Efficacy of non-surgical monotherapies for hidradenitis suppurativa: a systematic review and network meta-analyses of randomized trials

Abstract Objective We determined the relative efficacy of non-surgical monotherapies for hidradenitis suppurativa (HS). Methods Network meta-analyses were conducted to determine treatments’ surface under the cumulative ranking curve (SUCRA) value (i.e. an estimate that ranks efficacy); pairwise comparisons were conducted. Results and conclusions Ten trials were eligible for quantitative analyses; however, all did not have a common endpoint. Outcomes corresponded to pain severity, clinical response, quality of life and abscess count. For pain reduction, infliximab was ranked most efficacious (SUCRA = 94%) compared to bermekimab, anakinra and placebo; infliximab reduced pain more significantly (p < .05) than anakinra and then placebo. For the occurrence of clinical response, bimekizumab had the highest SUCRA (67%) relative to adalimumab, anakinra and placebo; bimekizumab was more efficacious than placebo (p < .05). For the quality of life in mild HS, Botox had the highest SUCRA (94%) compared to adalimumab and placebo; Botox was more efficacious than placebo (p < .05). For reduction in abscess count, oral tetracycline had the highest SUCRA (48%) compared to topical clindamycin and vehicle. Our work—being the first NMA study on non-surgical HS monotherapies—contributes to the comparative effectiveness literature for this condition.


Introduction
Hidradenitis suppurativa (HS) was first described in 1839 by the French surgeon Alfred Velpeau (1,2). It is a chronic and painful inflammatory skin condition of multifactorial etiology. According to the comprehensive literature review by Ingram (2020) (3), the prevalence of this disorder remains unresolved. Given that this condition occurs at sites that are rich in hair follicles, HS was believed to be linked to bacterial infection in the apocrine glands (4,5). However, the current body of evidence supports that HS is primarily an inflammatory disorder initiated by hair follicle dysfunction (1,(6)(7)(8)(9)(10).
While surgical intervention is a treatment option, numerous noninvasive modalities exist (1). Topical and systemic antibiotics have been used to treat HS; newer treatment modalities include immunomodulatory therapies such as inhibitors of tumor necrosis factor-alpha (TNF-a), interleukin-1 (IL-1) and selective phosphodiesterase-4 (PDE-4) (11). Adalimumab and infliximab are well-known TNF-a inhibitors whose efficacies have been determined in various randomized trials for HS; apremilast, a PDE-4 inhibitor, and the IL-1 inhibitor anakinra are among the HS therapies that have been investigated more recently (1,11,12). Results from single-arm studies have shown that interleukin-17 antagonists, such as brodalumab and secukinumab, are effective in treating HS (13)(14)(15).
There is a paucity of head-to-head evidence for the relative efficacy of non-surgical treatments for HS (16). We systematically reviewed the literature to identify randomized trials that evaluated the efficacy of non-surgical monotherapies for HS; data from these trials were used to determine the relative efficacy of such therapies using network meta-analyses (NMAs).
Data were extracted from tables, figures and textual content; extracted information included baseline characteristics and outcome data. More information pertaining to the data extraction process is provided in Appendix S3 (see Supporting Information). The quality of evidence was evaluated using the van Tulder scale (18).
Each network was depicted with a network plot, which can be defined as a diagram of nodes and edges. Each node represents a treatment; an edge, the line between two nodes, corresponds to a pairwise comparison of efficacy from direct evidence. Statistical consistency would be assessed if networks had closed loops (19).
Our NMAs were arm-based and were conducted using a Bayesian random-effects model with uniform priors; for each NMA, four Markov Chain Monte Carlo (MCMC) chains were used where each had 20,000 iterations. Pairwise comparisons of interventions within a given network were presented in league tables. For every network, we computed treatments' surface under the cumulative ranking curve (SUCRA) value; a treatment's SUCRA corresponds to their overall rank of efficacy. Statistical analyses were conducted with the software RStudio; the gemtc and BUGsnet packages were used (19,20). In all analyses, the alpha was set to 5%.

