Assessing the relative efficacy of interleukin-17 and interleukin-23 targeted treatments for moderate-to-severe plaque psoriasis: A systematic review and network meta-analysis of PASI response

Introduction New generation biologics, including interleukin (IL)-17 and IL-23 inhibitors, have delivered higher rates of skin clearance than older treatments in head-to-head studies. However, studies comparing these new biologics directly to one another are limited. Objectives To compare the short-term efficacy of available (or imminently available) biologic and non-biologic systemic therapies for treating patients with moderate-to-severe plaque psoriasis. Methods A systematic review was undertaken to identify randomised controlled trials evaluating biologic treatments, apremilast and dimethyl fumarate. MEDLINE, MEDLINE In-Process, Embase and the Cochrane Library were searched from the 1st January 2000 to 22nd November 2018. A Bayesian network meta-analysis (NMA) using a random-effects multinomial likelihood model with probit link and meta-regression to adjust for cross-trial variation in placebo responses compared the efficacy of interventions at inducing different levels of Psoriasis Area and Severity Index (PASI) response during the induction period. A range of sensitivity analyses was undertaken. Results Seventy-seven trials (34,816 patients) were included in the NMA. The base-case analysis showed that all active treatments were superior to placebo. IL-17 inhibitors, guselkumab and risankizumab were found to be more efficacious than tildrakizumab, ustekinumab, all TNF inhibitors and non-biologic systemic treatments at inducing all levels of PASI response. In addition, brodalumab, ixekizumab and risankizumab were significantly more efficacious than secukinumab; no significant difference was found in the comparison with guselkumab. The greatest benefit of brodalumab, ixekizumab, guselkumab, and risankizumab was seen for PASI 90 and PASI 100 response. Results were consistent across all analyses. Conclusions In the NMA brodalumab, ixekizumab, risankizumab and guselkumab showed the highest levels of short-term efficacy. There were differences in efficacy between treatments within the same class. Longer-term analyses are needed to understand differences between these drugs beyond induction in what is a life-long condition.


Introduction
Psoriasis is a chronic, immune-mediated skin disease which affects approximately 100 million people worldwide [1]. It is estimated that in most developed countries between 1.5 and 5% of the population suffer from psoriasis, with the incidence highest in people in their twenties, thirties and sixties [2]. It affects men and women of all ages, although it is rare in children. The most common type is plaque psoriasis (psoriasis vulgaris), characterized by patches of erythema covered in plaques that can range in colour from silvery-white to yellow [3]. Approximately 20% of patients have moderate-to-severe disease [4,5]. Psoriasis may affect patient's self-esteem and lead to low quality of life, comparable to that of patients with heart disease or diabetes [6-10]. Many factors must be considered when selecting the best treatment for patients with plaque psoriasis.
Biologic therapies are generally licensed for treatment of patients eligible for systemic therapy; however, in practice they may be considered after other therapies have failed. For example, in the UK initial management of such patients may involve conventional systemic therapies such as methotrexate, cyclosporine, or acitretin [11]. Only patients with moderateto-severe plaque psoriasis who do not respond to, or are intolerant of conventional systemic treatments are offered biologic systemic therapies, the phosphodiesterase 4 (PDE4) inhibitor apremilast or dimethyl fumarate (DMF) [4,[12][13][14]. In other countries, modern biologics may be prescribed earlier in the treatment pathway.
The first biologics used in psoriasis were anti-tumour necrosis factor (TNF) therapies, adalimumab, etanercept and infliximab [15]. Although their introduction transformed the treatment of the disease [16], in the majority of patients they do not provide complete skin clearance [17]. In addition, it is common for patients to discontinue anti-TNF treatment due to loss of response and adverse effects [18][19][20]. In 2009, the biologic therapy ustekinumab, which targets the interleukin (IL)-12/23 pathway, became available [21].
More recently, biologics targeting the IL-17 pathway have been introduced into clinical practice. The three approved IL-17 pathway inhibitors are secukinumab and ixekizumab, both IL-17A inhibitors, and brodalumab, which targets the IL-17-receptor A (IL-17RA) [22]. These treatments have been shown to be efficacious and safe and offer higher rates of complete or almost complete clearance compared to TNF inhibitors and ustekinumab [23][24][25]. Another class of systemic drugs potentially offering similarly high complete clearance rates target IL-23. These include guselkumab, tildrakizumab, and risankizumab [26][27][28][29]. Although data from from study investigators were not sought. The methodological quality of included studies was assessed by one independent reviewer and checked by another using the concise critical appraisal checklists in the National Institute for Health and Care Excellence (NICE) Single Technology Appraisal user guide [41].

