Defining management strategies for acute severe ulcerative colitis using predictive models: a simulation-modeling study

Danny Con; Peter De Cruz

doi:10.5217/ir.2023.00175

Con and De Cruz: Defining management strategies for acute severe ulcerative colitis using predictive models: a simulation-modeling study

Original Article

Intestinal Research 2024;ir-2023-00175.

Published online: May 7, 2024

DOI: https://doi.org/10.5217/ir.2023.00175

Defining management strategies for acute severe ulcerative colitis using predictive models: a simulation-modeling study

Danny Con¹

, Peter De Cruz^1,²

¹Department of Gastroenterology, Austin Health, Heidelberg, Australia

²Department of Medicine, Faculty of Medicine, Dentistry & Health Sciences, The University of Melbourne, Parkville, Australia

Correspondence to Peter De Cruz, Department of Gastroenterology, Austin Health, 145 Studley Road, Heidelberg, VIC 3084, Australia.
Tel: +61-3-9496-5717, Fax: +61-3-9496-3487, E-mail: ppdecruz@gmail.com

Received November 17, 2023 Revised January 29, 2024 Accepted February 15, 2024

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Background/Aims

Robust management algorithms are required to reduce the residual risk of colectomy in acute severe ulcerative colitis (ASUC) refractory to standard infliximab salvage therapy. The aim of this study was to evaluate the performance and benefits of alternative ASUC management strategies using simulated prediction models of varying accuracy.

Methods

This was a simulation-based modeling study using a hypothetical cohort of 5,000 steroid-refractory ASUC patients receiving standard infliximab induction. Simulated predictive models were used to risk-stratify patients and escalate treatment in patients at high risk of failing standard infliximab induction. The main outcome of interest was colectomy by 3 months.

Results

The 3-month colectomy rate in the base scenario where all 5,000 patients received standard infliximab induction was 23%. The best-performing management strategy assigned high-risk patients to sequential Janus kinase inhibitor inhibition and medium-risk patients to accelerated infliximab induction. Using a 90% area under the curve (AUC) prediction model and optimistic treatment efficacy assumptions, this strategy reduced the 3-month colectomy rate to 8% (65% residual risk reduction). Using an 80% AUC prediction model with only modest treatment efficacy assumptions, the 3-month colectomy rate was reduced to 15% (35% residual risk reduction). Overall management strategy efficacy was highly dependent on predictive model accuracy and underlying treatment efficacy assumptions.

Conclusions

This is the first study to simulate predictive model-based management strategies in steroid-refractory ASUC and evaluate their effect on short-term colectomy rates. Future studies on predictive model development should incorporate simulation studies to better understand their expected benefit.

Keywords: Infliximab; Tumor necrosis factor-alpha; Colectomy; Inflammatory bowel diseases

INTRODUCTION

Ulcerative colitis (UC), a type of inflammatory bowel disease, is an autoimmune condition affecting the colon and affects millions of individuals worldwide [1]. Patients with UC have a 20% lifetime risk of presenting with acute severe ulcerative colitis (ASUC), a potentially life-threatening complication that may require emergency colectomy [2]. Although the initial management of ASUC—involving intravenous corticosteroid use and the criteria for salvage infliximab—is well established, the optimal subsequent management in the event of infliximab failure is far less clear [2,3]. This is reflected by a 20% residual 3-month colectomy risk and 30% 12-month risk in patients who receive standard infliximab induction for ASUC [4]. Observational studies have suggested a possible benefit in accelerated infliximab induction in steroid-refractory ASUC [4,5], while controlled studies are still ongoing (NCT02770040, NCT03937609). Although sequential ciclosporin salvage was used occasionally in the event of infliximab failure historically, recent evidence has suggested a possible role for sequential Janus kinase inhibitor (JAK) inhibitors such as tofacitinib and upadacitinib [6-8]. However, despite the recent rapid expansion of inflammatory bowel disease therapeutics, it is important to acknowledge the difficulties and impracticalities of prospective controlled trials evaluating JAK inhibitors or other agents in infliximab-refractory ASUC, yet clinicians and patients alike face this important clinical problem on a regular basis. Thus, clinicians must rely on the utilization and synthesis of the limited available evidence to formulate a practical ASUC management strategy in the absence of prospective trials for the foreseeable future.

Precision medicine aims to tailor management strategies to the individual using clinical outcome prediction and robust management strategies [9-11]. An ideal strategy is to use a predictive model that risk stratifies patients based on their risk of infliximab failure to provide a criterion for alternative management options. Predictive models have historically been derived using basic statistical methods; however, advances in machine learning and artificial intelligence alongside the expansion of big data through genomics and microbiomics are likely to facilitate far more reliable and accurate models [12,13]. An ideal model would be capable of predicting response to infliximab, such that patients who demonstrate a partial or incomplete response due to pharmacokinetic failure (inadequate dosing) would benefit from additional infliximab dosing [5], while patients who demonstrate poor response due to alternative immunologically active pathways might benefit from out-of-class therapies such as JAK inhibition [6,7]. Put simply, a good predictive model should be able to identify the patients who would most likely benefit from one particular treatment over another. However, whether the use of a predictive model to guide decision-making actually reduces the risk of colectomy and improves outcomes remains unclear. Due to the same aforementioned impracticalities, prospective trials evaluating the use of predictive models in ASUC management are unlikely to occur in the near future. In light of this, simulation and modeling can be used to synthesize and summarize the limited available evidence to help guide real-world decision making.

The aim of this study was to evaluate the performance and benefits of emerging ASUC management strategies based on currently available evidence using a hypothetical cohort of steroid-refractory ASUC patients receiving infliximab salvage therapy.

METHODS

1. Study Design

This was a simulation-based modeling study using a hypothetical cohort of 5,000 patients with steroid-refractory ASUC receiving salvage infliximab. The 2 key components of management that were simulated included: (1) derivation of a predictive model to risk-stratify patients based on their probability of failing infliximab and (2) response to different clinical management strategies based on 3-month colectomy risk. The simulated predictive model was used to inform subsequent management decisions. Management strategies included accelerated infliximab induction and/or sequential JAK inhibition with tofacitinib and/or upadacitinib. Assumptions of efficacy of management strategies were based on published literature (Tables 1, 2). Management strategies were assessed by the estimated reduction in the residual 3-month colectomy risk. The 3-month time point was chosen due to its dependence on the efficacy of induction strategies, as opposed to later points such as 12 months which can be affected by subsequent maintenance regimens. Ethics approval was waived as no new data were generated from this study.

