Clinical assessment and prediction model construction for older patients with upper gastrointestinal bleeding

Clinical assessment and prediction model construction for older patients with upper gastrointestinal bleeding | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Clinical assessment and prediction model construction for older patients with upper gastrointestinal bleeding Yajie Li, Jingyuan Wang, Tianyi Wang, Qi Zhang, Xilong Ou This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6194668/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Objective To construct a clinical assessment prediction model for the safe discharge of older patients with upper gastrointestinal bleeding via machine learning algorithms. Methods A retrospective analysis was conducted on 1,032 older patients with upper gastrointestinal bleeding who were admitted through the emergency departments of Zhongda Hospital, Shanghai Changhai Hospital, Lishui Central Hospital, and Zhengzhou First Hospital from January 2018 to December 2020 to obtain relevant epidemiological, treatment, and prognostic data on Chinese older patients with upper gastrointestinal bleeding. This study aimed to clarify the comorbidities and medication history, onset conditions, causes, treatment interventions, and prognostic status of older patients with upper gastrointestinal bleeding, as well as the high-risk factors associated with death. A total of 218 older patients with upper gastrointestinal bleeding admitted to Zhongda Hospital from January 2021 to December 2022 who met the same criteria were subsequently selected as the validation group. A clinical assessment prediction model suitable for the older population was constructed via machine learning to predict safe discharge from upper gastrointestinal bleeding. Results The results revealed that the nine diagnostic variables of HB, SBP, INR, BUN, Alb, CR, HR, age, and CCI were used for algorithm modeling. The importance of the elements was consistent with the ranking in the machine learning algorithm. Among the machine learning algorithms, the random forest algorithm was the best clinical prediction model. For the validation group, the AUC of the RF model for the prediction of safe discharge was 0.889. The model prediction accuracy was 0.830 (0.781–0.880), the sensitivity was 0.868 (0.806–0.928), the specificity was 0.786 (0.705–0.867), the positive predictive value was 0.832 (0.765–0.898), and the negative predictive value was 0.828 (0.751–0.905), all of which were better than those of traditional scoring systems (CANUKA, AIMS65, MAP, ABC, and GBS). Conclusion Machine learning algorithms can form more accurate prediction models than traditional scoring systems. Upper gastrointestinal bleeding Older patients Safe discharge Data reduction Machine learning Introduction In recent years, the incidence of upper gastrointestinal bleeding (UGIB) in the United States has been reported to be 67 per 100,000 people, whereas in the United Kingdom, it is 134 per 100,000 people, with mortality rates ranging from 2% to 8.6% [1,2]. UGIB remains a leading cause of hospitalization and mortality among older individuals, with annual direct medical costs exceeding $1 billion in the U.S. [1,3]. The incidence of UGIB increases with age and is notably greater in older adults (197 per 100,000 for those aged 65–75 years and 425 per 100,000 for those over 75 years) [4]. According to the 2023 Statistical Bulletin on Civil Affairs Development released by China’s Ministry of Civil Affairs in August 2024, the population aged 60 and above in China reached 296.97 million by the end of 2023, accounting for 21.1% of the total population, with 216.76 million aged 65 and above (15.4% of the population). By international standards, China has entered a moderately aging society. It is projected that by 2030, the population aged 65 and above will approach 300 million. Currently, UGIB patients rely heavily on emergency interventions. In emergency settings, timely and appropriate risk stratification can guide endoscopic evaluation and therapeutic interventions for high-risk patients, thereby reducing mortality while enabling early discharge and outpatient management for low-risk patients to optimize resource utilization. Studies indicate that the application of the Glasgow–Blatchford score (GBS) for prognostic assessment may reduce UGIB hospitalization rates by 15–20% [5]. The Asia–Pacific Working Group consensus recommends managing UGIB through "early risk stratification" on the basis of prognostic factors, suggesting the identification of very low-risk patients for outpatient care [6]. Several risk scoring systems, such as the Rockall score (RS), AIMS65, and GBS, have been developed to predict outcomes, including mortality, rebleeding, need for hospital intervention, and transfusion requirements. However, research on these risk scores in older UGIB patients remains limited. CY Wang et al. reported that the RS accurately predicts rebleeding and mortality in older UGIB patients, yet the area under the ROC curve (AUC) remained below 0.8 [7]. Kalkan Ç et al. also demonstrated that the RS outperforms the GBS and AIMS65 in predicting mortality and rebleeding in older UGIB patients [8]. However, both studies had small sample sizes (341 and 335). Thus, the applicability of these risk scores in the older population lacks validation through large-scale clinical trials. The most recent guidelines for older UGIB patients were issued by the American Society for Gastrointestinal Endoscopy (ASGE) in 2013 [9], while China released its Expert Consensus on Emergency Diagnosis and Treatment of Older UGIB in December 2024 [10]. International consensus guidelines recommend risk scoring for UGIB patients, yet their precise role in the clinical assessment of older patients remains uncertain [11]. The newly published Chinese consensus recommends the GBS for predicting the need for interventions (endoscopic therapy, transfusion, surgery) and mortality risk in UGIB patients. However, the consensus acknowledges that the GBS, AIMS65, and RS each have strengths and limitations, as they target different clinical endpoints. Eugene Stolow et al. [12] noted that the GBS and AIMS65 may aid risk stratification but require validation through trials. Emergency physicians face challenges in rapidly determining the prognosis and intervention needs of UGIB patients. However, the applicability of existing risk scores in older UGIB patients remains unverified by multicenter, large-sample studies. Julian Prosenz et al. [13] reported only moderate-to-low physician adherence to the guidelines for nonvariceal UGIB. There is an urgent clinical need to establish a prognostic scoring system that addresses these gaps. Study objectives This multicenter study aims to do the following: Comprehensive data on comorbidities, concurrent medications, clinical presentation, etiology, treatment, and outcomes in older UGIB patients were collected to identify risk factors for safe discharge. A prognostic risk assessment model was developed on the basis of clinical and outcome data to stratify risk levels and guide therapeutic decisions. The model was validated via independent data from our institution to evaluate its predictive efficacy. 1. Materials and methods 1.1 General Information This study is a multicenter, noninterventional, real-world clinical study. The study included older patients diagnosed with nontraumatic upper gastrointestinal bleeding (UGIB) who were admitted through the emergency department from January 2018 to December 2020 at Zhongda Hospital Affiliated with Southeast University, Shanghai Changhai Hospital, Lishui Central Hospital, and Zhengzhou First Hospital, with hematemesis or melena as the primary manifestation, to form a construction cohort. In addition, older UGIB patients who met the above criteria from January 2021 to December 2022 at Zhongda Hospital Affiliated with Southeast University composed the validation cohort (2021ZDSYLL333–P01). Inclusion and exclusion criteria The inclusion criteria were as follows 1) aged ≥ 65 years; 2) presented with hematemesis, coffee-ground emesis, or melena as the primary complaint or chief medical history; and 3) hospital admission diagnosis of upper gastrointestinal bleeding (UGIB), including physician-documented diagnosis, admission records, and corresponding International Classification of Diseases, 10th Revision (ICD-10) codes. The exclusion criteria were as follows: 1) incomplete electronic medical records; 2) hospital-acquired UGIB during current admission; 3) patients transferred from other medical institutions; 4) unclassified gastrointestinal bleeding; and 5) UGIB secondary to mechanical trauma (e.g., nasogastric tube injury or foreign body ingestion). 1.1.1 Data collection The data collected in this study included the patients’ basic information (sex, age, hypertension, diabetes, liver diseases, kidney diseases, cardiovascular and cerebrovascular diseases, respiratory diseases, connective tissue diseases, blood system diseases, history of malignant tumors, etc.), history of relevant drug use (nonsteroidal drugs, antital platelet aggregation and anticoagulant drugs, etc.), clinical symptoms (hematemesis, hematochezia, syncope, etc.), vital signs and laboratory test indicators at the time of consultation (white blood cells, hemoglobin, hematocrit, platelets, blood urea nitrogen, creatinine, albumin, prothrombin time, fibrinogen, and international normalized ratio, etc.), and in-hospital treatment conditions (rebleeding, blood transfusion, interventional treatment, and prognosis, etc.). 1.1.2 Evaluation Results and Definition of Results Indicators The result of the evaluation is safe discharge. The definition of safe discharge [ 14 ] was as follows: lack of any of the following manifestations: 1) rebleeding; 2) need for blood transfusion; 3) need for therapeutic intervention, including endoscopic treatment, interventional treatment, and surgical treatment; and 4) death. Definition of rebleeding: After the first successful treatment, hematemesis and/or melena reappears, accompanied by the development of shock (pulse > 100 beats/min and/or systolic blood pressure < 100 mmHg) or a decrease in hemoglobin concentration exceeding \(\:2\hspace{0.33em}\text{g}/\text{d}\text{L}\) . The indications for blood transfusion are as follows: the hemoglobin level decreases to < 7 g/dL in ordinary patients and to < 8 g/dL[ 15 ] in patients at high risk of cardiovascular diseases. Endoscopic treatments include the injection of diluted adrenaline, titanium clip closure, or thermal coagulation therapy. Variceal bleeding is treated with a transjugular intrahepatic portosystemic shunt, band ligation, or injection of tissue glue. 1.2 Sample Size Estimation We used the software PASS23 (NCSS, LLC, Kaysville, Utah, USA) logistic regression for testing. Let \(\:\alpha\:=0.05\) , with a power of 0.8. According to previous findings from our research group, the incidence rate of endpoint events is approximately 50%. The OR value of low Hb is approximately 7.8, the OR value of increased BUN is approximately 2.1, the OR value of hypoproteinaemia is approximately 6.1, the OR value of hypotension is approximately 5.7, and the OR value of CCI > 2 is approximately 3.7. Combining the results of previous research, the \(\:{\text{R}}^{2}\) of this factor and several other covariates is 0.25–0.3. The percentages of \(\:\text{N}\) with \(\:\text{X}1\) are 65, 72, 6, 18.5, and 10, respectively. The sample sizes are 58, 389, 204, 322, and 200 cases, respectively. Taking the maximum value of 389 and considering a dropout rate of 10–20%, at least 467 patients were included in this study. 1.3 Statistical analysis The entire analysis was based on all the people in the cohort. All the collected data were analyzed via descriptive methods. SPSS 22.0 software and MedCalc 19.0 software were used for statistical analysis of the data. Count data are expressed as the number of cases \(\:\left(n,\text{\%}\right)\) , and the \(\:\chi\:2\) test was used for comparison. Data that conformed to a normal distribution are expressed as the mean \(\:\pm\:\) standard deviation \(\:\left(\overline{\text{x}}\pm\:\text{s}\right)\) , and the independent samples \(\:\text{t}\) test was used for comparisons between groups; measurement data that did not conform to a normal distribution are expressed as the median and quartiles \(\:\left[M\left(QL,QU\right)\right]\) , and the Mann‒Whitney U test was used for comparisons. This study was based on Python 3.x software. We then adopted machine learning algorithms, including KNN (K-nearest neighbor), GBoost (gradient boosted decision trees), MLP (multilayer perceptron), SVM (support vector machine), KSVM (kernel support vector machines), DT (decision tree), and RF (random forest) algorithms, to build models for the variable elements. The discriminatory power of the predictive outcomes was evaluated via receiver operating characteristic (ROC) curve analysis. Statistical significance was defined as a two-tailed P < 0.05. After the prognostic risk assessment model was constructed through the above steps, we calculated the AUROCs, sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) for safe discharge and verified the predictive efficacy of the new model in the validation cohort. 2. Results 2.1 Basic characteristics of patients In the cohort of this study, there were 673 male patients (65.2%), with an average age of 72.52±7.55 years. In terms of comorbidities, 248 patients (24.0%) had a history of liver cirrhosis, 202 (19.6%) had a history of malignant tumors, and 162 (15.7%) had a history of coronary heart disease. The Charlson Comorbidity Index (CCI) [16] was used to score comorbidities, and 313 patients (30.3%) had a CCI > 2. In terms of patients’ medication history, 165 patients (16.0%) took aspirin alone, 29 patients (2.8%) took clopidogrel or ticagrelor alone, and 33 patients (3.2%) were treated with dual anti-platelet drugs; 33 patients (3.2%) took anticoagulants, 10 patients (1%) took hormones, and 7 patients (0.7%) took nonsteroidal anti-inflammatory drugs. Among the 165 patients taking aspirin, a total of 133 underwent gastroscopy, and the vast majority of the causes were nonvariceal bleeding (129 cases, accounting for 97.0%), including 54 cases of digestive tract ulcers, 18 cases of digestive tract tumors, 23 cases of acute gastric mucosal lesions, and 7 cases of cardia rupture. Among the 29 patients taking clopidogrel or ticagrelor, a total of 16 underwent gastroscopy, including 6 patients with digestive tract ulcers, 4 patients with acute gastric mucosal lesions, and 2 patients with digestive tract tumors. The basic characteristics of all patients are detailed in Table 1. Table 1 Basic demographic characteristics of patients Variable Mean ± Standard Gender Male 673(65.2%) Female 359 (34.8%) Age 72.52±7.55 Comorbidities Cirrhosis of the liver 248(24.0%) Renal failure 85(8.2%) Malignant tumor 202(19.6%) Coronary heart disease 162 (15.7%) Heart failure 41(4.0%) Atrial fibrillation 59 (5.7%) Hypertension 502 (48.6%) Diabetes 245(23.7%) Chronic pulmonary disease 35(3.4%) Stroke 166(16.1%) CCI>2 313(30.3%) Previous medication history Aspirin 165(16.0%) Other antiplatelet drugs 29(2.8%) Dual antiplatelet drugs 33(3.2%) Anticoagulant drugs 33(3.2%) Non-steroidal anti-inflammatory drugs 7(0.7%) Hormone 10(1%) CCI ， Charlson Comorbidity Index 2.2 Admission status On admission, all patients had a systolic blood pressure of 124.07±20.63 mmHg and a heart rate of 80.91±14.13 beats per minute. Among all patients, 287 (27.8%) had a Glasgow Coma Scale (GCS) score of <14 at admission, including 91 (19.4%) in the safe discharge group and 196 (34.8%) in the nonsafe discharge group. The difference between the two groups was statistically significant, suggesting that the GCS score at admission may be correlated with prognosis. The vital signs, basic conditions and laboratory test results of all patients on admission are detailed in Table 2. Table 2 Conditions of all patients on admission Variable Mean ± Standard Hemodynamics Systolic Blood Pressure (mmHg) 124.07±20.63 Heart Rate (beats/min) 80.91±14.13 GCS ≥14 745（72.2%） <14 287（27.8%） ASA ≤2 549（53.2%） 3 312（30.2%） ≥4 171（16.6%） Laboratory examination Hemoglobin (g/L) 87.81±26.10 Albumin (g/L) 33.56±5.46 Urea nitrogen (mmol/L) 9.97±8.61 Creatinine (μmol/L) 95.43±88.38 INR 1.25±0.35 Hematocrit (%) 26.56±7.20 Platelets (*109/L) 175.41±99.53 Prothrombin time (s) 15.41±3.75 Activated partial thromboplastin time(s) 36.38±8.60 Fibrinogen (g/L) 2.64±1.61 D-dimer (μg/L) 2716.75±3510.34 Etiology Variceal bleeding 229（22.2%） Peptic ulcer 330（32.0%） Acute hemorrhagic erosion 139（13.5%） C Mallory-Weiss syndrome 21（2.0%） Digestive tract tumor 122（11.8%） GCS, Glasgow Coma Scale. ASA, American Society of Anesthesiologists Physical Status Classification. 