Network meta-analysis of PASI response
A hierarchical Bayesian NMA was used to combine results of all relevant studies allowing for all data from direct and indirect comparisons to be synthesised into a single set of effect sizes. The methods followed the recommended best practice for evidence synthesis [42].
The outcome of interest was the proportion of patients achieving each level of PASI response (PASI 50, PASI 75, PASI 90 and PASI 100) at the end of the induction phase (between 10 and 16 weeks). Data were analysed using an ordered probit model designed to estimate probabilities of achieving each level of PASI response [42]. PASI 50 responses in placebo arms of the included studies were utilised to inform the baseline event rates [43]. PASI response was modelled as a discrete dependent variable that takes ordered multinomial outcomes (PASI 50, PASI 75, PASI 90 and PASI 100). Cross-trial variation in placebo response rates has been observed in psoriasis [44] and is an indicator of between-study heterogeneity, which can introduce bias to synthesised comparisons of treatment outcomes [45][46][47]. Randomised trials in moderate-to-severe psoriasis vary by design, eligibility criteria, time to primary endpoint and concomitant emollient use, all of which could contribute to differences in placebo responses and consequently to relative effect sizes. Adjusting for differences in the placebo arms may account for heterogeneity between trials and could improve the model fit [42,[48][49][50]. This approach has previously been used in other NMAs comparing the efficacy of biological treatments for moderate-to-severe psoriasis [38,51] and has been recommended by NICE in recognition of the variability in placebo response across trials [52]. Analyses adjusting and not adjusting for cross-trial variation in placebo-arm responses were carried out using randomeffects assumptions and assessed for their goodness of fit. In accordance with NMA guidelines [49], fit was informed by the statistical significance of the regression coefficient and whether there was a reduction in between-trial heterogeneity. The model with the best fit was used to draw conclusions. Inconsistency between the direct and indirect evidence was assessed using a random-effects unrelated mean effects model [53]. Full details of the model used are shown in S2 Text.
WinBUGS version 1.4 [54] was used to perform all statistical analyses, using non-informative priors. After an initial burn-in of at least 20,000 simulations, convergence was confirmed through visual inspection the Brook-Gelman-Rubin diagnostic and history plots. Sampled parameters were then estimated using 20,000 simulations on three chains. Results were calculated as the absolute probabilities of response for each treatment and as risk ratios (RRs) for every pairwise comparison at each level of PASI response. Point estimates reflecting the median value are presented, along with 95% credible intervals (95% CrI), reflecting the range of true effects with 95% probability. Sensitivity analyses were performed to test the impact of excluding trials reporting <5% biologic exposed patients, using a different timepoint (12 weeks rather than 16 weeks) for secukinumab, and excluding studies with fewer than 50 patients per arm. PASI data inputs are reported in S3 Table.

Identification of trials
As shown in Fig 1, electronic database searches identified 6643 articles. A further fourteen were identified through searching conferences and reference lists of included studies, SRs and NMAs. After the removal of duplicates, 5047 titles and abstracts were screened, and subsequently 496 full texts were assessed for eligibility. Three-hundred and sixty-nine full text articles were excluded based on the population (n = 24), intervention (n = 42), study design (n = 174), outcomes (n = 105) and as duplicates (n = 24). References of excluded studies can be found in S4 Table. Eighty-three RCTs (reported in 127 articles) were included in the SR and 77 (including 34,816 patients) in the NMA. The trials not included in the NMA did not provide results at all [55][56][57] or in an appropriate format [58], or included a comparator that would not connect relevant interventions in the network [59,60].