2. Study Cohort

We generated a hypothetical cohort of 5,000 patients with steroid-refractory ASUC who receive standard induction of infliximab at 5 mg/kg at weeks 0, 2, and 6. All patients were then followed-up for up to 3 months. Each patient’s “true propensity” to fail infliximab was generated by applying a logistic transformation of a uniform random variable. The true propensity represents the unobserved or hidden probability that a patient will fail infliximab induction and require colectomy, with range between 0 and 1. The true propensity was calibrated to achieve a 3-month colectomy rate of 23% and a 12-month colectomy rate of 30% based on prior studies [4]. We used a Weibull distribution to generate the failure times [31], with a shape parameter of 0.7 (chosen heuristically) and a scale parameter proportional to the inverse of the true propensity.

3. Predictive Model

A predictive model for infliximab failure might include patient and disease factors such as C-reactive protein (CRP), albumin, lymphocyte count and endoscopy [32-34]. Predictive models can be useful if they can identify patients at high risk of failing standard dose infliximab, who might benefit from alternative salvage therapies. We simulated the derivation of a predictive model shortly after the first 5 mg/kg infliximab dose (between day 1 and day 7 post infliximab), which then guided subsequent management.

Traditionally, predictive models or scores can be created using standard statistical models such as Cox proportional hazards regression (for survival outcomes) or logistic regression (for binary outcomes). In both cases, candidate predictors (such as age, sex, disease severity markers, etc.) are entered into the statistical model, which generates beta coefficients (or weights) based on the importance of each predictor in explaining the outcome. The weighted linear combination of the predictors is then used as a predictive score. When applied to an individual patient, the predictive score generates a number. Usually, the higher the number is relative to other patients, the higher the risk of the predicted outcome. This predictive number can then be used to risk stratify patients, by assigning cutoff values and grouping patients based on the number. In this way, the predictive number should be highly correlated with the patient’s true risk of the outcome.

In order to simulate the derivation of a predictive model, first recall that each simulated patient was assigned a “true propensity” or true risk of failing infliximab and requiring colectomy. If the true propensity (which is simply a number) is used to create the predictive model, the model would nearly perfectly predict the outcome (with area under the curve [AUC] close to 100%) as the model and the true risk are perfectly correlated. In order to generate realistic predictive models, we needed to reduce the correlation between a theoretical model and the true propensity. Thus, the predictive model was simulated by introducing random noise into each patient’s true propensity to fail infliximab. Specifically, we added a normally distributed random number with mean 0 to the logit of each patient’s true propensity, where the standard deviation of the random number was calibrated to produce predictive scores of varying accuracies. The resulting number was treated as the predictive score. As more noise is introduced, the correlation between the predictive model and the true propensity decreases, thus decreasing the predictive utility and the AUC. We calibrated the predictive models to have AUCs of 70% (low), 80% (medium), and 90% (high). Current predictive models have modest AUCs between 70% and 80% [32-34], while a hypothetically strong model with 90% AUC might be achievable with the discovery of additional, novel biomarkers.

4. Management Strategies

The simulated management strategies were based on the use of the generated predictive model to risk stratify patients into 2 or 3 groups, where subsequent treatment depended on the risk category (Fig. 1). If the predictive model predicted a patient to have a high probability of failing standard infliximab induction, the patient’s therapy was escalated or changed according to the intervention strategy (Fig. 1). We tested a combination of accelerated infliximab induction and sequential JAK inhibition as subsequent therapies for the management strategy.

1) Strategy 1: Accelerated Infliximab Induction

Non-controlled studies have suggested a possible benefit in accelerated induction of infliximab in steroid-refractory ASUC (Table 1), while controlled studies are ongoing (NCT02770040, NCT03937609). The clinical rationale for a higher infliximab dose is based on high infliximab losses that occur in patients with a high inflammatory burden [5], such that a proportion of infliximab-failures are due to underdosing. If the latter theory holds, then a proportion of infliximab non-responders could achieve response and avoid colectomy if they receive accelerated (shortened dose interval or increased dose) induction. In practice, clinicians tend to select high-risk patients (reflected by high CRP and/or low albumin) for accelerated induction [4,18], hence observational studies cannot directly compare colectomy rates in patients receiving standard vs. accelerated induction, so the proposed benefits of accelerated dosing remain to be proven.

Strategy 1 relied on dividing the cohort into halves according to the predictive model after the first 5 mg/kg infliximab dose: the top 50% constituted the high-risk group, while the bottom 50% constituted the low-risk group. High-risk patients proceeded to receive accelerated infliximab induction, while low-risk patients continued standard induction. We assumed that the probability of responding to accelerated induction depended on the true propensity. That is, we assumed a high correlation between response to standard and accelerated infliximab. We based our efficacy assumptions on previously reported colectomy rates (Table 1). Sebastian et al. [4] suggested that accelerated induction led to a 45% residual risk reduction of colectomy by 12 months in a propensity score matched cohort (not to be confused with the “true propensity” for infliximab failure), while Shah et al. [18] suggested a 34% residual risk reduction after propensity score matching. In our simulations, we tested different efficacy assumptions corresponding to residual colectomy risk reductions of 57%, 35%, or 13%. That is, based on our cohort’s 77% 3-month colectomy-free rate of standard induction, we tested an optimistic 90% 3-month colectomy-free rate for accelerated induction (57% residual risk reduction), an intermediate 85% colectomy-free rate (35% residual risk reduction) and a pessimistic 75% colectomy-free rate (13% residual risk reduction) at 3 months.

We used independent Bernoulli trials to determine whether each patient required colectomy by 3 months, with probability of colectomy being proportional to the patient’s true propensity (after calibration to achieve the target efficacy assumption).

2) Strategy 2: Sequential JAK Inhibition

Recent non-controlled studies have suggested efficacy in sequential use of JAK inhibitors such as tofacitinib and upadacitinib after infliximab failure in ASUC (Table 2). Due to the differing mechanisms of actions between JAK inhibitors and anti-tumor necrosis factor (TNF) therapies such as infliximab, some patients who do not respond to infliximab salvage are theorized to have over-active alternate immunological pathways that continue to drive inflammation despite suppression of TNF [35]. In support of this possibility, JAK inhibitors have shown good efficacy in both UC and Crohn’s disease patients even with prior biologic experience [36].

