Efficacy and safety of dupilumab for the treatment of uncontrolled asthma: a meta-analysis of randomized clinical trials

We searched PubMed, Embase, the Cochrane Library and Chinese Biological Medicine (CBM) databases for articles published up to June 30, 2018, to identify all trials assessing dupilumab therapy for patients with uncontrolled asthma, using the search terms: “asthma” and “dupilumab”. Publication species was limited to humans. In addition, relevant review articles and their reference lists were checked manually.

Studies were considered eligible if they met the following criteria: 1) trials recruited adults/adolescents (≥ 12 years old) diagnosed with uncontrolled asthma, 2) participants received dupilumab therapy at any dose, 3) RCTs, and 4) RCTs reporting the following outcomes: lung function (FEV₁), he 5-item Asthma Control Questionnaire (ACQ-5) score, fractional exhaled nitric oxide (FE_NO), AM and PM asthma symptom scores, quality of life (AQLQ), severe exacerbation rate, or adverse events. Two investigators (XFX and MZ) independently screened all references according to the selection criteria. Disagreements were resolved through discussion.

The patients were diagnosed with asthma persisting for 1 year or more, according to the Global Initiative for Asthma 2009 or 2014 guidelines [13, 14]. Uncontrolled asthma was defined based on current treatment with a medium-to-high-dose inhaled glucocorticoid (fluticasone propionate at a total daily dose of ≥500 μg or equipotent equivalent), plus up to 2 additional controllers (e.g., a long-acting β2-agonist or leukotriene receptor antagonist); a forced expiratory volume in 1 s (FEV₁) ≤ 80% of the predicted normal value before bronchodilator use (or ≤ 90% of the predicted normal value in those 12–17 years of age); FEV₁ reversibility of at least 12% and 200 ml [15]; the 5-item Asthma Control Questionnaire (ACQ-5) score ≥ 1.5 (on a scale from 0 [no impairment] to 6 [maximum impairment]; the minimal clinically important difference is 0.5) [16]; and a worsening of asthma in the previous year that led to hospitalization, emergency medical care, or treatment with systemic glucocorticoids ≥3 days [15]. Asthma symptom scores are patient-reported measures of asthma symptoms, taken upon waking and in the evening. Their effects on activities (PM) and sleep (AM) were noted. These symptom scores range from 0 to 4, with higher scores indicating greater disruption [15]. A severe exacerbation was defined as a deterioration of asthma that required the use of systemic corticosteroids for at least 3 days, or hospital admission, or an emergency department visit because of asthma, treated with systemic corticosteroids [17]. The AQLQ is a patient-reported measure of the effect of asthma on quality of life; higher scores indicate a better quality of life; a global score is calculated and ranges from 0 to 7 [15].

The Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) statement was followed. Two authors (XFX and MZ) performed data extracted and recorded desirable information of each enrolled study in a standard form as recommended by Cochrane [18]. Any disagreement was resolved through discussion or adjudicated by a third author (HXW). In addition, we evaluated the risk of bias using the Cochrane Collaboration’s domains [19].

Intervention effects were presented using risk ratios (RR) and 95% confidence intervals (CIs) for dichotomous data and using mean differences (MD) and 95% CIs for continuous data. If a trial presented more than 2 intervention groups, we combined 2 or 3 groups into a single group according to the Cochrane handbook [19]. Heterogeneity was quantified by I² statistic and the χ² test. Random-effects model was performed in the presence of high heterogeneity (I² > 50%); otherwise, fixed-effects model was applied [20]. Meta-regression was used to explore potential sources of heterogeneity. Funnel plots with the Begg’s and Egger’s tests were used to test publication bias [21]. All statistical analysis was accomplished by an independent statistician using Review Manager (Version 5.3, The Cochrane Collaboration, Copenhagen) and Stata (Version 12.0, Stata Corporation, USA), and rendered statistical significance as P-value < 0·05.

Meta-analyses may cause type I errors because of sparse data and repetitive testing of accumulating data [22]. We performed trial sequential analysis (TSA) to assess the risk of type I errors, which can identify whether the evidence in a meta-analysis is reliable and authentic. If the cumulative z curve crosses the trial, the boundaries, and the required information size, the evidence is sufficient to reach a conclusion, and no further studies are needed. We calculated the required information size for FEV₁ using α = 0.05 (2-sided) and β = 0.20 (power of 80%). TSA version 0.9 beta (http://​www.​ctu.​dk/​tsa) was applied for the analyses [23].

The initial database search resulted in 419 articles. After reviewing the titles and abstracts, 43 were found to be relevant for further detailed evaluation. Of these 38 were excluded as the population was wrong (n = 24), there was no placebo control (n = 4), or data were unavailable (n = 10). Thus, 5 RCTs were included in the meta-analysis [15, 24‐27] (Fig. 1).