Study and patient characteristics
The main details of trials included in the NMA are summarised in Table 1  Assessing the relative efficacy of IL-17s and 1L-23s for psoriasis: A systematic review and NMA Assessing the relative efficacy of IL-17s and 1L-23s for psoriasis: A systematic review and NMA

Risk of bias
Overall, the risk of bias for most of the studies was judged to be low (details reported in S5 Table). Studies generally reported appropriate randomisation and allocation concealment. Patient characteristics appeared similar between treatment arms at baseline. Most of the studies were double-blind, however in some cases only investigator blinding was undertaken. Losses to follow-up appeared to be balanced between study arms, except for PRIME [79], where proportionally more patients were lost to follow-up in the FAE arm compared to secukinumab; Ohtsuki 2018 [83], where more patients dropped out of the placebo arm compared to guselkumab arms; and a phase II certolizumab trial [101,133], where drop-outs from the placebo arm were much higher than from certolizumab arms. The risk of reporting bias was considered low in all studies. Most studies used an intention-to-treat or modified intention-totreat approach, including all randomised patients. There was, however, variation in the method used to handle missing data: forty-one (53%) studies used non-responder imputation, 18 (23%) used last observation carried forward, four (5%) used multiple imputation methods, and the remaining fourteen (18%) did not report the method used.

Model fit
Goodness of fit was assessed for the model with placebo-adjustment compared to the model without adjustment. The estimated placebo arm adjustment coefficient was statistically significant (-0.676, 95% CrI: -0.824 to -0.507), indicating that the adjustment reduced unexplained heterogeneity and improved the model. In addition, the 95% CrI of the random effect, τ, was estimated to be from 60.4 to 271.9 in the adjusted model compared with 52.5 to 466.8 in the unadjusted model. The narrowing of the 95% CrI in the adjusted model demonstrates a reduction in the between-study heterogeneity being captured by the adjustment coefficient. No significant inconsistency was identified in the base-case or sensitivity analysis networks.  Table 2, the placebo-adjusted base-case analysis indicated all therapies to be superior to placebo. In comparison with placebo, licensed doses of ixekizumab, and brodalumab along with risankizumab and guselkumab have shown the greatest treatment effects at inducing all levels of PASI response. The effects of these treatments were particularly pronounced for PASI 100, for which the proportion achieving complete clearance was more than 350 times higher than with placebo. Compared with placebo, the number needed to treat (NNT) to achieve complete clearance was the lowest for ixekizumab (NNT = 3), risankizumab (NNT = 3), brodalumab (NNT = 3) and guselkumab (NNT = 3). Etanercept, apremilast and DMF were associated with the lowest probabilities of response and highest NNTs of 28, 50 and 55, respectively.

IL-17s versus IL-23s and IL-12/23
As shown in Fig 3, treatment with the IL-17RA brodalumab, the IL-17A ixekizumab and the IL-23s risankizumab and guselkumab produced similar proportions of patients achieving each level of PASI response, with any differences failing to reach statistical significance (PASI 75, 90 and 100 shown in Table 3). In the analysis risankizumab was significantly more efficacious than secukinumab, as were brodalumab and ixekizumab. Median point estimates suggest that guselkumab may be slightly more efficacious than secukinumab, but differences were not statistically significant. The NMA also showed all IL-17 targeted therapies, as well as guselkumab and risankizumab to be more efficacious than tildrakizumab, ustekinumab, adalimumab and certolizumab, which were all more efficacious than etanercept, apremilast and DMF. Risk ratios for each pairwise comparison at each level of PASI response are presented in S6 Table. In terms of NNT compared to ustekinumab, five patients would need to be treated with brodalumab, ixekizumab or risankizumab, six with guselkumab and eight with secukinumab for one more patient to achieve complete clearance. Compared to secukinumab, eleven, twelve and thirteen patients would need to be treated with ixekizumab, risankizumab and brodalumab, respectively, in order to generate an additional PASI 100 responder.

Additional analyses
The analysis that did not adjust for cross-trial variation in placebo responses produced results similar to the placebo-adjusted analysis, though the model fit was poorer, indicating that the results are at risk of heterogeneity-induced bias. For completeness, details of this analysis can be found in S7 Table. Thirty  [131]. For all other comparators, the results were similar to those of the placebo adjusted base-case. Details of this analysis can be found in S8 Table. A sensitivity analysis was carried out to test the effect of using 12-week rather than 16-week outcomes from three secukinumab trials ( To test for small study bias, a further sensitivity analysis was carried out excluding the eighteen Assessing the relative efficacy of IL-17s and 1L-23s for psoriasis: A systematic review and NMA trials with fewer than 50 patients per trial arm [73, 84, 85, 100, 134-147]. There were no major differences between the results of these sensitivity analyses and the base-case analysis.

