Predictive Value of Imaging-Based Classifications for Conservative Management Failure in Isolated Superior Mesenteric Artery Dissection

Abstract

Background

Isolated superior mesenteric artery dissection (ISMAD) is a relatively uncommon vascular disease with uncertain optimal management. Multiple imaging-based classifications have been developed, but their ability to predict conservative management failure remains unclear.

Objectives

This study aimed to compare the predictive performance of 8 ISMAD classification systems for conservative management failure.

Methods

This multicenter retrospective study included 156 patients with ISMAD from 3 centers in China (October 2009 to February 2023). The median computed tomography angiography follow-up time was 7.9 months (IQR: 3.7-24.3). Patients were classified using 8 systems and grouped as successful (108 of 156, 69.2%) or failed (48 of 156, 30.8%) conservative management. Performance was evaluated by model fit (Akaike Information Criterion, Bayesian Information Criterion), discrimination (C-statistic), net reclassification improvement, calibration, and decision curve analysis.

Results

Compared with the successful group, patients who failed conservative management had a longer duration of abdominal pain before admission (8 [IQR: 3-30] days vs 4 [IQR: 1-11] days; P = 0.0214) and more severe superior mesenteric artery degree of stenosis (0.83 [IQR: 0.69-1.00] vs 0.64 [IQR: 0.52-1.00]; P = 0.00544). Among all classifications, Tan’s classification showed good model fit (Akaike Information Criterion = 186; Bayesian Information Criterion = 226), the highest discrimination (C-statistic = 0.750; 95% CI: 0.670-0.830), and the greatest improvement (net reclassification improvement = 0.593; 95% CI: 0.244-0.903). It also demonstrated good calibration and yielded the highest net benefit across clinically relevant threshold probabilities in decision curve analysis.

Conclusions

Tan’s classification showed the best overall performance in predicting conservative management failure and may help guide timely interventions.

Central Illustration

Isolated superior mesenteric artery dissection (ISMAD) is characterized by dissection of the superior mesenteric artery (SMA) occurring independently of aortic dissection. For decades, it was considered a rare peripheral vascular disorder. In 1959, based on autopsy, Foord and Lewis¹ reported a prevalence of ISMAD of only 0.06%. To date, epidemiological studies on ISMAD remain scarce. With the widespread adoption of advanced imaging techniques, particularly computed tomography angiography (CTA), the number of diagnosed cases has increased markedly, especially in East Asian countries such as China, Japan, and South Korea.² According to recent systematic reviews, the mean age at onset of ISMAD is 54 to 55.7 years, and 80% to 89% of cases occurring in men.2, 3, 4, 5, 6, 7, 8, 9, 10

The clinical consequences of arterial dissection largely depend on the location of the affected vessel, particularly when arteries supplying vital organs are involved.11, 12, 13 ISMAD may compromise arterial integrity and mesenteric perfusion. Consequently, the clinical presentation of ISMAD ranges from asymptomatic incidental findings to acute abdominal pain, vomiting, or abdominal distension.⁵ Without timely intervention, a subset of patients may progress to life-threatening intestinal necrosis or even catastrophic complications such as SMA rupture.

Several meta-analyses of retrospective studies have demonstrated that conservative management can achieve substantial symptom relief in most patients with ISMAD.^2,4, 5, 6 The 2025 European Society for Vascular Surgery guidelines recommend that symptomatic patients with ISMAD be managed conservatively as first-line strategy.¹⁴ However, approximately 20% of patients eventually require further treatments, including endovascular interventions, open surgery, or even bowel resection.² In such cases, delayed optimal surgical treatment may result in unfavorable clinical outcomes. Consequently, early identification of patients at risk of conservative management failure is crucial for optimizing therapeutic strategies and improving prognosis in ISMAD.

CTA is currently the most widely used modality for diagnosing and assessing ISMAD.⁴ Prior studies have shown that specific anatomical features on CTA, such as the length of the dissection, the degree of stenosis, and the characteristics of entry and re-entry sites, are associated with the progression and severity of ISMAD.15, 16, 17 To characterize the anatomical features of ISMAD and facilitate risk stratification, at least 8 CTA-imaging–based classification systems have been proposed.^16,18, 19, 20, 21, 22, 23, 24 However, existing imaging studies have largely been limited to single-center cohorts with small sample sizes. Although some risk factors for ISMAD conservative management failure have been identified¹⁵, the prognostic value of these CTA-based classification systems has not been evaluated.

In this study, clinical information from ISMAD patients admitted to 3 medical centers in China between October 2009 and February 2023 was retrospectively collected. Using this multicenter cohort, the relationships between 8 existing ISMAD classification systems and the outcomes of conservative management were comprehensively examined. We aimed to evaluate the predictive performance of these classification systems, identify key risk factors for conservative management failure, and provide evidence to inform clinical decision making.