2.3 Comparison of the basic characteristics of patients in different groups We divided all patients in the construction group into two groups. Taking safe discharge as the outcome, the basic characteristics of all patients, admission status, and relevant results of endoscopic examination and intervention measures are shown in Table 3. In terms of comorbidities, 248 (24.0%) patients had a history of liver cirrhosis. Among them, 48 (10.2%) in the safe discharge group and 200 (35.5%) in the nonsafe discharge group had a history of liver cirrhosis. Only 48 patients with liver cirrhosis were safely discharged from the two groups, indicating that a history of liver cirrhosis may suggest a poor prognosis. A total of 202 (19.6%) patients had a history of malignant tumors, including 83 (17.7%) in the safe discharge group and 119 (21.1%) in the nonsafe discharge group. The comorbidities were scored via the Charlson Comorbidity Index (CCI) [26]. A total of 313 patients (30.3%) had a CCI > 2, including 65 (13.9%) in the safe discharge group and 248 (44.0%) in the non–safe discharge group. There was a significant difference between the two groups, suggesting that the CCI is correlated with the severity of the disease and patient prognosis. In terms of hemodynamics at admission, patients in the nonsafe discharge group had lower blood pressure (119.69±20.13 vs. 129.13±20.19) and a faster heart rate (81.96±14.67 vs. 79.21±13.22). The difference between the two groups was statistically significant, suggesting that the systolic blood pressure and heart rate at admission may be correlated with the severity of the disease and the prognosis. In terms of general laboratory tests, patients in the nonsafe discharge group had lower hemoglobin, platelet, and albumin levels (79.89±25.17 vs . 96.97±24.38, 168.40±109.29 vs. 187.26±2.23, 32.77±5.41 vs. 34.90±5.19) and higher blood urea nitrogen, international normalized ratio (INR), and PT values (10.57±9.71 vs. 9.50±7.37, 1.31±0.44 vs. 1.19±0.19, 15.54±4.41 vs. 15.18±2.23). The differences between the two groups were statistically significant ( P < 0.05), suggesting that hemoglobin, platelet, albumin, blood urea nitrogen, the international normalized ratio (INR), and PT are correlated with the severity of the disease and patient prognosis. Indicators that were not significantly different between the two groups, such as creatinine and ASA, were not included in the table. Table 3 Basic demographic characteristics and admission status of patients in the two groups Safe discharge group (469) Non safe discharge group (563) P Male/Female 312/157 361/202 0.430 Age (mean ± standard) 73.03±7.63 72.10±7.46 0.468 Comorbidities Cirrhosis 48(10.2%) 200(35.5%) <0.001 Renal failure 39 (8.3%) 46(8.2%) 0.933 Malignant tumor 83(17.7%) 119(21.1%) 0.166 Coronary heart disease 96(20.5%) 66(11.7%) <0.001 Heart failure 17(3.6%) 24(4.3%) 0.601 Atrial fibrillation 27 (5.8%) 32 (5.7%) 0.960 Hypertension 264 (56.3%) 238(42.3%) <0.001 Diabetes 122 (26.0%) 123 (21.8%) 0.117 Chronic lung disease 18(3.8%) 17(3.0%) 0.470 Stroke 99(21.1%) 67(11.9%) 2 65(13.9%) 248 (44.0%) <0.001 Etiology Variceal bleeding 44(9.4%) 185(32.8%) <0.001 Peptic ulcer 197(42%) 133 (23.6%) <0.001 Acute hemorrhagic erosion 88(18.8%) 51(9.1%) <0.001 Mallory-Weiss syndrome 11(2.3%) 10(1.8%) 0.519 Digestive tract tumors 44(9.4%) 78(13.9%) <0.05 Anticoagulant/antiplatelet drugs 145(30.9%) 97 (17.2%) <0.001 Systolic blood pressure (mmHg) 129.13±20.19 119.69±20.13 <0.01 Heart rate (beats/min) 79.21±13.22 81.96±14.67 <0.01 GCS score (<14) 91(19.4%) 196(34.8%) <0.001 Hemoglobin (g/L) 96.97±24.38 79.89±25.17 <0.05 Platelets (*10 9 /L) 187.26±2.23 168.40±109.29 <0.01 Albumin (g/L) 34.90±5.19 32.77±5.41 <0.01 Blood urea nitrogen (mmol/L) 9.50±7.37 10.57±9.71 <0.05 INR 1.19±0.19 1.31±0.44 <0.01 PT(s) 15.18±2.23 15.54±4.41 <0.01 PT, Prothrombin Time. INR, International Normalized Ratio 2.4 Etiological conditions A total of 890 patients underwent endoscopic examination. Among all etiologies, gastrointestinal ulcer bleeding accounted for the greatest proportion (32.0%), followed by variceal bleeding (22.2%), acute hemorrhagic erosion (13.5%), and gastrointestinal tumor bleeding (11.8%). In the safe discharge group, the most common etiology was gastrointestinal ulcer bleeding (42%), followed by acute hemorrhagic erosion (18.8%), whereas in the nonsafe discharge group, the leading etiology was varices (32.8%), followed by gastrointestinal ulcer bleeding (23.6%), suggesting that different etiologies may be correlated with the prognosis and severity of patients’ conditions. 2.5 Comparison of baseline data between the construction cohort and the validation cohort Table 4 Comparison of baseline data between the construction cohort and the validation cohort Derivation cohort (N=1032) Validate cohort (N=218) P Male/Female 673/359 159/59 <0.001 Age (mean ± standard) 72.52±7.55 77.54±8.04 <0.001 Discharged safely 469（45.4%） 84（38.5%） 2 313（30.3%） 44（20.1%） <0.001 Etiology Peptic ulcer 330（32.0%） 63（28.9%） <0.001 Variceal bleeding 229（22.2%） 20（9.2%） <0.001 Digestive tract tumors 122（11.8%） 17（7.8%） <0.001 Anticoagulant/antiplatelet drugs 242（23.4%） 74（33.9%） <0.001 Systolic blood pressure (mmHg) 124.07±20.63 125.55±21.47 0.319 Heart rate (beats/min) 80.91±14.13 83.29±17.52 <0.001 GCS score (<14) 287（27.8%） 23（10.6%） <0.001 Hemoglobin (g/L) 87.81±26.10 87.69±27.87 0.279 Albumin (g/L) 33.56±5.46 33.56±27.87 0.059 Urea nitrogen (mmol/L) 9.97±8.61 11.37±6.43 0.054 INR 1.25±0.35 1.18±0.29 0.225 2.6 Outcome indicators 2.6.1 Treatment status In the constructed cohort, 356 patients (34.5%) received red blood cell transfusions, 165 patients received fresh frozen plasma transfusions, and 56 patients received cryoprecipitate transfusions. A total of 271 patients (26.3%) underwent therapeutic interventions, among whom 238 underwent endoscopic treatment. Two patients experienced rebleeding after endoscopic treatment and received radiological interventional therapy, and 3 patients experienced rebleeding after endoscopic treatment and underwent emergency surgical operations. Fifteen patients received radiological interventional therapy, among whom 2 patients underwent endoscopic treatment failure, and 1 patient was transferred to an emergency surgical operation after the failure of interventional therapy. Twenty-four patients underwent emergency surgery. 2.6.2 ICU treatment In the constructed cohort, 97 patients (9.4%) were transferred to the ICU for continued treatment due to concurrent heart failure and multiple organ dysfunction, 45 of whom (46.4%) died. 2.6.3 Rebleeding In the construction cohort, 225 patients (21.8%) experienced rebleeding. Among them, 24 patients did not undergo endoscopy, 120 patients experienced variceal bleeding, and 81 patients experienced nonvariceal bleeding. The rebleeding rate in the variceal group was greater than that in the nonvariceal group. Among the 225 patients who experienced rebleeding, 34 died. 2.6.4 Death There were 74 cases (7.2%) of patient death in the construction cohort. Among them, 30 patients died of multiple organ failure, 19 patients died of acute gastrointestinal bleeding, 14 patients died of heart failure, and 11 patients died of combined pulmonary infection and septic shock. 2.7 Data analysis In this section, on the basis of the algorithms of KNN (K-nearest neighbor), GBoost (gradient-boosted decision trees), MLP (multilayer perceptron), SVM (support vector machine), KSVM (kernel support vector machines), DT (decision tree), and RF (random forest), models are constructed for variable factors to accurately predict whether older UGIB patients can be safely discharged from the hospital. These models are compared with existing scoring criteria to verify their effectiveness. Table 5 shows the AUROC of the models constructed by seven machine-learning algorithms for the prediction of safe discharge results in the validation group. Among them, the RF model has the largest AUC value, whereas the DT model has the smallest AUC value. In terms of the accuracy of predicting safe discharge, the SVM, KSVM, and RF algorithm models all have good accuracy; however, the prediction effects of KNN, GBoost, and MLP are average, and the prediction effect of DT is poor. Table 5 ROC curves of 7 machine - learning algorithms for predicting safe discharge in the validation group Algorithm AUROC (%，95% CI ) Sensitivity (%，95% CI ) Specificity (%，95% CI ) PPV (%，95% CI ) NPV (%，95% CI ) KNN 0.744(0.683-0.805) 82.4(75.0-89.7) 56.9(47.9-65.9) 62.7(4.5-70.9) 78.6(69.8-87.4) GBoost 0.763(0.716-0.810) 80.4(73.0-87.7) 58.5(49.1-67.9) 67.2(59.2-75.1) 73.8(64.4-83.2) MLP 0.779(0.710-0.848) 83.7(76.5-90.8) 58.8(49.7-67.8) 64.9(56.9-73.0) 79.8(71.2-88.3) SVM 0.853(0.788-0.918) 87.6(81.1-94.2) 59.5(50.8-68.2) 55.3(72.3-71.6) 85.7(78.2-93.2) K-SVM 0.826(0.785-0.867) 86.0(79.2-92.8) 56.3(47.5-68.2) 64.2(56.1-72.3) 83.3(75.4-91.3) DT 0.569(0.510-0.628) 68.7(59.6-77.8) 44.5(35.6-53.5) 50.8(44.3-59.2) 63.1(52.8-73.4) RF 0.889(0.858-0.920) 86.7(80.6-92.8) 78.6(70.5-86.7) 83.2(76.5-89.8) 82.8(75.1-90.5) Table 6 shows the comparison results of the AUROC for predicting safe discharge in the validation group between the model constructed by the RF algorithm and the models constructed by the other six machine learning algorithms. There were significant differences between the model constructed via RF and the models constructed via algorithms such as KNN, GBoost, MLP, DT, and K-SVM ( p <0. 01, p 0.05). Table 6 Comparison of ROC curves for predicting safe discharge in the validation group between the RF algorithm and the other six machine algorithms Algorithm RF ( P ) KNN <0.01 GBoost <0.01 MLP 0.05 KSVM <0.05 DT <0.01 2.8 Comparison of the advantages and disadvantages between the constructed scoring system and other existing scoring systems Table 7 shows the comparison between four existing international scoring systems for predicting safe discharge in the validation group and the scoring system constructed in this study. Among the four existing scoring systems internationally, the ABC has the largest AUC value. There was no statistically significant difference between the ABC and MAP methods and the GBS ( p >0.05), whereas there was a statistically significant difference between the ABC and ABC methods ( p <0.05) and the AIMS65 ( p <0.01). Among the four scoring systems, there were statistically significant differences between the AIMS65 and the other three systems in each group. However, there was no statistically significant difference between the MAP, ABC or GBS in terms of safe discharge among the groups ( p >0.05). Table 7 Comparison of the abilities of different scoring systems to predict safe discharge AUROC (%，95% CI ) Sensitivity (%，95% CI ) Specificity (%，95% CI ) PPV (%，95% CI ) NPV (%，95% CI ) MAP(ASH) 0.807(0.744-0.869) 83.4(76.3-89.4) 67.9(56.8-77.6) 80.6(75.1-85.1) 72.2(63.2-79.6) CANUKA 0.849(0.796-0.903) 80.6(72.9-86.9) 76.2(65.7-84.8) 84.4(78.5-88.9) 71.1(63.1-78.0) ABC 0.763(0.698-0.828) 75.4(67.2-82.4) 66.7(55.5-76.6) 78.3(72.4-83.2) 62.9(54.9-70.3) AIMS65 0.650(0.577-0.722) 49.3(40.5-58.0) 82.1(72.3-89.6) 81.5(72.9-87.8) 50.4(45.5-55.2) GBS 0.821(0.761-0.880) 66.4(57.8-74.3) 86.9(77.8-93.3) 89.0(82.2-93.4) 61.9(55.8-67.6) RF 0.889(0.858-0.920) 86.7(80.6-92.8) 78.6(70.5-86.7) 83.2(76.5-89.8) 82.8(75.1-90.5) Table 8 shows the comparison results of the AUC values between the model constructed by the RF algorithm and four existing scoring systems in predicting safe discharge in the validation group. For the population in the validation group, there were significant differences between the model constructed by RF and the scoring systems such as ABC, AIMS65, MAP, and GBS ( p <0.01 for ABC and AIMS65; p <0.05 for MAP and GBS). The model designed on the basis of the RF algorithm in this project has certain advantages. Table 8 Comparison of the ROC curves of four scoring systems and the RF algorithm in predicting the safe discharge of the validation group scoring systems RF（ P ） MAP（ASH） 0.05 ABC <0.01 AIMS65 <0.01 GBS <0.05 3. Discussion This study is a multicenter investigation of older patients with upper gastrointestinal bleeding (UGIB) in China. Among the patients in our multicenter study, the number of males was 1.8 times greater than that of females. The median age was 70 years, and one-third of the patients (361/1032) were over 74 years old. Stratified by etiology, the median age of patients with variceal bleeding was 67 years, with males accounting for 53.7% (123/229) of the patients. The median age of nonvariceal patients was 71 years, with males accounting for 69.7% (461/661) of the patients. Reports from other countries worldwide also show that gastrointestinal bleeding is more common in males [ 1 , 2 ]. In this study, 91% (939/1032) of the patients had comorbidities, 57.8% (596/1032) of whom had two or more comorbidities. Previous studies have noted that the severity of comorbidities in UGIB patients is correlated with the prognosis of the disease[ 17 ]. The Charlson Comorbidity Index (CCI) was proposed by Charlson et al. in 1987. Some studies have shown that a gradual increase in the CCI is correlated with the prognosis of the disease[ 18 ]. Zellmer S et al. reported that in patients with COVID-19 infection, an increased CCI was positively correlated with an increased risk of bleeding [ 19 ]. In this study, compared with patients with a CCI ≤ 2 points, among the 313 patients (30.3%) with a CCI > 2, only 65 patients (13.9%) were in the safe discharge group, whereas 248 patients (44.0%) were in the non–safe discharge group, which also confirmed the correlation between the presence of underlying comorbidities and safe discharge. In terms of medication history, 23.4% (242/1032) of our patients were continuously taking antiplatelet or anticoagulant drugs at the onset of the disease. Among the 330 patients with peptic ulcers, 32.7% (108/330) were taking the abovementioned drugs, including 22.7% (75/330) taking antiplatelet drugs alone, 4.2% (14/330) taking dual antiplatelet drugs, and 3.9% (13/330) taking anticoagulant drugs, suggesting that the use of antiplatelet or anticoagulant drugs may be related to peptic ulcers. The widespread use of antiplatelet/anticoagulant drugs is related to the comorbidities of patients. For example, in a multicenter real-world study in China, approximately 13.9% of patients had comorbid CHD [ 20 ], whereas 21.4% of our patients had comorbid CHD/AF. Studies have confirmed that among patients receiving nonsteroidal anti-inflammatory drugs, the proportion of upper gastrointestinal bleeding (UGIB) ranges from 2.4–12% [ 21 , 22 ]. A systematic review of 11 randomized controlled trials revealed that aspirin increased the risk of gastrointestinal bleeding by 60%[ 23 ]. Dual antiplatelet therapy can further increase the risk of UGIB. Multiple guidelines recommend performing emergency endoscopy (within 24 hours) on patients with upper gastrointestinal bleeding (UGIB), which is beneficial for identifying the cause and performing endoscopic hemostasis treatment if necessary [ 6 , 11 ]. However, among our patients, only 86.2% (890/1032) completed the endoscopic examination. Among the patients who did not undergo gastroscopy, 16.2% (23/142) died, and the mortality rate was much higher than the overall rate. This finding is similar to reports in the UK and China [ 20 , 24 ]. When multiple factors that lead to patients failing to complete the endoscopic examination are explored, the following key reasons may be involved. First, for certain specific groups, such as patients with dementia or stroke, since they may lack the cognitive ability and physical cooperation required for endoscopic examination, doctors often consider them unsuitable for such examinations. Second, some patients assessed as low risk were planned to undergo endoscopic examination after discharge, but owing to the difficulty in obtaining follow-up information after discharge, these data were not included in the statistical analysis. In addition, there are patients with hemodynamic instability or extremely critical conditions, and the risk of endoscopic examination is significantly increased, leading doctors to possibly postpone or avoid performing this examination. For patients with hemodynamic instability, although the risk of endoscopic examination is relatively high, considering that endoscopy is the preferred diagnostic and treatment method for UGIB, a study by McWhirter et al. revealed that emergency endoscopy is largely a safe procedure for older patients with acute UGIB [ 25 ]. If effective endoscopic intervention can be implemented in a timely manner for this high-risk group, theoretically, their mortality rate may be reduced. This hypothesis needs to be further verified and supported by larger-scale real-world