Strategy 2 again relied on halving the cohort based on the predictive model, where high-risk (top 50% risk) patients were switched to a JAK inhibitor after the first 5 mg/kg infliximab dose and low-risk patients (bottom 50% risk) continued on standard infliximab induction. Contrary to the correlation between response to accelerated and standard dose infliximab, we assumed the response to JAK inhibition to be independent of response to infliximab in our models. Therefore, response to JAK inhibition only depended on population-level estimates of response rates (and not related to the underlying true propensity to fail infliximab). A review of the evidence showed a wide range of 3-month colectomy rates often in small cohorts or case series (Table 2). Our review identified 168 patients with an overall crude 3-month colectomy rate of approximately 20%, representing a colectomy-free rate of 80%. Due to possible publication bias, we treated the 80% efficacy rate as an optimistic assumption. Additionally, we tested a more modest colectomy-free rate of 70% and a pessimistic rate of 60%.

We used independent Bernoulli trials to determine whether each patient required colectomy by 3 months, with probability of colectomy being equal to 20%, 30%, or 40% depending on the assumed efficacy of JAK inhibition.

3) Strategy 3: Accelerated Infliximab and JAK Inhibition

The final strategy utilized both accelerated infliximab induction as well as JAK inhibition. We hypothesized that patients who are underdosed on infliximab, but otherwise do not have a high true propensity of failing infliximab, would benefit from accelerated induction, while patients who have a high true propensity of failing infliximab would benefit from a switch to JAK inhibition. This algorithm relied on dividing the cohort into 3 groups: (1) a low-risk group who continued standard infliximab induction; (2) a medium-risk group who escalated to receive accelerated infliximab induction; and (3) a high-risk group who were switched to JAK inhibition.

We tested 2 risk-stratification methods. The first was to divide the cohort into risk quartiles, then use the top quartile (25%) to define the high-risk group who received sequential JAK inhibition and the second quartile (25%) to define the medium-risk group who received accelerated infliximab induction, leaving the bottom 2 quartiles (50%) to define the low-risk group who received standard infliximab induction. This resulted in half the cohort requiring treatment change. The second method was to divide the cohort into equal risk tertiles (33% in each risk group), where the top 33% were considered to be high-risk and received sequential JAK inhibition, while the middle 33% were considered to be medium-risk and received accelerated infliximab induction, resulting in two-thirds of the cohort requiring treatment change. We used independent Bernoulli trials using the same rules as strategies 1 and 2 to determine whether each patient required colectomy by 3 months.

5. Statistical Analysis

The Kaplan-Meier method was used to estimate survival curves. Receiver operator characteristic (ROC) analysis was used to assess discriminative performance of the theoretical predictive models, summarized by the AUC. All predictive models were simulated with differing AUCs of 90%, 80%, and 70%. We ran 1,000 simulations for each combination of predictive model accuracy, management strategy and efficacy assumption, where each simulation involved predictive model derivation, risk group allocation and colectomy status assignment based on the management strategy. The simulations were used to generate Monte Carlo confidence intervals (the 2.5th to 97.5th percentiles of the simulated colectomy rates) for the overall 3-month colectomy rate in each management strategy. All model building, simulation and analysis was performed using the R statistical programming language version 4.3.0 (RStudio 2023.06.1, Boston, MA, USA).

RESULTS

1. Study Cohort

Of 5,000 hypothetical steroid-refractory ASUC patients receiving standard dose salvage infliximab (5 mg/kg at weeks 0, 2, and 6), 759 patients required colectomy by 1 month, 1,156 patients required colectomy by 3 months and 1,514 patients required colectomy by 12 months, corresponding to 1-, 3- and 12-month colectomy rates of 16%, 23% and 30% (Fig. 2).

2. Predictive Models

Fig. 2 demonstrates the ROC curves of the simulated models predicting infliximab failure, calibrated to have average AUCs of 90%, 80%, and 70%. Table 3 demonstrates the 3-month colectomy rates after using the simulated predictive models to divide the baseline cohort in to different risk groups. After risk stratification into 2 groups using a highly accurate prediction model with 90% AUC, low-risk patients had 3-month colectomy rate of 2% compared to 44% for the high-risk group. Using a modest prediction model with 80% AUC instead, the 3-month colectomy rates for low and high-risk patients were 7% and 40%, while the respective rates for a less accurate 70% AUC model were 12% and 34% (Fig. 3).

Using the same models to stratify patients into 3 groups instead of 2, where the upper 50% risk patients are further divided into high risk (top quartile) and medium risk (second quartile), the 3-month colectomy rates for the medium and high-risk groups using the 90% AUC model were 19% and 69%. The colectomy rates for the 80% AUC model were 26% and 53% for the medium and high-risk groups, while the respective rates for the 70% AUC model were 26% and 42%.

Instead using the models to divide patients into equally sized risk tertiles, the 3-month colectomy rates for low, medium and high-risk patients using the 90% AUC model were 1%, 8% and 60% respectively. Using an 80% AUC model, the respective colectomy rates were 4%, 17%, and 48%, while the respective rates for a 70% AUC model were 9%, 21%, and 39%.

3. Management Strategies

1) Strategy 1: Accelerated Infliximab Induction

Table 4 demonstrates colectomy rates by treatment algorithm after 1,000 simulations. Assuming a 90% AUC for the prediction model and optimistic efficacy of accelerated induction (overall response rate of 90%), the 3-month colectomy rate was reduced to 11%, representing an overall residual risk reduction of 52%. Correspondingly, only 20% of the high-risk group required colectomy (55% residual risk reduction). Assuming a more modest 80% AUC for the prediction model and moderate efficacy of accelerated induction (overall response rate of 85%), the overall 3-month colectomy rate was reduced to 17% (29% residual risk reduction), while the colectomy rate in the high-risk group was 26% (35% residual risk reduction). Assuming a less accurate 70% AUC prediction model and a lower efficacy of accelerated induction (overall response rate of 80%), the overall 3-month colectomy rate was 21% (9% residual risk reduction), with a colectomy rate of 30% in the high-risk group (12% residual risk reduction).

