Gastrointestinal bleeding in patients on venoarterial extracorporeal membrane oxygenation: a multicenter observational study Authors: Zhongman Zhang, Yi Zhu, Gang Zhang, XiaoDong You, Yanbin Dong, Deliang Hu, Haijun Sun, Wei Li, Xufeng Chen Published: 2026 ABSTRACT AbstractBackgroundGastrointestinal (GI) bleeding is a common complication in patients treated with venoarterial extracorporeal membrane oxygenation (VA ECMO), yet comprehensive data on its epidemiology, clinical correlates, and prognostic implications remain inadequately characterized.MethodsVA ECMO-treated patients at the Emergency Medicine Center of the First Affiliated Hospital with Nanjing Medical University from January 2017 to December 2023 (analysis cohort) and Intensive Care Unit of The First People's Hospital of Suqian from January 2022 to June 2024 (external validation cohort) were enrolled in this study. In the analysis cohort, stabilized inverse probability of treatment weighting (SIPTW) was used to balance baseline characteristics. Least absolute shrinkage and selection operator (LASSO) regression and binary logistic regression were used for variable selection and predictive model construction, respectively.ResultsThe incidence of GI bleeding was 17.8% (62/349) in the analysis cohort and 12.0% (10/83) in the external validation cohort. After SIPTW adjustment, GI bleeding was significantly associated with higher 28-day mortality (HR = 2.25, 95%CI: 1.67–3.05, P < 0.001). Three independent predictors were identified: continuous renal replacement therapy (CRRT), baseline international normalized ratio (INR), and baseline calcium. The model yielded an area under the curve (AUC) of 0.797 (95%CI: 0.732–0.861) in the training set, 0.803 (95%CI: 0.706–0.900) in the internal validation set, and 0.753 (95%CI: 0.531–0.976) in the external validation cohort.ConclusionsGI bleeding is linked to elevated mortality risk in VA ECMO-treated patients. CRRT, baseline INR, and baseline calcium are independent predictors of GI bleeding in this population.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12876-026-04991-3.Keywords: VA ECMO, ECPR, Gastrointestinal hemorrhage, Cardiac arrest, Predictive model Introduction Recent studies have confirmed the benefit of venoarterial extracorporeal membrane oxygenation (VA ECMO) in patients with circulatory failure who did not benefit from conventional therapy [1, 2]. Despite improvements in ECMO technology, mortality remains high in patients treated with VA ECMO due to both the irreversible nature of primary disease and complications of VA ECMO [2–4]. Bleeding is the most common and potentially lethal complication and has been reported in 62% of VA ECMO-treated patients [4]. Bleeding during ECMO is independently associated with in-hospital mortality [5, 6]. This relationship may be accounted for by different bleeding sites [4, 7, 8]. According to data from the Extracorporeal Life Support Organization (ELSO) registry, gastrointestinal (GI) bleeding is the third most common bleeding complication in VA ECMO-treated patients, following only cannulation site and surgical bleeding. However, GI bleeding is associated with a substantially higher risk of in-hospital mortality (OR = 1.95 for GI bleeding) compared with other two bleeding sites [4]. As reported, GI bleeding occurs in 6.0% to 24.3% of patients treated with ECMO [3, 4, 7–9], and patients with GI bleeding were associated with a higher transfusion burden and increased mortality rates [10]. However, few studies have investigated the risk factors of GI bleeding in patients treated with VA ECMO [11, 12]. Therefore, this study aimed to determine the outcome and risk factors of GI bleeding in patients treated with VA ECMO. Methods Study design This multicenter observational study consisted of two independent cohorts. The analysis cohort (JSPH cohort) enrolled patients who received VA ECMO support at the Emergency Medicine Center of the First Affiliated Hospital with Nanjing Medical University from January 2017 to December 2023. The external validation cohort (SQFH cohort) included VA ECMO-treated patients admitted to the Intensive Care Unit (ICU) of The First People's Hospital of Suqian between January 2022 and June 2024. Inclusion criteria: (Ⅰ) reversible causes; (Ⅱ) ECMO duration > 24 h. Exclusion criteria: (Ⅰ) patients < 14 years old; (Ⅱ) GI bleeding pre-ECMO; (Ⅲ) pregnancy; (Ⅳ) data missing. GI bleeding was defined as upper or lower GI hemorrhage requiring packed red blood cell (PRBC) transfusion (> 3U PRBCs/24 h) and/or endoscopic intervention and/or hemostatic agent deployment [4]. Patients with GI bleeding during VA ECMO treatment were compared to those without GI bleeding. This