Methods

Study subjects

Data of patients with ISMAD were retrospectively collected from medical records between October 2009 and February 2023 from 3 major medical centers across China: Xiangya Hospital of Central South University, The First Medical Center of the Chinese PLA General Hospital, and West China Hospital of Sichuan University. Patients were consecutively included based on the following criteria: 1) a diagnosis of ISMAD confirmed by the presence of a dissected segment of the SMA on CTA or digital subtraction angiography, regardless of the presence of celiac artery dissection; and 2) symptoms such as abdominal pain, vomiting, diarrhea, and hematochezia present at initial clinical presentation. Exclusion criteria were as follows: 1) discharge diagnosis combined with aortic dissection, systemic vasculitis, or other autoimmune diseases; 2) mental illness or an inability to cooperate; 3) severe allergy to contrast medium; 4) radiological diagnosis from other hospitals without clear CTA findings from the 3 centers; and 5) emergency surgery without prior conservative management like bowel rest and intravenous fluids.

Clinical characteristics were systematically and retrospectively collected, including patient age, gender, duration of pain from onset to hospital admission, duration of in-hospital conservative therapy, and coexisting medical conditions (such as hypertension, diabetes, and smoking). In addition, white blood cell counts were recorded as indicators of abdominal infection severity. Admission CTAs were reviewed by 2 independent radiologists, who documented imaging features, including aortomesenteric angle, the distance from the SMA ostium to the beginning of the dissection, length of the dissection, diameter of dissection, degree of stenosis (defined as 1 – [true lumen diameter/normal SMA diameter]). In cases of disagreement, consensus was reached through discussion. Intraclass correlation coefficient was used to test the consistency of the measurement between the 2 radiologists. All ISMAD was further classified according to 8 radiographic classification systems as previously reported, including the classifications of Sakamoto et al¹⁸, Yun et al¹⁹, Zerbib et al²⁰, Luan and Li²¹, Li et al²², Xiong et al¹⁶, Yoo et al²³, and Tan et al (Supplemental Figures 1 to 8).²⁴

This study was approved by the ethics committees of 3 hospitals: Xiangya Hospital of Central South University, the First Medical Center of the Chinese PLA General Hospital, and West China Hospital of Sichuan University. All patients and their legal guardians provided informed consent.

Imaging assessment

The admission CTA images were reviewed to assess the association between image features and clinical outcomes. Mimics Medical 21.0 software (Materialise) was used for measurement and curved-planar reformat (CPR) of SMA to obtain parallel (vessel stretched) and perpendicular (orthogonal) views.^25,26 The following image measurements were made: 1) length of dissection, measured in parallel views after CPR; 2) aortomesenteric angle, measured in sagittal plane image, defined as the angle between the SMA trunk axis and the abdominal aorta axis; 3) distance from the SMA ostium to the beginning of the dissection, measured in perpendicular views after CPR; 4) diameter of dissection, measured in perpendicular views after CPR; and 5) degree of stenosis defined as 1 − (true lumen diameter/normal SMA diameter), measured in perpendicular views after CPR.

Treatment and outcome

All enrolled ISMAD patients initially received conservative treatment, including bowel rest, intravenous nutrition, blood pressure control, pain management, and selective antithrombotic therapy decided by the attending physician. Conservative management was considered unsuccessful if any of the following criteria were met: 1) persistent abdominal pain for 7 days or worsening symptoms within 24 hours during the initial hospitalization, as assessed by the attending physician; 2) ISMAD-related readmission; 3) new severe stenosis (defined as stenosis >80%), occlusion in the true lumen, or a dissecting aneurysm (defined as at least 1.5 times larger than normal arterial diameter) during the follow-up period; or 4) ISMAD-related death during conservative management or the follow-up period. The overall flowchart is shown in Figure 1.

Figure 1.

Study Flowchart

Flowchart of patient selection and study outcomes. Consecutive patients diagnosed with ISMAD between October 2009 and February 2023 were screened. Patients were excluded due to CTA not available (n = 43) and CTA not clear (n = 14) The final cohort was included in analysis and categorized according to clinical outcomes. The numbers of patients at each stage of screening and the distribution of outcomes are presented. CTA = computed tomography angiography; ISMAD = isolated superior mesenteric artery dissection.

Follow-up

Outpatient follow-up was conducted at 1, 3, and 6 months post discharge, and annually thereafter. Patients who did not undergo surgery underwent CTA examinations at 1 and 6 months post discharge and annually until complete vascular remodeling. Follow-up ended in February 2023.

