looking at several specific efficacy outcomes. No differences were found in these outcomes, namely pain, patient and physician global assessments of disease activity, tender and swollen joint counts, duration of morning stiffness, fatigue, and CRP (Table 3). Of note, there was much variation in the number of trials reporting the respective data.

Both trials investigating remission (Boers et al. and Wassenberg et al.; different definitions) found more cases in GC groups compared with placebo, and there was no heterogeneity (I²: 0%), but the RR CI crossed 1 (RR 1.4; 0.88–2.22). The RR for remission with GCs increased to 1.60 (1.25–2.05) when including BARFOT, which reported patients in remission defined by a DAS28 of <2.6.

Discussion

In this systematic review and meta-analysis of randomized double-blind placebo-controlled trials, we found no increased rate of AEs in low-dose GC treatment of RA over 2 years or more used in addition to any other treatment. Although the RoB was low in most included trials, the quality of evidence had to be downgraded by two levels due to heterogeneity and imprecision. Our primary outcome finding is in keeping with a systematic review and meta-analysis from 2009 [25] including trials of at least 1 year’s duration, which saw no increased withdrawals due to AEs [odds ratio (OR = 1.09)]. The number of AEs per patient-year (OR = 1.19) and sAEs (OR = 1.06) was also similar between groups.

We found an increase in infections, namely by 40%, compared with placebo. Of note, the only trial finding more infections with GCs is the GLORIA trial, which is the only trial conducted in an elderly population. Possibly, elderly patients might be more susceptible to infections than younger people; due to the small number of studies in our meta-analysis, we could not conduct a meta-regression analysis to investigate this hypothesis. Also, the risk increase of 40% must be interpreted in relation to the risk associated with anti-rheumatic treatment in general. In a large meta-analysis of trials investigating biologics in RA, the odds for serious infections were increased by 31% (OR 1.31) in patients treated with biologics compared with csDMARDs [26]. The (Peto) odds ratio of opportunistic infections was reported to be 1.79 in patients receiving bDMARDs in another systematic review and meta-analysis [27]. A study from the British Biologics Register for Rheumatoid Arthritis found a lower hazard ratio in patients treated with csDMARDs (hazard ratio 0.64) compared with biologics users [28]. Concerning Janus-Kinase inhibitors (JAKis), a meta-analysis of RCTs in RA yielded statistically significant pooled ORs of 1.43 (upadacitinib) and 1.52 (tofacitinib) for any infections of these drugs compared with placebo [29], but found no difference for serious infections [29, 30]. We conclude that the risk of infections seen with low-dose GCs could be on the same scale as with most biologic and targeted treatments.

Interestingly, other events typically associated with GC use—osteoporosis, fracture, cardiovascular disease, hypertension, and diabetes or glucose intolerance—were not more common in GC groups compared with placebo. We deem it likely that these GC-related AEs occur more commonly under high-dose treatment. Under low-dose treatment, however, potentially detrimental vascular and metabolic effects might possibly be compensated for by suppression of inflammation, thereby leading to a neutral net effect of GCs regarding the above-mentioned AEs.

Some readers might be surprised to see these results and those for other GC-related AEs, as there is observational evidence in RA finding an increase in various AEs in GC users. However, observational studies in GCs are prone to bias by indication. Higher disease activity increases the risk of AEs and the risk of receiving GCs at the same time. For example, in the recent COVID-19 Global Rheumatology Alliance Study [31], Strangfeld et al. aimed at identifying factors associated with COVID-19–related death in patients with inflammatory rheumatic diseases. In the original report, GCs at >10 mg/d prednisone equivalent were linked to a higher chance of death (OR: 1.69). For a later correspondence [32], the analyses were run again accounting for the interaction between disease activity and GCs. GCs were not associated with COVID-19–related death in the absence of moderate/high/severe disease activity. Further examples of bias by indication in GC research, which we avoided by including randomized trials only, can be found in the literature (e.g. [33]).