study was conducted in accordance with the Declaration of Helsinki and approved by the Medical Ethics Committee of The First Affiliated Hospital With Nanjing Medical University (No. 2023-SR-467) and The First People's Hospital of Suqian (KYSB20230181). Informed consent was waived. Patient management The Emergency Medicine Center of the First Affiliated Hospital with Nanjing Medical University houses a dedicated Extracorporeal Life Support (ECLS) Center with 17 intensive care beds. All VA ECMO patients in this center were managed in this specialized critical care environment with dedicated ECMO teams. In the both centers, unfractionated heparin (UFH) was the first-line anticoagulant; when UFH was contraindicated, argatroban was used as an alternative agent. The anticoagulation effect was titrated to maintain the activated partial thromboplastin time (APTT) at 1.5–2.5 times the normal reference range, with APTT monitoring performed every 6 h at the JSPH center and every 8 h at the SQFH center to adjust the dosage promptly. At the JSPH center, nutritional support for patients started within 24 h of ECMO cannulation. Enteral nutrition was preferred over parenteral nutrition (PN), with PN indicated in the presence of GI dysfunction (e.g., GI bleeding, gastric residual volume > 500 ml, abdominal distension). At the SQFH center, nutritional support is typically initiated 24 h subsequent to the procedure. In the absence of contraindications, enteral nutrition is the preferred mode of nutrition. A proton pump inhibitor (PPI) (omeprazole, 40 mg, qd) or a histamine-2 receptor antagonist (roxatidine, 75 mg, q12h) was routinely used in all patients. Clinicians determined treatments after GI bleeding, which usually included fasting, GI decompression, PN, and PPIs. Data collection For this study, we collected the following data: demographics, co-morbidities, pre-ECMO cardiac arrest, medication histories including anti-platelet and anti-coagulant therapies, survival after venoarterial ECMO (SAVE) score [13] (a comprehensive tool integrating key predictive factors: diagnosis, age, weight, pre-ECMO cardiac/respiratory/renal assessments, and pre-ECMO other organ failures), ECMO characteristics, adjunctive therapies, baseline laboratory parameters (the most recent parameters prior to ECMO cannulation) and the worst laboratory parameters during ECMO support (defined as the most recent parameters before the first GI bleeding event if GI bleeding occurred, or the worst parameters during the entire ECMO run if no GI bleeding event occurred), amounts of transfused PRBCs, fresh frozen plasma (FFP) and platelets (PLTs) (mL/kg/d) during ECMO treatment, and 28-day mortality. For patients who experienced multiple GI bleeding episodes during VA ECMO support, laboratory indicators measured closest to the onset of the first bleeding episode (but prior to its occurrence) were predefined as the reference standard for analysis. Laboratory parameters included platelet count, prothrombin time, APTT, fibrinogen, international normalized ratio, potassium, sodium, calcium, alanine aminotransferase, creatinine, urea nitrogen, total bilirubin, albumin. Statistical analysis Shapiro–Wilk test was performed to verify the distribution of continuous variables. The variables failed to meet the normality assumption were presented as median and interquartile range [M (Q1, Q3)] and analyzed using the Mann–Whitney U test. Data conforming to normal distribution are presented as mean ± standard deviation, and independent samples t-tests were employed to compare differences between the two groups. Categorical variables as frequencies and percentages. Between-group differences were assessed using the chi-square test or Fisher’s exact test. Missing data were imputed using multiple imputations (random forest) in the training set, the internal validation set, and the external validation set, respectively. Firstly, stabilized inverse probability of treatment weighting (SIPTW) was used to calculate the weights of GI bleeding and other potential variables related to 28-day mortality in the analysis cohort. Propensity scores were calculated using a multivariable logistic regression model, where GI bleeding was defined as the dependent variable, and the independent variables included baseline characteristics: SAVE score, sex, heart rate, and comorbidities (hypertension, diabetes mellitus, coronary heart disease, and stroke). The variables involved in SAVE score such as primary disease, age, assessments of organ function and so on, were not repeated. Combined with methodology and previous similar VA ECMO-related study [14, 15], an absolute standardized mean difference (SMD) value < 0.1 was defined