Results
Ten randomized trials were eligible for quantitative analyses ( Figure 1) (21)(22)(23)(24)(25)(26)(27)(28)(29); with the available data across these studies, we were able to construct six networks and details thereofincluding endpoint definition-are presented in Table 1. None of the 10 trials had a common measure of outcome; however, some studies had some endpoints in common.
In addition to controls (i.e. placebo), eight treatment modalities were considered across the six networks (adalimumab, anakinra, bimekizumab, botulinum toxin type b (Botox-B), clindamycin (topical), infliximab, MABp1 (also known as bermekimab), and tetracycline (oral). The common endpoints pertained to pain severity, clinical response, quality of life and abscess count ( Table 1). Characteristics of the eligible studies are detailed in Table 2; the earliest one was published by     Clemmensen (21). Quantitative evaluation of evidence quality is presented in Table 3.
To make each network homogenous, we used studies across which outcome(s) were measured at a given time point, and in patients with similar disease severity. Trials that met our eligibility criteria, but whose data were not included in our quantitative analyses, are listed in Table 4. Most of these studies were excluded because the endpoints therein were incongruent with those of our networks. For instance, the three endpoints Fajgenbaum et al. (2020) (36) used were 'patient satisfaction after 14 days', 'pain reduction after 5 days', and 'mean duration to the resolution of lesions' (Table 1). Ineligible (i.e. observational) studies that investigated the efficacy of HS monotherapies are listed in Table 5. As per the published literature, a repertoire of modalities were investigated with observational study designs (Table 5) Plots of the six networks are presented in Figure 2. Most of our networks did not have a closed loop, so analyses of inconsistency could not be undertaken. However the transitivity assumption was not violated.

Network meta-analyses
In Figure 3, the SUCRA values for treatments of each network are depicted; league tables for all six networks are presented in Figure 4.

Clinical response
Network IIa pertained to the simultaneous comparison of adalimumab, anakinra, bimekizumab and placebo in terms of occurrence of a clinical response-as per the Hidradenitis Suppurativa Clinical Response (HiSCR)-at 12 weeks of follow up-for moderate to severe HS (Table 1). A HiSCR is defined as the occurrence of an at least 50% reduction in the count of inflammatory lesions from baseline, in addition to the absence of new fistulas and abscesses (41).
Meta-analyzing data from trials that investigated bermekimab (formerly known as MABp1) and adalimumab in one network would arguably violate the transitivity assumption as subjects' eligibility for receiving bermekimab was predicated on their non-eligibility for adalimumab (27). Hence, we created 'network IIb' which-for the same outcome as network IIa-simultaneously compared anakinra, bermekimab and placebo (Table 1). Bimekizumab has the highest SUCRA in network IIa (67.38%), while bermekimab had the highest SUCRA in network IIb (82.8%) (Figure 3). Pairwise analyses showed that a HiSCR, on average, was significantly more likely to occur with adalimumab than with placebo (odds ratio¼ 3.3, p < .05) (Figure 4).

Pain severity
Network Ia (Figure 2.) pertained to the simultaneous comparison of infliximab, bermekimab, anakinra and placebo in terms of pain reduction at 8 weeks of follow up for moderate to severe HS (Table 1); pain was evaluated using the visual analogue scale Was the drop-out rate described and acceptable? Was the timing of the outcome assessment in all groups similar? Did the analysis include an intention-to-treat analysis? b This study was sponsored by UCB Biopharma SRL, and the date of the final analysis was March 28 2019. c MABp1 is also known as bermekimab (6). d Two trials constituted this study; the sample size of one trial was 307, and that of the other was 326.
Network Ib pertained to the simultaneous comparison of adalimumab, anakinra and placebo in terms of reduction in VAS pain scores at 12 weeks of follow up for moderate to severe HS (Table 1). Adalimumab had the highest SUCRA (67.4%) ( Figure  3), and pairwise analyses found no significant comparison ( Figure 4).

Quality of life
Network III pertained to the simultaneous comparison of Botox-B, adalimumab and placebo in terms of improvement in Dermatology Life Quality Index (DLQI) scores at 12 weeks of follow up (Table 1); the DLQI is a widely used measure for quality of life (42). Data for this network were obtained from Miller et al. (24) and Grimstad et al. (28); the former study included patients with Hurley stages II-III HS, while the latter included their counterparts with Hurley stage I -III HS (24,28). Hence, we conducted network meta-regression to adjust for disease severity; we created a variable for proportion of individuals who had Hurley stage I (where 1.0 corresponded to all individuals having stage I, 0 corresponds to all individuals having stage II-III).

Abscess reduction
Network IV pertained to the simultaneous comparison of oral tetracycline, topical clindamycin and vehicle (topical) in terms of reduction in mean number of abscesses at 4 weeks for moderate HS (Table 1). Oral tetracycline had the highest SUCRA (47.6%) (Figure 3), and pairwise analyses found no significant comparisons (Figure 4). IFX-1, placebo Results not finalized a 'Results unavailable': results were not available on the ClinicalTrials.gov website. b 'Incongruent endpoints': The endpoints were not concordant with that of any of our six networks (description of the networks, including endpoint definition, are provided in Table 1).