Statistical analysis

All statistical analyses were performed using R 4.4.3 software (R Foundation for Statistical Computing). Continuous variables were expressed as mean ± SD or median (IQR). The Student’s t-test or Wilcoxon rank-sum test was used to compare differences between groups. Categorical data were presented as frequencies (n, %), and differences were compared using Fisher’s exact test.

Univariable logistic regression analyses were performed within each classification system. In addition, univariable logistic regression analyses were performed for baseline clinical characteristics to explore potential clinical confounders. Multivariable logistic regression analyses adjusting for clinically relevant confounders were also performed.

Receiver operating characteristic curves and C-statistics were estimated based on probabilities derived from logistic regression analyses. Receiver operating characteristic and C-statistics were further calculated with 95% bootstrap CIs based on percentiles of the empirical bootstrap distribution with 2,000 iterations. Pairwise comparisons of C-statistics between the classification of Sakamoto et al¹⁸ and the other classification systems were performed using the DeLong test, and P values were adjusted using the Holm method. Continuous net reclassification improvement (NRI) was calculated using the classification of Sakamoto et al¹⁸ as the reference, with 1,000 bootstrap iterations. Model fit was compared using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The model with the lowest AIC and BIC was considered the best model fit. A P value of <0.05 was considered statistically significant.

Based on the fitted logistic regression models, calibration analysis was conducted for each classification system. Internal validation was performed with 1,000 bootstrap resamples to obtain bias-corrected estimates. The apparent and bias-corrected C-index, Brier score, calibration slope, calibration intercept, and maximum calibration error (E-max) were calculated. Bias-corrected calibration curves plotting predicted probabilities and observed proportions were generated for each classification.

Decision curve analysis was performed to evaluate the clinical utility of each classification system. Predicted probabilities were obtained from univariable logistic regression models for each classification. Net benefit was calculated across a range of threshold probabilities from 0 to 1, with particular focus on a clinically relevant range of 10% to 50%. The net benefit of each classification system was compared with the “treat all” and “treat none” strategies.

Results

Baseline characteristics of patients with ISMAD receiving conservative management

The baseline characteristics and imaging classifications of all enrolled patients are summarized in Table 1. A total of 156 patients with ISMAD were included, of whom 108 of 156 (69.2%) were successfully treated with conservative management, and 48 of 156 (30.8%) experienced treatment failure. The median CTA follow-up time was 7.9 months (IQR: 3.7-24.3).

Table 1.

Clinical Characteristics of Patients With ISMAD

	Successful Conservative Management Group (n = 108)	Failed Conservative Management Group (n = 48)	All Patients (N = 156)	P Value
Age, y	51.00 (47.00, 57.00)	52.50 (48.00, 57.25)	52.00 (47.00, 57.00)	0.724
Male	97 (89.8%)	45 (93.8%)	142 (91.0%)	0.799
Coexisting medical conditions
Smoking	50 (46.3%)	26 (54.2%)	76 (48.7%)	0.642
Hypertension	40 (37.0%)	14 (29.2%)	54 (34.6%)	0.521
Diabetes	4 (3.7%)	2 (4.2%)	6 (3.8%)	1.00
Clinical manifestations
Duration of pain before admission, d	4.00 (1.00, 11.00)	8.00 (3.00, 30.00)	4.00 (1.75, 15.00)	0.0214a
WBC count, ×109/L	6.62 (5.40, 8.70)	6.78 (5.33, 11.29)	6.69 (5.38, 9.10)	0.548
Radiographic findings
Aortomesenteric angle	79.24 (66.97, 87.94)	79.82 (62.95, 88.30)	79.24 (65.34, 87.94)	0.997
The distance from the SMA ostium to the first entry, mm	24.20 (18.91, 34.61)	22.65 (18.96, 31.16)	23.64 (18.91, 34.26)	0.840
Length of the dissection, mm	73.90 (50.14, 103.51)	81.00 (59.11, 116.39)	76.25 (53.11, 105.42)	0.479
Diameter of dissection, mm	10.98 (10.27, 12.41)	10.95 (10.21, 12.12)	10.98 (10.26, 12.20)	0.946
Degree of stenosis, %	0.64 (0.52, 1.00)	0.83 (0.69, 1.00)	0.69 (0.57, 1.00)	0.00544b
Follow-up
CTA follow-up duration, mo	5.7 (3.35, 14.2)	19.0 (8.4,33.2)	7.9 (3.7, 24.3)

Values are n (%), mean ± SD, or median (Q1, Q3). Degree of stenosis was calculated as 1 − (true lumen diameter/normal SMA diameter). Categorical variables were compared using Fisher’s exact test and continuous variables using the Mann-Whitney U test.