as adequate covariate balance. After weighing, cox proportional regression was used to determine the effect of GI bleeding on prognosis. Meanwhile, Kaplan–Meier survival curves were drawn. Secondly, patients in the analysis cohort were randomly allocated to the training and internal validation sets in a 7:3 ratio. In the training set, we used least absolute shrinkage and selection operator (LASSO) regression to shrink coefficients and mitigate overfitting. Variable selection was based on one standard error (λ. 1se) rule. A variance inflation factor (VIF) analysis was conducted to assess the presence of multicollinearity among the independent variables. A VIF value exceeding 5 was regarded as indicative of severe multicollinearity. Then, a nomogram using binary logistic regression was constructed to predict the occurrence of GI bleeding. The area under the receiver operating characteristic curve (equivalent to C-index) was used to determine the discriminatory capacity and calibration curves were drawn. Restricted cubic splines (RCS) were used to visualize the non-linear association between clinical parameters and the risk of GI bleeding. A two-sided P-value < 0.05 was considered statistically significant. Statistical analyses were performed with R studio software for windows (version 2023.12.1–402), using its «tableone», «rms», «mice», «survminer», «survival», «ggprism», «RISCA», «foreign», «survey», «caret», «WeightIt», «cobalt», «MatchIt», «ResourceSelection», «glmnet», «pROC», «verification» and «ggplot2» packages. Results Study population In the analysis cohort, 364 patients were treated with VA ECMO and 349 patients were eligible for final analysis, including 185 patients undergoing extracorporeal cardiopulmonary resuscitation (ECPR). The overall incidence of GI bleeding was 17.8% (n = 62). The most common symptom of GI bleeding was bleeding or a coffee-like colour in gastric tube (Table 1). In the external validation cohort, 83 patients were finally included for analysis. 10 patients developed GI bleeding. The characteristics of the study cohort are presented in Table 1. The median duration of VA ECMO support was 5 (2, 7) days in the analysis cohort and 6 (3, 9) days in the external validation cohort. The cumulative curve of GI bleeding events was shown in Fig. 1A. Table S1 presented the status of missing clinical data. Table 1Demographics and clinical characteristics of adult patients receiving VA ECMOVariableAnalysis cohort (n= 349)External validation cohort (n= 83)P valueDemographics Age (years)50 (36, 60)53 (38, 63)0.446 Male245 (70.2)64 (77.1)0.264 BMI (kg/m2)24.2 (22.3, 26.2)24.2 (20.8, 27.4)0.874Co-morbidity Hypertension109 (31.2)29 (34.9)0.603 DM66 (18.9)13 (15.7)0.596 CHD49 (14.0)11 (13.3)0.992 Stroke21 (6.0)2 (2.4)0.297 Chronic kidney disease9 (2.6)2 (2.4)1.000ECPR mode185 (53.0)40 (48.2)0.430Pre-ECMO cardiac arrest222 (63.6)40 (48.2)0.014Anti-platelet therapy28 (8.0)8 (9.6)0.797Anti-coagulant therapy10 (2.9)3 (3.6)0.722HR pre-ECMO (beats/minute)85 (30, 130)10 (29, 147)0.334MAP pre-ECMO (mmHg)51 (30, 71)52 (0, 74)0.686SAVE score−5 ± 7−6 ± 60.308Other supportive therapies MV312 (89.4)81 (97.6)0.018 CRRT208 (59.6)47 (56.6)0.711 IABP66 (18.9)7 (8.4)0.022 GI bleeding62 (17.8)10 (12.0)0.275symptoms of GI bleeding0.205 bleeding or coffee-like in gastric tubes36 (10.3)5 (6.0) melena4 (1.1)3 (3.6) hematochezia12 (3.4)1 (1.2) ≥ 2 types10 (2.9)1 (1.2)Data were presented as mean ± SD, median (Q1, Q3), or n (%)Abbreviations: VA ECMO Venoarterial extracorporeal membrane oxygenation, BMI Body mass index, DM Diabetes mellitus, CHD Coronary heart disease, ECPR Extracorporeal cardiopulmonary resuscitation, HR Heart rate, MAP Mean arterial pressure, MV Mechanical ventilation, CRRT Continuous renal replacement therapy, IABP Intra-aortic balloon pump, GI Gastrointestinal Demographics and clinical characteristics of adult patients receiving VA ECMO Data were presented as mean ± SD, median (Q1, Q3), or n (%) Abbreviations: VA ECMO Venoarterial extracorporeal membrane oxygenation, BMI Body mass index, DM Diabetes mellitus, CHD Coronary heart disease, ECPR Extracorporeal cardiopulmonary resuscitation, HR Heart rate, MAP Mean arterial pressure, MV Mechanical ventilation, CRRT Continuous renal replacement therapy, IABP Intra-aortic balloon pump, GI Gastrointestinal Timing of GI bleeding and survival curves. A Cumulative events of GI bleeding during VA ECMO. B Propensity-weighted Kaplan–Meier 28-day survival curves of GI bleeding and no-GI bleeding groups in all VA ECMO-treated patients. C Propensity-weighted Kaplan–Meier 28-day survival curves of GI bleeding and no-GI bleeding groups in ECPR-treated patients. Abbreviations: GI