studies. In this study, the most common cause of UGIB (upper gastrointestinal bleeding) was digestive tract ulcer bleeding (32.0%), which was similar to the proportion reported in previous studies [ 2 ]. The second most common cause was variceal bleeding (22.2%), which was similar to the proportion reported in a domestic multicenter real-world study [ 20 ]. The third most common cause in our study was acute erosive bleeding (13.5%), which was much lower than the proportion reported in European and American countries (approximately 25%). In our study, 23.1% (238/1032) of the patients underwent endoscopic treatment, among whom 60% had variceal bleeding. A total of 1.5% (15/1032) of the patients underwent radiological interventional treatment, and 2.3% (24/1032) of the patients underwent emergency surgical operations. In recent years, with the effective use of drugs and the rapid development of endoscopic and interventional technologies, the demand for emergency surgery among patients with digestive tract bleeding has significantly decreased [ 26 ]. In terms of rebleeding, the incidence rate among our patients was 21.8% (225/1032). The incidence rate was slightly higher among variceal patients than nonvariceal patients, and it was higher than that reported in Western countries (10.8%) [ 27 ]. Among the 225 patients who experienced rebleeding, 34 patients died (15.1%), indicating that rebleeding is an important risk factor for death. The mortality rate among our patients was 7.2% (74/1032), which was similar to the rate reported abroad (7.9%) [ 28 ]. A prospective study on digestive tract bleeding in Hong Kong revealed that among all deaths, only 18.9% were considered to be related to bleeding. Among older patients, most die from causes unrelated to bleeding, such as multiple organ failure (23.9%), pulmonary diseases (23.5%), and terminal malignancies (33.7%) [ 29 ]. In this study, we introduced a new prognostic evaluation indicator, safe discharge, which means having none of the following manifestations: 1. Rebleeding; 2. Requiring blood transfusion; 3. Require therapeutic interventions, including endoscopic treatment, interventional treatment, and surgical treatment; 4. Death. In this study, 45.4% (469/1032) of the patients were safely discharged, indicating that this group of patients could be treated in the outpatient department without occupying inpatient resources. A large-scale microcost study on UGIB in the UK revealed that the average cost per patient in the UK was £2458 [ 30 ]. A decrease in the number of inpatients will greatly reduce hospitalization costs and medical expenditures. Safe discharge from the hospital is an important determinant for evaluating whether patients with UGIB can be treated only on an outpatient basis. The subjects of this study were older patients with UGIB. In our study, machine learning algorithms revealed that even in the older population, age is still an influencing factor that cannot be ignored. This may be related to the fact that older patients have more comorbidities. This finding is consistent with the results of our previous study [ 31 ]: the older population is more likely to develop UGIB, and their prognosis is worse than that of the control group, which requires more attention. Machine learning models have been widely used in the evaluation of clinical disease risk and prognosis and have achieved good feedback. The latest research shows that the calculation effect of machine learning algorithm models is similar to that of deep neural networks [ 32 ]. This study is based on the KNN, GBoost, MLP, SVM, KSVM, DT, and RF algorithms in machine learning. Combined with simple and easily obtainable clinical parameters after dimensionality reduction, a prognosis evaluation of older UGIB patients is conducted to predict whether they can be safely discharged from the hospital. It also compares machine learning algorithm models with existing international prognosis prediction and evaluation criteria such as CANUKA, MAP (ASH), GBS, ABC, and AIMS65. On the basis of the test index elements, machine learning algorithms have more advantages than do existing international scoring systems. Among them, the algorithm models are significantly better than the existing scoring systems. Foreign studies have shown that the UGIB prognosis prediction model constructed by machine learning is significantly better than the traditional clinical risk scoring system [ 33 , 34 ]. In reference [ 35 ], a model was constructed using machine learning algorithms to predict the mortality of acute nonvariceal UGIB, and the results were even more optimistic than those evaluated by the artificial neural network model. A systematic review revealed that machine learning algorithm models are more effective at predicting rebleeding, mortality, and interventional treatment, and the AUC of the algorithm models can reach 0.80 − 0.90[ 36 ]. Deng Fuxing et al. [ 37 ] used a residual machine learning (ML) model to predict major bleeding in patients with upper gastrointestinal bleeding (UGIB) during their stay in the intensive care unit (ICU), and the area under the receiver operating characteristic curve (AUC) was as high as 0.96. In UGIB, machine learning models based on large international multicenter cohorts outperformed currently commonly used clinical risk scores (GBS, admission Rockall score, and AIMS65) in predicting composite outcomes (blood transfusion, hemostatic intervention, or death) in both internal and external validation References. This finding is consistent with the results of the model constructed via the random forest (RF) algorithm in this study to predict safe discharge. Moreover, the algorithm model had statistically significant advantages over the other scoring systems (MAP(ASH), GBS, ABC, and AIMS65). In this study, the algorithm model was also compared with the CANUKA score. Although the difference in the advantages of the algorithm model was not statistically significant, the algorithm model was superior to the CANUKA scoring system in both sensitivity and specificity for predicting safe discharge (0.867/0.786 vs . 0.806/0.762). In practice, risk stratification scores are used to guide clinical applications, and the threshold for maximizing sensitivity should be selected to minimize misdiagnosis. Therefore, the negative predictive value is a very important indicator. In this study, the negative predictive value of the model based on the RF algorithm was 0.828, whereas that of the CANUKA score was only 0.711. This machine learning model is superior to existing clinical risk assessment tools in identifying low-risk patients (who can be safely discharged) and can reduce the misdiagnosis rate. Therefore, constructing a model based on machine learning algorithms to conduct an early, simple, and effective assessment of UGIB patients is highly important for improving the prognosis of patients [ 38 ]. The risk stratification scoring system has been widely used in clinical practice. Multiple studies have shown that the use of the Glasgow-Blatchford score (GBS) seems to be able to safely identify patients with a low risk of death or those in need of intervention [ 39 , 40 ]. However, some studies have also suggested that the GBS has relatively low performance in predicting mortality or rebleeding [ 41 ]⋅. According to the guidelines, patients with a GBS score of ≤ 1 can be discharged from the emergency department with an outpatient treatment arrangement, as these patients rarely die or require hemostatic intervention [ 6 ]. Our previous study indicated [ 31 ] that when GBS ≤ 1 was used as the cutoff value, the sensitivity for the composite outcome of blood transfusion, hemostatic intervention, or death was 97.37%, and the specificity was 11.92%. A meta-analysis reported similar results for GBS ≤ 2 [ 42 ]. That is, for the composite outcome of recurrent upper gastrointestinal bleeding (UGIB), intervention, or death, when both the machine learning model and the GBS identified patients reaching the composite endpoint with 100% sensitivity at the cutoff score, the specificity value of the machine learning model was 26%, whereas that of the GBS was 12% [ 33 ]. When machine learning was used to construct an algorithm model to predict the composite endpoint of UGIB patients, the specificity increased from 12% (GBS) to 26%, indicating that the machine learning algorithm model can improve the identification of low-risk patients who can be safely discharged from the emergency department for outpatient management. Our research revealed that the clinical risk scoring machine learning model trained on the dimensionality-reduced detection index variables has excellent performance and is superior to the existing clinical risk scoring systems in identifying low-risk UGIB patients. Previous studies on machine learning algorithm models for UGIB have been limited by sample size, homogeneous patient cohorts, and a lack of external validation. For example, the largest known machine learning algorithm study on UGIB to date [ 43 ] used relevant data from 2,380 patients. The internal validation revealed that the neural network model improved the 30-day mortality rate according to the Rockall score, including the endoscopic examination results, but the external validation of this model was not conducted. This study evaluated outcome indicators related to prognosis. With safe discharge as the composite endpoint, this study aimed to screen patients at extremely low risk and manage them as outpatients. Unlike most previous studies, this study included a relatively large number of patients from multiple centers in China. We applied different machine learning algorithms to construct a prognostic evaluation model to assess its performance in modeling the same dataset and finally selected the optimal model. We also compared this model with existing international prognostic scoring systems to evaluate the prognostic prediction efficacy of the model. Finally, this study included both internal and external validation, which made the evaluation of the performance of the machine learning model more rigorous. This study has certain limitations. Owing to the limited time during the doctoral programme, the number of centers and the sample size included in this study were not large. We plan to conduct a multicenter prospective study in the future, including more samples, and further validate this model. Declarations Acknowledgements : Not applicable. Authors' contributions : YL and JW contributed equally to this work. YL, JW and XO conceived of the study. YL, JW, TW and QZ contributed to data collection. XO was responsible for data analyses. All authors contributed to interpretation of the results. YL and JW drafted the manuscript. All authors contributed to the refinements of the manuscript and approved the final manuscript for publication. XO is the guarantor of the manuscript. Funding ：No Funding. Availability of data and materials ： No. Ethics approval and consent to participate ：This manuscript reports a study involving human participants. This study was conducted in accordance with the Declaration of Helsinki. The ethical approval for the study was obtained from the Ethics Committee for Clinical Research of Zhongda Hospital, affiliated to Southeast University with the approval number 2021ZDSYLL333-P01 to which none of the authors was affiliated. This study used the medical records obtained from the past clinical diagnosis and treatment. Upon application, the Ethics committee agreed that informed consent was not required. Consent for publication ：Not applicable. Competing interests ：The authors declare no competing interests. References Wuerth BA, Rockey DC. Changing epidemiology of upper gastrointestinal hemorrhage in the last decade: A nationwide analysis[J]. Dig Dis Sci 2018, 63:1286–93. Button LA, Roberts SE, Evans PA, et al. Hospitalized incidence and case fatality for upper gastrointestinal bleeding from 1999 to 2007: a record linkage study[J]. Aliment Pharmacol Ther 2011, 33:64–76. Cryer BL, Wilcox CM, Henk HJ, et al. The Economics of upper gastrointestinal bleeding in a US managed- care setting: a retrospective, claims- based analysis[J]. J Med Econ 2010, 13:70–7. Laine L, Yang H, Chang S- C, et al. Trends for incidence of hospitalization and death due to GI complications in the United States from 2001 to 2009[J]. Am J Gastroenterol 2012, 107:1190–5; Laursen SB, Dalton HR, Murray IA, et al. Performance of new thresholds of the Glasgow Blatchford score in managing patients with upper gastrointestinal bleeding[J]. Clin Gastroenterol Hepatol 2015, 13:115–21. Sung JJ, Chiu PW, Chan FKL, et al. Asia- Paciffc working group consensus on non Variceal upper gastrointestinal bleeding: an update 2018[J]. Gut 2018, 67:1757–68. Wang C- Y, Qin J, Wang J, et al. Rockall score in predicting outcomes of elderly patients with acute upper gastrointestinal bleeding[J]. World J Gastroenterol 2013, 19:3466–72. Kalkan Ç, Soykan I, Karakaya F, et al. Comparison of three scoring systems for risk stratiffcation in elderly patients Wıth acute upper gastrointestinal bleeding[J]. Geriatr Gerontol Int 2017, 17:575–83. Early DS, Acosta RD, Chandrasekhara V, et al. Modiffcations in endoscopic practice for the elderly[J]. Gastrointestinal Endoscopy 2013, 78:1–7. Chinese Society of Emergency Medicine, Chinese Society of Emergency Medicine of the Chinese Geriatrics Society, Beijing Society of Emergency Medicine, et al. Expert Consensus on Emergency Diagnosis and Treatment of Upper Gastrointestinal Bleeding in the Elderly [J]. Journal of Clinical Emergency Medicine, 2025; 12(44): 1013-1022. Barkun AN, Almadi M, Kuipers EJ, et al. Management of Nonvariceal upper gastrointestinal bleeding: guideline recommendations from the International consensus group[J]. Ann Intern Med, 2019, 171(11):805-822. Stanley AJ, Laine L, Dalton HR, et al. International Gastrointestinal Bleeding Consortium. Comparison of risk scoring systems for patients presentingwith upper gastrointestinal bleeding: international multicenter prospective study[J]. BMJ, 2017,356:i6432. Stolow E, Moreau C, Sayana H, Patel S. Management of Non-Variceal Upper GI Bleeding in the Geriatric Population: An Update. Curr Gastroenterol Rep, 2021,23(4):5. Prosenz J, Stättermayer MS, Riedl F, et al. Adherence to guidelines in patients with nonvariceal upper gastrointestinal bleeding (UGIB) - results from a retrospective single tertiary center registry[J]. Scand J Gastroenterol. 