FormalPara Key Summary Points

Why carry out this study?

As the treatment armamentarium for moderate-to-severe atopic dermatitis (AD) continues to grow, the comparative efficacy of targeted systemic therapies for AD should be updated using the most recent phase 3 data. Network meta-analysis (NMA) is a useful tool for clinicians, payers, and healthcare providers to inform decision-making about various therapies when treating patients with moderate-to-severe AD.

The study analyzed 13 clinical trials for Eczema Area and Severity Index (EASI) improvement ≥ 90% from baseline (EASI-90), EASI improvement ≥ 75% from baseline (EASI-75), ≥ 4-point improvement on Pruritus Numerical Rating Scale from baseline (ΔNRS ≥ 4), and Investigator Global Assessment (IGA) score of 0 or 1 (clear or almost clear) with a reduction of ≥ 2 points from baseline (IGA 0/1) at the primary endpoint (week 12 or 16) and at earlier timepoints.

What was learned from the study?

The study found that upadacitinib 30 mg, upadacitinib 15 mg, and abrocitinib 200 mg appear to remain the most efficacious targeted systemic therapies across 12–16 weeks of therapy.

This NMA builds on work previously published in Silverberg et al., 2022, and continues to suggest that some targeted systemic treatment options provide greater efficacy across key disease domains, such as skin and itch responses. These updated findings can help healthcare providers evaluate the overall efficacy benefit of these treatments when personalizing a patient’s treatment plan. Other factors including safety, benefit–risk, and patient preferences should also be taken into account when personalizing a patient’s treatment plan.

Introduction

Network meta-analysis (NMA) serves as a useful tool for clinicians, payers, and healthcare providers by providing indirect comparisons between treatments when there is a paucity of head-to-head trials, such as in the disease space of moderate-to-severe atopic dermatitis (AD). However, as new therapies continue to be developed and new phase 3 data are made available, existing NMAs can quickly become outdated. As such, the objective of this study was to update the NMA presented in Silverberg et al. [1], which assessed the comparative efficacy of targeted systemic treatment without concomitant topical treatment, by including the latest phase 3 monotherapy data for lebrikizumab [2].

Methods

Data Source and Study Selection Criteria

Data from the two most recently published phase 3 monotherapy trials for lebrikizumab in moderate-to-severe AD, ADvocate1 (NCT04146363) and ADvocate2 (NCT04178967) [3, 4], were included in the analyses along with other eligible phase 3 or 4 randomized placebo-controlled trials identified through a systemic literature review in Silverberg et al. [1]. All eligible trials evaluated the targeted systemic therapies in adults with moderate-to-severe AD who had an inadequate response to treatment with topical corticosteroids (TCS) and/or topical calcineurin inhibitors (TCI), or for whom topical treatments were medically inadvisable. A total of six targeted systemic therapies with doses that are approved or could gain approval in the near future were included in the analyses, including abrocitinib, baricitinib, dupilumab, lebrikizumab, tralokinumab, and upadacitinib. Details of the methodology including data source, search strategy of Patient/population, Intervention, Comparison, Outcomes, and Study design (PICOS) criteria, are outlined in the primary and supplementary materials of Silverberg et al. [1]. Data from upadacitinib trials meeting search criteria were supplied by AbbVie. This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

Outcomes

Efficacy outcomes included the proportion of patients achieving Eczema Area and Severity Index (EASI) improvement ≥ 90% from baseline (EASI-90), EASI improvement ≥ 75% from baseline (EASI-75), ≥ 4-point improvement on Pruritus Numerical Rating Scale from baseline (ΔNRS ≥ 4), and Investigator Global Assessment (IGA) score of 0 or 1 (clear or almost clear) with a reduction of ≥ 2 points from baseline (IGA 0/1) [5–7]. Different doses were considered independent treatment options. Outcomes were evaluated at the primary endpoint timepoint of each trial (week 12 for abrocitinib, week 16 for all other therapies), as well as at weeks 4 and 8. Additionally, EASI-75 and ΔNRS ≥ 4 were compared at week 2, the earliest timepoint for which most treatments reported these outcome. Values for earlier timepoints that were only available in figures were extracted using DigitizeIt software (version 2.5) [8].