Concerning efficacy, this systematic review and meta-analysis showed effects of long-term GCs at a low dose on relevant clinical outcomes, one of which is a patient-reported one (HAQ). The difference of −0.09 in HAQ scores that we found is surprisingly similar to the difference of −0.1 found with adalimumab and MTX compared with MTX monotherapy in the 2-year double-blind PREMIER trial [34]. The DAS28 improvement achieved with biologic therapy seems a little higher than the one achieved with low-dose GCs, when comparing our results with those of the long-term double-blind COMET trial, which investigated etanercept in combination with MTX vs MTX monotherapy [35, 36]. Over 2 years, a quarter of patients remained randomized to MTX monotherapy and a quarter to combination therapy with etanercept. The difference in mean DAS28 change from baseline to 2 years was by 0.8 in favour of combination therapy.

In radiographic progression, we found an effect of long-term low-dose GCs when measured by Larsen scores. The studies’ results were homogeneous (I²: 0%), and 95% CIs were far from the line of no difference. In studies measuring radiographic progression by mSHS, however, CIs crossed the value of no difference (which is 0). This is due to the only study that used an unusual radiographic reading protocol and did not find an effect of long-term low-dose GCs on radiographic progression (Capell et al.) [24], leading to considerable heterogeneity in our meta-analysis of radiographic progression measured by mSHS, with an I² of 95%. The study of Capell et al. stands out as we had to combine the X-ray readings of two readers (which were reported separately) in a fixed-effects meta-analysis, to include the results in our overall meta-analysis of radiographic progression. The two readers differed greatly (by a factor of 6) in their ratings. We would have liked to look at the primary data as well to investigate the cause for this heterogeneity, but we were not able to retrieve the data despite enquiries. Interestingly, after including the BARFOT trial in a post hoc sensitivity analysis, 95% CIs for mSHS remained below 0.

Kirwan et al., in 2007, included trials with any measure of radiographic progression and summarized their results as percentage of maximum score. They found a statistically significant effect of GCs vs no GCs at both 1 and 2 years [37]. Another systematic review and meta-analysis looked at radiographic effects of GCs, csDMARDs, and bDMARDs and found similar effects for the three drug groups [38]. In summary, it is most likely that long-term low-dose GCs reduce radiographic progression in RA, although the effect was not statistically significant in our analyses (possibly because we included fewer studies than the above-mentioned ones as we were focusing on placebo-controlled RCTs).

There might be cost-saving effects if low-dose GCs are added to the DMARD treatment regimen in patients with active RA. In the CAMERA-II trial of 10 mg/d GCs vs no GCs, after observing patients for a median of 6.6 years, bDMARD initiation occurred in 31% vs 50% of patients, respectively [39]. A simple cost-effectiveness evaluation of adding 5 mg/d modified-release prednisone to the treatment of patients who have an indication for bDMARD treatment indicated cost-savings, as bDMARD initiation could be averted or postponed [40]. A cost-effectiveness evaluation is planned for the GLORIA trial and will provide further evidence.

Our study has several strengths. First, we only included double-blind trials. In GC research, a nocebo effect is likely to affect the occurrence of AEs, as GC-related AEs are generally well known and feared among clinicians and patients. By including double-blind trials only, we aimed at minimizing nocebo effects, but we had to exclude trials that would otherwise have been considered, e.g. the BARFOT trial [22]. In sensitivity analyses suggested in peer review, we included this trial. While our AE findings remained consistent in these analyses, prior insignificant efficacy results such as the meta-analysis of mSHS and remission became significant (in favour of GCs).

