The number of studies initially identified by the search are given in Fig. 1. Ninety-seven papers including relevant trials underwent full-text review, of which 11 papers covering 13 trials were included in the final analysis (Fig. 1). The studies are referenced here [7,8,9, 11,12,13,14,15,16,17,18] and are tabulated in Table 1. In the event, no disagreement over included studies occurred. The types of non-IPF entered into studies are summarised in Table 2.
Key to abbreviations for Table 2:
1. Idiopathic non-specific interstitial pneumonitis; 2. unclassifiable idiopathic interstitial pneumonitis; 3. Chronic hypersensitivity pneumonitis; 4. Rheumatoid arthritis related interstitial lung disease; 5. Mixed connective tissue disease associated ILD; 6. Systemic sclerosis-associated ILD.
Of the 13 trials included, all had data extracted on FVC changes, and 11 had data available on all-cause mortality (this was pooled for the two INPULSIS trials and the two CAPACITY trials, so the total of separate trials analysed for all-cause mortality reduced to nine). The main variables of these studies are summarised in Table 1. All were randomised controlled trials.
A consensus on bias assessment was achieved. Three studies were identified as having “some concerns” in one domain of five. The study of Maher et al. (2020) was rated as “some concerns” on domain 5 (selection of the reported result): this was because the primary outcome of home spirometry could not be used because of some impossible or implausible values, and the site spirometry (which is used in all the studies) is presented in this analysis instead. In Behr et al. (2021), the fact that a large quantity of data had to be imputed led to “some concerns” in domain 3 (missing outcome data: the algorithm gave the risk of bias as low). In Azuma et al. (2005), we assessed as having “some concerns” in domain 5 (selection of reported result), as the FVC data used was among several outcomes reported, as the primary outcome (Sp02 fall during 6MWT) had not proved significant. All other studies were rated as low risk of bias on all domains.
Qualitative review of studies
In several of the trials, co-administration of medication previously thought to have efficacy in IPF was noted. In the study by Huang et al. , all patients received N-acetyl cysteine (NAC) 600 mg three times daily. Evidence from an adequately powered RCT showed no effect of NAC on rate of FVC decline or mortality in IPF, so it is unlikely to have altered or biased the efficacy of anti-fibrotic therapy .
In the TOMORROW and INPULSIS trials, up to 15 mg daily of prednisolone was permitted: its use was well balanced between the treatment and control groups in all those trials, with about 21% of the study population on prednisolone therapy. In the INPULSIS trials, after 6 months, treating physicians could increase the dose of prednisolone, or add azathioprine, ciclosporin or cyclophosphamide: the numbers in which this happened are not given in the paper or its supplement.
In the study by Azuma et al., azathioprine was not permitted, but prednisone up to a dose of 10 mg per day was permitted: the frequency of oral corticosteroid use was not stated in the paper, but it is clarified that acute exacerbations were treated with high dose corticosteroids. In the study by Taniguchi et al., the same entry criteria were used, and this time the paper states that 8% of the pirfenidone group and 5% of the control group were on oral corticosteroids. In the CAPACITY trials, “other IPF therapy” was excluded. In Huang et al., prednisone up to 15 mg/day was permitted, and was a past or current medication, possibly also with azathioprine, in 74% of the pirfenidone group and 68% of the control group. In the ASCEND trial, no patients were receiving azathioprine, and prednisone was only permitted if required for conditions other than IPF: the number to whom that applied is not stated, but it is reasonable to suppose it was very low. In summary, of the studies of nintedanib in IPF, about 21% of the study population was on prednisolone up to 15 mg daily. Of the studies of pirfenidone in IPF, the majority of subjects were in the CAPACITY and ASCEND trials, where prednisolone was not permitted, so any effect of anti-inflammatory therapy is likely to be nugatory.
