Distinct predictive factors for acute severe ulcerative colitis in East Asian and Western populations

Abstract

Background

This study aimed to develop predictive models for acute severe ulcerative colitis (ASUC) using data from East Asian (EA) patients and to validate their performance in an Australia/New Zealand (ANZ) cohort.

Methods

A retrospective international study was conducted across 23 referral hospitals in EA and ANZ, enrolling consecutive ASUC patients between January 2015 and December 2022. Logistic regression analyses were used to construct predictive models for 1-year colectomy and non-response to corticosteroid therapy (NRS).

Results

Overall, 826 patients with ASUC were included (411 EA and 415 ANZ). Among EA patients, independent predictors of 1-year colectomy included female sex, prior exposure to tumor necrosis factor inhibitors, and admission albumin ≤3 g/dL. Independent predictors of NRS in the EA cohort were age at diagnosis  ≤37 years, baseline steroid use, admission albumin ≤2.5 g/dL, and the presence of extraintestinal manifestations. The EA-derived scoring systems demonstrated strong predictive performance within the EA cohort (1-year colectomy: P < .0001; NRS: P = .001) but showed limited utility in the ANZ cohort (P = .106 and P = .012). In contrast, European-developed models—including the French 1-year colectomy score and the ADMIT-ASC index for NRS—accurately predicted outcomes in the ANZ cohort (P = .007 and P < .0001) but had reduced predictive capacity in the EA cohort (P = .106 and P = .026). These findings were consistent in both cohorts following propensity score matching.

Conclusions

Predictive factors for ASUC differ substantially between EA and ANZ patients, highlighting the need for population-specific predictive tools.

What is already known? 

Many scoring systems for acute severe ulcerative colitis (ASUC) currently in use were developed based on data from Western cohorts.

What is new here? 

While the predictive accuracy of a scoring system developed using East Asian (EA) patients with ASUC was higher in EA patients, its performance was less accurate in patients from Australia/New Zealand (ANZ).

Conversely, existing predictive models of ASUC developed in European cohorts demonstrated better predictive accuracy for 1-year colectomy and non-response risk to corticosteroid in the ANZ cohort compared with the EA cohort.

How can this study help patient care? 

Given the distinct predictors of clinical outcomes observed in EA and ANZ patients with ASUC, management strategies for Asian patients should be specifically tailored to account for these differences, rather than relying solely on models derived from Western populations.

Introduction

Ulcerative colitis (UC) is a chronic inflammatory disorder that primarily affects the mucosal or submucosal layers of the colon. Due to its relapsing-remitting disease course, UC necessitates long-term management and significantly impacts patients’ quality of life.¹ Approximately 25% of patients with UC develop acute severe ulcerative colitis (ASUC) during the course of their disease.² ASUC is characterized by a severe disease flare accompanied by systemic inflammatory features such as fever, anemia, tachycardia, and elevated serum inflammatory markers, in addition to bloody diarrhea.³ Patients with ASUC often require hospitalization and intensified treatment, as the condition is associated with a substantial risk of colectomy (13%-23%) and mortality (1%).^4,5

Given the poor prognosis of ASUC, the ability to predict clinical outcomes, including the risk of colectomy and failure of intravenous (IV) corticosteroid therapy, is essential for guiding timely and effective management decisions.⁶ Existing predictive models for ASUC outcomes have been developed exclusively in Western populations.^5,7,8 However, emerging evidence highlights significant differences in the clinical characteristics of UC between Eastern and Western populations. Specifically, the risk of colectomy and non-response to corticosteroid (NRS) therapy is lower in Asian patients compared to their Western counterparts.^9–16 As such, predictive models developed in Western settings may not be directly applicable to Asian patients with ASUC. Currently, no predictive models for ASUC outcomes have been developed based on East Asian (EA) populations.

The objective of this study was to develop a predictive model for clinical outcomes in EA patients with ASUC and to evaluate its external validity in an Australia/New Zealand (ANZ) cohort as a Western comparator. Additionally, existing scoring systems developed in European countries were applied to both EA and ANZ patient cohorts, and their predictive performance was compared between these populations.