gastrointestinal, VA ECMO venoarterial extracorporeal membrane oxygenation, ECPR extracorporeal cardiopulmonary resuscitation Table S2 (see supplementary material) showed the comparisons of characteristics between GI bleeding and no-GI bleeding groups. Influence of GI bleeding on prognosis Before SIPTW, GI bleeding was significantly associated with higher 28-day mortality with a hazard ratio (HR) of 2.25 (95%CI: 1.67–3.05, P < 0.001). After SIPTW, the association remained significantly different with a HR of 2.07 (95%CI: 1.48–3.89, P < 0.001). Besides, results showed that in the ECPR subgroup, the incidence rate of GI bleeding was 19.6% (44/225), and the in-hospital mortality rate was 30.2% (68/225). GI bleeding was associated with a statistically significant increase in 28-day mortality (HR = 2.05, 95% CI: 1.41–2.98, P < 0.001). After SIPTW, GI bleeding still remained an independent risk factor for elevated 28-day mortality in ECPR patients (HR = 2.05, 95%CI: 1.36–3.10, P < 0.001). Comparisons of clinical features before and after SIPTW were summarized in Table S3 and S4 (see supplementary material). Kaplan–Meier survival probability curve showed patients with GI bleeding had a steeper decline in survival probability relative to those without GI bleeding (Fig. 1B&C). Among patients who developed GI bleeding, the median duration of EN interruption was 3 (1, 4) days, and the proportion of EN interruption time relative to the total VA ECMO support duration reached 80% (49%, 100%). Predictive model for GI bleeding Table S5 in supplementary material summarized the comparisons between the training (n = 245) and internal validation (n = 104) sets. In the training set, 37 parameters were included in the LASSO regression to select predictors for GI bleeding (Fig. 2). The results of collinearity diagnostics indicated that the VIF values of continuous renal replacement therapy (CRRT), baseline INR, and baseline calcium are all < 5 (Table S6). Finally, the above three variables were used for establishing the predictive model using binary logistic regression (Table 2). A nomogram was developed based on the results of logistic regression (Fig. 3). The variable filtering process of the lasso regression. A Lasso coefficient profiles of the candidate features; (B) the selection of optimal parameters (lambda) by tenfold cross-validation. Abbreviations: lasso least absolute shrinkage and selection operator Table 2Multivariate logistic regression resultsVariableCoefSEOR95%CIP valueLower limitUpper limitCRRT1.5740.5244.8281.87715.2340.003Baseline INR0.2590.0931.2961.0791.5670.005Baseline calcium−1.6930.7140.1840.0420.7020.018Abbreviations: Coef Regression coefficient, SE Standard error, OR Odds ratio, CI Confidence interval, CRRT Continuous renal replacement therapy, INR International normalized ratio Multivariate logistic regression results Abbreviations: Coef Regression coefficient, SE Standard error, OR Odds ratio, CI Confidence interval, CRRT Continuous renal replacement therapy, INR International normalized ratio Nomogram for GI bleeding in patients treated with VA ECMO. Abbreviations: GI gastrointestinal, VA ECMO venoarterial extracorporeal membrane oxygenation, INR international normalized ratio The AUC of the nomogram was 0.797 (95%CI: 0.732–0.861) in the training set, 0.803 (95%CI: 0.706–0.900) in the internal validation set, and 0.753 (95%CI: 0.531–0.976) in the external validation cohort (Fig. 4A-C). The P value of Spiegelhalter Z-test for the nomogram was 0.744 for the training set, 0.900 for the internal validation set, and 0.653 for the external validation cohort (Fig. 4D-F). Diagnostic performance and calibration of the predictive model for GI bleeding in VA ECMO-treated patients. A-C ROC curves showing the discriminatory ability of the model in the training cohort (A), internal validation cohort (B) and external validation cohort (C). D-F Calibration curves evaluating the agreement between the predicted probability and actual probability of GI bleeding in the training cohort (D), internal validation cohort (E) and external validation cohort (F). Abbreviations: ROC receiver operating characteristic, AUC area under the curve, CI confidence interval Fig. S2 (supplementary material) illustrated the non-linear correlations between baseline calcium, baseline INR, and the risk of GI bleeding. Fig. S3 (supplementary material) displayed the cumulative curves for GI bleeding events in patients with CRRT and those without CRRT. Additionally, we separately evaluated the predictive model in VA ECMO and ECPR groups (supplementary material, Fig. S4). Discussion Maintaining the delicate balance between thrombosis prevention and bleeding in patients undergoing