2) Strategy 2: Sequential JAK Inhibition

Assuming a 90% AUC prediction model and an optimistic efficacy of 80% for sequential JAK inhibitors, the overall 3-month colectomy rate was reduced to 11% (52% residual risk reduction), with a colectomy rate of 20% in the high-risk group (55% residual risk reduction). With a modest 80% AUC prediction model and 70% treatment efficacy, the overall 3-month colectomy rate was 18% (22% residual risk reduction) and the colectomy rate in the high-risk group was 30% (25% residual risk reduction). With a 70% AUC prediction model and a pessimistic 60% JAK inhibitor treatment efficacy, the overall 3-month colectomy rate increased to 26%, which was 13% higher than the baseline 3-month colectomy rate, while the colectomy rate in the high-risk group was 40% (18% higher than the base case).

3) Strategy 3: Accelerated Infliximab and JAK Inhibition

A similar overall residual risk reduction was achieved using both risk stratification methods (equal tertiles and upper 25% as high risk). Assuming a 90% AUC prediction model and the most optimistic treatment efficacy (90% response rate for accelerated infliximab, 80% response rate for sequential JAK inhibition), the overall 3-month colectomy rate was reduced to 8% (65% residual risk reduction); the high-risk colectomy rate was reduced to 20% (67% to 71% residual risk reduction). Assuming an 80% AUC prediction model with modest treatment efficacy, the overall 3-month colectomy rate was reduced to 15% (35% residual risk reduction). By instead assuming a 70% AUC prediction model with pessimistic treatment efficacy, the overall 3-month colectomy rate was reduced to 22% (4% residual risk reduction).

DISCUSSION

Predictive tools and novel management algorithms are required to reduce the residual risk of morbidity in ASUC [2]. In addition to prediction model development and further controlled studies in ASUC, a deeper understanding of the expected benefit of different management options is crucial for good decision-making. In this study, we have demonstrated the expected reduction in the residual risk of colectomy using different combinations of predictive models and management algorithms. Furthermore, we also demonstrated that in pessimistic circumstances, a poor predictive model with low sequential salvage efficacy can actually lead to an increase in the 3-month colectomy rate. Thus, the expected benefit of various management strategies can be weighed against healthcare costs and resource allocation in cost-benefit analyses to identify the strategy with the greatest overall utility. The best-performing management strategy divided steroid-refractory ASUC patients into 3 groups, where patients deemed to be at medium risk of infliximab failure were escalated to accelerated induction, while patients deemed to be at high risk of failure were switched to a JAK inhibitor instead. However, overall performance was highly dependent on the accuracy of the predictive model and the efficacy of salvage therapies, especially JAK inhibitors.

We have shown that the development of highly effective management strategies is expected to drastically reduce the residual risk of adverse outcomes in ASUC given appropriate conditions. When considering only modest predictive model accuracy assumptions (AUC 80%), in line with currently available predictive models [32-34], as well as modest treatment efficacy, the residual risk of colectomy was reduced by more than one-third from 23% to 15% using both accelerated infliximab induction and sequential JAK inhibition, as opposed to 17%-18% when using either strategy in isolation. Assuming a more accurate predictive model with 90% AUC can be developed in future, and with optimistic treatment efficacy assumptions, the combined accelerated infliximab and JAK inhibitor approach could reduce the 3-month colectomy rate by almost two-thirds to 8%, compared to 11% with either treatment alone. Our modeling also demonstrates that good outcomes can be achieved by both better predictive models as well as efficacious treatments. Despite the most pessimistic treatment efficacy, using a highly accurate model with 90% AUC could achieve an overall 3-month colectomy rate of 15% to 16% (30% to 35% residual risk reduction) using the combined accelerated infliximab/JAK inhibition strategy. Similarly, despite using a low accuracy model with 70% AUC but assuming the most optimistic treatment efficacy, the combined algorithm achieved a similar 3-month colectomy rate of 13% to 14% (39% to 43% residual risk reduction).

The concept of a hidden propensity for infliximab failure is useful to understand the evolving ASUC treatment paradigm. Patients with ASUC suffer a high inflammatory burden and consequently may suffer high infliximab losses [5,37]. Patients who lack response due to pharmacokinetic failure, where standard infliximab induction is unable to achieve adequate serum levels, would theoretically benefit from additional dosing. However, these patients must have a low propensity of failing infliximab, explained biologically by an inflammatory process that can be switched off by achieving an adequate level of TNF inhibition [35]. Patients who have accessory pathways of inflammation that are inadequately moderated by TNF inhibition, thereby having a high propensity for infliximab failure, are unlikely to benefit even from optimized infliximab dosing. These patients would benefit most from alternative, non-TNF-based therapies.

Although not explored in this study, the timing of predictive model development is a key factor that is often overlooked in model derivation [38]. Ultimately, a predictive model is only clinically useful if it can be used to inform decision-making. In the currently accepted ASUC management algorithm, the Truelove and Witt’s criteria is used to define ASUC (the first decision point), while the Oxford and Lichtiger indices are commonly used to define steroid-refractoriness (the second decision point), necessitating commencement of salvage infliximab [25]. Assuming the optimal starting dose of infliximab is known, the third key decision point is at some point after the initial infliximab dose, in order to predict or determine infliximab failure, necessitating treatment modification. Previously, sequential ciclosporin was the only alternative to emergency colectomy in the event of infliximab failure, which was associated with poor outcomes and a reported 12-month colectomy-free rate of only 42% [39]. With the advent of accelerated infliximab and JAK inhibition, this third decision point has become critically important, which has formed the basis of the management strategies that we have investigated in this study. Previous studies have identified that a higher CRP, lower albumin and lower lymphocyte count after infliximab salvage have predicted infliximab failure [33]. However, a robust predictive model has yet to be validated.

As precision medicine continues to advance, the likely eventuality is that a patient’s optimal therapy can be determined at the second decision point instead (at the time of steroid failure, prior to commencing infliximab). This could be achieved by advances in pharmacokinetic profiling (determining the patients who will have high infliximab losses), immunological profiling (determining patients with accessory pathways who will not respond to infliximab), genetic profiling using genomics, as well as machine learning approaches (black box prediction approach using big data) [40-44]. With the advent of newer biologic agents and small molecule therapies, truly robust management algorithms tailored to the individual will become possible, which will likely further reduce the residual risk of colectomy.