Methods

Patients

This retrospective, multinational, multicenter study was conducted across 14 referral hospitals in EAs (China, Japan, Korea, Taiwan) and 9 referral hospitals in ANZ. Electronic medical records of patients aged over 17 years, admitted between January 2015 and December 2022 for ASUC as defined by the modified Truelove and Witts’ criteria, were reviewed at each participating hospital. Patients referred specifically for colectomy or those with incomplete data were excluded from the study. Data collected included demographic characteristics (sex, age at diagnosis, and age at enrollment), prior hospitalization history, previous and current UC medications, albumin and C-reactive protein (CRP) levels at admission, presence of extraintestinal manifestations (EIM) at admission, initial ASUC treatment at admission, infections with cytomegalovirus (CMV) or Clostridioides difficile (C. diff), Mayo endoscopic score (MES) or UC endoscopic index of severity (UCEIS) at admission, the need for rescue therapy following corticosteroid therapy, and colectomy during hospitalization or within 1 year of discharge. CMV and C. difficile infections were defined based on treatment initiation. Diagnostic testing was performed according to local institutional protocols at each participating center; however, due to variability in diagnostic approaches and incomplete data, a uniform definition based on standardized diagnostic criteria could not be applied. Institutional Review Board approval was obtained at each participating hospital. Comparison of main disease outcomes including colectomy between these cohorts was previously reported.¹⁷

Main outcomes and predictive models

The primary outcomes were colectomy within 1 year after admission for ASUC and NRS. NRS was defined as the need for rescue medical therapy, including biologics, cyclosporine, or colectomy, during the index hospitalization. Decisions regarding escalation to rescue therapy, including biologics or cyclosporine, were made at the discretion of the treating physicians at each participating center, as standardized criteria were not uniformly available across sites. Patients who were not initially treated with IV corticosteroids were excluded from the NRS analysis. Predictive variables for modeling were selected using logistic regression analysis in the EA cohort. Predictive models for NRS were developed using Japanese and Korean patients as the development set (n = 248) and validated using Chinese and Taiwanese patients as the validation set (n = 92). Due to the limited number of colectomy cases available for multivariate analysis, the EA cohort could not be subdivided for model development and validation of 1-year colectomy outcomes. Consequently, model development and validation for 1-year colectomy were performed on the entire EA cohort. Score weights were assigned based on the odds ratios of selected parameters.¹⁸ These two models developed using the EA cohort were validated in the ANZ cohort.

Existing predictive models, including the French scoring system for 1-year colectomy⁵ and the Admission Model for the Intensification of Therapy in Acute Severe Colitis (ADMIT-ASC) index for NRS, were evaluated for comparison.⁷ The French scoring system includes previous treatment with tumor necrosis factor (TNF) inhibitors or thiopurines, C. diff infection, CRP, and albumin, while the ADMIT-ASC index incorporates CRP, albumin, and UCEIS scores. The performance of these models was compared between the EA and ANZ cohorts.

Statistical analysis

Continuous variables were presented as medians with interquartile ranges (IQRs) or means with standard deviations (SDs) and were compared using the Mann-Whitney U test or Student’s t-test, as appropriate. Categorical variables were reported as counts and percentages and were compared using the χ² test or Fisher’s exact test. Variables with a P-value <0.1 in univariate analysis were included in the multivariate logistic regression model. Different albumin thresholds were used for the two models based on multivariable analysis results. For the 1-year colectomy model in the EA cohort, a total of 16 events were observed. Given the limited number of events, the number of variables included in the multivariable logistic regression model was restricted to avoid model overfitting. Variables with a P-value <0.1 in univariate analysis were considered for inclusion, and only clinically relevant predictors were retained in the final model. For the development of the scoring system, point values were assigned based on the magnitude of the regression coefficients. As the selected predictors showed similar effect sizes in the multivariable analysis (Table S1), a simplified approach assigning one point to each variable was adopted to enhance clinical applicability. The discriminative performance of the prediction models was assessed using receiver operating characteristic (ROC) curve analysis, and the area under the curve (AUC) was calculated to quantify model discrimination in both the EA and ANZ cohorts. A one-to-one propensity score matching (PSM) analysis using the nearest-neighbor method was performed to balance baseline characteristics between the two cohorts. A caliper width of 0.2 was applied to specify the maximum allowable difference in propensity scores between matched pairs. Variables that differed significantly between groups—primarily demographic factors such as age at admission, disease duration, family history of inflammatory bowel disease (IBD), and prior steroid use—were included in the matching process. Current therapies and laboratory parameters were excluded from PSM, as they were considered potential predictors of the primary outcomes. A P-value <0.05 was considered statistically significant. Data preparation and analyses were conducted using R software, version 4.4.1.