Discussion
Our NMA evaluating PASI response among licensed treatments for moderate-to-severe plaque psoriasis included 77 RCTs and a total of 34,816 patients. Data were available on all interventions set out in the inclusion criteria. The base-case analysis showed that all active treatments were superior to placebo. A comparison of PASI response between the classes of drugs with the highest efficacy, IL-17, IL-23 and IL-12/23 targeted treatments, was also undertaken. This showed that IL-17 inhibitors (brodalumab, ixekizumab and secukinumab) were more efficacious than tildrakizumab and ustekinumab and not significantly different from guselkumab. Compared to risankizumab, there was no significant difference in our analysis for brodalumab and ixekizumab, however secukinumab was less efficacious. Comparisons within the IL-17 inhibitor class showed that ixekizumab and brodalumab were not significantly different from one another, but that both were more efficacious than secukinumab. When comparisons were Assessing the relative efficacy of IL-17s and 1L-23s for psoriasis: A systematic review and NMA made among the IL-23 inhibitors, risankizumab and guselkumab were both more efficacious than tildrakizumab, but were not significantly different from one another. The greatest benefits of IL-17 inhibitors and IL-23 inhibitors guselkumab and risankizumab were seen for PASI 90 and PASI 100 response. Since 2017, numerous NMAs have been published comparing the efficacy of various treatments in moderate to severe plaque type psoriasis [23, 24, 38, [148][149][150][151][152][153], and some have been critically appraised in the literature [154]. This largely reflects the dynamic development of the field, both in terms of treatment availability and new clinical trials being carried out. However, none of the currently published NMAs includes all currently licensed treatments in this patient population. In addition to NMAs, several pairwise meta-analyses have been undertaken comparing the efficacy of treatments in moderate-to-severe plaque psoriasis [155][156][157]. However, pairwise meta-analyses can only provide limited information and do not allow for a formal comparison of all available treatments. Valid relative efficacy comparisons between multiple treatments can only be made in an NMA [158,159].
To our knowledge, this NMA provides the most recent and comprehensive comparison of available treatments for chronic plaque psoriasis. Specifically, it is the first to include the recently published clinical trial data for risankizumab and certolizumab pegol. Though many of the most recent clinical trials in moderate-to-severe psoriasis include a head-to-head comparison of targeted therapies, there are still many comparisons that have not been observed. In their absence, the network meta-analytic approach gives the most reliable estimate of comparative efficacy [159]. A median risk ratio higher than 1 indicates that the intervention is more efficacious than the comparator at inducing a given level of PASI response. https://doi.org/10.1371/journal.pone.0220868.t003 Assessing the relative efficacy of IL-17s and 1L-23s for psoriasis: A systematic review and NMA This NMA was based on a systematic review of RCTs evaluating licensed treatment doses. We followed a protocol designed for the systematic review; however, this was not registered online. English-language publications were searched, which has the potential to introduce bias [160,161]; however, we believe this was unlikely to have had a substantial impact on our findings. A recent systematic review [23] that did not apply language restrictions identified no additional studies that would have been relevant to our NMA.
Although we were interested in evaluating licensed treatments, we included trials with any intervention or dose as the comparator. Although other recent NMAs chose to restrict inclusion to licensed doses only [25], we believe that such a broad approach allowed the construction of an extensive network of evidence and ensured that all relevant evidence was included and uncertainty in efficacy estimates was reduced [158].
As we set out to provide the best efficacy estimates to inform clinical practice, our analysis was dose-specific. Some recent NMAs have analysed a broad range of treatments, however the conclusions that can be drawn from these may be limited by their design. In particular, some of these analysis pool together both licensed and unlicensed doses of treatments [23,24]. Although this has the advantage of including all the available information, interpretation of results may be challenging, and may lead to underestimation of the efficacy of treatments available to patients if evidence for subtherapeutic doses is included. Therefore, we analysed different doses separately to provide information we believe to be most relevant to clinical practice. Other recent NMAs [148,149] have analysed a narrower range of treatments which makes a comprehensive assessment of the available treatments for plaque psoriasis impossible and conclusions limited in their relevance.