This study’s strengths include a robust simulation-modeling approach to answer the study’s aim. Through simulation, a realistic cohort of steroid-refractory ASUC patients could be generated, and predictive models of various accuracies in combination with different management algorithms could be tested. The study is limited by its reliance on the efficacy assumptions made of different treatments. Due to the lack of high-quality evidence, the assumptions on the efficacy of accelerated infliximab induction and JAK inhibition might be inaccurate. In particular, due to the low quantity of retrospective data available, especially with regards to JAK inhibition (Table 2), the true efficacy of these treatments may be overestimated due to publication bias. In order to mitigate this risk, we explored multiple levels of assumed efficacy, including pessimistic assumptions, to produce a robust modeling study. Furthermore, in terms of external validity of the findings from our study, the colectomy rates used to generate the hypothetical cohort are based only on published studies (Table 1), thus our results may only be applicable to ASUC cohorts of similar baseline characteristics as those reported from external studies from Table 1. Further, our models assume an independence between propensity to respond to infliximab and propensity to respond to JAK inhibition, whereas in reality, they may be correlated (albeit with a low correlation). If so, our results for JAK inhibition may be optimistic. Future modeling-simulation studies should be performed when there is more clarity regarding treatment efficacy and response behaviors. Finally, our simulated predictive models had fairly symmetric ROC curves (favoring neither false positives nor false negatives). In practice, derived predictive models may have skewed ROC curves, which may require choosing risk thresholds at different locations to better risk stratify patients (for example, at the 60% percentile as opposed to the 50% percentile).

In conclusion, this is the first study to model novel management algorithms in steroid-refractory ASUC and simulate their effect on long-term colectomy rates. A management algorithm incorporating both accelerated infliximab induction as well as sequential JAK inhibition appears to be the most robust and achieves the greatest residual colectomy risk reduction by 3 months. However, management strategies incorporating a predictive model for risk stratification should ideally be evaluated within controlled trials before their use can be routinely recommended in clinical practice. Optimizing outcomes in ASUC relies on both accurate predictive models as well as efficacious treatment alternatives. Future studies on predictive model development should consider incorporating simulation studies to better understand their expected benefit.

ADDITIONAL INFORMATION

Funding Source

The authors received no financial support for the research, authorship, and/or publication of this article.

Conflict of Interest

Con D declares no conflicts of interest. De Cruz P has received research funding from Janssen, AbbVie, Takeda, Shire and Ferring. De Cruz P has been on advisory boards and received speaker fees from Janssen, AbbVie, Takeda, Celltrion, Shire, Ferring and Pfizer. De Cruz P is supported by a National Health and Medical Research Council (NHMRC) Emerging Leader Fellowship.

Data Availability Statement

Not applicable.

Author Contributions

Conceptualization; Formal analysis; Investigation; Software: Con D. Supervision: De Cruz P. Visualization: Con D. Writing - original draft: Con D. Writing - review & editing: Con D, De Cruz P. Approval of final manuscript: all authors.

Fig. 1.

Simulated management algorithms represented by decision trees. (A) Base case: standard infliximab (IFX) induction for all patients. (B) Simulation algorithm 1: accelerated IFX induction for high-risk patients. (C) Simulation algorithm 2: sequential Janus kinase inhibitor (JAK) inhibition for high-risk patients. (D) Simulation algorithm 3: accelerated IFX induction for medium-risk patients and sequential JAK inhibition for high-risk patients. ASUC, acute severe ulcerative colitis.

Fig. 2.

Results of simulation modeling. (A) Kaplan-Meier survival curve for hypothetical study cohort receiving standard infliximab induction (n=5,000). (B) Receiver operator characteristic curves of first 50 simulations of predictive score generation with target area under the curve (AUC) thresholds of 90%, 80% and 70%, highlighting example curves (randomly selected).

Fig. 3.

Kaplan-Meier survival curves of study cohort, risk-stratified by prediction scores. (A) Score with 90% area under the curve (AUC), upper 25% as high risk, middle 25% as medium risk. (B) Score with 90% AUC, upper 33% as high risk, middle 33% as medium risk. (C) Score with 80% AUC, upper 25% as high risk, middle 25% as medium risk. (D) Score with 80% AUC, upper 33% as high risk, middle 33% as medium risk. (E) Score with 70% AUC, upper 25% as high risk, middle 25% as medium risk. (F) Score with 70% AUC, upper 33% as high risk, middle 33% as medium risk.

Table 1.

Studies Comparing SD Infliximab Induction and AD Infliximab Induction, Including Propensity Score Matched Patients

Study (year)	SD infliximab		AD infliximab		Propensity score matching
Study (year)	No.	Colectomy-free rate	No.	Colectomy-free rate	Propensity score matching
Syal et al. (2021) [14]	29	86% (3 mo); 79% (12 mo)	34	85% (3 mo); 80% (12 mo)	No
Gibson et al. (2019) [15]	87	62% (6 mo)	58	67% (6 mo)	No
Sebastian et al. (2019) [4]	102	80% (3 mo); 72% (12 mo)	29	76% (3 mo); 69% (12 mo)	Yes: 12 mo colectomy-free rate 42% in SD vs. 69% in AD (53% residual risk reduction)
Chao et al. (2019) [16]	37	95% (3 mo)	35	86% (3 mo)	Partial, specific results not described
Nalagatla et al. (2019) [17]	132	86% (3 mo); 73% (12 mo)	81	80% (3 mo); 72% (12 mo)	No
Shah et al. (2018) [18]	120	72% (12 mo)	26	73% (12 mo)	Yes: 12 mo colectomy-free rate 71% in SD vs. 81% in AD (34% residual risk reduction)
Gibson et al. (2015) [19]	35	63% (1 mo)	15	93% (1 mo)	No

SD, standard dose; AD, accelerated dose.

Table 2.

