Methods

Study retrieval

Studies were sought in PubMed, Embase and CINAHL by JNH and SMMV. A list of relevant keywords for disease (RA, SLE, PMR, COPD, asthma and IBD) and treatment (low- to medium-dose GCs) was compiled and checked by experts (JWGJ and JWJB). Keywords, including words of the title, abstract and Mesh were combined using Boolean operators (AND, OR) (see Appendix 1 in online supplement).

Study selection

The selection of potentially relevant retrieved publications by JNH and SMMV was based on applying the following criteria to the title, abstract, full text or all:

• Study population: adults with inflammatory diseases—notably RA, SLE, PMR, COPD, asthma and IBD—who were treated with GCs. Studies involving patients who received GCs for another disease were excluded.

• GC dose: the dose of GCs had to be low to medium (⩽30 mg prednisolone equivalent during the study except for the first month when a high dose (⩽60 mg) was allowed as in the COBRA trial).2 Studies were excluded if patients had previously used long-term GCs (⩾3 months) or if any GC had been used within 3 months before study onset. If only a group of the whole study population used a suitable dose, then only this part of the population was used for analysis, if stratified data were reported for this group.

• Type of publication: only full papers on original patient data reporting AE of GC treatment were considered for further appraisal.

• Type of study: (randomised) trials and follow-up studies were considered for further appraisal if follow-up had been clearly assessed and the duration of the study was ⩾1 month (see fig 2 for description of the several designs of included studies).

• Type of AE reporting: only studies reporting dichotomous AE outcomes were included.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Quality of studies. The following criteria were used for quality appraisal of the selected studies: (a) Did the study predefine adverse events (AE)? (b) Standardised AE scoring protocol? (c) Did the study detail the number of missing data and the reason of missing? Each study could score 1 point per criterion (Y, yes; N, no) up to a maximum of 3 points. Design of included studies: (1) randomised trials comparing glucocorticoids (GCs) with placebo or non-GCs: (1a) all patients use identical co-medication (disease-modifying antirheumatic drug (DMARD), inhalation GC); (1b) different patients use different co-medication (DMARD). (2) Randomised trial comparing two medication groups (DMARDs vs placebo, GC vs another type of GC, antibiotic vs other antibiotic, bisphosphonate vs placebo or calcium): (2a) all patients received GCs; (2b) a subgroup of patients received GCs. (3) Longitudinal cohorts (prospective/inception/retrospective including follow-up): (3a) all patients received GCs; (3b) a subgroup of patients received GCs. IBD, inflammatory bowel disease; PMR, polymyalgia rheumatica; RA, rheumatoid arthritis; N, number of patients using GCs.

Disagreements in selection were resolved by discussion; selection was based on full consensus.

Quality appraisal

The following criteria were used for quality appraisal of the selected studies:

• Standardised AE scoring protocol: Did the study use an AE protocol? How many AE were monitored during the study? How often were AE scored?

• Predefined AE: Did the study predefine AE? How many AE were predefined?

• Description of missing data: Did the study detail the number of missing data and the reason for them being missing?

Each study could score 1 point per criterion, up to a maximum of 3 points.

The quality of the articles was critically appraised and AE data were summarised by two assessors (JNH and SMMV) independently of each other. Disagreements regarding study quality were resolved by discussion; results are based on full consensus.

Data extraction

Data from studies were considered useful for meta-analysis if the following could be extracted:

• Type and number of AE: AE that did not fit one of the predefined AE list (box 1) were listed as “other”.

• Number of patient-years (py) (duration of follow-up (years) × number of patients).

• Gender, age.

• Type of GC and mean dosage of GC: low GC dose was defined as dosages of 0–7.5 mg and medium dose as >7.5–30 mg prednisolone equivalent; short follow-up was defined as ⩾1–6 months and long follow-up as >6 months.

• Number of missing data (ie, patients dropping out from each study).

Data analyses

The AE rate of GC users, defined as AE per patient-year, was calculated by pooling the data of all retrieved follow-up studies using the Comprehensive Meta-Analysis software (Biostat, Engelwood, New Jersey, USA). Differences in the pooled AE rates were explored by disease, study quality and dosage; a Mantel-Haenszel approach was used to control for confounding.

