The search strategy and the study selection process are described in the PRISMA flowchart (supplementary figure A1). We included 18 randomised and two [37, 38] quasi-randomised controlled trials with 2140 participants that fulfilled the eligibility criteria (table 1 and supplementary table A1).

Characteristics of the included trials

Most trials (n=14) recruited hospitalised participants. Outpatient [54] and emergency department [39] settings were each evaluated in one trial. Clinical setting was not clarified in the remaining four trials. Blinding was double in six trials and single in four trials, whereas another one was open-labelled and nine did not report on blinding. Concomitant asthma was explicitly excluded in most trials (n=14). Eight trials excluded participants with a smoking index <10 pack-years [53, 54, 58] or <20 pack-years [42, 47, 48, 50, 56]. We noted no significant baseline differences between treatment arms, apart from two trials [48, 58] (supplementary table A1). Participants were followed-up for a median 10 days, with a range from 24 h [39] to 12 months [41].

Duration of exacerbation symptoms prior to enrolment was generally not specified, apart from three trials [47, 50, 64] with a mean duration ranging between 2.4 and 11 days. Four trials set an upper limit of 24 h [53], 7 days [43, 54] or 14 days [50] of symptoms. Exacerbations were not specifically eosinophilic, but rather unselected; one trial [56] recruited exclusively patients with infective bacterial exacerbation. Participants had not received corticosteroids for the presenting exacerbation in 16 trials, five of which explicitly excluded both systemic and high-dose inhaled corticosteroids (table 1). Another trial [54] prohibited high-dose inhaled corticosteroids at study entry, but all participants had received oral corticosteroids prior to randomisation as ambulatory treatment for the exacerbation. Withdrawal rates were high or imbalanced in six trials, up to 29.3% for inhaled [66] and up to 24.5% for systemic corticosteroids [45], being >10% higher for inhaled than systemic corticosteroid arms in two trials [50, 66].

Corticosteroids were administered for a median of 7 days, ranging from 1 to 15 days (table 1). The inhaled corticosteroids arms received budesonide mostly via a nebuliser (n=17 trials, median dose 8 mg·day−1, range 1.5–8 mg·day−1) or via inhaler devices (n=2, dose 800–1280 μg·day−1), but one trial used both in different arms [53]. Two other trials [45, 50] initially administered corticosteroids via a nebuliser, and afterwards via an inhaler device. Systemic corticosteroids, namely prednisone, prednisolone, methylprednisolone and hydrocortisone, were administered at various dosages consistent with routine clinical practice, intravenously (n=8), orally (n=6), both (in different treatment arms, n=3) [57, 59, 61] or allowed either (n=1) [38]. Two trials [45, 56] initially offered intravenous and later oral corticosteroids. Concurrent treatment for the exacerbation included bronchodilators with or without methylxanthines, antibiotics, mucolytics, antitussives, oxygen supplementation, fluid and electrolyte support.

All trials were deemed at a high risk of bias, mostly due to lack of blinding with potential bias in outcome measurement, unavailable data or per protocol analysis (figure 1).

Risk-of-bias table.

Figure 2 and supplementary figure A2 feature the forest plots with the effect estimates for all meta-analyses. The GRADE evidence profile for the main clinically relevant outcomes is shown in table 2.

Forest plots featuring the overall effect estimates for the main outcomes. a) Treatment failure during the intervention. b) Breathlessness, mean change (pre-treatment to post-treatment). c) Breathlessness, post-treatment score. d) Serious adverse events during treatment. e) Any adverse event during treatment. f) Health-related quality of life, post-treatment score. CS: corticosteroids; M-H: Mantel–Haenszel method; IV: inverse variance method; SMD: standardised mean difference.

Evidence profile according to Grading of Recommendations Assessment, Development and Evaluation methodology for clinical outcomes up to treatment completion and at longest follow-up

Treatment success, defined as a dichotomous measure of the overall outcome of the exacerbation, was not assessed in any trial. Treatment failure, defined as lack of efficacy, deterioration or need for treatment intensification, had a similar occurrence between groups during treatment, as shown in five trials with 569 participants (relative risk 1.75, 95% CI 0.76–4.02, I2=0%, low certainty) [38, 48, 50, 53, 54].

Breathlessness was evaluated in eight trials with 721 participants, where inhaled corticosteroids showed similar impact to systemic corticosteroids. Specifically, change from pre-treatment values was captured in the Borg, modified Borg or a nonvalidated scale in three trials [47, 54, 64]. Treatment effects were similar between groups (standardised mean difference, SMD −0.11, 95% CI −0.36–0.15, I2=0%, 239 participants, low certainty). The post-treatment scores on the modified Medical Research Council and modified Borg scales did not differ significantly between groups, according to three trials, including one of the above [38, 47, 48] (SMD −0.18, 95% CI −0.41–0.05, I2=0%, 293 participants, low certainty). Finally, two further trials (146 participants) reported similar impact across treatment groups, but did not provide specific numerical data to pool [42, 58].