Results

Baseline characteristics of patients

A total of 411 and 415 patients were included from EA and ANZ, respectively. The EA cohort comprised patients from China (53 [12.9%]), Japan (63 [15.3%]), Korea (229 [55.7%]), and Taiwan (66 [16.1%]). The ANZ cohort included patients from Australia (377 [90.8%]) and New Zealand (38 [9.2%]). Patients in the ANZ cohort were younger at both diagnosis and hospital admission compared to those in the EA cohort. The proportions of prior corticosteroid use (48.0% vs 24.8%, P < .001), current corticosteroid use (41.1% vs 29.9%, P < .001), and previous biologics and/or thiopurine therapy (48.0% vs 24.8%, P < .001) were significantly higher in the ANZ cohort. Family history of IBD was more observed in the ANZ cohort (P < .001). Disease duration was longer in ANZ patients (P = .018). Conversely, the EA cohort showed a greater prevalence of E3 extent (pancolitis, P < .001) and MES 3 at admission (P = .001). CMV infection was more found in the EA patients (P < .001). Hemoglobin level was lower in the EA cohort (P < .001) while albumin level was lower in the ANZ cohort (P = .016). Baseline characteristics of the patients are summarized in Table 1.

Table 1.

Baseline characteristics of patients from EAST Asia and Australia/New Zealand.

	East Asia, n = 411	ANZ, n = 415	P-value
Female	169 (41.1)	198 (47.7)	0.066
Age at diagnosis, year, mean ± SD	39.6 ± 13.6	34.5 ± 16.8	<0.001
Age at diagnosis, year, median (IQR)	38 (26-52)	29.5 (22-45)	<0.001
Age at admission, year,	43.4 ± 16.2	39.5 ± 16.6	0.001
Family history of IBD	10 (2.4)	58 (16.7)	<0.001
Smoking			0.430
Never	291 (70.8)	276 (66.8)
Ex-	83 (20.2)	98 (23.7)
Current	37 (9)	39 (9.4)
Charlson Index ≥1	77 (18.7)	96 (23.2)	0.137
Disease duration, year	1 (0-5)	2 (0-7)	0.018
ASUC within 3 months after diagnosis	120 (29.2)	113 (27.3)	0.596
Previous UC-related admission	147 (35.8)	153 (37)	0.777
Previous steroid use	153 (37.2)	248 (59.9)	<0.001
Previous anti-TNF or thiopurine	21 (24.8)	199 (48)	<0.001
Current steroid use	123 (29.9)	170 (41.1)	0.001
Extent at admission			<0.001
E1 (Proctitis)	7 (1.7)	29 (7.3)
E2 (Left sided)	115 (28.3)	160 (40.5)
E3 (Extensive)	285 (70)	206 (52.2)
EIM at admission	31 (7.5)	21 (5.1)	0.463
C. diff infection	30 (7.3)	17 (4.1)	0.051
CMV infection	45 (10.9)	6 (1.5)	<0.001
Additional TW criteria, number	2 (3-1)	2 (2-1)	0.131
Hemoglobin at admission, g/dL	11.5 (10-13)	12.6 (11.2-13.8)	<0.001
C-reactive protein at admission, mg/dL	5.1 (2.1-10.8)	4.9 (2-10.5)	0.698
Albumin at admission	3.3 (2.9-3.7)	3.2 (2.6-3.7)	0.016
MES at admission			0.001
1	0	8 (2.1)
2	102 (25.2)	119 (31.5)
3	303 (74.8)	251 (66.4)

Additional Truelove-Witts (TW) criteria include body temperature >37.8°C, pulse rate >90 beats/min, hemoglobin <10.5 g/dL, and CRP level >3 mg/dL. Values are expressed as median (interquartile range), mean ± standard deviation, or n (%) as applicable.

Abbreviations: IBD, inflammatory bowel disease; ASUC, acute severe ulcerative colitis; TNF, tumor necrosis factor; EIM, extraintestinal manifestation; C. diff, Clostridioides difficile; CMV, cytomegalovirus; TW, Truelove-Witts; MES, Mayo endoscopic score; IQR, interquartile range; SD, standard deviation.

Short-term outcomes were also compared between the two cohorts. The 30-day colectomy rate was significantly lower in the EA cohort compared with the ANZ cohort (1.5% vs 14.2%, P < .001). In contrast, in-hospital mortality was low in both groups and did not differ significantly between cohorts (0.2% vs 0%, P = .996).