Feasibility Assessment

NMA feasibility was assessed per Cope et al. [9]. Network connectivity of included trials was checked. Relevant study and baseline patient characteristics, including potential treatment effect modifiers (age, gender, duration of disease, EASI, IGA, itch), and mean placebo outcomes were compared to identify potential sources of cross-study heterogeneity. [10–12]

Statistical Analysis

NMAs were conducted in a generalized linear model (GLM) framework using Bayesian Markov Chain Monte Carlo (MCMC) simulations with four chains of 100,000 posterior iterations each. Models with additional posterior iterations were considered if a baseline risk-adjusted model did not converge with the default number of iterations but the associated baseline risk-adjustment regression parameter was significant. NMAs were run in JAGS (version 4.3.0) via R statistical software (R Foundation for Statistical Computing, Vienna, Austria; version 4.0.2) using the bnma package (version 1.4.0) [13–16]. Convergence was assessed with the Brooks–Gelman–Rubin method using the Potential Scale Reduction Factor (PSRF). Fixed effects, random effects, and baseline risk-adjusted models were evaluated and NMA consistency assumptions were checked to identify the best-fitting models [16, 17]. Placebo-unadjusted response rates, numbers needed to treat (NNT), odds ratios, and Surface Under the Cumulative RAnking curve (SUCRA) scores were estimated [17]. Statistical significance was determined by examining calculated odds ratio 95% credible intervals excluding 1.

Results

Study Inclusion

A total of 13 unique trials encompassing 7105 patients in 32 arms were included with the following targeted therapy study arms: abrocitinib 100 mg daily, abrocitinib 200 mg daily, baricitinib 2 mg daily, baricitinib 4 mg daily, dupilumab 600 mg followed by 300 mg once every 2 weeks, lebrikizumab 500 mg at weeks 0 and 2 followed by 250 mg once every 2 weeks, tralokinumab 600 mg followed by 300 mg once every 2 weeks, upadacitinib 15 mg daily, and upadacitinib 30 mg daily (Table 1).

Table 1 Overview of studies for the network meta-analysis
Full size table

Network Meta-analysis

All trials were placebo controlled (Fig. 1) and generally comparable based on enrollment inclusion and exclusion criteria and potential treatment effect modifiers (Table 1). Some differences were observed in response rates across placebo arms. Baseline risk-adjusted models were estimated to account for this heterogeneity. Model consistency checks and fit diagnostics supported fixed-effects models as the best fitting, most parsimonious models for all efficacy outcomes evaluated except for EASI-75 at week 4; ΔNRS ≥ 4 at weeks 2, 4, and 8; and IGA 0/1 at weeks 4 and 8 where evidence supported the use of fixed-effects models adjusted for baseline risk.

Fig. 1
figure 1

Network meta-analysis diagram. Network displayed above is for primary endpoint analysis. The ΔNRS ≥ 4 network of the week 2 analysis is identical to the above except without ECZTRA 1 and ECZTRA 2 (tralokinumab) as these trials did not report ΔNRS ≥ 4 at week 2. The EASI-75 network of the week 2 analysis is identical except with pooled SOLO 1 and SOLO 2 data as reported in Thaçi et al. [18]. EASI Eczema Area and Severity Index, EASI-75 EASI improvement ≥ 75% from baseline, NRS Numerical Rating Scale, ΔNRS ≥ 4 Pruritus Numerical Rating Scale reduction of ≥ 4 points from baseline

Full size image

Results at the primary endpoint timepoint are presented in Table 2. Following the inclusion of lebrikizumab, the results largely aligned with those observed in Silverberg et al. [1]. The odds ratios of all targeted therapies remained statistically more efficacious relative to placebo across all outcomes. The EASI-90 response rate was highest for upadacitinib 30 mg, followed by abrocitinib 200 mg, upadacitinib 15 mg, dupilumab, and abrocitinib 100 mg, with the newly added lebrikizumab ranked seventh; in EASI-75, the response rate was highest for upadacitinib 30 mg, followed by abrocitinib 200 mg, upadacitinib 15 mg, lebrikizumab, dupilumab, and abrocitinib 100 mg (Fig. 2). For ΔNRS ≥ 4, upadacitinib 30 mg also had the highest response rate, followed by abrocitinib 200 mg, upadacitinib 15 mg, lebrikizumab, and dupilumab (Fig. 3). Finally, the IGA 0/1 response rate was highest for upadacitinib 30 mg, followed by upadacitinib 15 mg, abrocitinib 200 mg, dupilumab, and lebrikizumab (Fig. 3). Response rates, NNT, and SUCRA score rankings for all efficacy outcomes at the primary endpoint timepoint are also shown in Figs. S1, S2, and S4, respectively. Response rate rankings were the same as NNT and SUCRA score rankings.

