Ulcerative colitis (UC) affects nearly 1.5 million individuals in the United States¹. It is characterized by a protracted course, leading to disease-related hospitalization or surgery in over one-third of patients^{2, 3}. Many patients with UC develop moderate-to-severe disease activity, requiring initiation of advanced immunosuppressive therapy^{3, 4}. The past two decades have witnessed significant progress in the development of these therapies, with the approval by the United States Food and Drug Administration (FDA) of six different therapeutic mechanisms, ten distinct drugs, and the anticipated approval of guselkumab⁵ and risankizumab⁶ in 2024. The increase in the number of available treatments has brought to the forefront the importance of appropriate positioning and sequencing of treatments and therapeutic mechanisms. Head-to-head trials in moderately-to-severely active UC to help guide these clinical decisions have been sparse and are limited to the VARSITY⁷ (vedolizumab versus adalimumab) and VEGA⁸ (guselkumab versus golimumab) trials among approved therapies. Consequently, relative efficacy is often inferred through network meta-analyses (NMAs), which combine direct evidence from head-to-head trials and indirect evidence from placebo-controlled trials. Evidence from these NMAs have informed the 2020 American Gastroenterological Association (AGA) Clinical Guidelines on the management of moderate-to-severe UC^{9, 10}. Although NMAs have been published recently for moderate-to-severe UC^11–13, there have recently been substantial additions to the evidence base, including the availability of phase 3 efficacy data for mirikizumab¹⁴, guselkumab⁵, risankizumab⁶, and etrasimod¹⁵, evidence of efficacy for subcutaneous formulations of infliximab and vedolizumab¹⁶, and head-to-head comparison data between guselkumab and golimumab⁸. Prior NMAs have not critically appraised the available evidence which can inform clinical guidelines. Moreover, there has been limited data on the efficacy of different therapies stratified by prior exposure to biologic therapy, a key prognostic factor. Finally, there is discordance between options available for first-line therapy between other regions and the United States, where Janus kinase (JAK) inhibitors are approved by the FDA for use only after failure of tumor necrosis factor α (TNF) antagonists, necessitating different algorithms in these regions.

To inform a comprehensive 2024 Update to the AGA living guidelines on the management of moderate-to-severe UC, we conducted systematic reviews and NMAs to inform the comparative efficacy of different advanced therapies in biologic-naïve and biologic-exposed patients with moderate-to-severe UC. We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for NMA to appraise the confidence in estimates.

This systematic review was performed using an a priori established protocol, and is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) extension statement for systematic reviews incorporating NMAs for health care interventions.¹⁷ We followed good research practices outlined in the International Society for Pharmacoeconomics and Outcomes Research report on interpreting indirect treatment comparisons and NMA for health-care decision-making.¹⁸

The therapeutic armamentarium for the treatment of moderate-to-severe UC has expanded significantly with the approval of three new therapeutic mechanisms – JAK inhibitors, selective IL-23 antagonists, and S1P modulators. In this updated systematic review and NMA, which is being used to inform an update to the AGA Clinical Guidelines for the management of moderate-to-severe UC, we examined direct and indirect evidence from 35 trials to inform positioning of treatments within our therapeutic algorithm. We performed a thorough critical appraisal of the body of evidence using GRADE for NMA, contextualizing the absolute magnitude of benefit after setting a minimal important difference of 5% to confirm higher efficacy of one medication over another. We make several key observations which will decisively inform positioning of newer therapies for the management of moderate-severe UC: (a) as first-line treatment, in countries outside the United States where JAK inhibitors can be used prior to failure of other advanced therapies, upadacitinib was the most efficacious in inducing clinical remission, achieving remission in ~50% patients; (b) as first-line therapy excluding JAK inhibitors as guided by the FDA approval in the United States, risankizumab and ozanimod were ranked highest in achieving clinical remission; and (c) in all regions, upadacitinib was the most effective second-line agent in biologic-exposed patients. Interestingly, in the treat-through maintenance trials, vedolizumab and etrasimod were more effective than TNF-antagonists, supporting the efficacy of anti-lymphocyte trafficking mechanisms in the treatment of bio-naive moderate-to-severe UC.

Our findings update the results of recent NMAs^{11–13, 45} by three important methodologic additions. First, we added data on several additional therapies with available phase 3 efficacy data since the prior analyses including one S1P modulator (etrasimod) and three anti-IL-23 agents (guselkumab, risankizumab, and mirikizumab). Second, with the increasing use of biosimilars and alternate routes of administration, we also included data from trials of subcutaneous infliximab and vedolizumab. Third, we were able to further strengthen the network with addition of a second direct-head-to-head trial in UC, comparing guselkumab to golimumab. In addition, prior NMAs included 5 trials of etrolizumab, an anti-integrin, which did not meet its primary efficacy endpoint and is thus unlikely to be approved for use at this time, which may have influenced relative positioning. Also distinct from prior NMAs, we separated our analyses including and excluding JAK inhibitors as first line agents for the treatment of moderate-to-severe UC.