Observational Studies on JAK Inhibitors in Steroid-Refractory ASUC

Study (year)	Tofacitinib dose	No.	Colectomy-free rate	Setting
Uzzan et al. (2021) [20]	TOF 10 mg BD	55	79% (3 mo); 74% (6 mo)	Hospitalized refractory UC or first line salvage for ASUC due to prior biologic experience
Berinstein et al. (2021) [21]	TOF 10 mg BD or 10 mg TDS	40	85% (3 mo)	First line salvage for ASUC due to prior biologic experience
Eqbal et al. (2023) [22]	TOF 10 mg TDS	11	82% (3 mo); 82% (12 mo)	Sequential therapy after IFX failure
Constant et al. (2022) [23]	TOF 10 mg BD or 10 mg TDS	11	73% (3 mo)	First line salvage for ASUC due to prior biologic experience or sequential therapy after IFX failure
Xiao et al. (2022) [24]	TOF 10 mg BD or 10 mg TDS	8	63% (3 mo)	First line salvage for ASUC due to prior biologic experience or sequential therapy after IFX failure
Komeda et al. (2023) [25]	TOF 10 mg BD	8	88% (unclear timing)	First line salvage in biologic naïve ASUC
Malakar et al. (2023) [26]	TOF 10 mg BD	8	88% (3 mo)	First line salvage for ASUC due to prior biologic experience
Honap et al. (2020) [27]	TOF 10 mg BD	7	43% (3 mo)	First line salvage for ASUC due to prior biologic experience or sequential therapy after IFX failure
Gilmore et al. (2022) [28]	TOF 10 mg TDS	5	80% (3 mo)	Sequential therapy after IFX failure or early post-IFX induction (ASUC < 7 day completing IFX induction)
Kotwani et al. (2020) [29]	TOF 10 mg BD	4	100% (3 mo)	First line salvage for ASUC due to prior biologic experience
Gilmore et al. (2023) [6]	UPA 45 mg OD	6	100% (3 mo); 83% (4 mo)	First line salvage for ASUC due to prior biologic experience
Zinger et al. (2023) [8]	UPA 45 mg OD	4	75% (3 mo)	Sequential therapy after IFX failure
Ali et al. (2023) [30]	UPA 45 mg OD	1	100% (6 mo)	First line salvage for ASUC due to prior biologic experience

Indication for TOF and UPA included a combination of first line salvage (in both biologic experienced and naïve patients) as well as sequential (second line salvage) therapy after IFX failure (total 168 patients with a crude combined 3-month colectomy-free rate of approximately 80%).

JAK, Janus kinase inhibitor; ASUC, acute severe ulcerative colitis; TOF, tofacitinib; UPA, upadacitinib; BD, twice daily; OD, once daily; TDS, three times daily; UC, ulcerative colitis; IFX, infliximab.

Table 3.

Colectomy Rates of Simulated Cohort According to Risk Stratification Using Predictive Models of Varying Utility

Risk group	90% AUC		80% AUC		70% AUC
Risk group	n (95% CI)	%	n (95% CI)	%	n (95% CI)	%
Dividing by halves
Low risk (lower 50%)	46 (39-53)	2	165 (148-182)	7	304 (283-326)	12
High risk (upper 50%)	1,110 (1,103-1,117)	44	991 (974-1,008)	40	852 (830-873)	34
Dividing by tertiles
Low risk (lower 33%)	20 (16-24)	1	64 (53-76)	4	154 (137-172)	9
Medium risk (middle 33%)	134 (119-147)	8	287 (268-308)	17	351 (327-373)	21
High risk (upper 33%)	1,002 (989-1,016)	60	804 (783-824)	48	651 (629-676)	39
Dividing by quartiles
Low risk (bottom 50%)	46 (39-53)	2	165 (148-183)	7	303 (282-326)	12
Medium risk (next 25%)	243 (226-260)	19	323 (302-344)	26	325 (304-346)	26
High risk (upper 25%)	867 (851-883)	69	667 (647-690)	53	528 (505-548)	42

This shows a baseline 3-month colectomy rate of 23% (n=1,156) in the whole cohort of 5,000 acute severe ulcerative colitis patients receiving standard infliximab induction.

AUC, area under the curve; CI, confidence interval.

Table 4.

Results of Simulation Study Using Hypothetical Cohort of 5,000 Steroid-Refractory ASUC Patients Requiring Standard Dose IFX Salvage, with 23% Requiring Colectomy by 3 Months

Treatment strategy & risk stratification	Risk score AUC	Treatment efficacy	3-mo colectomy rate
			Low-risk group		Medium-risk group		High-risk group		Total (entire cohort)
			n (95% CI)	%	n (95% CI)	%	n (95% CI)	%	n (95% CI)	%
Accelerated IFX in high-risk patients, standard IFX in low-risk patients; Low risk 50%, high risk 50%	90%	IFX 90%	46 (39-53)	2	-	-	505 (483-525)	20	551 (527-572)	11
		IFX 85%	46 (39-53)	2	-	-	738 (717-759)	30	784 (761-807)	16
		IFX 80%	46 (39-53)	2	-	-	982 (962-1,003)	39	1,029 (1,008-1,049)	21
	80%	IFX 90%	166 (149-182)	7	-	-	452 (432-474)	18	620 (594-642)	12
		IFX 85%	164 (148-182)	7	-	-	660 (638-682)	26	826 (803-847)	17
		IFX 80%	165 (149-182)	7	-	-	874 (852-896)	35	1,039 (1,017-1,061)	21
	70%	IFX 90%	304 (282-325)	12	-	-	387 (366-410)	15	692 (667-713)	14
		IFX 85%	304 (283-325)	12	-	-	566 (542-587)	23	869 (847-892)	17
		IFX 80%	304 (282-326)	12	-	-	747 (722-771)	30	1,051 (1,029-1,073)	21
Switch to JAK inhibitor in high-risk patients, standard IFX in low-risk patients; Low risk 50%, high risk 50%	90%	JAK 80%	46 (39-53)	2	-	-	501 (469-532)	20	546 (513-579)	11
		JAK 70%	46 (39-53)	2	-	-	750 (711-788)	30	795 (756-835)	16
		JAK 60%	46 (39-53)	2	-	-	1,000 (960-1,041)	40	1,046 (1,007-1,088)	21
	80%	JAK 80%	165 (150-182)	7	-	-	501 (467-534)	20	666 (630-703)	13
		JAK 70%	165 (149-183)	7	-	-	750 (713-788)	30	916 (875-955)	18
		JAK 60%	165 (148-183)	7	-	-	1,000 (959-1,042)	40	1,165 (1,121-1,209)	23
	70%	JAK 80%	304 (282-325)	12	-	-	499 (468-534)	20	803 (766-843)	16
		JAK 70%	303 (281-325)	12	-	-	751 (711-786)	30	1,054 (1,007-1,092)	21
		JAK 60%	304 (281-325)	12	-	-	1,000 (957-1,038)	40	1,303 (1,255-1,350)	26
JAK inhibitor in upper high-risk, accelerated IFX in medium-risk, standard IFX in low-risk patients; Low risk 50%, medium risk 25%, high risk 25%	90%	IFX 90%/JAK 80%	46 (39-53)	2	87 (75-101)	7	250 (228-275)	20	384 (356-412)	8
		IFX 85%/JAK 70%	46 (39-53)	2	139 (125-154)	11	376 (350-402)	30	562 (530-592)	11
		IFX 80%/JAK 60%	45 (39-53)	2	212 (195-232)	17	500 (473-529)	40	758 (725-793)	15
	80%	IFX 90%/JAK 80%	165 (149-182)	7	136 (119-152)	11	249 (226-273)	20	552 (519-582)	11
		IFX 85%/JAK 70%	165 (150-183)	7	204 (185-221)	16	374 (348-401)	30	743 (708-778)	15
		IFX 80%/JAK 60%	165 (149-183)	7	281 (259-300)	22	499 (471-527)	40	944 (907-979)	19
	70%	IFX 90%/JAK 80%	304 (281-324)	12	143 (126-159)	11	250 (227-274)	20	696 (665-729)	14
		IFX 85%/JAK 70%	304 (281-326)	12	209 (189-229)	17	375 (349-402)	30	888 (849-925)	18
		IFX 80%/JAK 60%	304 (283-326)	12	283 (262-304)	23	500 (473-529)	40	1,087 (1,049-1,124)	22
JAK inhibitor in upper high-risk, accelerated IFX in medium-risk, standard IFX in low-risk patients; Low risk 33%, medium risk 33%, high risk 33%	90%	IFX 90%/JAK 80%	20 (16-25)	1	39 (29-48)	2	333 (307-360)	20	391 (363-419)	8
		IFX 85%/JAK 70%	20 (16-25)	1	64 (53-77)	4	501 (469-531)	30	585 (553-617)	12
		IFX 80%/JAK 60%	20 (16-24)	1	106 (93-122)	6	666 (632-700)	40	793 (757-829)	16
	80%	IFX 90%/JAK 80%	64 (53-75)	4	114 (100-130)	7	334 (307-360)	20	514 (480-545)	10
		IFX 85%/JAK 70%	65 (53-76)	4	174 (156-191)	10	501 (468-529)	30	739 (701-772)	15
		IFX 80%/JAK 60%	64 (53-76)	4	245 (226-266)	15	667 (636-699)	40	976 (939-1,017)	20
	70%	IFX 90%/JAK 80%	155 (138-172)	9	149 (132-167)	9	334 (308-361)	20	638 (604-674)	13
		IFX 85%/JAK 70%	156 (140-172)	9	221 (201-240)	13	501 (470-530)	30	876 (840-914)	18
		IFX 80%/JAK 60%	155 (137-172)	9	300 (280-325)	18	667 (632-702)	40	1,122 (1,083-1,165)	22