Prediction model for 1-year colectomy

The 1-year colectomy rate was significantly lower in the EA cohort compared to the ANZ cohort (16/411, 3.9% vs 94/415, 22.7%, P < .001, Figure 1A). In a multivariate analysis using data from the EA cohort, female sex (odds ratio [OR] 3.399, 95% confidence interval [CI] 1.137-10.161, P = .028), albumin level ≤ 3 g/dL at admission (OR 3.874, 95% CI, 1.301-11.539, P = .015), and prior anti-TNF therapy (OR 3.586, 95% CI, 1.043-12.332, P = .043) were identified as independent predictors of 1-year colectomy (Table 2). Based on these findings, a scoring system was developed, assigning one point to each variable, resulting in a total score range of 0-3. In the EA cohort, the predicted probabilities of 1-year colectomy for scores of 0, 1, 2, and 3 were 0.7%, 3.1%, 9.7%, and 28.6%, respectively. The scoring model demonstrated acceptable discriminatory ability within the EA cohort (P < .0001). However, when applied to the ANZ cohort, the model was unable to accurately predict colectomy risk (P = .106). In the ANZ cohort, the observed 1-year colectomy rates for scores of 0, 1, 2, and 3 were 16.8%, 20.6%, 29.5%, and 33.3%, respectively (Figure 2A). ROC analysis demonstrated that the EA-derived model showed good discrimination in the EA cohort (AUC 0.75), but reduced performance in the ANZ cohort (AUC 0.58, Figure 3A).

Figure 1.

Patient flow for colectomy (A) and non-response to steroids (B). Chi-square test: P < .001 for comparisons of colectomy and non-response to steroids between groups. ANZ: Australia and New Zealand.

Table 2.

Multivariate analysis of predictive factors for 1-year colectomy in EAST Asian patients.

	Univariate			Multivariate
	Non-colectomy (n = 395)	Colectomy (n = 16)	P-value	Odds ratio	95% CI	P-value
Female	158 (40)	11 (68.8)	0.035	3.399	1.137-10.161	0.028
Age at diagnosis (mean ± SD)	39.5 ± 16.2	42.4 ± 19.6	0.486
Family history of IBD	10 (2.5)	0	0.999
Smoking			0.301
Never	227 (70.1)	14 (87.5)
Ex-	82 (20.8)	1 (6.3)
Current	36 (9.1)	1 (6.3)
Disease extent at admission			0.447
E1 (Proctitis)	7 (1.8)	0
E2 (Left sided)	113 (28.6)	2 (12.5)
E3 (Extensive)	271 (68.6)	14 (87.5)
Unknown	7 (1.8)	0
Disease duration <1 month	117 (29.6)	3 (18.8)	0.416
Previous admission	138 (34.9)	9 (56.3)	0.109
EIM	29 (7.3)	2 (12.5)	0.344
C-reactive protein ≥3 mg/dL	263 (66.6)	12 (75)	0.595
Albumin ≤3 g/dL	139 (35.2)	11 (68.8)	0.014	3.874	1.301-11.539	0.015
CMV treatment	42 (10.6)	3 (18.8)	0.401
C. diff treatment	29 (7.3)	1 (6.3)	0.999
MES = 3	288 (73)	14 (87.5)	0.378
UCEIS ≥4	370 (95.1)	14 (87.5)	0.198
UCEIS ≥7	153 (39.3)	9 (56.3)	0.199
Previous steroid use	145 (36.7)	8 (50)	0.3
Previous anti-TNF	33 (8.4)	4 (25)	0.046	3.586	1.043-12.332	0.043
Previous thiopurine	43 (10.9)	0	0.393
Current steroid use	121 (30.6)	2 (12.5)	0.165

Abbreviations: IBD, inflammatory bowel disease; TNF, tumor necrosis factor; EIM, extraintestinal manifestation; C.diff, Clostridioides difficile; CMV, cytomegalovirus; MES, Mayo endoscopic score; UCEIS, ulcerative colitis endoscopic index of severity; SD, standard deviation.

Figure 2.

Proportion of 1-year colectomy (A) and non-response to steroids (B) following acute severe ulcerative colitis, based on the prediction model developed using the East Asian cohort. ANZ: Australia and New Zealand.

Figure 3.

Receiver operating characteristic (ROC) curves for prediction of 1-year colectomy (A) and non-response to corticosteroid therapy (B) using the East Asia (EA)-derived model in the EA and Australia/New Zealand (ANZ) cohorts. The model demonstrated better discrimination in the EA cohort compared with the ANZ cohort.

Prediction model for NRS therapy

A total of 116 patients who were not initially treated with IV corticosteroids were excluded from the NRS analysis, leaving 340 patients in the EA cohort and 370 patients in the ANZ cohort (Figure 1B). The NRS rate in the EA cohort was significantly lower compared to the ANZ cohort (25% vs 58.9%, P < .001, Figure 1B). For the development of the prediction model within the EA cohort, data from Japanese and Korean patients (n = 248) were used as the development dataset, with validation conducted using an independent dataset of Chinese and Taiwanese patients (n = 92) and the ANZ cohort (n = 370). In the development cohort, four factors emerged as independent predictors of NRS: age at diagnosis ≤ 37 years (OR 2.151, 95% CI, 1.133-4.085, P = .019), the presence of EIM (OR 3.129, 95% CI, 1.164-8.414, P = .024), albumin ≤ 2.5 g/dL at admission (OR 3.228, 95% CI, 1.370-7.605, P = .007), and current corticosteroid use (OR 2.199, 95% CI 1.071-4.513, P = .032) (Table 3).¹⁸ Based on the odds ratios, all factors had ORs ≥ 2.0, and a simplified scoring system was developed where one point was assigned to each factor, resulting in a score range of 0-4.