In regions without restrictions on the use of JAK inhibitors, our findings demonstrate that upadacitinib was more efficacious than nearly all other available therapies except ozanimod and risankizumab for induction of clinical remission. In the upadacitinib induction trials, the rate of clinical remission observed were 33% and 26%. The low placebo rates of 4% and 5% in these clinical trials further strengthened the evidence in support of superior efficacy of this drug, yielding an absolute benefit of 29% and 22% in achieving clinical remission, a margin that was greater than has been observed for other advanced therapies in UC. Data from multiple real-world cohorts has supported these high rates of remission and response with upadacitinib^{46, 47}. In the United States, a boxed warning placed by the FDA on this therapeutic class requires prior failure of, or contraindications to, TNF antagonists before the use of any JAK inhibitors. This decision by the FDA was informed by a post-marketing surveillance study of adults 50 years or age or older with rheumatoid arthritis and at least one cardiovascular risk factor treated with tofacitinib, a non-selective JAK inhibitor, in which both an increased risk of major adverse cardiovascular events (MACE) and malignancy compared to anti-TNF therapy were observed, particularly with the 10 mg BID dose of tofacitinib⁴⁸. The EMA, while not recommending against first-line use, advises caution in use in older adults and smokers. Post-hoc analyses of trials of tofacitinib in UC and observational studies did not identify a clear increase in risk of venous thromboembolism or MACE in this population compared to anti-TNF therapy^49–51. In a subsequent NMA, the risk of cardiovascular events was not higher with JAK inhibitors compared to TNF antagonists or anti-interleukin therapy among patients with IBD⁵². Given the superiority of upadacitinib over other therapies in most clinical scenarios, it is important to prioritize future studies of safety of this therapeutic class in those with IBD, and additionally whether the safety concerns apply equally to both selective and non-selective JAK inhibitors, as well as across specific indications for their use.

In biologic-naïve patients excluding use of JAK inhibitors as first line, our NMA demonstrated the superiority of several agents (infliximab, vedolizumab, risankizumab, ozanimod) over adalimumab. These findings are different from several recent NMAs which did not show a difference between the two treatments. This is an important observation as adalimumab is one of the most commonly prescribed biologics for UC⁵³, influenced by both physician and patient preferences and payor restrictions. While efficacy differences on indirect comparisons are only one factor in determining therapeutic positioning of available therapies, it is important that any positioning restriction reflect treatment effectiveness and the impact on patient outcomes to ensure the highest-value care.

Our study also highlights the importance of prior treatment exposure on subsequent therapy selection in moderate-to-severe UC. Absent data on infliximab or golimumab, the most efficacious treatment in biologic-exposed patients was upadacitinib, which demonstrated superior efficacy despite 30–35% of patients in its clinical trials having failed two or more prior biologics. Intriguingly, the difference in relative efficacy of a drug in biologic-exposed vs. biologic-naïve patients was significantly greater for some drugs, most notably for vedolizumab and ozanimod in our analysis. Ozanimod was highly efficacious in inducing clinical remission in biologic naïve-patients and superior to both adalimumab and mirikizumab. In contrast, it failed to achieve superiority over placebo in inducing clinical remission in biologic-exposed patients. This observation is consistent with results from the pivotal clinical trial where both in biologic-naïve and in patients with only one prior biologic exposure, ozanimod had rates of clinical remission of 23% and 17% compared to 7% and 8% in placebo⁵⁴. However, in those who were exposed to 2 or more biologics, ozanimod was no better than placebo in achieving clinical remission after induction (4% vs. 3%). The exact mechanism(s) of this drop in efficacy in a biologic exposed population remains to be robustly determined, but our findings highlight the critical determinative nature of this as a prognostic variable in therapy selection.

Both this study and prior NMAs demonstrate that the efficacy of different drugs within the same therapeutic class may not be comparable. We found infliximab to be superior to adalimumab, upadacitinib to be superior to both tofacitinib and filgotinib, and risankizumab to be superior to mirikizumab in biologic-naïve patients. The mechanism(s) for a potential difference in efficacy within agents in the same class is unclear and may be driven by both methodological considerations related to trial design and biologic plausibility. Multiple observational studies have compared the effectiveness of infliximab and adalimumab in real-world cohorts, demonstrating a benefit favoring infliximab in many but not all studies^{55, 56}. The mechanism(s) for this difference in efficacy does not appear to be related to either the weight-based dosing of infliximab or induction or maintenance dose of adalimumab, as no incremental evidence of greater efficacy with high-dose adalimumab induction was observed in the SERENE-UC trials⁵⁷. The greater JAK selectivity of upadacitinib may allow for a better dosing profile without compromising safety leading to superior efficacy^{58, 59}. This efficacy difference was also observed in real-world cohort studies, supporting our conclusion⁴⁶.