ISMAD = isolated superior mesenteric artery dissection; SMA = superior mesenteric artery; WBC = white blood cell.

^aP < 0.05.

^bP < 0.01.

The median age was 52.0 years (IQR: 47.0-57.0), and 142 of 156 (91.0%) were men. Age, sex, smoking status, hypertension, diabetes, and white blood cell counts were comparable between groups. However, patients with conservative management failure had a longer duration of pain before admission (8.0 [3.0-30.0] days vs 4.0 [1.0-11.0] days; P = 0.0214).

Radiographic findings, including the aortomesenteric angle, distance from the SMA ostium to the first entry, length of dissection, and diameter of dissection, showed no significant differences. In contrast, SMA stenosis was more severe among patients with treatment failure (median 0.83, IQR: 0.69-1.00 vs median 0.64, IQR: 0.52-1.00); P = 0.00544).

Interobserver variability

Interobserver agreement for radiographic measurements was good to excellent, with intraclass correlation coefficients ranging from 0.80 to 0.92 (Supplemental Table 1).

Association of subtypes within each classification system with conservative management failure

Associations between subtypes of imaging classification system and conservative management failure in ISMAD were assessed by univariate logistic regression (Table 2). The first subtype of each classification system (Type I or Type A) served as the reference to calculate the ORs. The subtypes with the highest ORs in each system are illustrated in Figure 2.

Table 2.

Univariate Logistic Regression of Classification Systems and Conservative Management Failure in ISMAD

Classification	n (%)	OR (95% CI)	P Value
Sakamoto et al18
I	30 (19.23)	Reference
II	52 (33.33)	2.816 (1.029-7.706)	0.0438a
III	59 (37.82)	1.120 (0.400-3.135)	0.829
IV	15 (9.62)	0.505 (0.091-2.801)	0.435
Yun et al19
I	30 (19.23)	Reference
IIa	42 (26.92)	2.464 (0.868-6.998)	0.0903
IIb	67 (42.95)	0.868 (0.310-2.433)	0.788
III	17 (10.90)	3.696 (1.034-13.214)	0.0443a
Zerbib et al20
I	30 (19.23)	Reference
II	42 (26.92)	2.464 (0.868-6.998)	0.0903
IIIa	12 (7.69)	1.095 (0.231-5.195)	0.909
IIIb	41 (26.28)	0.924 (0.300-2.843)	0.891
IV	10 (6.41)	0.821 (0.141-4.800)	0.827
V	1 (0.64)	NA	NA
VIa	17 (10.90)	3.696 (1.034-13.214)	0.0443a
VIb	3 (1.92)	NA	NA
Luan and Li21
A	7 (4.49)	Reference
B	44 (28.21)	2.800 (0.307-25.524)	0.361
C	73 (46.79)	1.821 (0.205-16.201)	0.591
D	32 (20.51)	6.000 (0.647-55.661)	0.115
Li et al22
I	28 (17.95)	Reference
IIa	18 (11.54)	1.833 (0.484-6.948)	0.373
IIb	22 (14.10)	3.667 (1.072-12.547)	0.0384a
IIc	8 (5.13)	6.111 (1.125-33.193)	0.036a
IIIa	31 (19.87)	0.705 (0.189-2.628)	0.603
IIIb	21 (13.46)	1.467 (0.396-5.426)	0.566
IIIc	7 (4.49)	2.750 (0.479-15.794)	0.257
IVa	10 (6.41)	0.917 (0.153-5.508)	0.924
IVb	3 (1.92)	NA	NA
IVc	2 (1.28)	3.667 (0.199-67.652)	0.382
V	6 (3.85)	3.667 (0.584-23.026)	0.166
Xiong et al16
I	30 (19.23)	Reference
II	42 (26.92)	2.464 (0.868-6.998)	0.0903
IIIa	38 (24.36)	0.742 (0.228-2.411)	0.62
IIIb	9 (5.77)	1.643 (0.324-8.333)	0.549
IIIc	7 (4.49)	1.314 (0.208-8.319)	0.772
IV	13 (8.33)	0.597 (0.106-3.363)	0.559
V	17 (10.90)	3.696 (1.034-13.214)	0.0443a
Yoo et al23
I	27 (17.31)	Reference
Is	3 (1.92)	NA	0.994
II	23 (14.74)	1.524 (0.452-5.138)	0.497
IIs	29 (18.59)	3.516 (1.136-10.884)	0.0292a
III	10 (6.41)	NA	0.991
IIIs	4 (2.56)	1.429 (0.213-9.581)	0.713
IV	41 (26.28)	0.529 (0.146-1.913)	0.332
IVs	20 (12.82)	1.587 (0.499-5.048)	0.434
Tan et al24
I	27 (17.31)	Reference
ILS	0	NA	NA
IL	3 (1.92)	NA	NA
IS	0	NA	NA
II	23 (14.74)	1.524 (0.452-5.138)	0.497
IILS	11 (7.05)	5.000 (1.116-22.408)	0.0355a
IIL	1 (0.64)	NA	NA
IIS	17 (10.90)	3.214 (0.890-11.603)	0.0746
III	32 (20.51)	0.529 (0.146-1.913)	0.332
IIILS	7 (4.49)	3.810 (0.678-21.419)	0.129
IIIL	8 (5.13)	1.714 (0.323-9.109)	0.527
IIIS	13 (8.33)	0.857 (0.182-4.042)	0.846
IV	8 (5.13)	NA	NA
IVLS	1 (0.64)	NA	NA
IVL	2 (1.28)	NA	NA
IVS	4 (2.56)	NA	NA