Serious adverse events were monitored in two trials during treatment, showing a similar rate in both arms (relative risk 1.47, 95% CI 0.56–3.88, I2=0%, 246 participants, low certainty) [50, 66]. Longer follow-up was assessed in one trial that did not reveal between-group differences (3 months follow-up, relative risk 0.63, 95% CI 0.24–1.66, 113 participants) [66].

Health-related quality of life showed similar improvement with both inhaled and systemic corticosteroids across six trials with 885 participants. These used the St George's Respiratory Questionnaire (SGRQ) [38], SGRQ-C [53], the COPD Assessment Test [41, 48], the clinical COPD questionnaire (CCQ) [54] and modified Guyatt measure [59] to capture post-treatment scores. When pooling data with the SMD, we noted a balanced effect in both groups (SMD 0.0, 95% CI −0.16–0.16, I2=22%, moderate certainty). The number of patients with improvement in SGRQ did not appear to differ between groups, as shown in one of the aforementioned studies (relative risk 1.07, 95% CI 0.93–1.23, 125 participants, very low certainty) [38].

The overall symptom score, comprising breathlessness, cough and wheezing, with or without sputum, was similar in both groups post-treatment (SMD 0.00, 95% CI −0.91–0.90, I2=90%, 200 participants, very low certainty), according to two trials using a nonvalidated score [37] or the symptoms subscore of the SGRQ-C [53]. Improvement in post-treatment scores was reportedly similar in the Ställberg et al. [54] (109 participants, CCQ symptoms subscore) and Djordjevic et al. [42] (60 participants, score not described) trials. Cough [48, 54] and sputum [48] were scarcely reported, without any significant between-group difference.

The length of hospital stay was assessed in seven trials with 1069 participants. We found similar hospitalisation duration in both treatment groups (MD −0.59, 95% CI −1.88–0.70, I2=89%, three trials with 613 participants, low certainty) [38, 41, 48]. Three other trials reported narratively consistent results (326 participants) [50, 58, 63]. The median hospital stay ranged from 7 to 11.3 days and from 6 to 13 days among patients receiving inhaled and systemic corticosteroids, respectively [38, 41, 48, 50]. Prolonged hospitalisation, reaching 10 days [50] or more [38, 45], was proportionate in both groups (relative risk 1.03, 95% CI 0.73–1.45, I2=27%, three trials with 340 participants, low certainty). The number of patients who were discharged by day 5 from recruitment, was documented in one study [38], with a tendency to favour inhaled over systemic corticosteroids (relative risk 1.78, 95% CI 1.00–3.14, 125 participants, very low certainty).

Among patients recruited at a primary health centre or the emergency department, hospital admission was more frequent in systemic corticosteroid group (relative risk 0.70, 95% CI 0.53–0.93, I2=0%, two trials with 160 participants, low certainty) [39, 54]. Among hospitalised patients, indication for mechanical ventilation or admission to the ICU did not differ between groups (relative risk 0.82, 0.17–3.89, I2=0%, four trials, 388 participants, low certainty) [38, 50, 51, 58].

The risk of future exacerbations was assessed in five trials with 513 participants and was similar between groups. The number of patients experiencing an exacerbation within 3 months did not differ between groups (relative risk 1.08, 95% CI 0.50–2.33, one trial with 109 participants, very low certainty) [54]. Two trials with 184 participants recorded a similar number of total exacerbations between groups during follow-up for 1 month [45] or 3 months [54] after discharge (relative event rate 1.02, 95% CI 0.62–1.70, I2=0%, low certainty). Another trial followed-up 126 patients for 12 months [41], demonstrating a similar rate (MD −0.08, 95% CI −0.31–0.15, very low certainty) and mean time to next exacerbation in months (MD 0.46, 95% CI −0.75–1.67, very low certainty) in both treatment groups. Mean time to next exacerbation up to 3 months was also similar between groups according to another trial with no numerical data to pool [54]. The risk for severe exacerbations leading to hospital admission did not differ between groups either (relative risk 1.05, 95% CI 0.08–14.42, I2=60%, two trials with 203 participants, low certainty; relative event rate 0.88, 95% CI 0.29–2.69, I2=3%, two trials with 184 participants, low certainty) [38, 45, 48, 54].

Mortality did not differ between groups at up to 10 days of treatment (relative risk 0.29, 95% CI 0.01–7.03, three trials with 328 participants) [45, 50, 66] and up to 3 months follow-up (relative risk 0.32, 95% CI 0.03–3.03, I2=0%, two trials with 188 participants, low certainty) [45, 66].