2023,58(8):856-862. Oakland K, Jairath V, Uberoi R, et al. Derivation and validation of a novel risk score for safe discharge after acute lower gastrointestinal bleeding: a Modeling study[J]. Lancet Gastroenterol Hepatol, 2017,2(9):635-643. Villanueva C, Colomo A, Bosch A, et al. Transfusion strategies for acute upper gastrointestinal bleeding[J]. N Engl J Med, 2013, 368(1):11-21. Armitage JN, van der Meulen JH, Royal College of Surgeons Comorbidity Consensus Group. Identifying Comorbidity in surgical patients using administrative data with the Royal college of Surgeons Charlson score[J]. BrJ Surg, 2010, 97(5):772-781. Ahmed A, Stanley AJ. Acute upper gastrointestinal bleeding in elderly individuals:etiology, diagnosis and treatment[J]. Drugs Aging, 2012, 29(12):933-940. Charlson ME, Carrozzino D, Guidi J, et al. Charlson Comorbidity Index: A Critical Review of Clinimetric Properties[J]. Psychother Psychosom, 2022,91(1):8-35. Zellmer S, Hanses F, Muzalyova A, et al. Gastrointestinal bleeding and endoscopic findings in critically and noncritically ill patients with corona virus disease 2019(COVID-19): Results from Lean European Open Survey onSARS-CoV-2(LEOSS) and COKA registries[J]. United Eur Gastroenterol J,2021, 9(9):1081-1090. Cryer B, Li C, Simon LS, et al. GI-REASONS: A novel 6-month, prospective, randomized, open-label, blinded endpoint (PROBE) trial[J]. Am J Gastroenterol, 2013, 108(3):392–400. Lanas A, Boers M, Nuevo J. Gastrointestinal events in at-risk patients starting nonsteroidal anti-inflammatory drugs (NSAIDs) for rheumatic diseases:the EVIDENCE study of European routine practice[J]. Ann Rheum Dis, 2015, 74(4):675-681. Elwood PC, Morgan G, Galante J, et al. Systematic review and meta-analysis of randomized trials to ascertain fatal gastrointestinal bleeding events attributable to preventive low-dose aspirin:No evidence of increased risk[J].PLoS One, 2016, 11(11):e0166166. Gu L,Xu F,Yuan J.Comparison of AIMS65,Glasgow-Blatchford and Rockall scoring approaches in predicting the risk of in-hospital death among emergency hospitalized patients with upper gastrointestinal bleeding: a retrospective observational study in Nanjing, China[J].BMC Gastroenterol, 2018,18(1): 98-106. McWhirter A, Mahmood S, Mensah E, et al. Evaluating the Safety and Outcomes of Esophagogastroduodenoscopy in Elderly Patients Presenting With Acute Upper Gastrointestinal Bleeding[J]. Cureus, 2023;15(10):e47116. Loffroy R, Rao P, Ota S, et al. Embolization of acute nonvariceal upper gastrointestinal hemorrhage resistant to endoscopic treatment: results and predictors of recurrent bleeding[J]. Cardiovasc Intervent Radiol, 2010,33(6):1088-1100. Laursen SB, Stanley AJ, Laine L, Schaffalitzky de Muckadell OB. Rebleeding in peptic ulcer bleeding - a nationwide cohort study of 19,537 patients[J]. Scand J Gastroenterol, 2022, 57(12):1423-1429. Chaudhary S, Mackay D, Pell JP, et al. Upper gastrointestinal bleeding in Scotland 2000-2015: trends in demographics, etiology and outcomes[J]. Aliment Pharmacol Ther, 2021, 53(3):383-389. Sung JJ, Tsoi KK, Ma TK, et al. Causes of mortality in patients with peptic ulcer bleeding: a prospective cohort study of 10,428 cases[J]. Am J Gastroenterol, 2010, 105(1):84-89. Campbell HE, Stokes EA, Bargo D, et al. Costs and quality of life associated with acute upper gastrointestinal bleeding in the UK: cohort analysis of patients in a cluster randomized trial[J]. BMJ Open, 2015, 5(4):e007230. Li Y, Lu Q, Song M, et al Novel risk score for acute upper gastrointestinal bleeding in elderly patients: a single-center retrospective study[J]. BMJOpen, 2023, 13(6):e072602. Sheridan RP, Wang WM, Liaw A, et al. Extreme gradient boosting as a method for quantitative structureactivity relationships[J]. Journal of chemical information and modeling, 2016, 56(12):2353-2360. Shung DL, Au B, Taylor RA, et al. Validation of a machine learning model that outperforms clinical risk scoring systems for upper gastrointestinalbleeding[J]. Gastroenterology, 2020,158(1):160-167. Nigam GB, Murphy MF, Travis SPL, et al. Machine learning in the assessment and management of acute gastrointestinal bleeding[J]. BMJ Med, 2024,19;3(1):e000699. Ungureanu BS, Gheonea DI, Florescu DN, et al. Predicting mortality in patients with nonvariceal upper gastrointestinal bleeding using machine- learning[J]. Frontiers in medicine, 2023,10:1134835. Shung D, Simonov M, Gentry M, et al. Machine learning to predict outcomes in patients with acute gastrointestinal bleeding: a systematic review[J]. Digestive diseases and sciences, 2019,64(8):2078-2087. Deng F, Cao Y, Wang H, et al. Prognosis of major bleeding based on residual variables and machine learning for critical patients with upper gastrointestinal bleeding: A multicenter study[J]. J Crit Care, 2025,85:154923. Pai AK, Fox VL. Gastrointestinal bleeding and management[J]. Pediatric clinics of north america, 2017,64(3): 543-561. Rivieri S, Carron PN, Schoepfer A, et al. External validation and comparison of the Glasgow-Blatchford score, modified Glasgow-Blatchford score, Rockall score and AIMS65 score in patients with upper gastrointestinal bleeding: a cross-sectional observational study in Western Switzerland[J]. Eur J Emerg Med, 2023,30(1):32-39. Pognonec C, Dirhoussi Z, Cury N, et al. External validation of Glasgow-Blatchford, modified Glasgow-Blatchford and CANUKA scores to identify low-risk patients with upper gastrointestinal bleeding in emergency departments: a retrospective cohort study[J]. Emerg Med J, 2023,40(6):451-457. Kozai L, Tan A, Nebrejas K, et al. Comparative diagnostic utility of Rockall and Glasgow-Blatchford scores in nonvariceal upper gastrointestinal bleeding: a systematic review and meta-analysis[J]. Eur J Gastroenterol Hepatol, 2025,37(2):161-166. Boustany A, Alali AA, Almadi M, et al. Pre-Endoscopic Scores Predicting Low-Risk Patients with Upper Gastrointestinal Bleeding: A Systematic Review and Meta-Analysis[J]. J Clin Med, 2023,12(16):5194. Rotondano G, Cipolletta L, Grossi E, et al. Artificial neural networks accurately predict mortality in patients with nonvariceal upper GI bleeding[J]. Gastrointestinal Endoscopy, 2011, 73(2):218–226. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 04 May, 2025 Editor assigned by journal 29 Apr, 2025 Editor invited by journal 10 Apr, 2025 Submission checks completed at journal 09 Apr, 2025 First submitted to journal 09 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6194668","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":451970703,"identity":"23f81d2e-dca2-4e35-b982-c74f96d2ce6f","order_by":0,"name":"Yajie Li","email":"","orcid":"","institution":"Zhongda Hospital, Southeast University","correspondingAuthor":false,"prefix":"","firstName":"Yajie","middleName":"","lastName":"Li","suffix":""},{"id":451970704,"identity":"42b82f17-6e85-4cca-9547-22a6737ed16a","order_by":1,"name":"Jingyuan Wang","email":"","orcid":"","institution":"Changhai Hospital of Second Military Medical University, Naval Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jingyuan","middleName":"","lastName":"Wang","suffix":""},{"id":451970709,"identity":"a2efe689-db7e-430d-bf8c-92140ef85e20","order_by":2,"name":"Tianyi Wang","email":"","orcid":"","institution":"Southeast University","correspondingAuthor":false,"prefix":"","firstName":"Tianyi","middleName":"","lastName":"Wang","suffix":""},{"id":451970712,"identity":"8b9270a9-ff87-48f5-87ea-e4d2a103eafe","order_by":3,"name":"Qi Zhang","email":"","orcid":"","institution":"Southeast University","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Zhang","suffix":""},{"id":451970714,"identity":"4c2fa730-b173-47e6-a147-04eba9025082","order_by":4,"name":"Xilong Ou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIie2PMWrDMBSGFQT2IqpVvoVKoHQw9lWeEGjy0DGjsrhLDuBj9AgvfbSTD+AhQybPgiwZMtRuyGprLFTf8POG/+PnMZZI/FHwPIXkHDHosopTYIrivTXH7s3ZuJlZ0X2/JRE+N36tLVXzjLA7VX4ATaVGznL6+lhSis4BQj/afQdAjT49MeHcsKTowSKalixXgJMycqbEy4pi/K+SKePpVdPGryv3lUoIYsRilOIwzr8QqLxlx4N2Nlv7RebNNoQd1TXJS7jeykrm9L2oPDD+cWUx9Zk6tphIJBL/kB+Z31Y2J5DzwwAAAABJRU5ErkJggg==","orcid":"","institution":"Zhongda Hospital, Southeast University","correspondingAuthor":true,"prefix":"","firstName":"Xilong","middleName":"","lastName":"Ou","suffix":""}],"badges":[],"createdAt":"2025-03-10 10:53:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6194668/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6194668/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82206377,"identity":"1e4f0764-d387-4878-9369-9fb623e730fb","added_by":"auto","created_at":"2025-05-07 17:36:45","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1084472,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6194668/v1/09146aaa-7faf-45cc-b3a4-9c7c5556b093.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Clinical assessment and prediction model construction for older patients with upper gastrointestinal bleeding","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIn recent years, the incidence of upper gastrointestinal bleeding (UGIB) in the United States has been reported to be 67 per 100,000 people, whereas in the United Kingdom, it is 134 per 100,000 people, with mortality rates ranging from 2% to 8.6% [1,2]. UGIB remains a leading cause of hospitalization and mortality among older individuals, with annual direct medical costs exceeding $1 billion in the U.S. [1,3]. The incidence of UGIB increases with age and is notably greater in older adults (197 per 100,000 for those aged 65–75 years and 425 per 100,000 for those over 75 years) [4]. According to the \u003cem\u003e2023 Statistical Bulletin on Civil Affairs Development\u003c/em\u003e released by China’s Ministry of Civil Affairs in August 2024, the population aged 60 and above in China reached 296.97 million by the end of 2023, accounting for 21.1% of the total population, with 216.76 million aged 65 and above (15.4% of the population). By international standards, China has entered a moderately aging society. It is projected that by 2030, the population aged 65 and above will approach 300 million. Currently, UGIB patients rely heavily on emergency interventions. In emergency settings, timely and appropriate risk stratification can guide endoscopic evaluation and therapeutic interventions for high-risk patients, thereby reducing mortality while enabling early discharge and outpatient management for low-risk patients to optimize resource utilization. Studies indicate that the application of the Glasgow–Blatchford score (GBS) for prognostic assessment may reduce UGIB hospitalization rates by 15–20% [5]. The Asia–Pacific Working Group consensus recommends managing UGIB through \"early risk stratification\" on the basis of prognostic factors, suggesting the identification of very low-risk patients for outpatient care [6]. Several risk scoring systems, such as the Rockall score (RS), AIMS65, and GBS, have been developed to predict outcomes, including mortality, rebleeding, need for hospital intervention, and transfusion requirements.\u003c/p\u003e\n\u003cp\u003eHowever, research on these risk scores in older UGIB patients remains limited. CY Wang et al. reported that the RS accurately predicts rebleeding and mortality in older UGIB patients, yet the area under the ROC curve (AUC) remained below 0.8 [7]. Kalkan Ç et al. also demonstrated that the RS outperforms the GBS and AIMS65 in predicting mortality and rebleeding in older UGIB patients [8]. However, both studies had small sample sizes (341 and 335). Thus, the applicability of these risk scores in the older population lacks validation through large-scale clinical trials. The most recent guidelines for older UGIB patients were issued by the American Society for Gastrointestinal Endoscopy (ASGE) in 2013 [9], while China released its \u003cem\u003eExpert Consensus on Emergency Diagnosis and Treatment of Older UGIB\u003c/em\u003e in December 2024 [10].\u003c/p\u003e\n\u003cp\u003eInternational consensus guidelines recommend risk scoring for UGIB patients, yet their precise role in the clinical assessment of older patients remains uncertain [11]. The newly published Chinese consensus recommends the GBS for predicting the need for interventions (endoscopic therapy, transfusion, surgery) and mortality risk in UGIB patients. However, the consensus acknowledges that the GBS, AIMS65, and RS each have strengths and limitations, as they target different clinical endpoints. Eugene Stolow et al. [12] noted that the GBS and AIMS65 may aid risk stratification but require validation through trials. Emergency physicians face challenges in rapidly determining the prognosis and intervention needs of UGIB patients. However, the applicability of existing risk scores in older UGIB patients remains unverified by multicenter, large-sample studies. Julian Prosenz et al. [13] reported only moderate-to-low physician adherence to the guidelines for nonvariceal UGIB. There is an urgent clinical need to establish a prognostic scoring system that addresses these gaps.\u003c/p\u003e\n\u003ch3\u003eStudy objectives\u003c/h3\u003e\n\u003cp\u003eThis multicenter study aims to do the following:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eComprehensive data on comorbidities, concurrent medications, clinical presentation, etiology, treatment, and outcomes in older UGIB patients were collected to identify risk factors for safe discharge.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eA prognostic risk assessment model was developed on the basis of clinical and outcome data to stratify risk levels and guide therapeutic decisions.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe model was validated via independent data from our institution to evaluate its predictive efficacy.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"1. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.1 General Information\u003c/h2\u003e \u003cp\u003eThis study is a multicenter, noninterventional, real-world clinical study. The study included older patients diagnosed with nontraumatic upper gastrointestinal bleeding (UGIB) who were admitted through the emergency department from January 2018 to December 2020 at Zhongda Hospital Affiliated with Southeast University, Shanghai Changhai Hospital, Lishui Central Hospital, and Zhengzhou First Hospital, with hematemesis or melena as the primary manifestation, to form a construction cohort. In addition, older UGIB patients who met the above criteria from January 2021 to December 2022 at Zhongda Hospital Affiliated with Southeast University composed the validation cohort (2021ZDSYLL333\u0026ndash;P01).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eInclusion and exclusion criteria\u003c/h3\u003e\n\u003cp\u003e \u003cstrong\u003eThe inclusion criteria were as follows\u003c/strong\u003e \u003cp\u003e1) aged\u0026thinsp;\u0026ge;\u0026thinsp;65 years; 2) presented with hematemesis, coffee-ground emesis, or melena as the primary complaint or chief medical history; and 3) hospital admission diagnosis of upper gastrointestinal bleeding (UGIB), including physician-documented diagnosis, admission records, and corresponding International Classification of Diseases, 10th Revision (ICD-10) codes.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eThe exclusion criteria were as follows: 1) incomplete electronic medical records; 2) hospital-acquired UGIB during current admission; 3) patients transferred from other medical institutions; 4) unclassified gastrointestinal bleeding; and 5) UGIB secondary to mechanical trauma (e.g., nasogastric tube injury or foreign body ingestion).\u003c/p\u003e\n\u003ch3\u003e1.1.1 Data collection\u003c/h3\u003e\n\u003cp\u003eThe data collected in this study included the patients\u0026rsquo; basic information (sex, age, hypertension, diabetes, liver diseases, kidney diseases, cardiovascular and cerebrovascular diseases, respiratory diseases, connective tissue diseases, blood system diseases, history of malignant tumors, etc.), history of relevant drug use (nonsteroidal drugs, antital platelet aggregation and anticoagulant drugs, etc.), clinical symptoms (hematemesis, hematochezia, syncope, etc.), vital signs and laboratory test indicators at the time of consultation (white blood cells, hemoglobin, hematocrit, platelets, blood urea nitrogen, creatinine, albumin, prothrombin time, fibrinogen, and international normalized ratio, etc.), and in-hospital treatment conditions (rebleeding, blood transfusion, interventional treatment, and prognosis, etc.).\u003c/p\u003e\n\u003ch3\u003e1.1.2 Evaluation Results and Definition of Results Indicators\u003c/h3\u003e\n\u003cp\u003eThe result of the evaluation is safe discharge. The definition of safe discharge [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] was as follows: lack of any of the following manifestations: 1) rebleeding; 2) need for blood transfusion; 3) need for therapeutic intervention, including endoscopic treatment, interventional treatment, and surgical treatment; and 4) death.\u003c/p\u003e \u003cp\u003eDefinition of rebleeding: After the first successful treatment, hematemesis and/or melena reappears, accompanied by the development of shock (pulse\u0026thinsp;\u0026gt;\u0026thinsp;100 beats/min and/or systolic blood pressure\u0026thinsp;\u0026lt;\u0026thinsp;100 mmHg) or a decrease in hemoglobin concentration exceeding \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:2\\hspace{0.33em}\\text{g}/\\text{d}\\text{L}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe indications for blood transfusion are as follows: the hemoglobin level decreases to \u0026lt;\u0026thinsp;7 g/dL in ordinary patients and to \u0026lt;\u0026thinsp;8 g/dL[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] in patients at high risk of cardiovascular diseases.\u003c/p\u003e \u003cp\u003eEndoscopic treatments include the injection of diluted adrenaline, titanium clip closure, or thermal coagulation therapy. Variceal bleeding is treated with a transjugular intrahepatic portosystemic shunt, band ligation, or injection of tissue glue.\u003c/p\u003e\n\u003ch3\u003e1.2 Sample Size Estimation\u003c/h3\u003e\n\u003cp\u003eWe used the software PASS23 (NCSS, LLC, Kaysville, Utah, USA) logistic regression for testing. Let \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\alpha\\:=0.05\\)\u003c/span\u003e\u003c/span\u003e, with a power of 0.8. According to previous findings from our research group, the incidence rate of endpoint events is approximately 50%. The OR value of low Hb is approximately 7.8, the OR value of increased BUN is approximately 2.1, the OR value of hypoproteinaemia is approximately 6.1, the OR value of hypotension is approximately 5.7, and the OR value of CCI\u0026thinsp;\u0026gt;\u0026thinsp;2 is approximately 3.7. Combining the results of previous research, the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\text{R}}^{2}\\)\u003c/span\u003e\u003c/span\u003e of this factor and several other covariates is 0.25\u0026ndash;0.3. The percentages of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{N}\\)\u003c/span\u003e\u003c/span\u003e with \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{X}1\\)\u003c/span\u003e\u003c/span\u003e are 65, 72, 6, 18.5, and 10, respectively. The sample sizes are 58, 389, 204, 322, and 200 cases, respectively. Taking the maximum value of 389 and considering a dropout rate of 10\u0026ndash;20%, at least 467 patients were included in this study.