Classification I/A served as the reference.

NA = not available; Other abbreviation as in Table 1.

^aP < 0.05. ORs and P values were not calculated for subtypes with very small numbers of patients because the estimates would be unstable and potentially misleading.

Figure 2.

High-Risk ISMAD Subtypes Across Classifications

Schematic overview of the highest-risk subtypes associated with failure of conservative management across 8 classification systems for ISMAD. The right-sided vertical structure represents the abdominal aorta, and the left-sided branch represents the SMA. Arrows are used to illustrate the patent false lumen with entry and/or re-entry, and shaded regions represent areas of arterial occlusion or thrombus. Subtypes associated with increased risk were identified based on univariate logistic analyses. Despite differences in classification criteria, these subtypes share common morphological characteristics, suggesting potential value in risk stratification and clinical decision making. SMA = superior mesenteric artery; other abbreviation as in Figure 1.

In the classification of Sakamoto et al,¹⁸ Type II (patent false lumen without re-entry) showed the highest OR for conservative management failure, which was statistically significant (OR: 2.816; 95% CI: 1.029-7.706; P = 0.0438). The Yun et al¹⁹ Type III, Zerbib et al²⁰ Type Via, and Xiong et al¹⁶ Type V, all representing occlusion of the SMA, showed the highest and significant ORs in each classification system (OR: 3.696; 95% CI: 1.034-13.214; P = 0.0443). Among the remaining systems, Luan and Li’s²¹ Type D (dissection extending distally to the ileocolic or distal ileal artery) yielded the highest OR (OR: 6.000; 95% CI: 0.647-55.661) without statistical significance. In contrast, the Li et al²² Type IIc (thrombosed false lumen with total occlusion of the true lumen) showed a significant association (OR: 6.111; 95% CI: 1.125-33.193; P = 0.036). Furthermore, the Yoo et al²³ Type IIs (patent false lumen without re-entry, accompanied by a long segment of dissection and/or true lumen stenosis) and the Tan et al²⁴ Type IILS (patent false lumen without re-entry, accompanied by a long segment of dissection and true lumen stenosis) both demonstrated the highest and statistically significant ORs for conservative management failure (OR: 3.516; 95% CI: 1.136-10.884; P = 0.0292; and OR: 5.000; 95% CI: 1.116-22.408; P = 0.0355, respectively).

Univariate analysis of clinical characteristics

Univariate logistic regression analyses were performed to evaluate the associations between baseline clinical characteristics and conservative management failure (Supplemental Table 2). None of the clinical characteristics examined showed a statistically significant association with the outcome (P > 0.05).

Adjusted associations of subtypes with conservative management failure

After adjustment for confounders, several subtypes remained associated with conservative management failure. The Tan et al²⁴ classification type IILS was significantly associated with failure (OR: 4.789; 95% CI: 1.010-22.714; P = 0.049). In the Li et al²² classification, types IIb and IIc were also associated with failure (OR: 3.814; 95% CI: 1.071-13.585; P = 0.039; and OR: 6.108; 95% CI: 1.040-35.886; P = 0.045, respectively). In the Yoo et al²³ classification, type IIs showed significant association (OR: 3.590; 95% CI: 1.100-11.719; P = 0.034). The detailed multivariable logistic regression results for all subtypes and covariates are presented in Supplemental Table 3.

Predictive performance comparison of 8 ISMAD classification systems

As shown in Table 3, the Tan et al²⁴ classification had the lowest AIC (186) and favorable BIC (226), indicating the best overall model fit, followed by Yoo et al²³ (AIC 188, BIC 213). In contrast, Li et al²² (AIC 198, BIC 232) and Zerbib et al²⁰ (AIC 196, BIC 221) showed comparatively poorer fit.