ECMO remains challenging. Data from the Extracorporeal Life Support Organization (ELSO) revealed that bleeding events in VA ECMO are nearly twice as common as thrombotic events (62.1% vs. 37.9%) [4]. A total of 432 VA ECMO patients from two centers were included, with an overall GI bleeding incidence of 16.7% (72/432). Our study found that GI bleeding often occurred in the early stage of VA ECMO support, which is consistent with the finding of Hu et al. [16] that "most bleeding events occurred within the first 3 days of ECMO therapy". A major cause of mucosal injury of the digestive tract was hypoperfusion of the digestive organs prior to VA ECMO. Sympathetic nerve excitation and increased catecholamine secretion due to circulatory failure result in vascular constriction and blood redistribution, diverting blood supply from the digestive tract and skin to vital organs, such as the brain. Following VA ECMO, reperfusion can exacerbate damage to digestive epithelial cells, as well as other life-sustaining systems, hypoxia and coagulopathy further aggravate this alteration [17, 18].Furthermore, we found that GI bleeding significantly impaired the outcome of VA ECMO-treated patients. On the one hand, fluid management is challenging for patients with GI bleeding who are likely to need blood products. As previously reported, patients on ECMO often face challenges in achieving adequate caloric intake due to multiple factors such as hemodynamic instability and GI dysfunction [19, 20], and early EN was associated with reduced in-hospital mortality [21]. Our study revealed that GI bleeding further exacerbated this issue by disrupting EN delivery. These findings provide direct evidence that GI bleeding-induced EN interruption poses a substantial challenge to energy supply in this vulnerable patient population, which may contribute to the observed worse prognosis. Previous studies have identified several risk factors for lethal bleeding among patients undergoing ECMO, which can be divided into the following categories: primary disease-related (renal failure, high APACHE III score, ECMO following surgery, and older age), clot-forming material deficiencies (lower PLT count and low fibrinogen level), and treatment-related (high APTT and long duration of ventilation) [4, 9, 22–25]. Three studies have investigated the riskfactors associated with GI bleeding in adults on ECMO; however, the findings were notably inconsistent [11, 12, 26]. The study of Sahri Kim et al. encompassed 131 patients with venovenous ECMO and 353 patients with VA ECMO, the majority of whom were on ECMO for an extended period, with a median duration of 28 days for patients with GI bleeding and 12 days for those without GI bleeding [26]. Stern et al. solely included patients with upper GI bleeding, and their definition of ECMO-related GI bleeding differed from that of mainstream studies [11].Stern et al. revealed that ECPR is a risk factor for GI bleeding. In our study, although the incidence of GI bleeding was slightly higher in the ECPR group (21.1%, 39/185) than in the VA ECMO group (14.0%, 23/164), the difference did not meet the threshold of statistical difference. In the study conducted by de Roux Quentin et al., a history of peptic ulcer was the only independent risk factor of upper GI bleeding [12]. A key strength of this work was the construction of a predictive model for GI bleeding using LASSO regression and binary logistic regression. The model demonstrated good discrimination and calibration in both the training and internal validation sets. Although patients with GI bleeding exhibit worse laboratory indicators during ECMO, these factors were not incorporated into the final predictive model. Most GI bleeding events occurred within 72 h after initiating ECMO, especially in the ECPR group (Fig. 1A and supplementary material, Fig. S5). These findings indicate that GI bleeding may be attributed to patients' underlying conditions, and ischemia–reperfusion injury after ECMO may worsen GI tract damage [27]. 