Discussion
We determined the relative efficacies of non-surgical monotherapies for the treatment of HS. We found no common endpoint across the 10 trials included in quantitative analyses-a finding that is congruent with the fact that core outcome sets (COSs) for HS were non-existent when majority of these trials were published (17). Since our analyses excluded data from non-randomized studies, our simultaneous comparison did not include HS treatments investigated in single-arm studies, case series and retrospective cohort studies. The current study has some limitations. Firstly, our systematic search excluded evidence from the non-English language literature; most published systematic reviews have this language limitation (43). Secondly, the NMA technique currently does not incorporate quality of evidence into to its estimation of rank probabilities (a treatment's SUCRA is computed from rank probabilities) (44). Clinical decisions are more informed when based on ranking information, evidence quality, and point estimates (e.g. odds ratio, mean difference, etc.) with statistical evidence (i.e. p-values, standard errors, or confidence intervals) (45).
Given that diminished quality of life (QoL) and skin pain are consequences of HS (46), treatments that address pain and wellbeing are extremely important to persons with HS (47). Moreover, the recently published COSs for HS include domains for pain and QoL (17). We determined the relative efficacies of HS treatments on pain management as it was measured using the VAS, a measurement device used for quantifying various subjective experiences (48). The use of this instrument is widespread in studies of other health conditions (49)(50)(51)(52).
We simultaneously compared the efficacy of HS treatments on patients' assessment of their life quality as per the DLQI. Like the VAS, the DLQI has been used as a primary outcome in numerous meta-analyses and has also been evaluated for reliability and validity. Furthermore, the DLQI is not specific for HS (42,48,53,54).
We found that oral tetracycline ranked more efficacious than topical clindamycin for HS management insofar as reducing number of abscesses; this supports clinicians' belief of systemic antibiotics being more effective than their topical counterparts in HS care. This is also congruent with the 2018 British Association of Dermatologists guidelines for HS (55); this guideline recommends treating HS with oral tetracyclines such as lymecycline and doxycycline. North American guidelines also recommend antibiotics as a treatment modality (56). Hendricks et al. (2019) compared nine international HS management guidelines and found that all nine-including the 2018 British Association of Dermatologists HS guidelines-recommend oral tetracycline as a first-line option (57).
Adalimumab is approved by the Food and Drug Administration (FDA) for treatment of HS (58); this TNF-a inhibitor is also approved by the European Medicine Agency (EMA) (59). Of the 10 trials that were included in our quantitative analyses, the one by Kimball et al. (34)-where adalimumab was investigated-had the highest quality score as per the van Tulder scale. Adalimumab has been recommended as a first-line biologic in at least nine clinical guidelines including the 2015 European Academy of Dermatology and Venerology; across these guidelines, infliximab has been recommended as a second-line therapy (57). Guidelines' recommendation of infliximab as a second-line therapy-while adalimumab is a first-line option-can be due to the fact that randomized-controlled trials for infliximab are fewer, in addition to being of smaller sample sizes (57). Various clinical guidelines for HS management-including the 2018 British Association of Dermatologists-suggest that more evidence regarding the efficacy of anakinra is essential for consideration of it as a first-, second-, or third-line treatment option.
We simultaneously compared the efficacy of various non-surgical monotherapies for HS; all the included studies did not have a common outcome; however, various endpoints were common in some trials. Therefore, we were able to determine the relative efficacy of non-surgical HS monotherapies as per endpoints pertaining to pain severity, clinical response, quality of life and reduction in abscess count. For pain reduction, infliximab had the highest SUCRA (94%) compared to bermekimab (47%), anakinra (34%) and placebo (23%); the mean difference in pain reduction for infliximab vs. anakinra, and infliximab vs. placebo were 30 and 33, respectively. For occurrence of clinical response (as per a HiSCR), bimekizumab had the highest SUCRA (67%) when compared to adalimumab (64%), anakinra (49%) and placebo (19%); the odds of having a clinical response was significantly higher with adalimumab than with placebo (odds ratio¼ 3.3, p < .05). For quality of life in mild HS, Botox-B had the highest SUCRA (94%) compared to adalimumab (43%) and  Table 1. The geometry of the network was represented through network plots. Such plots are characterized by nodes and edges (i.e. the lines between two nodes that represent a comparison from direct evidence). Nodes correspond to an intervention. The number of studies contributing to direct evidence is listed in the edges.   Table 1 for more details of the networks). The league tables for Networks IIa and IIb correspond to pairwise comparisons in terms of the log odds ratio (and 95% credible intervals in parentheses). For all the other networks, pairwise comparisons correspond to mean differences (and 95% credible intervals in parentheses). Each cell compares the column against the row. Alpha was set to 5%, and cells with " ÃÃ " correspond to statistically significant results (i.e. p < .05). Abbreviations: Botox-B: botulinum toxin type b. placebo (12%); Botox-B had significantly better-quality scores than placebo (mean difference ¼ 8.40, p < .05). Our work is the first to determine the relative efficacy of HS monotherapies; our findings contribute to the comparative effectiveness literature for HS treatments. Our results support the conduct of future randomized trials with consistent endpoints, and larger sample sizes. Such trials would eventually allow for more statistically powered meta-analyses which, in turn, could permit clinicians to more confidently make therapeutic decisions in HS care.

Disclosure statement
No potential conflict of interest was reported by the author(s).