Our NMA suggested upadacitinib, tofacitinib, and ustekinumab to be superior to vedolizumab in those with prior biologic failures. This is also consistent with recent observational data. Buisson et al. conducted a propensity-matched analysis of vedolizumab and tofacitinib in those with a prior anti-TNF failure. While there was no difference in clinical remission rates at week 16 between both agents, vedolizumab was inferior to tofacitinib in achieving endoscopic improvement and histologic healing⁶⁰. A large Dutch registry-based study similarly demonstrated superior efficacy of tofacitinib over vedolizumab⁶¹. In contrast, while a US insurance-based study demonstrated lower persistence of tofacitinib over vedolizumab, this lacked granular data to discern differences in treatment efficacy⁶². There are few real-world studies comparing ustekinumab to vedolizumab in biologic-exposed patients with UC. However, our observation of superiority of ustekinumab over vedolizumab is consistent with the signal observed in the phase 3 clinical trials. In the GEMINI 1 trial, the magnitude of clinical benefit of vedolizumab over placebo in TNF antagonist-naïve patients was 16.5% compared to only 6.5% in TNF antagonist-exposed patients. In contrast, in the UNIFI studies, ustekinumab achieved a similar benefit over placebo in both TNF antagonist-naïve (delta=8.5%) and TNF antagonist-exposed individuals (delta=11.5%)⁶³.

There are several limitations to our work. The direct and indirect evidence was informed by RCT data of FDA-approved doses of each agent. Modification of dosing in real-world practice may alter comparative efficacy and was not included in our models. For example, early proactive trough assessment and dose optimization for TNF antagonist therapy, combination immunomodulator therapy for infliximab, and extended induction for tofacitinib or upadacitinib have been shown to incrementally increase rates of remission over therapy with the standard induction dose alone^{64, 65}. Second, while our analysis stratified by prior biologic exposure status, it is acknowledged that with the increasing complexity of our therapeutic armamentarium, this comparison will become more nuanced. Apart from TNF antagonists, S1P modulators, JAK inhibitors, and IL-23 inhibitors, all have multiple agents within the same therapeutic class. Prior failure of a drug within that class may have specific implications for other drugs within that class but not other mechanisms of action. In addition, trials of subsequent agents within a therapeutic class often exclude patients previously exposed to that mechanism of action. In many of the early clinical trials, most prior biologic exposure comprised of TNF antagonist failure. In more recent studies, a small fraction entering the trials may have failed vedolizumab but not other mechanisms. How these prior exposures to available biologic or small molecule therapies affect subsequent treatment efficacy remains to be established. Third, we focused on induction of clinical remission as the critical outcome of interest. Other outcomes including corticosteroid-free remission are clinically meaningful. However, there is significant heterogeneity between trials in how concomitant steroid use is handled. Almost all induction trials require stable doses of steroids during the trial, but studies differed by whether steroid taper was mandated during maintenance, and the pace of such tapers. Consequently, we elected not to use steroid-free remission as an outcome for our analysis. We did not separately stratify by both concomitant immunomodulator use and prior biologic exposure due to lack of statistical power for this comparison given wide range of immunomodulator use and that most patients entering the trials had failed immunomodulator or 5-aminosalicylate therapy prior to study entry. Similarly, both the paucity of data (particularly in older trials) and heterogeneity in definition between studies precluded comparisons of endoscopic remission (endoscopic subscore=0) or histologic healing. These may increasingly be relevant for future analyses but it will remain a challenge to integrate such data for newer agents with the absence of ascertainment of these endpoints for older trials. We also selected a threshold of an absolute difference of 5% in defining superiority of one agent over the other. Anchored against the approximately 25% and 15% clinical remission rates for biologic-naïve and exposed patients respectively, these represent a 20% and 33% increase in efficacy which was felt to be a meaningful difference. In future studies, engaging patients and providers to ascertain what a clinically meaningful difference might be would be helpful to refine this threshold. Therapy choice is influenced not just by efficacy but also safety of treatments under consideration. We did not perform an NMA of safety between different therapies for several reasons. Serious adverse events including infections and malignancy risk were low with all examined treatments within clinical trials, limiting statistical power for meaningful comparison between treatments. Larger observational cohorts with longer duration of follow-up have been more meaningful in informing comparative safety in this population. For example, such studies have demonstrated lower rates of infection with vedolizumab when compared to TNF antagonists, and similar risks with TNF antagonist and JAK inhibitor treatment^{50, 66}. More than efficacy, decisions about safety are more nuanced incorporating patient characteristics such as age, race, ethnicity, geography, comorbidity, concomitant treatments (including steroid use), and frailty. Consequently, NMAs may be less applicable for extrapolation of safety at the level of individual patients.

In conclusion, our NMA highlights upadacitinib as the most effective therapy for moderately-to-severely active UC in both bio-naïve and bio-exposed populations, demonstrating the urgent importance of studies of long-term safety of this treatment in patients with UC to inform the need for usage restriction. We also identify within-class differences in treatment efficacy across different therapeutic mechanisms and highlight the importance of prior biologic exposure as a key determinant of treatment selection in moderate-to-severe UC.