Table 3.

Multivariate analysis of predictive factors for non-response to corticosteroid therapy in East Asian patients.

	Univariate			Multivariate
	Responder (n = 179)	Non-responder (n = 69)	P-value	Odds ratio	95% CI	P-value
Female	72 (40.2)	37 (53.6)	0.064	0.575	0.311-1.062	0.077
Age at diagnosis ≤37	80 (44.7)	42 (60.9)	0.024	2.151	1.133-4.085	0.019
Family history of IBD	3 (1.7)	3 (4.3)	0.352
Smoking			0.832
Never	110 (61.5)	44 (63.8)
Ex-	48 (26.8)	16 (23.2)
Current	21 (11.7)	9 (13)
Disease extent at admission			0.488
E1 (proctitis)	3 (1.7)	1 (10.4)
E2 (left sided)	55 (30.7)	16 (23.9)
E3 (extensive)	121 (67.6)	52 (75.4)
Disease duration <1 month	60 (33.5)	14 (20.3)	0.045	0.846	0.380-1.88	0.681
Previous admission	44 (24.6)	28 (40.6)	0.019	1.084	0.515-2.278	0.832
EIM	9 (5)	12 (17.4)	0.004	3.129	1.164-8.413	0.024
CRP ≥ 10 mg/dL	46 (25.7)	23 (33.3)	0.269
Alb ≤ 2.5 g/dL	18 (10.1)	14 (20.3)	0.036	3.228	1.370-7.605	0.007
C. diff infection treatment	14 (7.8)	8 (11.6)	0.332
MES = 3	122 (68.2)	55 (79.7)	0.085	1.240	0.592-2.598	0.569
UCEIS ≥ 4	165 (92.2)	67 (97.1)	0.248
UCEIS ≥ 7	57 (31.8)	27 (39.1)	0.297
Previous steroid use	56 (31.3)	35 (50.7)	0.004	1.758	0.776-3.981	0.176
Anti-TNF or thiopurine exposure (previous or current)	35 (19.6)	22 (31.9)	0.044	1.002	0.451-2.225	0.996
Current 5-ASA	128 (71.5)	46 (66.7)	0.536
Current thiopurine	21 (11.7)	12 (17.4)	0.296
Current steroid	32 (17.9)	23 (33.3)	0.011	2.199	1.071-4.513	0.032

Abbreviations: IBD, inflammatory bowel disease; TNF, tumor necrosis factor; EIM, extraintestinal manifestation; C. diff, Clostridioides difficile; MES, Mayo endoscopic score; UCEIS, ulcerative colitis endoscopic index of severity; ASA, aminosalicylic acid.

In the validation cohort of Chinese and Taiwanese patients, the NRS percentages for scores of 0, 1, 2, and 3/4 were 8%, 9.1%, 40%, and 66.7%, respectively, indicating strong discriminative ability for NRS risk in EA patients (P = .001). However, in the ANZ cohort, the scoring model showed reduced predictive utility; NRS rates for scores of 0, 1, 2, and 3/4 were 51%, 53.1%, 67.2%, and 80%, respectively (P = .012, Figure 2B). Notably, over half of the patients with a score of 0 in the ANZ cohort experienced NRS, rendering the scoring system less effective as a predictive model in this population. Similarly, the EA-derived model showed moderate discrimination in the EA cohort (AUC 0.69) and lower performance in the ANZ cohort in ROC analysis (AUC 0.60, Figure 3B).

Performance of the European Indices for ASUC outcomes

The predictive performance of the French 1-year colectomy scoring system was evaluated in both cohorts. This model considers previous anti-TNF or thiopurine use, C. diff infection, CRP >3 mg/dL at admission, and albumin <3 g/dL at admission, with each factor contributing one point (score range: 0-4).⁵ In the EA cohort, the colectomy percentages for scores of 0, 1, 2, and 3/4 were 1.3%, 2%, 6.5%, and 8.3%, respectively, demonstrating limited predictive value (P = .106). Conversely, in the ANZ cohort, the model exhibited stronger discriminative ability, with 1-year colectomy rates of 7.9%, 17.9%, 25.7%, and 34.5% for scores of 0, 1, 2, and 3/4, respectively (P = .007, Figure 4A).