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e1.3 Statistical analysis\u003c/h2\u003e \u003cp\u003eThe entire analysis was based on all the people in the cohort. All the collected data were analyzed via descriptive methods. SPSS 22.0 software and MedCalc 19.0 software were used for statistical analysis of the data. Count data are expressed as the number of cases \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left(n,\\text{\\%}\\right)\\)\u003c/span\u003e\u003c/span\u003e, and the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\chi\\:2\\)\u003c/span\u003e\u003c/span\u003e test was used for comparison. Data that conformed to a normal distribution are expressed as the mean \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\pm\\:\\)\u003c/span\u003e\u003c/span\u003e standard deviation \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left(\\overline{\\text{x}}\\pm\\:\\text{s}\\right)\\)\u003c/span\u003e\u003c/span\u003e, and the independent samples \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{t}\\)\u003c/span\u003e\u003c/span\u003e test was used for comparisons between groups; measurement data that did not conform to a normal distribution are expressed as the median and quartiles \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\left[M\\left(QL,QU\\right)\\right]\\)\u003c/span\u003e\u003c/span\u003e, and the Mann‒Whitney U test was used for comparisons.\u003c/p\u003e \u003cp\u003eThis study was based on Python 3.x software. We then adopted machine learning algorithms, including KNN (K-nearest neighbor), GBoost (gradient boosted decision trees), MLP (multilayer perceptron), SVM (support vector machine), KSVM (kernel support vector machines), DT (decision tree), and RF (random forest) algorithms, to build models for the variable elements. The discriminatory power of the predictive outcomes was evaluated via receiver operating characteristic (ROC) curve analysis. Statistical significance was defined as a two-tailed P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003eAfter the prognostic risk assessment model was constructed through the above steps, we calculated the AUROCs, sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) for safe discharge and verified the predictive efficacy of the new model in the validation cohort.\u003c/p\u003e \u003c/div\u003e"},{"header":"2. Results","content":"\u003ch1\u003e2.1 Basic characteristics of patients\u003c/h1\u003e\n\u003cp\u003eIn the cohort of this study, there were 673 male patients (65.2%), with an average age of 72.52\u0026plusmn;7.55 years. In terms of comorbidities, 248 patients (24.0%) had a history of liver cirrhosis, 202 (19.6%) had a history of malignant tumors, and 162 (15.7%) had a history of coronary heart disease. The Charlson Comorbidity Index (CCI) [16] was used to score comorbidities, and 313 patients (30.3%) had a CCI \u0026gt; 2.\u003c/p\u003e\n\u003cp\u003eIn terms of patients\u0026rsquo; medication history, 165 patients (16.0%) took aspirin alone, 29 patients (2.8%) took clopidogrel or ticagrelor alone, and 33 patients (3.2%) were treated with dual anti-platelet drugs; 33 patients (3.2%) took anticoagulants, 10 patients (1%) took hormones, and 7 patients (0.7%) took nonsteroidal anti-inflammatory drugs. Among the 165 patients taking aspirin, a total of 133 underwent gastroscopy, and the vast majority of the causes were nonvariceal bleeding (129 cases, accounting for 97.0%), including 54 cases of digestive tract ulcers, 18 cases of digestive tract tumors, 23 cases of acute gastric mucosal lesions, and 7 cases of cardia rupture. Among the 29 patients taking clopidogrel or ticagrelor, a total of 16 underwent gastroscopy, including 6 patients with digestive tract ulcers, 4 patients with acute gastric mucosal lesions, and 2 patients with digestive tract tumors. The basic characteristics of all patients are detailed in Table 1.\u003c/p\u003e\n\u003cp\u003eTable 1 Basic demographic characteristics of patients\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMean \u0026plusmn; Standard\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e673(65.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e359 (34.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e72.52\u0026plusmn;7.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"11\" valign=\"top\"\u003e\n \u003cp\u003eComorbidities\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCirrhosis of the liver\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e248(24.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRenal failure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85(8.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMalignant tumor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e202(19.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCoronary heart disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e162 (15.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHeart failure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e41(4.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAtrial fibrillation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e59 (5.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e502 (48.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDiabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e245(23.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eChronic pulmonary disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e35(3.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eStroke\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e166(16.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCCI\u0026gt;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e313(30.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"6\" valign=\"top\"\u003e\n \u003cp\u003ePrevious medication history\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAspirin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e165(16.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOther antiplatelet drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e29(2.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDual antiplatelet drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33(3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnticoagulant drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33(3.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNon-steroidal anti-inflammatory drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e7(0.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHormone\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10(1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eCCI\u003c/em\u003e\u003cem\u003e，\u003c/em\u003e\u003cem\u003e\u0026nbsp;Charlson Comorbidity Index\u003c/em\u003e\u003c/p\u003e\n\u003ch1\u003e2.2 Admission status\u003c/h1\u003e\n\u003cp\u003eOn admission, all patients had a systolic blood pressure of 124.07\u0026plusmn;20.63 mmHg and a heart rate of 80.91\u0026plusmn;14.13 beats per minute. Among all patients, 287 (27.8%) had a Glasgow Coma Scale (GCS) score of \u0026lt;14 at admission, including 91 (19.4%) in the safe discharge group and 196 (34.8%) in the nonsafe discharge group. The difference between the two groups was statistically significant, suggesting that the GCS score at admission may be correlated with prognosis. The vital signs, basic conditions and laboratory test results of all patients on admission are detailed in Table 2.\u003c/p\u003e\n\u003cp\u003eTable 2 Conditions of all patients on admission\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 435px;\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003eMean \u0026plusmn; Standard\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003eHemodynamics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eSystolic Blood Pressure (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e124.07\u0026plusmn;20.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eHeart Rate (beats/min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e80.91\u0026plusmn;14.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003eGCS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u0026ge;14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e745（72.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u0026lt;14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e287（27.8%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003eASA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u0026le;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e549（53.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e312（30.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003e\u0026ge;4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e171（16.6%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003eLaboratory \u0026nbsp;examination\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eHemoglobin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e87.81\u0026plusmn;26.10\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eAlbumin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e33.56\u0026plusmn;5.46\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eUrea nitrogen (mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e9.97\u0026plusmn;8.61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eCreatinine (\u0026mu;mol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e95.43\u0026plusmn;88.38\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e1.25\u0026plusmn;0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eHematocrit (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e26.56\u0026plusmn;7.20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003ePlatelets (*109/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e175.41\u0026plusmn;99.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eProthrombin time (s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e15.41\u0026plusmn;3.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eActivated partial thromboplastin time(s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e36.38\u0026plusmn;8.60\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eFibrinogen (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e2.64\u0026plusmn;1.61\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eD-dimer (\u0026mu;g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e2716.75\u0026plusmn;3510.34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003eEtiology\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eVariceal bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e229（22.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003ePeptic ulcer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e330（32.0%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eAcute hemorrhagic erosion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e139（13.5%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eC Mallory-Weiss syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e21（2.0%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 186px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 249px;\"\u003e\n \u003cp\u003eDigestive tract tumor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e122（11.8%）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eGCS, Glasgow Coma Scale. ASA, American Society of Anesthesiologists Physical Status Classification.\u003c/em\u003e\u003c/p\u003e\n\u003ch1\u003e2.3 Comparison of the basic characteristics of patients in different groups\u003c/h1\u003e\n\u003cp\u003eWe divided all patients in the construction group into two groups. Taking safe discharge as the outcome, the basic characteristics of all patients, admission status, and relevant results of endoscopic examination and intervention measures are shown in Table 3.\u003c/p\u003e\n\u003cp\u003eIn terms of comorbidities, 248 (24.0%) patients had a history of liver cirrhosis. Among them, 48 (10.2%) in the safe discharge group and 200 (35.5%) in the nonsafe discharge group had a history of liver cirrhosis. Only 48 patients with liver cirrhosis were safely discharged from the two groups, indicating that a history of liver cirrhosis may suggest a poor prognosis. A total of 202 (19.6%) patients had a history of malignant tumors, including 83 (17.7%) in the safe discharge group and 119 (21.1%) in the nonsafe discharge group. The comorbidities were scored via the Charlson Comorbidity Index (CCI) [26]. A total of 313 patients (30.3%) had a CCI \u0026gt; 2, including 65 (13.9%) in the safe discharge group and 248 (44.0%) in the non\u0026ndash;safe discharge group. There was a significant difference between the two groups, suggesting that the CCI is correlated with the severity of the disease and patient prognosis.\u003c/p\u003e\n\u003cp\u003eIn terms of hemodynamics at admission, patients in the nonsafe discharge group had lower blood pressure (119.69\u0026plusmn;20.13 \u003cem\u003evs.\u0026nbsp;\u003c/em\u003e129.13\u0026plusmn;20.19) and a faster heart rate (81.96\u0026plusmn;14.67\u003cem\u003e\u0026nbsp;vs.\u003c/em\u003e 79.21\u0026plusmn;13.22). The difference between the two groups was statistically significant, suggesting that the systolic blood pressure and heart rate at admission may be correlated with the severity of the disease and the prognosis.\u003c/p\u003e\n\u003cp\u003eIn terms of general laboratory tests, patients in the nonsafe discharge group had lower hemoglobin, platelet, and albumin levels (79.89\u0026plusmn;25.17 \u003cem\u003evs\u003c/em\u003e. 96.97\u0026plusmn;24.38, 168.40\u0026plusmn;109.29 \u003cem\u003evs.\u0026nbsp;\u003c/em\u003e187.26\u0026plusmn;2.23, 32.77\u0026plusmn;5.41 \u003cem\u003evs.\u0026nbsp;\u003c/em\u003e34.90\u0026plusmn;5.19) and higher blood urea nitrogen, international normalized ratio (INR), and PT values (10.57\u0026plusmn;9.71 \u003cem\u003evs.\u003c/em\u003e 9.50\u0026plusmn;7.37, 1.31\u0026plusmn;0.44 \u003cem\u003evs.\u003c/em\u003e 1.19\u0026plusmn;0.19, 15.54\u0026plusmn;4.41 \u003cem\u003evs.\u003c/em\u003e 15.18\u0026plusmn;2.23). The differences between the two groups were statistically significant (\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.05), suggesting that hemoglobin, platelet, albumin, blood urea nitrogen, the international normalized ratio (INR), and PT are correlated with the severity of the disease and patient prognosis. Indicators that were not significantly different between the two groups, such as creatinine and ASA, were not included in the table.\u003c/p\u003e\n\u003cp\u003eTable 3 Basic demographic characteristics and admission status of patients in the two groups\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSafe discharge group (469)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNon safe discharge group (563)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMale/Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e312/157\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e361/202\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.430\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAge (mean \u0026plusmn; standard)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e73.03\u0026plusmn;7.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e72.10\u0026plusmn;7.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.468\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eComorbidities\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCirrhosis\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e48(10.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e200(35.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRenal failure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e39 (8.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e46(8.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.933\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMalignant tumor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e83(17.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e119(21.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.166\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCoronary heart disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96(20.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e66(11.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHeart failure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17(3.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e24(4.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.601\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAtrial fibrillation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e27 (5.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e32 (5.