Table 2 Odds ratios versus placebo, NNT, response rate, and SUCRA scores, at week 2 and primary endpoint timepoint (NMA fixed-effects results)
Full size table
Fig. 2
figure 2

EASI-75 and EASI-90 absolute response rate estimates for moderate-to-severe atopic dermatitis (primary endpoint timepoint).  EASI Eczema Area and Severity Index, EASI-75 EASI improvement ≥ 75% from baseline, EASI-90 improvement ≥ 90% from baseline

Full size image
Fig. 3
figure 3

IGA 0/1 versus ΔNRS ≥ 4 absolute response rate estimates for moderate-to-severe atopic dermatitis (primary endpoint timepoint). IGA Investigator Global Assessment for Atopic Dermatitis, IGA 0/1 score of 0 or 1 (clear or almost clear) and reduction of ≥ 2 points from baseline, ΔNRS ≥ 4 Pruritus Numerical Rating Scale reduction of ≥ 4 points from baseline

Full size image

The odds ratios of pairwise comparison across all outcomes at the primary endpoint timepoint were computed, with statistical differences observed between some targeted therapies (Table S2). For EASI-90, EASI-75, and ΔNRS ≥ 4, upadacitinib 30 mg remained statistically more efficacious than all other therapies except for abrocitinib 200 mg. For IGA 0/1, upadacitinib 30 mg remained statistically more efficacious than all other therapies. Upadacitinib 15 mg was statistically more efficacious than abrocitinib 100 mg (IGA 0/1), baricitinib 2 mg (all outcomes), baricitinib 4 mg (EASI-90, EASI-75, IGA 0/1), dupilumab (EASI-90, EASI-75, IGA 0/1), lebrikizumab (EASI-90, EASI-75, IGA 0/1), and tralokinumab (all outcomes). Abrocitinib 200 mg was statistically more efficacious than abrocitinib 100 mg (all outcomes), baricitinib 2 mg (all outcomes), baricitinib 4 mg (EASI-75), dupilumab (EASI-75), lebrikizumab (EASI-90, EASI-75), and tralokinumab (all outcomes). No statistical differences were observed between upadacitinib 15 mg and abrocitinib 200 mg across the outcomes examined.

The results at week 2 for EASI-75 and ΔNRS ≥ 4 are presented in Table 2. The week 2 response rate in EASI-75 was highest for upadacitinib 30 mg, followed by upadacitinib 15 mg, abrocitinib 200 mg, and baricitinib 4 mg. The response rate in ΔNRS ≥ 4 was again highest for upadacitinib 30 mg, followed by abrocitinib 200 mg, upadacitinib 15 mg, and abrocitinib 100 mg. Dupilumab and the newly added lebrikizumab were found to rank lower in EASI-75 (seventh and eighth respectively) and ΔNRS ≥ 4 (sixth and eighth respectively). Odds ratios indicate that all therapies at week 2, except for tralokinumab in EASI-75 (data not available for ΔNRS ≥ 4), were statistically more efficacious than placebo (Table S3). Results at other early timepoints are available in the Supplementary Materials (Fig. S6, Table S5–S6).

Discussion

This NMA, updated with recent phase 3 data of placebo-controlled trials, found that monotherapy with upadacitinib 30 mg continued to have the highest efficacy at the primary endpoint timepoint across outcomes, followed by abrocitinib 200 mg (second in EASI-90, EASI-75, and ΔNRS ≥ 4, third in IGA 0/1) and upadacitinib 15 mg (second in IGA 0/1, third in the other outcomes). The newly added lebrikizumab in general had similar efficacy response rates with dupilumab, abrocitinib 100 mg, and baricitinib 4 mg in this NMA. Baricitinib 2 mg and tralokinumab were generally ranked lower across all efficacy outcomes. All targeted therapies were superior to placebo at the primary endpoint timepoint. In early timepoints assessed, upadacitinib 30 mg remained the top-performing treatment, generally followed by upadacitinib 15 mg and abrocitinib 200 mg across weeks 2, 4, and 8, as assessable by outcome. It is important to note that these presented treatment efficacies are to be understood as meta-estimates derived from estimated relative treatment effects (i.e., estimates produced by NMA and not by a randomized head-to-head trial). These estimates are also impacted by differences in placebo efficacy (i.e., baseline risk) across trials, though baseline risk-adjusted NMA models for many outcomes and timepoints did not produce a significant baseline risk regression parameter thus not suggesting the need to adjust for placebo efficacy.