We enrolled 5 studies with 3369 patients (Table 1). The sample sizes ranged from 52 to 632 subjects. A single intervention group (dupilumab 300 mg qw and 300 mg q2w) was presented in 2 trials, and the remaining studies included 2 or more interventions (dupilumab 200 mg q2w, 200 mg q4w, 300 mg q2w, 300 mg q4w). Outcome reporting varied among the trials. FEV₁ was reported in 5 studies [15, 24‐27]. Severe asthma exacerbations rate was reported in 4 trials [15, 24‐26]. ACQ-5 scores, FE_NO, and AM and PM asthma symptom scores were reported in 3 trials [15, 26, 27]. AQLQ was reported in 2 trials [15, 26].

TSA found that the optimal sample size required for reliable detection of a reasonable effect of dupilumab treatment on FEV₁ of asthma was 1882 participants; the number of patients included in our study far exceeded this number. TSA presented that the cumulative z curve crossed both the conventional boundary and the sequential monitoring boundary, which indicated that the cumulative evidence is reliable and authentic. Therefore, further studies were not needed (Fig. 2).

We assessed severity measures for efficacy including lung function (FEV₁), ACQ-5 score, FE_NO, AM and PM asthma symptom scores, severe asthma exacerbation rate, and quality of life (AQLQ). Overall, our studies illustrated significant improvement in the efficacy in the treatment of asthma with the use of dupilumab in terms of all clinical indexes. The pooled analyses indicated that dupilumab treatment significantly improved FEV₁ (SMD = 4.29, 95% CI: 2.78 to 5.81, P < 0.001). However, there was a high level of heterogeneity (I² = 99%, P < 0.001) (Fig. 3), with no evidence of publication bias (Egger’s test P = 1.00; Begg’s test P = 0.61). In terms of ACQ-5 scores and FE_NO, our meta-analysis showed that dupilumab administration significantly decreased the ACQ-5 score and FE_NO (SMD = − 4.95, 95% CI: − 7.30 to − 2.60, P < 0.001; SMD = − 2.40, 95% CI: − 3.64 to − 1.17, P < 0.001, respectively), with high heterogeneity (I² = 100%, P < 0.001) (Figs. 4 and 5), but no publication bias (Egger’s test P = 0.73; Begg’s test P = 0.52; Egger’s test P = 0.68; Begg’s test P = 0.58, respectively). Similar efficacy was observed in terms of the AM and PM asthma symptom score. Compared with placebo, dupilumab resulted in a greater reduction in the AM and PM asthma symptom scores (SMD = − 5.09, 95% CI: − 6.40 to − 3.77, P < 0.001; SMD = − 4.92, 95% CI: − 5.98 to − 3.86, P < 0.001, respectively), with high heterogeneity (I² = 99%, P < 0.001; I² = 98%, P < 0.001, respectively) (Figs. 6 and 7), and no publication bias (Egger’s test P = 0.73; Begg’s test P = 0.51). Additionally, dupilumab treatment was associated with a significant reduction in severe asthma exacerbation risk (RR = 0.73; 95% CI: 0.67 to 0.79; P < 0.001) (Fig. 8), without heterogeneity among studies (I² = 0%, P = 0.59), or publication bias (Egger’s test P = 0.98; Begg’s test, P = 0.62). For AQLQ, our analysis demonstrated that dupilumab treatment significantly increased AQLQ scores (SMD = 4.39, 95% CI: 1.44 to 7.34, P = 0.004), with high heterogeneity (I² = 100%, P < 0.001) (Fig. 9), and no publication bias (Egger’s test, P = 0.27; Begg’s test, P = 0.07).

All studies presented adverse events and dupilumab was relatively tolerated. The overall frequency of adverse events was similar in dupilumab-treated (79.2%) and placebo-treated (78.6%) patients (RR = 1.0; 95% CI: 0.96 to 1.04; P = 0.94), with no heterogeneity (I² = 0%, P = 0.59) (Fig. 10) and no publication bias (Egger’s test P = 1.00; Begg’s test, P = 0.69). The incidence of serious adverse events was relatively low in the dupilumab treatment group (2–8.7%). Common adverse events were injection-site reactions, upper respiratory tract infection, headache, nasopharyngitis, bronchitis, and sinusitis (Table 2).

All trials had a low risk of bias in terms of the 6 domains (Fig. 11). The publication bias of trials was estimated by visual inspection of the funnel plot and Egger’s test, and there was no publication bias (Fig. 12, P = 1.00). According to the funding source, subgroup analyses were performed to assess sponsorship biases. Unfortunately, all studies were funded by a pharmaceutical company (Sanofi or Regeneron Pharmaceuticals). Thus, the results should be expounded with caution. In order to examine the source of heterogeneity, subgroup analyses were conducted for FEV₁ and ACQ-5 scores. The studies were stratified in terms of the effects model, region of trial, sample size, treatment duration, blood eosinophil count, and sources of funding, but no significant difference was found between groups in terms of FEV₁ and ACQ-5 scores (Table 3).