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.960\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e264 (56.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e238(42.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDiabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e122 (26.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e123 (21.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.117\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eChronic lung disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e18(3.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17(3.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.470\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eStroke\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e99(21.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e67(11.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCCI \u0026gt; 2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e65(13.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e248 (44.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEtiology\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVariceal bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e44(9.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e185(32.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePeptic ulcer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e197(42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e133 (23.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcute hemorrhagic erosion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e88(18.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e51(9.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMallory-Weiss syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11(2.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10(1.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.519\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDigestive tract tumors\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e44(9.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e78(13.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnticoagulant/antiplatelet drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e145(30.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97 (17.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSystolic blood pressure (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e129.13\u0026plusmn;20.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e119.69\u0026plusmn;20.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHeart rate (beats/min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e79.21\u0026plusmn;13.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e81.96\u0026plusmn;14.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eGCS score (\u0026lt;14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e91(19.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e196(34.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHemoglobin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.97\u0026plusmn;24.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e79.89\u0026plusmn;25.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePlatelets (*10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e187.26\u0026plusmn;2.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e168.40\u0026plusmn;109.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAlbumin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e34.90\u0026plusmn;5.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e32.77\u0026plusmn;5.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBlood urea nitrogen (mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.50\u0026plusmn;7.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e10.57\u0026plusmn;9.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.19\u0026plusmn;0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.31\u0026plusmn;0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePT(s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.18\u0026plusmn;2.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e15.54\u0026plusmn;4.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003ePT, Prothrombin Time. INR, International Normalized Ratio\u003c/em\u003e\u003c/p\u003e\n\u003ch1\u003e2.4 Etiological conditions\u003c/h1\u003e\n\u003cp\u003eA total of 890 patients underwent endoscopic examination. Among all etiologies, gastrointestinal ulcer bleeding accounted for the greatest proportion (32.0%), followed by variceal bleeding (22.2%), acute hemorrhagic erosion (13.5%), and gastrointestinal tumor bleeding (11.8%). In the safe discharge group, the most common etiology was gastrointestinal ulcer bleeding (42%), followed by acute hemorrhagic erosion (18.8%), whereas in the nonsafe discharge group, the leading etiology was varices (32.8%), followed by gastrointestinal ulcer bleeding (23.6%), suggesting that different etiologies may be correlated with the prognosis and severity of patients\u0026rsquo; conditions.\u003c/p\u003e\n\u003ch1\u003e2.5 Comparison of baseline data between the construction cohort and the validation cohort\u003c/h1\u003e\n\u003cp\u003eTable 4 Comparison of baseline data between the construction cohort and the validation cohort\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDerivation cohort (N=1032)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eValidate cohort (N=218)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMale/Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e673/359\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e159/59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAge (mean \u0026plusmn; standard)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e72.52\u0026plusmn;7.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e77.54\u0026plusmn;8.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDischarged safely\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e469（45.4%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e84（38.5%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCCI\u0026gt;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e313（30.3%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e44（20.1%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003eEtiology\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePeptic ulcer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e330（32.0%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e63（28.9%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eVariceal bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e229（22.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e20（9.2%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDigestive tract tumors\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e122（11.8%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e17（7.8%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnticoagulant/antiplatelet drugs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e242（23.4%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e74（33.9%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSystolic blood pressure (mmHg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e124.07\u0026plusmn;20.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e125.55\u0026plusmn;21.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.319\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHeart rate (beats/min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e80.91\u0026plusmn;14.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e83.29\u0026plusmn;17.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eGCS score (\u0026lt;14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e287（27.8%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e23（10.6%）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHemoglobin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e87.81\u0026plusmn;26.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e87.69\u0026plusmn;27.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.279\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAlbumin (g/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33.56\u0026plusmn;5.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e33.56\u0026plusmn;27.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.059\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUrea nitrogen (mmol/L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e9.97\u0026plusmn;8.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e11.37\u0026plusmn;6.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.054\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eINR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.25\u0026plusmn;0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.18\u0026plusmn;0.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.225\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch1\u003e2.6 Outcome\u0026nbsp;indicators\u003c/h1\u003e\n\u003ch1\u003e2.6.1 Treatment\u0026nbsp;status\u003c/h1\u003e\n\u003cp\u003eIn the constructed cohort, 356 patients (34.5%) received red blood cell transfusions, 165 patients received fresh frozen plasma transfusions, and 56 patients received cryoprecipitate transfusions. A total of 271 patients (26.3%) underwent therapeutic interventions, among whom 238 underwent endoscopic treatment. Two patients experienced rebleeding after endoscopic treatment and received radiological interventional therapy, and 3 patients experienced rebleeding after endoscopic treatment and underwent emergency surgical operations. Fifteen patients received radiological interventional therapy, among whom 2 patients underwent endoscopic treatment failure, and 1 patient was transferred to an emergency surgical operation after the failure of interventional therapy. Twenty-four patients underwent emergency surgery.\u003c/p\u003e\n\u003ch1\u003e2.6.2 ICU\u0026nbsp;treatment\u003c/h1\u003e\n\u003cp\u003eIn the constructed cohort, 97 patients (9.4%) were transferred to the ICU for continued treatment due to concurrent heart failure and multiple organ dysfunction, 45 of whom (46.4%) died.\u003c/p\u003e\n\u003ch1\u003e2.6.3\u0026nbsp;Rebleeding\u003c/h1\u003e\n\u003cp\u003eIn the construction cohort, 225 patients (21.8%) experienced rebleeding. Among them, 24 patients did not undergo endoscopy, 120 patients experienced variceal bleeding, and 81 patients experienced nonvariceal bleeding. The rebleeding rate in the variceal group was greater than that in the nonvariceal group. Among the 225 patients who experienced rebleeding, 34 died.\u003c/p\u003e\n\u003ch1\u003e2.6.4 Death\u003c/h1\u003e\n\u003cp\u003eThere were 74 cases (7.2%) of patient death in the construction cohort. Among them, 30 patients died of multiple organ failure, 19 patients died of acute gastrointestinal bleeding, 14 patients died of heart failure, and 11 patients died of combined pulmonary infection and septic shock.\u003c/p\u003e\n\u003ch1\u003e2.7 Data\u0026nbsp;analysis\u003c/h1\u003e\n\u003cp\u003eIn this section, on the basis of the algorithms of KNN (K-nearest neighbor), GBoost (gradient-boosted decision trees), MLP (multilayer perceptron), SVM (support vector machine), KSVM (kernel support vector machines), DT (decision tree), and RF (random forest), models are constructed for variable factors to accurately predict whether older UGIB patients can be safely discharged from the hospital. These models are compared with existing scoring criteria to verify their effectiveness. Table 5 shows the AUROC of the models constructed by seven machine-learning algorithms for the prediction of safe discharge results in the validation group. Among them, the RF model has the largest AUC value, whereas the DT model has the smallest AUC value. In terms of the accuracy of predicting safe discharge, the SVM, KSVM, and RF algorithm models all have good accuracy; however, the prediction effects of KNN, GBoost, and MLP are average, and the prediction effect of DT is poor.\u003c/p\u003e\n\u003cp\u003eTable 5 ROC curves of 7 machine - learning algorithms for predicting safe discharge in the validation group\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"578\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eAlgorithm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ePPV\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.744(0.683-0.805)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e82.4(75.0-89.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e56.9(47.9-65.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e62.7(4.5-70.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e78.6(69.8-87.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.763(0.716-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e80.4(73.0-87.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e58.5(49.1-67.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e67.2(59.2-75.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e73.8(64.4-83.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eMLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.779(0.710-0.848)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e83.7(76.5-90.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e58.8(49.7-67.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e64.9(56.9-73.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e79.8(71.2-88.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.853(0.788-0.918)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e87.6(81.1-94.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e59.5(50.8-68.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e55.3(72.3-71.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e85.7(78.2-93.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eK-SVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.826(0.785-0.867)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e86.0(79.2-92.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e56.3(47.5-68.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e64.2(56.1-72.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e83.3(75.4-91.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.569(0.510-0.628)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e68.7(59.6-77.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e44.5(35.6-53.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e50.8(44.3-59.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e63.1(52.8-73.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.889(0.858-0.920)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e86.7(80.6-92.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e78.6(70.5-86.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e83.2(76.5-89.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e82.8(75.1-90.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 6 shows the comparison results of the AUROC for predicting safe discharge in the validation group between the model constructed by the RF algorithm and the models constructed by the other six machine learning algorithms. There were significant differences between the model constructed via RF and the models constructed via algorithms such as KNN, GBoost, MLP, DT, and K-SVM (\u003cem\u003ep\u003c/em\u003e\u0026lt;0.\u003cem\u003e01, p\u003c/em\u003e\u0026lt;0.05), whereas the difference between the model constructed via RF and the model constructed via the SVM algorithm was not statistically significant (\u003cem\u003ep\u003c/em\u003e\u0026gt;0.05).\u003c/p\u003e\n\u003cp\u003eTable 6 Comparison of ROC curves for predicting safe discharge in the validation group between the RF algorithm and the other six machine algorithms\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAlgorithm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRF (\u003cem\u003eP\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMLP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch3\u003e2.8 Comparison of the advantages and disadvantages between the constructed scoring system and other existing scoring systems\u003c/h3\u003e\n\u003cp\u003eTable 7 shows the comparison between four existing international scoring systems for predicting safe discharge in the validation group and the scoring system constructed in this study. Among the four existing scoring systems internationally, the ABC has the largest AUC value. There was no statistically significant difference between the ABC and MAP methods and the GBS (\u003cem\u003ep\u003c/em\u003e\u0026gt;0.05),\u0026nbsp;whereas\u0026nbsp;there\u0026nbsp;was\u0026nbsp;a statistically significant difference between\u0026nbsp;the\u0026nbsp;ABC and ABC\u0026nbsp;methods\u0026nbsp;(\u003cem\u003ep\u003c/em\u003e\u0026lt;0.05)\u0026nbsp;and\u0026nbsp;the\u0026nbsp;AIMS65\u0026nbsp;(\u003cem\u003ep\u003c/em\u003e\u0026lt;0.01). Among the four scoring systems, there\u0026nbsp;were\u0026nbsp;statistically significant differences between\u0026nbsp;the\u0026nbsp;AIMS65 and the other three systems in each group. However, there\u0026nbsp;was\u0026nbsp;no statistically significant difference between\u0026nbsp;the\u0026nbsp;MAP, ABC\u0026nbsp;or\u0026nbsp;GBS in\u0026nbsp;terms of\u0026nbsp;safe discharge\u0026nbsp;among the groups\u0026nbsp;(\u003cem\u003ep\u003c/em\u003e\u0026gt;0.05).