The numbers of observed colectomies per group are presented as median (95% CI) and the colectomy rates are calculated as a percentage using the median value. Risk stratification models and treatment strategies are each analyzed with 1,000 simulations.

Response defined as proportion of patients avoiding colectomy by 3 months. Standard IFX induction with assumed overall 3-month colectomy-free rate of 77%; accelerated IFX response rate tested at 80%, 85% and 90%, using the same propensity of response as standard IFX; JAK inhibitor response rate tested 60%, 70% and 80%, where propensity to respond to JAK is assumed to be independent of propensity to respond to IFX.

ASUC, acute severe ulcerative colitis; AUC, area under the curve; CI, confidence interval; IFX, infliximab; JAK, Janus kinase inhibitor.

REFERENCES

1. Wang R, Li Z, Liu S, Zhang D. Global, regional and national burden of inflammatory bowel disease in 204 countries and territories from 1990 to 2019: a systematic analysis based on the Global Burden of Disease Study 2019. BMJ Open 2023;13:e065186.

2. Seah D, De Cruz P. Review article: the practical management of acute severe ulcerative colitis. Aliment Pharmacol Ther 2016;43:482-513.

3. Choy MC, Seah D, Faleck DM, et al. Systematic review and meta-analysis: optimal salvage therapy in acute severe ulcerative colitis. Inflamm Bowel Dis 2019;25:1169-1186.

4. Sebastian S, Myers S, Argyriou K, et al. Infliximab induction regimens in steroid-refractory acute severe colitis: a multicentre retrospective cohort study with propensity score analysis. Aliment Pharmacol Ther 2019;50:675-683.

5. Hindryckx P, Novak G, Vande Casteele N, et al. Review article: dose optimisation of infliximab for acute severe ulcerative colitis. Aliment Pharmacol Ther 2017;45:617-630.

6. Gilmore R, Tan WL, Fernandes R, An YK, Begun J. Upadacitinib salvage therapy for infliximab-experienced patients with acute severe ulcerative colitis. J Crohns Colitis 2023;17:2033-2036.

7. Steenholdt C, Dige Ovesen P, Brynskov J, Benedict Seidelin J. Tofacitinib for acute severe ulcerative colitis: a systematic review. J Crohns Colitis 2023;17:1354-1363.

8. Zinger CH, Ringel Y, Eitan M, et al. Upadacitinib for acute severe ulcerative colitis. Inflamm Bowel Dis 2023;29:1667-1669.

9. Vieujean S, Louis E. Precision medicine and drug optimization in adult inflammatory bowel disease patients. Therap Adv Gastroenterol 2023;16:17562848231173331.

10. Cheah E, Huang JG. Precision medicine in inflammatory bowel disease: individualizing the use of biologics and small molecule therapies. World J Gastroenterol 2023;29:1539-1550.

11. Con D, Parthasarathy N, Bishara M, et al. Development of a simple, serum biomarker-based model predictive of the need for early biologic therapy in Crohn’s disease. J Crohns Colitis 2021;15:583-593.

12. Con D, van Langenberg DR, Vasudevan A. Deep learning vs conventional learning algorithms for clinical prediction in Crohn’s disease: a proof-of-concept study. World J Gastroenterol 2021;27:6476-6488.

13. Stafford IS, Gosink MM, Mossotto E, Ennis S, Hauben M. A systematic review of artificial intelligence and machine learning applications to inflammatory bowel disease, with practical guidelines for interpretation. Inflamm Bowel Dis 2022;28:1573-1583.

14. Syal G, Robbins L, Kashani A, et al. Hypoalbuminemia and bandemia predict failure of infliximab rescue therapy in acute severe ulcerative colitis. Dig Dis Sci 2021;66:199-205.

15. Gibson DJ, Doherty J, McNally M, et al. Comparison of medium to long-term outcomes of acute severe ulcerative colitis patients receiving accelerated and standard infliximab induction. Frontline Gastroenterol 2019;11:441-447.