Serving as an update to Silverberg et al. [1] this analysis is the first to our knowledge to incorporate recently available phase 3 monotherapy data for lebrikizumab in a NMA. While multiple NMAs evaluating efficacy and/or safety across treatments in moderate-to-severe AD may be found in the recent literature, several key methodological considerations should be highlighted when interpreting results across different NMA publications. Trial inclusion criteria is one such consideration. Some NMA publications include both phase 2 and phase 3 trials [19–22]. While inclusion of phase 2 trials allows for a wider breadth of data inputs in the NMA, inclusion may also introduce greater variability in estimated treatment effects owing to the smaller sample sizes generally present in phase 2 trials. The NMA presented in this manuscript included only phase 3 trials to minimize potential treatment effect heterogeneity. Other published NMAs may also limit trial inclusion to only specific classes of targeted systemic therapies, such as only Janus kinase (JAK) inhibitors [21, 23]. To capture therapies and associated doses most likely to be of clinical relevance, the NMA presented here included classes of targeted systemic therapies approved for or which could gain approval in the near future (namely, lebrikizumab as assessed in associated phase 3 trials) for treatment of moderate-to-severe AD. Some NMA publications additionally synthesize data across monotherapy (i.e., targeted systemic therapy or placebo without concomitant TCS or TCI use) and combotherapy (i.e., targeted systemic therapy or placebo with concomitant TCS or TCI use) trials [19, 21, 24]. Including both monotherapy and combotherapy trials in NMA without further adjustment may introduce heterogeneity in patient characteristics. Furthermore, inclusion of both monotherapy and combotherapy trials may conflict with the core NMA assumption of a common comparator to connect therapies within the network, as the efficacy of placebo arms across trials may notably differ with and without the concomitant use of TCS or TCI. In contrast, the analyses presented within this manuscript limited trial inclusion criteria to placebo-controlled monotherapy trials and considered baseline risk-adjusted models to mitigate these potential concerns.

While methodological differences across NMA publications are notable, conclusions across publications are generally similar, with upadacitinib 30 mg, upadacitinib 15 mg, and abrocitinib 200 mg consistently identified as the most efficacious therapies.

Recent clinical literature suggests establishing a treat-to-target framework for AD, as utilized in other immunological disease spaces, may be beneficial in optimizing patient outcomes [25]. To develop such a treat-to-target framework, consistent and optimal treatment targets then need to be established to guide treatment decisions. Research from panel consensus of clinicians and patient qualitative interviews suggests use of higher thresholds of response to achieve minimal disease activity (MDA), such as EASI-90 and peak/worst pruritus NRS 0 or 1 [26, 27]. The MDA concept seeks to establish a shared decision making approach of clinician-reported and patient-reported targets which addresses patient-centric treatment goals. The EASI-90 outcome assessed in this NMA serves as an optimal target of a clinician-reported outcome and may be of particular interest for medical decision-makers. Additional treatment efficacy data on other MDA criteria (e.g., peak/worst pruritus NRS 0 or 1) are required to enable a more complete comparison of efficacy in achieving optimal treatment targets across therapies in AD; this may be a promising area for future research.

Rapidity of treatment effect may also be an important factor in treatment selection. Research has shown that patient satisfaction was higher for treatments with faster onset of symptom improvement [28]. Therapies that provide rapid and greater efficacy across multiple disease domains and outcomes may provide better alignment with the expectations of personalized treatment and benefit a greater number of patients. From this updated NMA, as well as the previous work presented in Silverberg et al. [1], there appears to be a potential class effect in rapidity of response, with JAK inhibitors displaying numerically greater response rates at week 2 in assessed outcomes than biologic therapies. While NMA results are sensitive to baseline risk at the assessed early timepoints and should be interpreted with caution, upadacitinib 30 mg is consistently observed as the therapy with the greatest response rate, generally followed by other JAK inhibitors, which in turn are generally followed by biologic therapies regardless of model selection. These findings thus may help healthcare providers evaluate efficacy across therapies when developing a patient’s personalized treatment plan.