\u003c/p\u003e\n\u003cp\u003eTable 7 Comparison of the abilities of different scoring systems to predict safe discharge\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"580\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ePPV\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003cp\u003e(%，95%\u003cem\u003eCI\u003c/em\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eMAP(ASH)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.807(0.744-0.869)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e83.4(76.3-89.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e67.9(56.8-77.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e80.6(75.1-85.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e72.2(63.2-79.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eCANUKA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.849(0.796-0.903)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e80.6(72.9-86.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e76.2(65.7-84.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e84.4(78.5-88.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e71.1(63.1-78.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eABC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.763(0.698-0.828)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e75.4(67.2-82.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e66.7(55.5-76.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e78.3(72.4-83.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e62.9(54.9-70.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eAIMS65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.650(0.577-0.722)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e49.3(40.5-58.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e82.1(72.3-89.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e81.5(72.9-87.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e50.4(45.5-55.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eGBS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.821(0.761-0.880)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e66.4(57.8-74.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e86.9(77.8-93.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e89.0(82.2-93.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e61.9(55.8-67.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 113px;\"\u003e\n \u003cp\u003e0.889(0.858-0.920)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e86.7(80.6-92.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e78.6(70.5-86.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e83.2(76.5-89.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 98px;\"\u003e\n \u003cp\u003e82.8(75.1-90.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 8 shows the comparison results of the AUC values between the model constructed by the RF algorithm and four existing scoring systems in predicting safe discharge in the validation group. For the population in the validation group, there were significant differences between the model constructed by RF and the scoring systems such as ABC, AIMS65, MAP, and GBS (\u003cem\u003ep\u003c/em\u003e\u0026lt;0.01 for ABC and AIMS65; \u003cem\u003ep\u003c/em\u003e\u0026lt;0.05 for MAP and GBS). The model designed on the basis of the RF algorithm in this project has certain advantages.\u003c/p\u003e\n\u003cp\u003eTable 8 Comparison of the ROC curves of four scoring systems and the RF algorithm in predicting the safe discharge of the validation group\u003c/p\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003escoring systems\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRF（\u003cem\u003eP\u003c/em\u003e）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003eMAP（ASH）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003eCANUKA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u0026gt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003eABC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003eAIMS65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u0026lt;0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 125px;\"\u003e\n \u003cp\u003eGBS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u0026lt;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"3. Discussion","content":"\u003cp\u003eThis study is a multicenter investigation of older patients with upper gastrointestinal bleeding (UGIB) in China. Among the patients in our multicenter study, the number of males was 1.8 times greater than that of females. The median age was 70 years, and one-third of the patients (361/1032) were over 74 years old. Stratified by etiology, the median age of patients with variceal bleeding was 67 years, with males accounting for 53.7% (123/229) of the patients. The median age of nonvariceal patients was 71 years, with males accounting for 69.7% (461/661) of the patients. Reports from other countries worldwide also show that gastrointestinal bleeding is more common in males [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In this study, 91% (939/1032) of the patients had comorbidities, 57.8% (596/1032) of whom had two or more comorbidities. Previous studies have noted that the severity of comorbidities in UGIB patients is correlated with the prognosis of the disease[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. The Charlson Comorbidity Index (CCI) was proposed by Charlson et al. in 1987. Some studies have shown that a gradual increase in the CCI is correlated with the prognosis of the disease[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Zellmer S et al. reported that in patients with COVID-19 infection, an increased CCI was positively correlated with an increased risk of bleeding [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In this study, compared with patients with a CCI\u0026thinsp;\u0026le;\u0026thinsp;2 points, among the 313 patients (30.3%) with a CCI\u0026thinsp;\u0026gt;\u0026thinsp;2, only 65 patients (13.9%) were in the safe discharge group, whereas 248 patients (44.0%) were in the non\u0026ndash;safe discharge group, which also confirmed the correlation between the presence of underlying comorbidities and safe discharge.\u003c/p\u003e \u003cp\u003eIn terms of medication history, 23.4% (242/1032) of our patients were continuously taking antiplatelet or anticoagulant drugs at the onset of the disease. Among the 330 patients with peptic ulcers, 32.7% (108/330) were taking the abovementioned drugs, including 22.7% (75/330) taking antiplatelet drugs alone, 4.2% (14/330) taking dual antiplatelet drugs, and 3.9% (13/330) taking anticoagulant drugs, suggesting that the use of antiplatelet or anticoagulant drugs may be related to peptic ulcers. The widespread use of antiplatelet/anticoagulant drugs is related to the comorbidities of patients. For example, in a multicenter real-world study in China, approximately 13.9% of patients had comorbid CHD [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], whereas 21.4% of our patients had comorbid CHD/AF. Studies have confirmed that among patients receiving nonsteroidal anti-inflammatory drugs, the proportion of upper gastrointestinal bleeding (UGIB) ranges from 2.4\u0026ndash;12% [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. A systematic review of 11 randomized controlled trials revealed that aspirin increased the risk of gastrointestinal bleeding by 60%[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Dual antiplatelet therapy can further increase the risk of UGIB.\u003c/p\u003e \u003cp\u003eMultiple guidelines recommend performing emergency endoscopy (within 24 hours) on patients with upper gastrointestinal bleeding (UGIB), which is beneficial for identifying the cause and performing endoscopic hemostasis treatment if necessary [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, among our patients, only 86.2% (890/1032) completed the endoscopic examination. Among the patients who did not undergo gastroscopy, 16.2% (23/142) died, and the mortality rate was much higher than the overall rate. This finding is similar to reports in the UK and China [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. When multiple factors that lead to patients failing to complete the endoscopic examination are explored, the following key reasons may be involved. First, for certain specific groups, such as patients with dementia or stroke, since they may lack the cognitive ability and physical cooperation required for endoscopic examination, doctors often consider them unsuitable for such examinations. Second, some patients assessed as low risk were planned to undergo endoscopic examination after discharge, but owing to the difficulty in obtaining follow-up information after discharge, these data were not included in the statistical analysis. In addition, there are patients with hemodynamic instability or extremely critical conditions, and the risk of endoscopic examination is significantly increased, leading doctors to possibly postpone or avoid performing this examination. For patients with hemodynamic instability, although the risk of endoscopic examination is relatively high, considering that endoscopy is the preferred diagnostic and treatment method for UGIB, a study by McWhirter et al. revealed that emergency endoscopy is largely a safe procedure for older patients with acute UGIB [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. If effective endoscopic intervention can be implemented in a timely manner for this high-risk group, theoretically, their mortality rate may be reduced. This hypothesis needs to be further verified and supported by larger-scale real-world studies.\u003c/p\u003e \u003cp\u003eIn this study, the most common cause of UGIB (upper gastrointestinal bleeding) was digestive tract ulcer bleeding (32.0%), which was similar to the proportion reported in previous studies [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The second most common cause was variceal bleeding (22.2%), which was similar to the proportion reported in a domestic multicenter real-world study [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. The third most common cause in our study was acute erosive bleeding (13.5%), which was much lower than the proportion reported in European and American countries (approximately 25%). In our study, 23.1% (238/1032) of the patients underwent endoscopic treatment, among whom 60% had variceal bleeding. A total of 1.5% (15/1032) of the patients underwent radiological interventional treatment, and 2.3% (24/1032) of the patients underwent emergency surgical operations. In recent years, with the effective use of drugs and the rapid development of endoscopic and interventional technologies, the demand for emergency surgery among patients with digestive tract bleeding has significantly decreased [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn terms of rebleeding, the incidence rate among our patients was 21.8% (225/1032). The incidence rate was slightly higher among variceal patients than nonvariceal patients, and it was higher than that reported in Western countries (10.8%) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Among the 225 patients who experienced rebleeding, 34 patients died (15.1%), indicating that rebleeding is an important risk factor for death.\u003c/p\u003e \u003cp\u003eThe mortality rate among our patients was 7.2% (74/1032), which was similar to the rate reported abroad (7.9%) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. A prospective study on digestive tract bleeding in Hong Kong revealed that among all deaths, only 18.9% were considered to be related to bleeding. Among older patients, most die from causes unrelated to bleeding, such as multiple organ failure (23.9%), pulmonary diseases (23.5%), and terminal malignancies (33.7%) [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this study, we introduced a new prognostic evaluation indicator, safe discharge, which means having none of the following manifestations: 1. Rebleeding; 2. Requiring blood transfusion; 3. Require therapeutic interventions, including endoscopic treatment, interventional treatment, and surgical treatment; 4. Death. In this study, 45.4% (469/1032) of the patients were safely discharged, indicating that this group of patients could be treated in the outpatient department without occupying inpatient resources. A large-scale microcost study on UGIB in the UK revealed that the average cost per patient in the UK was \u0026pound;2458 [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. A decrease in the number of inpatients will greatly reduce hospitalization costs and medical expenditures.\u003c/p\u003e \u003cp\u003eSafe discharge from the hospital is an important determinant for evaluating whether patients with UGIB can be treated only on an outpatient basis. The subjects of this study were older patients with UGIB. In our study, machine learning algorithms revealed that even in the older population, age is still an influencing factor that cannot be ignored. This may be related to the fact that older patients have more comorbidities. This finding is consistent with the results of our previous study [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]: the older population is more likely to develop UGIB, and their prognosis is worse than that of the control group, which requires more attention.\u003c/p\u003e \u003cp\u003eMachine learning models have been widely used in the evaluation of clinical disease risk and prognosis and have achieved good feedback. The latest research shows that the calculation effect of machine learning algorithm models is similar to that of deep neural networks [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. This study is based on the KNN, GBoost, MLP, SVM, KSVM, DT, and RF algorithms in machine learning. Combined with simple and easily obtainable clinical parameters after dimensionality reduction, a prognosis evaluation of older UGIB patients is conducted to predict whether they can be safely discharged from the hospital. It also compares machine learning algorithm models with existing international prognosis prediction and evaluation criteria such as CANUKA, MAP (ASH), GBS, ABC, and AIMS65. On the basis of the test index elements, machine learning algorithms have more advantages than do existing international scoring systems. Among them, the algorithm models are significantly better than the existing scoring systems. Foreign studies have shown that the UGIB prognosis prediction model constructed by machine learning is significantly better than the traditional clinical risk scoring system [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. In reference [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], a model was constructed using machine learning algorithms to predict the mortality of acute nonvariceal UGIB, and the results were even more optimistic than those evaluated by the artificial neural network model. A systematic review revealed that machine learning algorithm models are more effective at predicting rebleeding, mortality, and interventional treatment, and the AUC of the algorithm models can reach 0.80\u0026thinsp;\u0026minus;\u0026thinsp;0.90[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDeng Fuxing et al. [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e] used a residual machine learning (ML) model to predict major bleeding in patients with upper gastrointestinal bleeding (UGIB) during their stay in the intensive care unit (ICU), and the area under the receiver operating characteristic curve (AUC) was as high as 0.96. In UGIB, machine learning models based on large international multicenter cohorts outperformed currently commonly used clinical risk scores (GBS, admission Rockall score, and AIMS65) in predicting composite outcomes (blood transfusion, hemostatic intervention, or death) in both internal and external validation References. This finding is consistent with the results of the model constructed via the random forest (RF) algorithm in this study to predict safe discharge. Moreover, the algorithm model had statistically significant advantages over the other scoring systems (MAP(ASH), GBS, ABC, and AIMS65). In this study, the algorithm model was also compared with the CANUKA score. Although the difference in the advantages of the algorithm model was not statistically significant, the algorithm model was superior to the CANUKA scoring system in both sensitivity and specificity for predicting safe