As our NMA only included data up to 16 weeks obtained from clinical trials, it does not provide an estimate of long-term effectiveness in a trial or real-world clinical setting. Psoriasis is a chronic condition, therefore a better understanding of the comparative effectiveness of biological drugs beyond induction would be beneficial. The randomised evidence is however limited, although some trials have reported 52-week results [30, 35, 36, 65, 69, 75, 174,175] and researchers have synthesised some of these data [176]. Existence of such data will facilitate future evaluation of these treatments using an NMA approach.
A further limitation of the analysis was the inclusion of only PASI response rates. Although these are the most commonly used efficacy outcomes in psoriasis clinical trials, they do not provide a complete understanding of patient wellbeing. Statistical analysis of health-related quality of life (HRQoL) and safety data may also be helpful in clinical decision making. One of the most recent evaluations of HRQoL and tolerability in psoriasis suggested that IL-17 and IL-23 inhibitors provide the most benefit in terms of quality of life, while the occurrence of serious adverse events is not significantly more frequent than in patients treated with placebo [23, 24].
In psoriasis, placebo response rates have been observed to vary considerably between trials [44], reflecting differences in study design and patient characteristics. In line with these observations, we have identified variation in factors such as patient age, baseline PASI scores, or prior exposure to systemic therapy. To account for the potential heterogeneity introduced by these differences, our analysis presented the results of two statistical models: one that adjusted for variation in placebo-arm response and one that did not. Consistent with three previous NMAs [25, 38, 51], the estimated placebo arm adjustment coefficient was statistically significant, indicating that the adjustment reduced unexplained heterogeneity and improved the model. Cameron et al. explored several covariates in meta-regression analyses, including placebo-response adjustment, prior biologic use, body weight, disease duration, age, race and baseline outcome scores. In their analysis of PASI responses, the authors also found the placebo-adjusted model to have the best fit and argued that it likely accounted for multiple effect modifiers simultaneously.
Our results are broadly similar to those of other recently published NMAs [23-25, 38] in showing that all active treatments are superior to placebo and that IL-17 (in particular brodalumab and ixekizumab) and IL-23 inhibitors (in particular guselkumab and risankizumab) are the most efficacious in inducing PASI response. All analyses show that achieving complete and near complete clearance is more likely with these new treatments. We can also compare our results to those of the ECLIPSE study which has recently been reported in Johnson & Johnson press material [37]. Both our analysis and the current reports for ECLIPSE suggest there is no significant difference between secukinumab and guselkumab at the end of a 12-week induction period. Future work should focus on incorporating this new head-to-head comparison once it is published.
We believe our NMA adds to the current body of evidence, evaluating the efficacy of licensed doses of all treatments that are or will soon be available to patients. This NMA is the first to include risankizumab, and the recently licensed certolizumab pegol in its licensed dose. As our searches have been carried out recently, we have also included a number of recent trials that were not previously included in any NMA [72, 73, 79, 81-84].
With PASI 90 and PASI 100 becoming attainable treatment goals, consensus is growing that dermatologists should expect more highly effective therapies for their psoriasis patients [177,178], especially as these have been shown to translate to improved HRQoL [179,180].

Conclusions
The results of the NMA indicated that brodalumab, ixekizumab, risankizumab and guselkumab showed the highest levels of efficacy. It is not possible to make conclusions at drug class level, as there are differences between treatments within the same class, some significant. In our analysis the IL-17RA inhibitor brodalumab and IL-17A inhibitor ixekizumab as well as the IL-23 targeted risankizumab were found to be significantly more efficacious than secukinumab. The NMA showed that across all levels of PASI response, all IL-17 targeted therapies, risankizumab and guselkumab outperformed tildrakizumab and ustekinumab as well as anti-TNFs such as adalimumab, certolizumab pegol and etanercept. These effects were particularly pronounced for PASI 90 and PASI 100. Our results are broadly consistent with previously published NMAs.