The above studies were in IPF. Those below are the four studies in non-IPF, where anti-inflammatory therapy was administered to some patients in all the studies reviewed. In the study by Maher et al., subjects were stated to be not on oral corticosteroids or azathioprine, and 5% were on mycophenolate in both pirfenidone and control groups. In the study by Behr et al., for the pirfenidone group 27% were on monotherapy with oral corticosteroids, 36% were on combination therapy with oral corticosteroids and another agent, usually azathioprine (17% of the treatment group) or mycophenolate, with similar figures for the control group. In the INBUILD study, up to 20 mg prednisolone was permitted: at baseline, 17.2% of subjects on nintedanib and 17.8% of control subjects were on at least one anti-inflammatory drug, mainly non-biologics such as methotrexate, with less than 1% on prednisolone and 0% on azathioprine. Over the course of the study period, anti-inflammatory therapy was added or altered in 10.8% of the nintedanib group and 21.1% of the control group (mainly oral corticosteroids). In the Distler study of patients with Systemic Sclerosis, 48.3% of the nintedanib group were on mycophenolate, and 8% were on methotrexate, with similar figures for both drugs in the placebo group. Patients who had recently received oral corticosteroids or azathioprine were excluded.
Effects of antifibrotics on standardised changes in FVC for IPF and non-IPF considered together
A meta-analysis was performed on included trials assessing the effect of antifibrotic therapy on the rate of decline in FVC, using a standardised effect size to allow comparison of studies using different metrics for the rate of decline in FVC. The total number of subjects in trials in IPF was 2872 (of whom 1564 took the active medication), and the total number of subjects in trials in non-IPF was 1292, of whom 647 took active medication. Of the studies in IPF, 3 were with nintedanib and 6 were with pirfenidone. Of the studies in non-IPF, 2 were with nintedanib and 2 were with pirfenidone (Table 1). The forest plot is shown in Fig. 2.
The overall mean effect size is − 0.306 (SE 0.033), p < 0.001. The Cochrane Q is 13.9, df 12, p = 0.306, so no statistically significant heterogeneity was shown. A sensitivity analysis for the mean effect size was performed using one study removed and re-analysis: the mean effect size showed little variation, with a range of − 0.286 to − 0.328. A funnel plot of the studies did not show any evidence of publication bias (Fig. 3).
A further analysis was performed between two groups of studies (those on IPF and those on non-IPF) using a mixed effects model with the difference between the two groups treated as a fixed effect. The forest plot for that analysis is shown in Fig. 4. The mean effect size in the IPF studies was − 0.305 (SE 0.043) (p < 0.001) and in the non-IPF studies the figures were − 0.307 (SE 0.063) (p < 0.001). The Q statistic for the group difference was 0.001, df 1, with p = 0.979: there was no evidence of any difference between the two groups for standardised rate of FVC decline.
Although when all studies are considered together, no significant heterogeneity was shown, we can analyse the data as if there were heterogeneity, and calculate a prediction interval for the distribution of true effects: this gives an interval for the true effect size (disregarding sampling error) which would be found in 95% of studies from a comparable universe of similar studies, using either pirfenidone or nintedanib, and performed in patients with IPF or non-IPF. That interval, expressed as a standardised difference between treatment and control groups is − 0.18 to − 0.43.
Effects of nintedanib and pirfenidone analysed separately on FVC change for IPF and non-IPF considered together
A further analysis was performed between two groups of studies, namely those where nintedanib was the treatment and those where pirfenidone was the treatment. This was performed over all studies in the meta-analysis, so both IPF and non-IPF. Again, this used a mixed effects model with the difference between the two groups treated as a fixed effect. The forest plot for that analysis is shown in Fig. 5. The mean effect size in the nintedanib studies was − 0.340 (SE 0.052) (p < 0.001) and in the pirfenidone studies the figures were − 0.266 (SE 0.045) (p < 0.001). The Q statistic for the group difference was 1.139, df 1, with p = 0.286: there was no statistically significant evidence of any difference between the two groups by anti-fibrotic treatment given.
Effects of antifibrotics on all-cause mortality in patients with IPF and non-IPF considered together
A meta-analysis of 9 studies where all-cause mortality was stated in the publications used in this paper was performed. The forest plot of that analysis is given in Fig. 6. The overall mean risk ratio for all-cause mortality was 0.701 in favour of anti-fibrotic therapy (confidence intervals 0.539 to 0.911), p = 0.008. The Q value was 5.613, df 8, p = 0.69, so no evidence of heterogeneity was evident. A further analysis was performed grouping the studies by disease type: IPF (n = 5) versus non-IPF (n = 4). The estimate for risk ratio (95% interval) for the IPF studies was 0.637 (0.469 to 0.866), p = 0.004. The estimate for risk ratio (95% interval) for the non-IPF studies was 0.908 (0.547 to 1.508), p = 0.71. An all-cause mortality benefit in the non-IPF group was not demonstrated.