Adverse events tended to be more frequent among patients receiving systemic versus inhaled corticosteroids (relative risk 0.80, 95% CI 0.64–1.0, I2=0%, six trials with 480 participants, moderate certainty [43, 50, 51, 56, 63, 66]. Another trial reported narratively similar occurrence between treatment groups [58]. The risk for adverse events at longest follow-up (3 months) was similar in both groups (relative risk 1.14, 95% CI 0.64–2.03, 113 participants) [66]. Hyperglycaemia affected more patients during treatment with systemic than inhaled corticosteroids (relative risk 0.28, 95% CI 0.14–0.53, I2=15%, nine trials with 1114 participants, moderate certainty) [41, 50, 56–59, 61, 63, 64]. This was corroborated by a further trial [48] recording a 2.9-fold higher rate (p=0.026) in patients treated with systemic versus inhaled corticosteroids (78 participants). Oral fungal infection was more frequent in the inhaled treatment group (relative risk 7.90, 95% CI 1.82–34.30, I2=0%, four trials with 431 participants, low certainty) [37, 58, 61, 66]. Conversely, the rate of any fungal infection (local or systemic) did not differ between groups (relative risk 0.39, 95% CI 0.14–1.11, one trial with 410 participants, low certainty) [41].

The partial pressure of oxygen in arterial blood (PaO2, in mmHg) showed similar mean change from pre-treatment with both corticosteroid routes (MD −4.33, 95% CI −10.36–1.69, I2=67%, two trials with 283 participants [41, 64]; narrative in another trial with 82 participants [45]). There was a tendency for higher post-treatment values with systemic corticosteroids, which were not clinically important (MD −1.33, 95% CI −2.64–−0.02, I2=34%, six trials with 695 participants) [37, 41, 45, 59, 61, 64]. Notably, PaO2 measurements were not performed explicitly under room air conditions in all participants [37, 41, 50, 59, 61], thus rendering results uninformative. When excluding these trials in a sensitivity ad hoc analysis, we found no between-group difference (post treatment values MD −1.25, 95% CI −3.34–0.84, I2=0%, two trials with 122 participants) [45, 64]. Three other trials with 156 participants [51, 56, 58] reported nonsignificant between-group differences, with no numerical data to be included in the meta-analysis. The partial pressure of carbon dioxide in the arterial blood (in mmHg) mean change from pre-treatment was similar between groups (MD 0.60, 95% CI −1.84–3.03, I2=47%, two trials with 283 participants) [41, 64]. However, one trial that was not pooled reported a mean change that exceeded 3.9 mmHg, favouring systemic over inhaled corticosteroids (82 participants); however, this difference is not clinically significant [45]. Post-treatment values did not differ significantly across six trials with 695 participants (MD −0.03, −1.01–0.94, I2=34%) [37, 41, 45, 59, 61, 64]. Two further studies that were not pooled [51, 56] did not find any between-group difference either (70 participants). Both inhaled and systemic corticosteroids showed similar impact on oxygen saturation post-treatment values across six trials (535 participants) [38, 39, 45, 48, 51, 58] using either pulse oximetry or arterial blood measurements (MD 0.28, −0.28–0.85, I2=54%, four trials with 369 participants) [38, 39, 45, 48].

Disease progression was not assessed in line with the ERS COPD exacerbations core outcome set [31] in any trial. One trial with 98 participants [50] reported on FEV1 mean change from baseline disease state, with similar results between groups on the third treatment day. Mean change from pre-treatment generally did not differ between groups (mean change in percentage predicted values: MD 0.10, 95% CI −3.13–3.33, one trial with 109 participants that assessed FEV1 % before and after treatment [66]; mean change MD 0.03 L, 95% CI −0.05–0.11 L, I2=73%, three trials with 188 participants [41, 47, 64]; narrative in two trials with 200 participants [45, 53]). Post-treatment FEV1 values were similar in both groups (MD 0.18% pred, 95% CI −2.69–3.04% pred, I2=85%, eight trials with 804 participants) [38, 43, 45, 48, 53, 54, 59, 61], but there was a tendency in favour of systemic corticosteroids for measurements in litres (MD −0.05 L, 95% CI −0.12–0.03 L, I2=68%, seven trials with 703 participants) [37, 38, 41, 47, 54, 57, 61]. Two additional trials that we could not include in our meta-analysis, due to insufficient numerical data, did not yield between-group differences [56, 58]. Evaluation of FVC yielded similar results across the treatment groups [38, 45, 48, 53, 57, 58].

Treatment adherence was quantified in one trial documenting an average compliance that reached 89.9% and 96.7% in the inhaled and systemic corticosteroid groups, respectively [54].

Worsening of symptoms after initial treatment, activities of daily living, development of pneumonia or resistant bacteria were not evaluated in any trial.