Figure 4.

Proportion of 1-year colectomy (A) and non-response to steroids (B) following acute severe ulcerative colitis, according to the French colectomy model (A) and the ADMIT-ASC index (B). ANZ: Australia and New Zealand.

The ADMIT-ASC index, a four-point model for predicting NRS, was also applied to both cohorts. This model incorporates CRP ≥ 10 mg/dL at admission, albumin ≤ 2.5 g/dL at admission, and a UCEIS score ≥ 4 or ≥ 7.⁷ Due to missing UCEIS data, the MES = 3 was used as a substitute endoscopic factor, resulting in a scoring range of 0-3 points. In the EA cohort, the NRS rates for scores of 0, 1, 2, and 3 were 15.9%, 22.8%, 33.3%, and 43.8%, respectively (P = .026). In the ANZ cohort, NRS rates were as follows: 41.5%, 59.8%, 70.9%, and 78.6% for scores of 0, 1, 2, and 3, respectively (P < .0001, Figure 4B).

Results of PSM analysis

After PSM, 298 patients and 247 patients were included from each cohort for the colectomy and NRS analyses, respectively (Tables S2 and S3). The EA colectomy model demonstrated good discriminative ability for predicting 1-year colectomy risk within the EA cohort (P < .0001); however, its performance did not generalize to the ANZ cohort (P = .506) (Figure S1A). Similarly, the EA NRS model accurately predicted NRS risk in the EA cohort (P < .001), but not in the ANZ cohort (P = .120) (Figure S1B).

Conversely, the French 1-year colectomy scoring system effectively differentiated colectomy risk in the ANZ cohort (P = .048), but showed no significant predictive value in the EA cohort (P = .167) (Figure S2A). Likewise, the ADMIT-ASC index predicted NRS risk more accurately in the ANZ cohort (P = .016) than in the EA cohort (P = .454) (Figure S2B).

Discussion

This retrospective, international, multicenter study on EA patients with ASUC identified female sex, admission albumin levels ≤3 g/dL, and prior anti-TNF use as predictive factors for 1-year colectomy. Conversely, an age at diagnosis ≤37 years, the presence of EIM, admission albumin levels ≤2.5 g/dL, and current corticosteroid use were predictive factors for NRS. Notably, the scoring system derived from these variables demonstrated reduced predictive accuracy for disease outcomes in ANZ patients compared to EA patients. However, predictive models of ASUC originally developed in European countries appeared to better differentiate the risks of 1-year colectomy and NRS in the ANZ cohort than in the EA cohort, suggesting distinctive clinical features of ASUC between Eastern and Western populations. These findings were consistent in both cohorts following PSM analysis. The substantial baseline differences between the EA and ANZ cohorts, particularly in prior biologic exposure and corticosteroid use, may have influenced clinical outcomes. Although we applied PSM to minimize these differences, residual confounding cannot be completely excluded.

The observed disparities in ASUC prognosis between Eastern and Western populations may have a genetic basis. A Danish nationwide genome-wide association study identified the HLA-DRB1*01:03 genotype as a marker of severe UC, associated with colectomy, hospitalization, and systemic corticosteroid use.¹⁹ This genotype is predominantly found in individuals of Western European descent and is rare in Asian populations, providing a plausible explanation for the better clinical outcomes observed in EA patients with ASUC.²⁰ However, this hypothesis requires validation through prospective comparative studies between Eastern and Western ASUC cohorts. As a younger age at diagnosis is a known poor prognostic factor for UC,²¹ the differing ages at diagnosis between groups may contribute to the variations in prognosis. In our study, the median age at diagnosis in the EA cohort was significantly higher than in the ANZ cohort (38 vs 29 years, P < .001; Table 1). Patients in European studies, such as the French one-year colectomy and the ADMIT-ASC index for NRS, had a similar age at diagnosis (median 29-30 years) to the ANZ patients, but were younger than the EA patients. However, in the current study, the results maintained after controlling age between groups using PSM analysis (Figures S1 and S2). Factors such as access to tertiary hospitals and variations in the level of ASUC care provided at primary and secondary hospitals between Eastern and Western regions may be additional contributors. Further research is needed to clarify these aspects.