Table 3.

Comparative Performance of 8 Classification Systems in Predicting Failed Conservative Management in Patients With ISMAD

First Author	AIC	BIC	C-Statistic (95% CI)	Difference in C-Statistic from Sakamoto’s Classification	Unadjusted P Value	Adjusted P Value (FDR)
Sakamoto et al18	191	203	0.639 (0.549-0.728)
Yun et al19	190	202	0.645 (0.553-0.737)	0.006	0.883	0.883
Zerbib et al20	196	221	0.654 (0.562-0.746)	0.015	0.674	0.883
Luan and Li21	192	205	0.629 (0.538-0.719)	−0.010	0.874	0.883
Li et al22	198	232	0.689 (0.599-0.779)	0.050	0.188	0.438
Xiong et al16	195	216	0.665 (0.574-0.755)	0.026	0.465	0.814
Yoo et al23	188	213	0.698 (0.613-0.783)	0.059	0.125	0.438
Tan et al24	186	226	0.750 (0.670-0.830)	0.111	0.0141a	0.0986

AIC = Akaike Information Criterion; BIC = Bayesian information criterion; FDR = false discovery rate; other abbreviation as in Table 1.

^aP < 0.05.

As shown in Table 3 and Figure 3, Tan et al’s²⁴ classification showed the highest discriminative ability (C-statistic 0.750; 95% CI: 0.670-0.830), followed by the Yoo et al’s²³ classification (C-statistic 0.698; 95% CI: 0.613-0.783). However, when using Sakamoto et al’s¹⁸ classification as the reference, no classification showed a statistically significant difference in C-statistics (all P > 0.05).

Figure 3.

ROC Analysis of ISMAD Classifications

Receiver operating characteristic (ROC) curves derived from univariable logistic regression models evaluating 8 classification systems for predicting failure of conservative management in ISMAD. (A) Combined ROC curves of all classification systems. (B to I) ROC curves for individual classifications, including the systems of Sakamoto et al,¹⁸ Yun et al,¹⁹ Zerbib et al,²⁰ Luan and Li,²¹ Li et al,²² Xiong et al,¹⁶ Yoo et al,²³ and Tan et al.²⁴ Tan’s classification demonstrated the highest area under the curve (AUC); however, no statistically significant differences in AUC were observed between any classification and the Sakamoto et al¹⁸ classification. Abbreviation as in Figure 1.

As shown in Table 4, the Tan et al²⁴ classification had the largest NRI (0.593; 95% CI: 0.244-0.903), improving both event (NRI+ 0.333; 95% CI: 0.058-0.591) and nonevent (NRI− 0.259; 95% CI: 0.076-0.430) reclassification. Yoo et al²³ also demonstrated favorable reclassification (NRI 0.440; 95% CI: 0.126-0.768; NRI+ 0.458; NRI− −0.019), whereas other classifications showed smaller or mainly nonevent improvements.

Table 4.

Continuous NRI of Each Classification System Compared With the Sakamoto et al¹⁸ Classification

First Author	NRI (95% CI)	NRI+ (95% CI)	NRI− (95% CI)
Yun et al19	−0.028 (−0.330 to 0.318)	−0.250 (−0.519 to 0.026)	0.222 (0.037-0.394)
Zerbib et al20	0.162 (−0.109 to 0.444)	−0.542 (−0.778 to 0.295)	0.704 (0.560-0.822)
Luan and Li21	0.157 (−0.182 to 0.510)	−0.0833 (−0.373 to 0.204)	0.241 (0.053-0.430)
Li et al22	0.417 (0.096-0.756)	0.250 (−0.022 to 0.526)	0.167 (−0.020 to 0.356)
Xiong et al16	0.111 (−0.216 to 0.450)	−0.333 (−0.592 to −0.060)	0.444 (0.275-0.619)
Yoo et al23	0.440 (0.126-0.768)	0.458 (0.209-0.700)	−0.019 (−0.207 to 0.162)
Tan et al24	0.593 (0.244-0.903)	0.333 (0.058-0.591)	0.259 (0.0762-0.430)

NRI = net reclassification improvement.

Bias-corrected calibration curves for each classification are shown in Figure 4. Calibration was generally satisfactory across classifications, with the Tan et al²⁴ classification showing the closest agreement between predicted and observed risks. Bias-corrected C-indices ranged from 0.603 to 0.656, with the Yoo et al²³ and Tan et al²⁴ classifications demonstrating the highest discriminatory ability. Brier scores were low across systems (0.205-0.214), indicating reasonable overall predictive accuracy. Bias-corrected calibration slopes and intercepts were close to 1 and 0, respectively, reflecting good agreement between predicted and observed probabilities. Maximum calibration errors (E-max) were small (0.069-0.106), indicating minimal local deviation from perfect calibration. Detailed results are provided in Supplemental Table 4.