16. Chao CY, Al Khoury A, Aruljothy A, et al. High-dose infliximab rescue therapy for hospitalized acute severe ulcerative colitis does not improve colectomy-free survival. Dig Dis Sci 2019;64:518-523.

17. Nalagatla N, Falloon K, Tran G, et al. Effect of accelerated infliximab induction on short- and long-term outcomes of acute severe ulcerative colitis: a retrospective multicenter study and meta-analysis. Clin Gastroenterol Hepatol 2019;17:502-509.

18. Shah SC, Naymagon S, Panchal HJ, Sands BE, Cohen BL, Dubinsky MC. Accelerated infliximab dosing increases 30-day colectomy in hospitalized ulcerative colitis patients: a propensity score analysis. Inflamm Bowel Dis 2018;24:651-659.

19. Gibson DJ, Heetun ZS, Redmond CE, et al. An accelerated infliximab induction regimen reduces the need for early colectomy in patients with acute severe ulcerative colitis. Clin Gastroenterol Hepatol 2015;13:330-335.

20. Uzzan M, Bresteau C, Laharie D, et al. Tofacitinib as salvage therapy for 55 patients hospitalised with refractory severe ulcerative colitis: a GETAID cohort. Aliment Pharmacol Ther 2021;54:312-319.

21. Berinstein JA, Sheehan JL, Dias M, et al. Tofacitinib for biologic-experienced hospitalized patients with acute severe ulcerative colitis: a retrospective case-control study. Clin Gastroenterol Hepatol 2021;19:2112-2120.

22. Eqbal A, Hilley P, Choy M, Srinivasan A, de Cruz P. Outcomes out to 12 months after sequential use of high-dose tofacitinib following infliximab in acute severe ulcerative colitis. Intern Med J 2023;53:1497-1500.

23. Constant BD, Baldassano R, Kirsch J, Mitchel EB, Stein R, Albenberg L. Tofacitinib salvage therapy for children hospitalized for corticosteroid- and biologic-refractory ulcerative colitis. J Pediatr Gastroenterol Nutr 2022;75:724-730.

24. Xiao Y, Benoit N, Sedano R, et al. Effectiveness of tofacitinib for hospitalized patients with acute severe ulcerative colitis: case series. Dig Dis Sci 2022;67:5213-5219.

25. Komeda Y, Kono M, Kashida H, et al. Successful initial tofacitinib treatment for acute severe ulcerative colitis with steroid resistance: a case series. Ann Gastroenterol 2023;36:97-102.

26. Malakar S, Kothalkar S, Shamsul Hoda U, Ghoshal UC. Tofacitinib in steroid-refractory acute severe ulcerative colitis: a retrospective analysis. Cureus 2023;15:e45416.

27. Honap S, Pavlidis P, Ray S, et al. Tofacitinib in acute severe ulcerative colitis: a real-world tertiary center experience. Inflamm Bowel Dis 2020;26:e147-e149.

28. Gilmore R, Hilley P, Srinivasan A, Choy M, De Cruz P. Sequential use of high-dose tofacitinib after infliximab salvage therapy in acute severe ulcerative colitis. J Crohns Colitis 2022;16:166-168.

29. Kotwani P, Terdiman J, Lewin S. Tofacitinib for rescue therapy in acute severe ulcerative colitis: a real-world experience. J Crohns Colitis 2020;14:1026-1028.

30. Ali NM, Shehab MA. Upadacitinib as a rescue therapy in acute severe ulcerative colitis: a case report and review of the literature. Am J Case Rep 2023;24:e940966.

31. Zhang Z. Parametric regression model for survival data: Weibull regression model as an example. Ann Transl Med 2016;4:484.

32. Choy MC, Seah D, Gorelik A, et al. Predicting response after infliximab salvage in acute severe ulcerative colitis. J Gastroenterol Hepatol 2018;33:1347-1352.

33. Con D, Andrew B, Nicolaides S, van Langenberg DR, Vasudevan A. Biomarker dynamics during infliximab salvage for acute severe ulcerative colitis: C-reactive protein (CRP)-lymphocyte ratio and CRP-albumin ratio are useful in predicting colectomy. Intest Res 2022;20:101-113.

34. Grant RK, Jones GR, Plevris N, et al. The ACE (albumin, CRP and endoscopy) index in acute colitis: a simple clinical index on admission that predicts outcome in patients with acute ulcerative colitis. Inflamm Bowel Dis 2021;27:451-457.

35. Rosen MJ, Minar P, Vinks AA. Review article: applying pharmacokinetics to optimise dosing of anti-TNF biologics in acute severe ulcerative colitis. Aliment Pharmacol Ther 2015;41:1094-1103.

36. Sandborn WJ, Su C, Sands BE, et al. Tofacitinib as induction and maintenance therapy for ulcerative colitis. N Engl J Med 2017;376:1723-1736.

37. Battat R, Hemperly A, Truong S, et al. Baseline clearance of infliximab is associated with requirement for colectomy in patients with acute severe ulcerative colitis. Clin Gastroenterol Hepatol 2021;19:511-518.

38. Con D, Vasudevan A, van Langenberg DR. Predictive scores in acute severe ulcerative colitis: which, what, and when are the decision points we should target? Clin Gastroenterol Hepatol 2022;20:e344-e345.

39. Weisshof R, Ollech JE, El Jurdi K, et al. Ciclosporin therapy after infliximab failure in hospitalized patients with acute severe colitis is effective and safe. J Crohns Colitis 2019;13:1105-1110.

40. Con D, Vasudevan A. Real-world guidance from artificial intelligence? Predicting outcomes of inflammatory bowel disease using machine learning. Dig Dis Sci 2022;67:4604-4605.

41. Choy MC, Visvanathan K, De Cruz P. An overview of the innate and adaptive immune system in inflammatory bowel disease. Inflamm Bowel Dis 2017;23:2-13.

42. Kantasiripitak W, Wicha SG, Thomas D, et al. A model-based tool for guiding infliximab induction dosing to maximize longterm deep remission in children with inflammatory bowel diseases. J Crohns Colitis 2023;17:896-908.

43. Liu XY, Tang H, Zhou QY, et al. Advancing the precision management of inflammatory bowel disease in the era of omics approaches and new technology. World J Gastroenterol 2023;29:272-285.

44. Stidham RW, Takenaka K. Artificial intelligence for disease assessment in inflammatory bowel disease: how will it change our practice? Gastroenterology 2022;162:1493-1506.