Low albumin levels at admission and prior use of TNF inhibitors were common predictors of 1-year colectomy risk in both the French colectomy model⁵ and this study’s scoring system. Hypoalbuminemia, a negative acute phase reactant associated with poor therapeutic response and increased colectomy risk, has been consistently incorporated into ASUC scoring systems.^5,7,22–25 Additionally, given that infliximab is a widely used salvage therapy in ASUC, it is unsurprising that prior exposure to or failure of anti-TNF therapy serves as a predictor of colectomy. While C. difficile infection was identified as a predictor of 1-year colectomy in the French study,⁵ it was not included in this model. Although C. difficile infection is associated with adverse outcomes in UC in Western countries,^26,27 it did not appear to influence disease progression in Asian UC patients. A prospective Korean study demonstrated no significant differences in mortality, length of hospital stay, or colectomy rates between relapsed UC patients with and without C. difficile infection.²⁸ The lower prevalence of the hypervirulent C. difficile ribotype 027 strain in Korea compared to Western countries may account for this disparity.^29,30 Interestingly, female sex was identified as a risk factor for 1-year colectomy in the EA cohort, contrasting with the established association of male sex with colectomy risk in UC.³¹ This discrepancy may reflect differences in surgical thresholds between genders across ethnicities. While genetic and environmental factors likely contribute to the differing demographic and clinical characteristics of Asian and Western UC patients,³² further studies are warranted to validate these findings in Asian populations.

In addition to an initially low albumin level, age at diagnosis under 37 years, concurrent corticosteroid uses at admission, and the presence of EIMs were identified as predictors of NRS in the EA cohort. Despite differences in study design, findings from the Swiss UC cohort similarly demonstrated that a younger age at diagnosis and the presence of EIMs were independent factors associated with poor clinical outcomes.³³ Additionally, an Indian study on ASUC identified EIMs as a predictor for colectomy.³⁴ Prolonged use of oral corticosteroids has been previously reported to correlate with NRS at admission.^24,35 Consistent with our findings, a recent Japanese study on ASUC observed an increased risk of NRS in patients with concomitant corticosteroid use at the time of registration compared to those not using corticosteroids.³⁶

CRP has been recognized as a predictor of one-year colectomy and NRS risk in Western patients with ASUC.^5,7 However, CRP was not included in our models for the EA cohort. Notably, other Korean studies on UC patients found no significant difference in baseline CRP levels between responders and non-responders (or colectomy and non-colectomy groups), although follow-up CRP levels showed significant prognostic value for disease outcomes.^10,25 Similarly, initial endoscopic activity—an essential predictor of poor clinical outcomes in Western models⁷—was not significant in the EA cohort. While the underlying cause is unclear, we hypothesize that the prognostic value of these initial activity markers may be offset by more favorable clinical outcomes in EA patients with ASUC, including lower colectomy rates and higher responsiveness to corticosteroid therapy.

We selected the French 1-year colectomy model⁵ and the ADMIT-ASC index for NRS⁷ because they utilized variables assessed on the first day of admission, aligning closely with the design of our study. Moreover, these studies were conducted during the biologic era, making them more comparable to our study population.

The primary limitation of this study was its retrospective design, which resulted in some data being unavailable. For example, the UCEIS was not evaluated in all patients, necessitating the use of the MES instead. Therefore, the ADMIT-ASC index was applied using a modified definition, substituting MES = 3 for UCEIS ≥ 7. This modification may have affected the model’s performance and limits direct comparison with the original index. Additionally, no blood or fecal samples were stored from the patients, precluding the analysis of genetic and microbial characteristics that could have provided further insights into the observed differences between EA and ANZ patients with ASUC. In addition, CMV and C. difficile infections were defined based on treatment initiation, which may have introduced misclassification bias. Treatment decisions may vary across centers and may not always reflect confirmed infection. Furthermore, heterogeneity in diagnostic testing approaches and incomplete data limited the use of standardized diagnostic criteria across all sites. Furthermore, due to the lack of a validation cohort for assessing the 1-year colectomy risk, the scoring system developed using the EA cohort requires validation in independent cohorts of EA patients with ASUC. The relatively small number of colectomy events in the EA cohort (n = 16) may have limited the stability of the multivariable model and increased the risk of overfitting. In addition, because the 1-year colectomy model was evaluated within the same cohort used for its development, the reported performance may be subject to optimistic bias. Therefore, the results should be interpreted with caution and require validation in independent cohorts.

In conclusion, the predictive models developed from the EA cohort appear to be less accurate in predicting disease outcomes for the ANZ cohort compared to the EA cohort. In contrast, the French colectomy model and the British ADMIT-ASC index demonstrated superior performance in the ANZ cohort relative to the EA cohort. These findings underscore the importance of tailoring management strategies for Asian patients with ASUC to account for the differing predictors of outcomes observed in comparison to Western patients.