Figure 4.

Calibration Analysis of ISMAD Classifications

Bias-corrected calibration curves derived from univariable logistic regression models for 8 classification systems predicting failure of conservative management in ISMAD. (A to H) Calibration curves for the systems of Sakamoto et al,¹⁸ Yun et al,¹⁹ Zerbib et al,²⁰ Luan and Li,²¹ Li et al,²² Xiong et al,¹⁶ Yoo et al,²³ and Tan et al,²⁴ respectively. Overall calibration was satisfactory across classifications, with predicted risks showing good agreement with observed outcomes. The Tan et al²⁴ classification demonstrated the closest agreement. Discrimination was modest, with bias-corrected C-indices ranging from 0.603 to 0.656. Calibration slopes and intercepts were close to ideal values, and Brier scores and E-max indicated minimal prediction error. Abbreviation as in Figure 1.

Decision curve analysis is presented in Figure 5. Across a clinically relevant threshold probability range of 10% to 50%, the Tan et al²⁴ classification demonstrated the highest net benefit, followed by the Yoo et al²³ and Li et al²² classifications, whereas the remaining systems showed relatively lower net benefit. Overall, all classification systems provided greater net benefit than the “treat all” and “treat none” strategies within this range. Detailed results are provided in Supplemental Table 5.

Figure 5.

DCA of ISMAD Classifications

Decision curve analysis (DCA) comparing 8 classification systems for predicting failure of conservative management in ISMAD. Net benefit is plotted against threshold probabilities ranging from 0% to 100%. The “treat all” and “treat none” strategies are included as reference lines. Within the clinically relevant threshold range of 10% to 50%, the Tan et al²⁴ classification shows higher net benefit compared with other systems. Abbreviation as in Figure 1.

Discussion

Most patients with ISMAD may achieve symptom relief after conservative treatment, which is recommended as the first-line treatment for ISMAD by guideline.¹⁴ However, approximately 20% of patients continue to experience symptoms after conservative management and require further surgical intervention.^2,15 In this retrospective multicenter study, we compared 8 imaging-based classification systems for predicting outcomes of conservative management in ISMAD. Among them, the Tan et al²⁴ classification demonstrated the best overall performance across discrimination, risk stratification, calibration, and clinical utility as assessed by decision curve analysis. The overall study design and key findings are summarized in the Central Illustration. These insights highlight its potential value in guiding clinical decision making.

Central Illustration.

Study Design and Key Findings

A multicenter retrospective cohort of patients with ISMAD was analyzed to compare 8 imaging-based classification systems for predicting failure of conservative management. Performance was evaluated using discrimination, calibration, reclassification, and decision curve analysis. Among all systems, the Tan et al²⁴ classification demonstrated the best overall performance. These findings support its potential role in guiding risk assessment and clinical decision making. Created in BioRender, Wang W, (2026), https://BioRender.com/zisbirr. ISMAD = isolated mesenteric artery dissection; SMA = superior mesenteric artery.

In ISMAD patients, symptomatic cases account for 81% to 93.3%, with acute abdominal pain being the most common presenting symptom.^2,4,7, 8, 9 However, the underlying mechanism is not fully understood. Some studies suggested that abdominal pain in ISMAD patients resulted from bowel ischemia caused by the dissection.27, 28, 29 Other studies suggest that the dissection itself is a cause of abdominal pain. Kim et al³⁰ reported that although patients experienced rapid symptomatic improvement, follow-up angiography of part of the patients showed worsened imaging findings inconsistent with their clinical course. Moreover, perivascular inflammation has been identified in patients with acute coronary syndrome and vertebral artery dissection.^31,32 Therefore, it was inferred that the pain of ISMAD is possibly a result of perivascular inflammation that may stimulate visceral nerve plexus.^19,24,33 In our study, long-segment dissection, a patent false lumen without re-entry, and severe SMA stenosis or occlusion were identified as potential risk factors for failed conservative management, consistent with previous reports.^6,15,17,34 Taken together with mechanisms proposed in earlier studies, these findings support that abdominal pain in ISMAD is multifactorial, arising from a combination of ischemia and irritation of the visceral nerve plexus. Furthermore, patients with prolonged duration of pain before admission may have already experienced more severe vascular or intestinal injury, which may render these patients less responsive to conservative management. Given the potential role of inflammation in abdominal pain of ISMAD, we examined white blood cell count as an inflammatory marker but found no significant difference. Currently, there is still a lack of reliable serum biomarkers with sufficient sensitivity and specificity for the early detection of intestinal ischemia and injury,³⁵ further underscoring the importance of imaging as the primary tool for evaluating ISMAD.