'CRRT' and 'baseline INR' in our predictive model align with two risk factors of clinically significant GI bleeding identified in the SUP-ICU study: 'utilization of renal replacement therapy' and 'co-existing coagulopathy' [18]. Besides, more attention should be paid to hypocalcemia. Previous studies have seldom linked serum calcium to ECMO-related bleeding. However, as an indispensable element in the coagulation cascade, deficiency in ionized calcium has been consistently proven to be associated with major bleeding in various disorders [28–30]. Owing to the incomplete data on ionic calcium, we opted to utilize serum total calcium instead for a preliminary exploration. Our findings indicated a negative correlation between serum total calcium levels and the risk of GI bleeding (Fig. S2). We intend to prospectively collect data on serum total and ionic calcium to further investigate the relationship between hypocalcemia and ECMO-related bleeding. Routine assessment of serum calcium and INR levels is recommended for patients scheduled to receiving VA ECMO. For patients with a prolonged treatment course prior to VA ECMO, correction of pre-existing hypocalcemia and optimization of baseline coagulation function may help mitigate the risk of GI bleeding during subsequent VA ECMO support. Our study also has some limitations which warrant mention. Studies of risk factors for GI bleeding in patients with ECMO are rare. Thus, we may have missed some potential risk factors due to a lack of previous studies to draw on. As a retrospective observational research, this study is restricted by incomplete data collection, and accurate identification of the etiologies leading to EN interruption in patients free from GI bleeding remains challenging. Furthermore, despite the inclusion of an external validation cohort, the limited sample size resulted in relatively wide confidence intervals. Therefore, additional validation in larger independent external cohorts is still required to verify the performance of this predictive model. Conclusions GI bleeding is a common complication among patients treated with VA ECMO and is associated with poor outcomes. CRRT, baseline INR, and baseline calcium were independently associated with GI bleeding in this study. The predictive model showed good diagnostic performance. Supplementary Information Supplementary Material 1. Acknowledgements Authors would like to thank all clinical staff at Extracorporeal Life Support Center of The First Affiliated Hospital with Nanjing Medical University. The authors would like to express their gratitude to EditSprings (https://www.editsprings.cn) for the expert linguistic services provided. Abbreviations Activated clotting time Activated partial thromboplastin time Continuous renal replacement therapy Extracorporeal membrane oxygenation Extracorporeal cardiopulmonary resuscitation Extracorporeal Life Support Organization Fresh frozen plasma Gastrointestinal International normalized ratio Least absolute shrinkage and selection operator Odds ratio Platelet Proton pump inhibitor Packed red blood cell Restricted cubic splines Survival after venoarterial ECMO Stabilized inverse probability of treatment weighting Stress-related mucosal disease Venoarterial Authors’ contributions Zhongman Zhang: Writing – original draft, Validation, Methodology, Supervision, Investigation.Yi Zhu: Writing – original draft, Validation, Methodology, Supervision. Gang Zhang: Writing – original draft, Methodology, Software. Xiaodong You: Writing – review & editing, Visualization, Data curation. Yanbin Dong: Writing – review & editing, Data curation. Deliang Hu: Writing – review & editing, Resources, Investigation. Haijun Sun: Writing – review & editing, Resources, Project administration, Conceptualization. Wei Li: Writing – review & editing, Resources, Project administration, Conceptualization. Xufeng Chen: Writing – review & editing, Resources, Project administration, Funding acquisition, Conceptualization. Funding This work was supported by Jiangsu Province Capability Improvement Project through Science, Technology and Education-Jiangsu Provincial Medical Key Discipline (ZDXK202213), Jiangsu Province Hospital Excellent Young and Middle-aged Talent Support Program Project (YNRCZN002) and Suqian Municipal Science and Technology Project-2023 Special Research Project of Suqian First Hospital(SY202313). Data availability The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study was approved by the Ethics Committee of The First Affiliated Hospital with Nanjing Medical University (No. 2023-SR-467) and The First People's Hospital of Suqian (KYSB20230181). Since the study was a retrospective study, the need for informed consent was waived by the Ethics Committee of The First Affiliated Hospital with Nanjing Medical University and The First People's Hospital of Suqian. This study was conducted in compliance with the Declaration of Helsinki. This study was carried out in compliance with the strengthening the reporting of observational studies in epidemiology (STROBE) and the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Zhongman Zhang and Yi Zhu contributed equally as the first authors. Contributor Information Haijun Sun, Email: Haijun6793100@163.com. Wei Li, Email: liwei5477@jsph.org.cn. Xufeng Chen, Email: cxfyx@njmu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.