Bridging Gaps in Evidence-based Clinical Practice in Inflammatory Bowel Diseases: Observational Comparative Effectiveness Research for the Win

Combination therapy with tumor necrosis factor (TNF)-α antagonists and immunomodulators like thiopurines or methotrexate has been recommended by the American Gastroenterological Association over TNFα-antagonist monotherapy for induction of clinical remission in patients with moderate-to-severe Crohn’s disease (CD). These guidelines do not make clear recommendations regarding combination therapy for maintenance of remission, however.¹ The evidence for this was derived from the landmark SONIC trial, in which biologic- and immunomodulator-naive patients with recently diagnosed CD (median disease duration, 2.2y) treated with combination therapy of infliximab and azathioprine achieved greater rates of both steroid-free remission and mucosal healing by week 26, compared to monotherapy with infliximab or azathioprine². While SONIC was not designed to study maintenance of remission, continued benefit was noted in exploratory analyses up to 50 weeks. However, these findings were not confirmed in COMMIT (Combination of Maintenance Methotrexate-Infliximab Trial), which failed to show benefit of combination infliximab and methotrexate for maintenance of remission³. Similar negative findings were noted in DIAMOND (Deep Remission of Immunomodulator and Adalimumab Combination Therapy for Crohn’s Disease) which assessed adalimumab and thiopurines for induction and maintenance of corticosteroid-free clinical remission over biologic monotherapy.^{3, 4}

In contrast to CD, there are no longer-term clinical trials comparing biologic-based combination therapy and biologic monotherapy in patients with moderate-severe ulcerative colitis (UC); UC-SUCCESS (Infliximab, Azathioprine, or Infliximab + Azathioprine for Treatment of Moderate to Severe Ulcerative Colitis) demonstrated higher rates of achieving clinical remission, but no endoscopic remission, at week 16 with combination therapy. However, the trial was terminated prior to recruitment into the planned maintenance study due to observed infusion reactions with intermittent infliximab therapy in an unrelated trial.⁵ Retrospective cohort studies, typically conducted at referral-centers and thus prone to selection bias, have likewise failed to consistently demonstrate a benefit of combination therapy for clinical, biochemical, and endoscopic outcomes over a short duration in both CD and UC. These studies have also not been able to examine patient-important outcomes like hospitalization and surgery over a longer duration of follow-up.⁶

In this issue of Clinical Gastroenterology and Hepatology, Targownik and colleagues used an elegant observational comparative effectiveness design in a population-based cohort of patients with IBD to bridge this gap.⁷ In a cohort of 852 patients with CD and 303 patients with UC in Manitoba, Canada, the authors compared the effectiveness of combination therapy to TNFα-antagonist monotherapy. Using a novel “intent-to-use” definition of exposure to combination therapy (either new prescription of immunomodulators within 30 days before or after starting a TNFα-antagonist, or continued refills of immunomodulators after starting a TNFαantagonist in prevalent immunomodulator users), the investigators attempted to mimic an intention to treat model typically used in randomized controlled trials. Upfront combination therapy in new users of TNFα-antagonists with CD was associated with a clinically meaningful 47% lower risk of hospitalization and 37% lower risk of needing biologic switching when compared to those not using anti-TNF monotherapy, along with numerically lower risks of intestinal respective surgery and need for corticosteroids. These benefits were seen regardless of the type of TNFα-antagonist (infliximab or adalimumab) or immunomodulator (thiopurines or methotrexate) used. In contrast, investigators failed to observe a significantly higher effectiveness of upfront combination therapy over TNFα-antagonist monotherapy in patients with UC. These findings were likely impacted by their smaller sample size.