Results

Search and selection

Using this search strategy, 6097 studies were retrieved (see Appendix 1 in online supplement), which were loaded into an electronic bibliographic management system (Reference Manager). The results of retrieval and selection of studies are shown in fig 1 and Appendix 2 (in online supplement). From the 6097 studies, 28 met the inclusion criteria and AE data were extracted from these. No study on obstructive pulmonary diseases (asthma, COPD) fitted the selection criteria so only studies on RA, PMR and IBD were evaluated for this review.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Results of retrieval and selection of studies. At the search, several double hits were found.

In some studies part of the population did not fit our inclusion criteria. For instance, certain studies evaluated the anti-osteoporosis effect of bisphosphonates against placebo in a population that used GCs. We then looked at AE of the placebo or control groups on GC therapy in these studies, but we did not include the group on bisphosphonates.

Quality of studies

Figure 2 shows the studies categorised according to their quality. The mean (SD) quality of the studies was 2.2 out of 3, and this did not differ substantially between the different subpopulations—ie, disease (RA, PMR and IBD), low- to medium-dose GCs (⩽7.5 mg or 7.5–30 mg), or short/long-term follow-up (⩽6 months or >6 months) (fig 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Disease: IBD, inflammatory bowel disease; PMR, polymyalgia rheumatica; RA, rheumatoid arthritis. Follow-up: short, study duration of ⩾1 month and ⩽6 months; long, study duration >6 months. The number for “follow-up” of the overall groups represents the mean follow-up (months) corrected for the number of patients per individual study. Glucocorticoid (GC) type: 1502 patients used prednisone (P), 685 budesonide (B), 90 beclometasone (B), 105 fluticasone (F). Dose: dose of glucocorticoid in mg. The number for “dose” of the overall groups represents the mean dose of prednisone equivalent (mg) corrected for the number of patients per individual study. The relative potency of budesonide is 0.83.12 No relative potency was known to the authors for beclometasone and fluticasone, which are mainly used topically (enema or aerosol) but, since these drugs were not more efficacious in the included studies,13 14 it was assumed that the dosages of these compounds were of equal or less potency than equal dosages of prednisone. Quality: as described in fig 2. Design of included studies: as described in fig 2. N, number of patients using GCs.

Study characteristics

The quality and characteristics of all studies and of the different subgroups are shown in table 1. The 28 included studies reported on AE data of 2382 patients using GCs, representing 39 subgroups. The mean number of predefined AE and the percentage of drop-outs did not differ across studies.

The characteristics of patients included in studies of IBD, generally with short-term follow-up, were different from those in studies of RA/PMR, generally with long-term follow-up (table 1). The majority of studies were randomised trials, most of which compared the effect of GCs with that of placebo or other medication such as disease-modifying antirheumatic drugs (DMARDs) in addition to standard therapy. Typically, studies on RA used low-dose GCs and had long-term follow-up, while most studies on IBD used both low and medium GC doses and had a short-term follow-up. Compared with long-term studies of RA and PMR, AE in short-term IBD studies were scored more frequently (eg, every 0.5–1 month, see table 1), which will automatically result in higher AE rates.

Adverse event rates

The reported AE rate for GC users over all studies was 150 events per 100 py (95% confidence interval (CI) 132 to 169). The rates of the various AE were comparable, but psychological and behavioural AE were the highest (table 2). Comparison of low and medium dosages did not show dose dependency of any of the AE.

Figure 3 shows the data per diagnosis subgroup. In patients with RA using GCs (14 studies, 796 patients) the AE rate was 43/100 py (95% CI 30 to 55). Shorter term follow-up and higher study quality or design resulted in higher reported AE rates. In patients with PMR using GCs (4 studies, 167 patients) the AE rate was 80/100 py (95% CI 15 to 146). Patients with PMR most often reported gastrointestinal, endocrine and metabolic, cardiovascular and infectious AE. Higher dosages of GCs resulted in higher AE rates in studies of PMR of comparable quality. Among patients with IBD on GCs (10 studies, 1419 patients) the AE rate was 555/100 py (95% CI 391 to 718). These patients most often reported gastrointestinal and neurological AE. Studies with short-term follow-up, most of which were randomised trials in patients with IBD, reported higher AE rates than those with a long-term follow-up.