Treatment choice is also based on the evaluation of safety and benefit–risk. However, assessing safety outcomes via NMA may not always be appropriate due to notable limitations of currently available data. In AD, clinical trials feasible to include in NMA (i.e., those that contain similar patient populations at baseline as well as a common comparator arm to connect the network) usually have a relatively short follow-up period of 12–16 weeks. Data available beyond these primary endpoint timepoints often lack a placebo arm or follow a re-randomized design in which patients previously exposed to a targeted systemic therapy are re-assigned treatment regimens, thus potentially violating the common comparator assumption of NMA. Adverse events are also often reported as broad groups of events (e.g., any adverse event) that may be heterogeneously defined across trials. While some groups of adverse events likely of clinical interest (e.g., serious adverse events, adverse events leading to treatment discontinuation, major adverse cardiovascular events, malignancy excluding nonmelanoma skin cancer, infections) are often reported, the incidences of these events are often too low during the initial follow-up period to facilitate a well-powered indirect comparison across treatments. Therefore, NMA results on safety outcomes should be interpreted with caution. A compilation of safety outcomes commonly reported across trials utilized in this manuscript are contained in Table S9 of the Supplementary Materials for reference.

Limitations

Limitations to this study include variability in the primary endpoint timepoint (12 weeks for abrocitinib and 16 weeks for the other therapies) used across trials. Low response rates in early timepoints, especially at week 2, led to wide credible intervals and some instances of model convergence issues. NMA of long-term timepoints was deemed nonfeasible due to notable heterogeneity in trial designs across therapies and lack of a common comparator arm. Cross-trial differences regarding outcome methodologies as discussed in Silverberg et al. [1] hold for analyses presented in this manuscript, namely nuances regarding EASI and IGA scale definitions, including differences in response definition language of the scale, or, specifically in abrocitinib trials, lack of patient scalp, palms, and soles components of the scale [5]. Different imputation methods were used across trials for missing data, including nonresponse imputation or multiple imputation due to the coronavirus disease 2019 (COVID-19) pandemic. Other sources of heterogeneity in the analyzed trials include prior corticosteroid exposure and inclusion of adolescent patients. This updated analysis continued to exclude trials of patients receiving TCS or TCI in combination with targeted therapies for AD. While this approach minimizes heterogeneity across trials, a separate analysis is thus required to assess the comparative efficacy of targeted therapies in combination with TCS or TCI. Efficacy data stratified by race and/or geographic location is not widely available or reported for trials included in these analyses and thus could not be directly assessed. As more finely stratified efficacy data become available, this may be a notable topic of further research.

There appears to be some heterogeneity in placebo response rates across trials, though baseline risk-adjusted models accounting for this heterogeneity did not provide a better fit to the data for primary endpoint timepoints across all outcomes assessed. In earlier timepoint analyses, baseline risk-adjusted models were selected for ΔNRS ≥ 4, EASI-75, and IGA 0/1 at some timepoints based on model fit. While maintaining a significant regression parameter, baseline risk-adjusted models failed to converge for ΔNRS ≥ 4 at week 2 following inclusion of lebrikizumab data using the default number of posterior iterations; increasing the number of posterior iterations allowed the fixed effects baseline risk-adjusted model to successfully converge. Baseline risk-adjusted models for EASI-90 at week 4 also failed to converge while displaying a significant baseline-adjustment regression parameter, though increasing the number of posterior iterations did not result in convergence. Trial observed response rates at earlier timepoints were lower than at the primary timepoint, thus earlier timepoint analyses were more likely to present challenges in model convergence. Relatively wider credible intervals were also produced in earlier timepoint analyses.

Finally, this NMA focused on select efficacy outcomes and did not evaluate other items such as overall symptom severity, quality of life, or safety outcomes. Safety information for the therapies should be studied carefully, with attention to the risk–benefit of each treatment option and with an understanding of the potential limitations of indirect treatment comparisons. Future research in this area is thus warranted to better understand the risk–benefit profile of these therapies, as well as how such profiles may relate to the experiences of physicians and preferences of patients, to better inform decision-making processes between providers and patients.

Conclusions

In this NMA updated to contain the latest phase 3 placebo-controlled data, among targeted systemic therapies for AD used as monotherapy, upadacitinib 30 mg remains to be the most efficacious targeted therapy, followed by abrocitinib 200 mg and upadacitinib 15 mg, after 12 or 16 weeks of therapy. In addition to efficacy, other factors including safety, benefit–risk, and patient preferences should also be taken into account when personalizing a patient’s treatment plan.