At least 8 classification systems of ISMAD have been proposed to link ISMAD imaging features with clinical outcomes.^16,18, 19, 20, 21, 22, 23, 24 Earlier ISMAD classification systems primarily focused on morphology, whereas more recent ones (Li et al,²² Yoo et al,²³ Tan et al²⁴) integrate imaging features associated with conservative management failure, including long-segment dissection, patent false lumen without re-entry, and severe SMA stenosis or occlusion.¹⁵ Our analysis suggested that cases with 1 or more of these risk features were more prone to conservative management failure, with risk further increased when multiple factors occurred (Tan et al²⁴ type IILS). This emphasizes the importance of integrating multiple imaging indicators rather than relying on a single morphologic feature.

The integration of multiple imaging risk factors associated with conservative management failure may account for the relatively higher C-statistic observed for the Tan et al²⁴ classification in our study. We further evaluated whether this advantage translated into improved patient stratification and clinical utility. Reclassification analysis indicated that the Tan et al²⁴ classification modestly improved risk stratification compared with the Sakamoto et al¹⁸ system, although continuous NRI may overestimate predictive performance in small cohorts.³⁶ The Tan et al²⁴ classification also demonstrated relatively better calibration performance. Decision curve analysis further supported its clinical value, as the Tan et al²⁴ classification achieved the highest net benefit across a wide range of threshold probabilities compared with other systems. The Sakamoto et al¹⁸ classification was selected as the reference model given its historical role as the earliest and most widely used imaging-based classification for ISMAD.^2,18 Collectively, these results suggest that the Tan et al²⁴ classification may offer superior discrimination and more meaningful risk stratification, supporting its potential utility in guiding clinical decision making.

It is noteworthy that despite limited predictive performance in our study, simpler classifications such as the Sakamoto et al¹⁸ and Yun et al¹⁹ remain widely used due to ease of application.^2,5 This reflects an important trade-off between predictive performance and clinical usability, as complex systems may hinder routine adoption in busy practice. More recent classification systems, such as those proposed by Li et al,²² Yoo et al,²³ and Tan et al,²⁴ incorporate a larger number of subtypes and imaging risk factors, which may improve prognostic discrimination but also increase complexity. From a practical perspective, an ideal ISMAD classification should balance simplicity with predictive accuracy, describing morphological features while stratifying disease severity.²⁴ Future research should explore incorporating additional imaging markers and relevant clinical variables to further enhance risk stratification. One promising direction is the integration of functional assessments, such as computational fluid dynamics (CFD). As powerful tools for studying vascular diseases,^37,38 CFD offers a promising approach for future risk assessment of ISMAD. Recent studies have highlighted its potential in improving clinical assessment and risk stratification.^39,40 Integration of CFD-derived functional indicators with anatomical classification systems holds promise for more precise management of ISMAD.

Study limitations

First, the 3 participating centers are large tertiary hospitals located in metropolitan areas, which could introduce selection bias. This may also partly explain the higher rate of conservative management failure observed in our cohort. Second, the follow-up duration in the conservative management group was relatively short. Therefore, long-term disease progression, delayed radiographic changes, and late interventions could not be adequately evaluated. Third, as ISMAD is a relatively rare disease, the sample size in this study was limited, which may have influenced statistical power. Future prospective multicenter studies with long-term follow-up and large sample size are needed to further investigate the predictive value of different classifications for long-term ISMAD outcomes.

Conclusions

Imaging features, including patent false lumen without re-entry, narrowed or occluded true lumen, and long-segment dissection were associated with conservative management failure in ISMAD patients. The classification of Tan et al²⁴ showed the greatest potential for identifying patients at higher risk of conservative management failure, which may provide valuable guidance for early intervention strategies.

Data Availability

The data that support this study are available from the corresponding author on reasonable request.

Funding Support and Author Disclosures

This study is supported by the National Natural Science Foundation of China (grant nos.82070491 and 82470507), the National Science Foundation for Young Scientists of China (grant no.82200536), the Provincial Natural Science Foundation of Hunan (no. 2022JJ40812), and Clinical Research Foundation of National Clinical Research Centre for Geriatric Disorder of Xiangya Hospital (No.2021LNJJ11) and The Scientific Research Program of FuRong Laboratory (No. 2025PT5001). The authors have reported that they have no relationships relevant to the contents of this paper to disclose.