These data, from a well-described population-based cohort, provide compelling evidence demonstrating the real-world effectiveness of a strategy of upfront combination therapy over TNFα-antagonist monotherapy in reducing healthcare utilization in patients with CD beyond induction therapy. This is an important addition to our current body of knowledge, as the gap between efficacy estimates derived from clinical trials and effectiveness in everyday practice is well described.⁸ Several factors may contribute to difficulty in replicating efficacy estimates from clinical trials in the real world, perhaps the most prominent of which is the idealistic design of clinical trials. In order to isolate the effects of a particular therapy being studied, researchers apply strict inclusion and exclusion criteria regarding cohort composition, treatment utilization, and outcome measurement. While this is critical for accurately measuring how efficacious a therapy is when testing new compounds or preparing for regulatory body approval, the application of these criteria significantly limits generalizability: Prior research has demonstrated that only 31% of patients with moderate-to-severely active IBD were eligible for trials at a tertiary center.⁹ Patients in clinical practice may have significantly greater variation in age, disease phenotype and severity, prior and concomitant medication use and complications, and comorbidities. Additionally, trials often have a shorter duration of follow-up due to significant costs and are unable to study direct patient-important outcomes like surgery and hospitalization.

While trials of novel therapies targeting different disease pathways in IBD continue to be warranted, there is a clear paucity of active head-to-head comparative efficacy trials and pragmatic phase IV trials (of approved therapies) comparing treatment strategies. Such well-designed trials can reduce inherent bias, but face several significant challenges, as they are more complicated, expensive, and risky for industry sponsors, and budgets for these often exceed what may be feasible for public funding. Observational comparative effectiveness research methods, such as those used in this study, will be vital for bridging the gap between efficacy and effectiveness and informing long-term comparative safety of new therapies and treatment strategies in IBD, as well as informing cost-effectiveness in the real world.^10–16

Observational datasets also have several inherent limitations that must be addressed. Some of these may be overcome through the use of advanced statistical methods and analyses, while others would need to be acknowledged and interpreted appropriately with critical evidence synthesis. One foremost challenge is identifying like-for-like patients for comparative effectiveness research, such that both patients have comparable levels of disease severity and distribution of key effect modifiers; if a clinical trial could be performed, then patients could be randomly assigned to either intervention. Propensity score methods, which conceptually evaluate and compare the likelihood (or propensity) of being prescribed a particular treatment approach, can be an important statistical approach to closely replicate true randomization.¹⁷ The propensity score is the probability of treatment assignment conditional on observed baseline characteristics, generated through multinomial logistic regression. There are various applications of propensity score methods that can be applied to assist in identifying like-for-like patients. These approaches have been used in observational comparative effectiveness research in IBD, comparing effectiveness of different TNFα-antagonists¹⁰ and TNF-α antagonists vs. chronic corticosteroids,¹¹ and comparing safety of TNFα-antagonists and immunomodulators.^12–14 While propensity scores model the risk of exposures, an alternative approach called disease risk scores models the risk of outcome.¹⁸ Disease risk scores model the contribution of independent predictors to the risk of a specific outcome, allowing for multivariable reduction into a single composite measure. They are particularly helpful with newer therapies, in which provider preferences may evolve rapidly, and outcomes may be rare.

Observational datasets are also likely to have unmeasured or incompletely measured confounders, which may bias findings that cannot be accounted for by propensity score methods. In these instances, sensitivity analyses using different approaches such as the array method proposed by suggested by Schneeweiss et al may be useful.¹⁹ In this approach, if an incompletely measured (for example, disease severity) or unmeasured confounder (for example, smoking) may preferentially result in prescription of one agent over another, and may influence risk of outcomes, then the probabilistic relationship between the two agents in different scenarios of confounder distribution may be generated to allow conservative interpretation.

Finally, observational data are also prone to misclassification of exposures and outcomes, both of which can influence effectiveness estimates. Linking diverse datasets, such as electronic medical records, with administrative claims databases and patient registries, through privacy-preserving methods, may help minimize the potential impact of misclassification and missingness.²⁰

In summary, as demonstrated in this novel cohort study, well-designed observational comparative effectiveness research will play an increasingly important role in plugging holes, bridging gaps between trials and real-world, and directly informing clinical practice in IBD in the years to come.