Another strength of this systematic review and meta-analysis is methodological rigour [41]. We published a research protocol before performing the study. The study selection was conducted separately by two reviewers, and three databases were searched, all in line with recommendations from the Cochrane Handbook [12]. The RoB and quality of evidence assessments were also conducted separately by two reviewers. We were able to include up-to-date results of the largest long-term GC RCT to date, the GLORIA trial [15]. Yet another strength is that the results from trials on GCs were found to be of good generalizability to real-world patients [42]. The GLORIA trial even purposely studied the medication in elderly and multimorbid patients [43]. This is in contrast to most trials in RA, in which elderly patients are underrepresented [44], and also in contrast to most trials investigating biologic agents in RA [45]. Furthermore, we had access to the individual participant data of GLORIA [15] and of the Arthritis and Rheumatism Council Low-Dose Glucocorticoid Study [16] and were able to gather additional information from the Low-Dose Prednisolone Trial [17].

A limitation of our study is the estimation of AEs per patient-year. By assuming that all patients finished the trials, the duration of exposure is likely to have been overestimated and the frequency of AEs underestimated. Unfortunately, for most trials (except GLORIA and Kirwan’s trial), there was no data on the exact duration of exposure to treatment. A weakness with regard to methodology is that data extraction was conducted by one reviewer only (the first author), mainly because of personnel constraints associated with the COVID-19 pandemic. Furthermore, apart from GLORIA, all trials were powered to study efficacy and were probably underpowered to detect harm. However, research synthesis may increase power to overcome this issue [11, 12]. Also, most studies had a duration of 2 years, and there were no studies with a duration of >3.5 years. Consequently, no statement can be made concerning ‘very-long-term’ safety and efficacy. Furthermore, there was heterogeneity between trials in several outcomes, which reduced the confidence in some estimates according to the GRADE approach. Finally, two included trials were >25 years old, and AE recording and reporting may have been less accurate in the past.

Conclusion

Data from RCTs indicate that long-term low-dose GCs are not associated with an increased risk of AEs as compared with placebo in the treatment of RA, apart from a higher risk of infections. The quality of this evidence ranges from very low to moderate, mainly because of imprecision and inconsistency, and clinically relevant harm cannot be fully excluded. The benefit–risk ratio might nevertheless be reasonable for using add-on long-term low-dose GCs, taking into account the moderate to high quality evidence for disease-modifying properties, i.e. benefits in disease activity, function, and radiographic progression. These findings are especially important, since the 2021 ACR guideline for RA advises against any use of GCs on the basis of ‘very low’ (short-term use) to ‘moderate’ (long-term use) quality evidence.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

F.B. and A.P. are willing to examine all requests for the data and materials used in this study for a period of 5 years from the date of this publication.

Funding

This project is part of the GLORIA project and trial (Glucocorticoid low-dose outcome in rheumatoid arthritis study (gloriatrial.org; registered on https://clinicaltrials.gov/; identifier NCT02585258) and has received funding from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under grant agreement No. 634886. The Section for Biostatistics and Evidence-Based Research, The Parker Institute, (S.M.N. and R.C.) is supported by a core grant from The Oak Foundation (OCAY-18–774-OFIL), a group of philanthropic organizations giving grants to not-for-profit organizations around the world.

Disclosure statement: A.P., Z.B., R.C., S.M.N., J.K. and L.H. have no conflicts of interest to disclose. F.B. reports receiving consultancy fees, honoraria and travel expenses or grant support from Horizon Pharma, Grünenthal, AstraZeneca, Mundipharma, Pfizer and Roche. F.B., R.C., J.A.P.S., L.H. and M.B. were collaborators in the GLORIA trial. J.K. was the Principal Investigator of the Arthritis and Rheumatism Council Low-Dose Glucocorticoid Study. S.W. was the Principal Investigator of the Low-Dose Prednisolone Trial. A.P. and S.M.N. were involved in satellite projects associated with the GLORIA trial. M.B. received consultancy fees from Novartis. S.W. received honoraria, travel expenses and consultancy fees from AbbVie, BMS, Galapagos, Gilead, Lilly, MSD, Mylan, Pfizer, Sanofi, and UCB. C.D. has received consulting/speaker’s fees from AbbVie, Eli Lilly, Janssen, Novartis, Pfizer, Roche, Galapagos and Sanofi, all unrelated to this manuscript.