Discussion

To our knowledge, this is the first study to present a pooled analysis of AE of only low- to medium-dose GCs (up to 30 mg prednisone equivalent) in patients with inflammatory diseases (RA, PMR and IBD) using longer term (1 month) GCs. Although dosages up to 30 mg may seem high, the actual dosages used in the studies included in this review were much lower (table 1).

Overall, GC use results in an AE rate of 150/100 py. The AE rate depends to an extent on the quality of the study but primarily on the population (43/100 py in RA, 80/100 py in PMR and 555/100 py in IBD). Psychological, behavioural and gastrointestinal AE were most often reported. Studies of patients with IBD were often clinical trials of short duration with accurate and frequent documentation of all types of AE. In studies of patients with RA scoring AE in an accurate and predefined manner, the generally longer follow-up may have resulted in the lower AE rates. A limited number of relatively high quality studies were found on AE in patients with PMR using GC. As a substantial number of the included studies did not report on the frequency of scoring AE, it was not possible to compute AE rates adjusted for frequency of scoring.

Our results show that the study design blocks direct comparison of GC-related AE rates in patients with different diseases. In fig 2 the frequency of side effects does not seem to be higher in patients enrolled in high-quality trials, suggesting that the difference between the diseases was the most important factor (and not the quality of the study). The differences between AE rates for the different diseases studied could be related to patient and study characteristics; for example, studies of patients with IBD were often short, used higher GC doses next to other medications and, most importantly, AE were measured very frequently (table 1) resulting in a high rate of AE. In PMR doses were also higher but the study duration was longer, co-medication was more limited and disease symptoms were possibly less prone to be interpreted as AE. The large differences in the occurrence of AE between the study populations (RA, PMR and IBD) are therefore most probably due to the fact that the study characteristics of these populations varied substantially. Our review did not provide other clues or alternative explanations than that disease-specific influences caused the large difference in the occurrence of AE between RA, PMR and IBD.

Previous studies8 9 10 have shown a dose and time relation for more serious long-term GC-related AE such as osteoporotic fractures. In this systematic review we report on commonly reported GC-related AE, most of which may interfere with compliance with GC use. The low AE rate in patients with RA using long-term low doses of GCs confirms the modest toxicity profiles previously described.8 Nevertheless, it is contraintuitive that medium-dose long-term GC use seems to cause AE only as frequently as low-dose short-term GC use. However, both the dose and duration of GC treatment were related to the different disease populations with different baseline risks for AE, different study designs and different comedication and comorbidities. It is therefore difficult to unravel the direct relation between GC treatment and AE.

Although included in our search strategy, we found no studies on patients with obstructive pulmonary diseases (asthma, COPD) that fit our selection criteria. This probably reflects clinical practice where these patients are treated with oral GCs only for short periods (<1 month) to overcome disease exacerbations.

It should be noted that the main goal of this meta-analysis was to examine rates for GC-related AE. Since we did not want to make aetiological or casual inferences (ie, relative risks or odds ratio on AE from the reported data), we did not need a reference population. The primary aim of most studies included in this review was to demonstrate the beneficial effects of GCs, not to measure harm.

In conclusion, the occurrence of GC-related AE largely depends on the disease in the study population. The overall mean rate of AE with GC use was 150/100 py, varying from 43/100 py in RA and 80/100 py in PMR to 555/100 py in IBD.

Recommendations on the use of long-term GCs have recently been published for rheumatic diseases.11 Adherence to these recommendations will most likely help to reduce the occurrence of GC-related AE in any disease. However, more thorough research on the safety of long-term GC use is needed to establish a more exact risk ratio of GCs. The risk-benefit ratio of GCs is an important issue for future studies and could help to create new targets for drug development (eg, selective GC receptor agonists). To achieve this, the development of a core set of AE assessment tools for GC-related AE is needed, with directives on systematic and accurate scoring of predefined AE. In the development of these tools, patient involvement is considered to be crucial.