Glossary

Abbreviations

ADMIT-ASC

Admission Model for the Intensification of Therapy in Acute Severe Colitis

ANZ

Australia/New Zealand

ASA

aminosalicylate

ASUC

acute severe ulcerative colitis

C. diff

Clostridioides difficile

CI

confidence interval

CMV

cytomegalovirus

CRP

C-reactive protein

EA

East Asia

EIM

extraintestinal manifestation

NRS

non-response to corticosteroid therapy

MES

Mayo endoscopic score

UCEIS

ulcerative colitis endoscopic index of severity

IQR

interquartile ranges

IRB

institutional Review Board

IV

intravenous

OR

odds ratio

SD

standard deviation

TNF

tumor necrosis factor

Author contributions

E.S.K. conceptualized the study, analyzed data and wrote the manuscript. D.H.K., S.J.K., S.H.P., H.S.L., S.K.K., H.S.K., J.L., K.O.K., B.I.J., Y.J.L., E.M.S., D.S.K., C.C.L., J.W.Y.M., J.L., Q.C., S.C.W., W.C.L., W.H.H., S.S., R.G.F., R.G., Y.K.A., J.B., K.R., L.R.C., A.V., J.Z., A.B., B.N., K.L., C.H., L.M.L.B., E.W., E.W., N.C., T.S., G.M., G.M., K.G., M.G. and M.S. collected the data. T.K. conceptualized, designed, and supervised the study. All authors have read the manuscript and offered critical comments.

Supplementary material

Supplementary data is available at Inflammatory Bowel Diseases online.

Funding

This study was supported by an International Joint Research Grant of the Korean Association for the Study of Intestinal Diseases for 2022 (2022-3). This research was supported by Kyungpook National University Research Fund, 2025.

Conflicts of interest

E.S.K. has served as an advisory board member or speaker for Abbive, Eli Lilly, J&J Pharmaceuticals, Bristol-Myers Squibb, Takeda, Pfizer, Samsung Bioepis, Celltrion, and Ferring Pharmaceuticals. S.S. has served as an advisory board member or speaker for AbbVie, Alimentiv, Eli Lilly, Janssen Pharmaceutical K.K., Gilead Sciences, Inc., JIMRO Co., Ltd., KISSEI Pharmaceutical Co., Ltd., Kyorin Pharmaceutical Co., Ltd., Mitsubishi Tanabe Pharma Corporation, EA Pharma Co., Takeda Pharmaceutical Co., Ltd., Nippon Kayaku Co., Ltd., and Zeria Pharmaceutical Co., Ltd. and has received research grants from Gilead Sciences, Bristol-Myers Squibb, and Ferring Pharmaceuticals. T.K. reports serving as an advisory board member, consultant, or speaker for AbbVie GK, Alfresa Pharma Corporation, Alimentiv, Bristol Myers Squibb, Celltrion, Covidien, EA Pharma, Eli Lilly, Ferring Pharmaceuticals, Gilead Sciences, Janssen Pharmaceutical K.K., JIMRO, Kissei Pharmaceutical, Kowa, Kyorin Pharmaceutical, Mitsubishi Tanabe Pharma Corporation, Mochida Pharmaceutical, MSD, Nippon Kayaku, Pfizer, Sanofi, Sekisui Medical, Spyre, Takeda Pharmaceutical, TOPPAN Holdings, and Zeria Pharmaceutical and receiving research funding from AbbVie GK, Alfresa Pharma Corporation, Bristol Myers Squibb, EA Pharma, Gilead Sciences, Helmsley Charitable Trust, JIMRO, Kyorin Pharmaceutical, Miyarisan Pharmaceutical, Mochida Pharmaceutical, Nippon Kayaku, Pfizer, Sekisui Medical, Samsung Medison, Takeda Pharmaceutical, and Zeria Pharmaceutical. AV has served as an advisory board member for AbbVie and has received speaker or consulting fees from AbbVie, Ferring Pharmaceuticals and Pfizer. Mayur Garg has served on the advisory board of AbbVie, Pfizer and Ferring, and has received speaker fees, research or travel grants from AbbVie, Celltrion, Dr Falk, Ferring, Fresenius Kabi, Janssen, Pfizer, and Takeda. Craig Haifer has received speaker fees, educational support and been an advisory board member with Pfizer, Takeda, Ferring, Dr Falk, BiomeBank, Celtrion and AbbVie. Other authors do not have conflict of interest. Jing Liu has served as speaker for Takeda and Ferring Pharmaceuticals. Qian Cao has served as an advisory board member for Bristol-Myers Squibb, Janssen Pharmaceuticals, has received research funding from Takeda, Janssen Pharmaceuticals.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.