discharge (0.867/0.786 \u003cem\u003evs\u003c/em\u003e. 0.806/0.762). In practice, risk stratification scores are used to guide clinical applications, and the threshold for maximizing sensitivity should be selected to minimize misdiagnosis. Therefore, the negative predictive value is a very important indicator. In this study, the negative predictive value of the model based on the RF algorithm was 0.828, whereas that of the CANUKA score was only 0.711. This machine learning model is superior to existing clinical risk assessment tools in identifying low-risk patients (who can be safely discharged) and can reduce the misdiagnosis rate. Therefore, constructing a model based on machine learning algorithms to conduct an early, simple, and effective assessment of UGIB patients is highly important for improving the prognosis of patients [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe risk stratification scoring system has been widely used in clinical practice. Multiple studies have shown that the use of the Glasgow-Blatchford score (GBS) seems to be able to safely identify patients with a low risk of death or those in need of intervention [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. However, some studies have also suggested that the GBS has relatively low performance in predicting mortality or rebleeding [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u0026sdot;. According to the guidelines, patients with a GBS score of \u0026le;\u0026thinsp;1 can be discharged from the emergency department with an outpatient treatment arrangement, as these patients rarely die or require hemostatic intervention [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Our previous study indicated [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] that when GBS\u0026thinsp;\u0026le;\u0026thinsp;1 was used as the cutoff value, the sensitivity for the composite outcome of blood transfusion, hemostatic intervention, or death was 97.37%, and the specificity was 11.92%. A meta-analysis reported similar results for GBS\u0026thinsp;\u0026le;\u0026thinsp;2 [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. That is, for the composite outcome of recurrent upper gastrointestinal bleeding (UGIB), intervention, or death, when both the machine learning model and the GBS identified patients reaching the composite endpoint with 100% sensitivity at the cutoff score, the specificity value of the machine learning model was 26%, whereas that of the GBS was 12% [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. When machine learning was used to construct an algorithm model to predict the composite endpoint of UGIB patients, the specificity increased from 12% (GBS) to 26%, indicating that the machine learning algorithm model can improve the identification of low-risk patients who can be safely discharged from the emergency department for outpatient management. Our research revealed that the clinical risk scoring machine learning model trained on the dimensionality-reduced detection index variables has excellent performance and is superior to the existing clinical risk scoring systems in identifying low-risk UGIB patients.\u003c/p\u003e \u003cp\u003ePrevious studies on machine learning algorithm models for UGIB have been limited by sample size, homogeneous patient cohorts, and a lack of external validation. For example, the largest known machine learning algorithm study on UGIB to date [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e] used relevant data from 2,380 patients. The internal validation revealed that the neural network model improved the 30-day mortality rate according to the Rockall score, including the endoscopic examination results, but the external validation of this model was not conducted. This study evaluated outcome indicators related to prognosis. With safe discharge as the composite endpoint, this study aimed to screen patients at extremely low risk and manage them as outpatients. Unlike most previous studies, this study included a relatively large number of patients from multiple centers in China. We applied different machine learning algorithms to construct a prognostic evaluation model to assess its performance in modeling the same dataset and finally selected the optimal model. We also compared this model with existing international prognostic scoring systems to evaluate the prognostic prediction efficacy of the model. Finally, this study included both internal and external validation, which made the evaluation of the performance of the machine learning model more rigorous.\u003c/p\u003e \u003cp\u003eThis study has certain limitations. Owing to the limited time during the doctoral programme, the number of centers and the sample size included in this study were not large. We plan to conduct a multicenter prospective study in the future, including more samples, and further validate this model.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e: Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e: YL and JW contributed equally to this work.\u0026nbsp;YL, JW and XO conceived of the study. YL, JW, TW and QZ contributed to data collection. XO was responsible for data analyses. All authors contributed to interpretation of the results. YL and JW drafted the manuscript. All authors contributed to the refinements of the manuscript and approved the final manuscript for publication. XO is the guarantor of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e：No Funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003eNo.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e：This manuscript reports a study involving human participants. This study was conducted in accordance with the Declaration of Helsinki. The ethical approval for the study was obtained from the Ethics Committee for Clinical Research of Zhongda Hospital, affiliated to Southeast University with the approval number 2021ZDSYLL333-P01 to which none of the authors was affiliated. This study used the medical records obtained from the past clinical diagnosis and treatment. Upon application, the Ethics committee agreed that informed consent was not required.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e：Not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e：The authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWuerth BA, Rockey DC. Changing epidemiology of upper gastrointestinal hemorrhage in the last decade: A nationwide analysis[J]. Dig Dis Sci 2018, 63:1286\u0026ndash;93.\u003c/li\u003e\n\u003cli\u003eButton LA, Roberts SE, Evans PA, et al. Hospitalized incidence and case fatality for upper gastrointestinal bleeding from 1999 to 2007: a record linkage study[J]. Aliment Pharmacol Ther 2011, 33:64\u0026ndash;76. \u003c/li\u003e\n\u003cli\u003eCryer BL, Wilcox CM, Henk HJ, et al. The Economics of upper gastrointestinal bleeding in a US managed- care setting: a retrospective, claims- based analysis[J]. J Med Econ 2010, 13:70\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eLaine L, Yang H, Chang S- C, et al. Trends for incidence of hospitalization and death due to GI complications in the United States from 2001 to 2009[J]. Am J Gastroenterol 2012, 107:1190\u0026ndash;5; \u003c/li\u003e\n\u003cli\u003eLaursen SB, Dalton HR, Murray IA, et al. Performance of new thresholds of the Glasgow Blatchford score in managing patients with upper gastrointestinal bleeding[J]. Clin Gastroenterol Hepatol 2015, 13:115\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eSung JJ, Chiu PW, Chan FKL, et al. Asia- Paciffc working group consensus on non Variceal upper gastrointestinal bleeding: an update 2018[J]. Gut 2018, 67:1757\u0026ndash;68.\u003c/li\u003e\n\u003cli\u003eWang C- Y, Qin J, Wang J, et al. Rockall score in predicting outcomes of elderly patients with acute upper gastrointestinal bleeding[J]. World J Gastroenterol 2013, 19:3466\u0026ndash;72. \u003c/li\u003e\n\u003cli\u003eKalkan \u0026Ccedil;, Soykan I, Karakaya F, et al. Comparison of three scoring systems for risk stratiffcation in elderly patients Wıth acute upper gastrointestinal bleeding[J]. Geriatr Gerontol Int 2017, 17:575\u0026ndash;83. \u003c/li\u003e\n\u003cli\u003eEarly DS, Acosta RD, Chandrasekhara V, et al. Modiffcations in endoscopic practice for the elderly[J]. Gastrointestinal Endoscopy 2013, 78:1\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eChinese Society of Emergency Medicine, Chinese Society of Emergency Medicine of the Chinese Geriatrics Society, Beijing Society of Emergency Medicine, et al. Expert Consensus on Emergency Diagnosis and Treatment of Upper Gastrointestinal Bleeding in the Elderly [J]. Journal of Clinical Emergency Medicine, 2025; 12(44): 1013-1022.\u003c/li\u003e\n\u003cli\u003eBarkun AN, Almadi M, Kuipers EJ, et al. Management of Nonvariceal upper gastrointestinal bleeding: guideline recommendations from the International consensus group[J]. Ann Intern Med, 2019, 171(11):805-822.\u003c/li\u003e\n\u003cli\u003eStanley AJ, Laine L, Dalton HR, et al. International Gastrointestinal Bleeding Consortium. Comparison of risk scoring systems for patients presentingwith upper gastrointestinal bleeding: international multicenter prospective study[J]. BMJ, 2017,356:i6432. \u003c/li\u003e\n\u003cli\u003eStolow E, Moreau C, Sayana H, Patel S. Management of Non-Variceal Upper GI Bleeding in the Geriatric Population: An Update. Curr Gastroenterol Rep, 2021,23(4):5.\u003c/li\u003e\n\u003cli\u003eProsenz J, St\u0026auml;ttermayer MS, Riedl F, et al. Adherence to guidelines in patients with nonvariceal upper gastrointestinal bleeding (UGIB) - results from a retrospective single tertiary center registry[J]. Scand J Gastroenterol. 2023,58(8):856-862. \u003c/li\u003e\n\u003cli\u003eOakland K, Jairath V, Uberoi R, et al. Derivation and validation of a novel risk score for safe discharge after acute lower gastrointestinal bleeding: a Modeling study[J]. Lancet Gastroenterol Hepatol, 2017,2(9):635-643.\u003c/li\u003e\n\u003cli\u003eVillanueva C, Colomo A, Bosch A, et al. Transfusion strategies for acute upper gastrointestinal bleeding[J]. N Engl J Med, 2013, 368(1):11-21.\u003c/li\u003e\n\u003cli\u003eArmitage JN, van der Meulen JH, Royal College of Surgeons Comorbidity Consensus Group. Identifying Comorbidity in surgical patients using administrative data with the Royal college of Surgeons Charlson score[J]. BrJ Surg, 2010, 97(5):772-781.\u003c/li\u003e\n\u003cli\u003eAhmed A, Stanley AJ. Acute upper gastrointestinal bleeding in elderly individuals:etiology, diagnosis and treatment[J]. Drugs Aging, 2012, 29(12):933-940.\u003c/li\u003e\n\u003cli\u003eCharlson ME, Carrozzino D, Guidi J, et al. Charlson Comorbidity Index: A Critical Review of Clinimetric Properties[J]. Psychother Psychosom, 2022,91(1):8-35. \u003c/li\u003e\n\u003cli\u003eZellmer S, Hanses F, Muzalyova A, et al. Gastrointestinal bleeding and endoscopic findings in critically and noncritically ill patients with corona virus disease 2019(COVID-19): Results from Lean European Open Survey onSARS-CoV-2(LEOSS) and COKA registries[J]. United Eur Gastroenterol J,2021, 9(9):1081-1090.\u003c/li\u003e\n\u003cli\u003eCryer B, Li C, Simon LS, et al. GI-REASONS: A novel 6-month, prospective, randomized, open-label, blinded endpoint (PROBE) trial[J]. Am J Gastroenterol, 2013, 108(3):392\u0026ndash;400. \u003c/li\u003e\n\u003cli\u003eLanas A, Boers M, Nuevo J. Gastrointestinal events in at-risk patients starting nonsteroidal anti-inflammatory drugs (NSAIDs) for rheumatic diseases:the EVIDENCE study of European routine practice[J]. Ann Rheum Dis, 2015, 74(4):675-681.\u003c/li\u003e\n\u003cli\u003eElwood PC, Morgan G, Galante J, et al. Systematic review and meta-analysis of randomized trials to ascertain fatal gastrointestinal bleeding events attributable to preventive low-dose aspirin:No evidence of increased risk[J].PLoS One, 2016, 11(11):e0166166.\u003c/li\u003e\n\u003cli\u003eGu L,Xu F,Yuan J.Comparison of AIMS65,Glasgow-Blatchford and Rockall scoring approaches in predicting the risk of in-hospital death among emergency hospitalized patients with upper gastrointestinal bleeding: a retrospective observational study in Nanjing, China[J].BMC Gastroenterol, 2018,18(1): 98-106.\u003c/li\u003e\n\u003cli\u003eMcWhirter A, Mahmood S, Mensah E, et al. Evaluating the Safety and Outcomes of Esophagogastroduodenoscopy in Elderly Patients Presenting With Acute Upper Gastrointestinal Bleeding[J]. Cureus, 2023;15(10):e47116.\u003c/li\u003e\n\u003cli\u003eLoffroy R, Rao P, Ota S, et al. Embolization of acute nonvariceal upper gastrointestinal hemorrhage resistant to endoscopic treatment: results and predictors of recurrent bleeding[J]. Cardiovasc Intervent Radiol, 2010,33(6):1088-1100.\u003c/li\u003e\n\u003cli\u003eLaursen SB, Stanley AJ, Laine L, Schaffalitzky de Muckadell OB. Rebleeding in peptic ulcer bleeding - a nationwide cohort study of 19,537 patients[J]. Scand J Gastroenterol, 2022, 57(12):1423-1429.\u003c/li\u003e\n\u003cli\u003eChaudhary S, Mackay D, Pell JP, et al. Upper gastrointestinal bleeding in Scotland 2000-2015: trends in demographics, etiology and outcomes[J]. Aliment Pharmacol Ther, 2021, 53(3):383-389.\u003c/li\u003e\n\u003cli\u003eSung JJ, Tsoi KK, Ma TK, et al. Causes of mortality in patients with peptic ulcer bleeding: a prospective cohort study of 10,428 cases[J]. Am J Gastroenterol, 2010, 105(1):84-89.\u003c/li\u003e\n\u003cli\u003eCampbell HE, Stokes EA, Bargo D, et al. Costs and quality of life associated with acute upper gastrointestinal bleeding in the UK: cohort analysis of patients in a cluster randomized trial[J]. BMJ Open, 2015, 5(4):e007230. \u003c/li\u003e\n\u003cli\u003eLi Y, Lu Q, Song M, et al Novel risk score for acute upper gastrointestinal bleeding in elderly patients: a single-center retrospective study[J]. BMJOpen, 2023, 13(6):e072602. \u003c/li\u003e\n\u003cli\u003eSheridan RP, Wang WM, Liaw A, et al. Extreme gradient boosting as a method for quantitative structureactivity relationships[J]. Journal of chemical information and modeling, 2016, 56(12):2353-2360.\u003c/li\u003e\n\u003cli\u003eShung DL, Au B, Taylor RA, et al. Validation of a machine learning model that outperforms clinical risk scoring systems for upper gastrointestinalbleeding[J]. Gastroenterology, 2020,158(1):160-167.\u003c/li\u003e\n\u003cli\u003eNigam GB, Murphy MF, Travis SPL, et al. Machine learning in the assessment and management of acute gastrointestinal bleeding[J]. BMJ Med, 2024,19;3(1):e000699.\u003c/li\u003e\n\u003cli\u003eUngureanu BS, Gheonea DI, Florescu DN, et al. Predicting mortality in patients with nonvariceal upper gastrointestinal bleeding using machine- learning[J]. Frontiers in medicine, 2023,10:1134835.\u003c/li\u003e\n\u003cli\u003eShung D, Simonov M, Gentry M, et al. Machine learning to predict outcomes in patients with acute gastrointestinal bleeding: a systematic review[J]. Digestive diseases and sciences, 2019,64(8):2078-2087.\u003c/li\u003e\n\u003cli\u003eDeng F, Cao Y, Wang H, et al. Prognosis of major bleeding based on residual variables and machine learning for critical patients with upper gastrointestinal bleeding: A multicenter study[J]. J Crit Care, 2025,85:154923.\u003c/li\u003e\n\u003cli\u003ePai AK, Fox VL. Gastrointestinal bleeding and management[J]. Pediatric clinics of north america, 2017,64(3): 543-561.\u003c/li\u003e\n\u003cli\u003eRivieri S, Carron PN, Schoepfer A, et al. External validation and comparison of the Glasgow-Blatchford score, modified Glasgow-Blatchford score, Rockall score and AIMS65 score in patients with upper gastrointestinal bleeding: a cross-sectional observational study in Western Switzerland[J]. Eur J Emerg Med, 2023,30(1):32-39.\u003c/li\u003e\n\u003cli\u003ePognonec C, Dirhoussi Z, Cury N, et al. External validation of Glasgow-Blatchford, modified Glasgow-Blatchford and CANUKA scores to identify low-risk patients with upper gastrointestinal bleeding in emergency departments: a retrospective cohort study[J]. Emerg Med J, 2023,40(6):451-457.\u003c/li\u003e\n\u003cli\u003eKozai L, Tan A, Nebrejas K, et al. Comparative diagnostic utility of Rockall and Glasgow-Blatchford scores in nonvariceal upper gastrointestinal bleeding: a systematic review and meta-analysis[J]. Eur J Gastroenterol Hepatol, 2025,37(2):161-166.\u003c/li\u003e\n\u003cli\u003eBoustany A, Alali AA, Almadi M, et al. Pre-Endoscopic Scores Predicting Low-Risk Patients with Upper Gastrointestinal Bleeding: A Systematic Review and Meta-Analysis[J]. J Clin Med, 2023,12(16):5194.\u003c/li\u003e\n\u003cli\u003eRotondano G, Cipolletta L, Grossi E, et al. Artificial neural networks accurately predict mortality in patients with nonvariceal upper GI bleeding[J]. Gastrointestinal Endoscopy, 2011, 73(2):218\u0026ndash;226.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Upper gastrointestinal bleeding, Older patients, Safe discharge, Data reduction, Machine learning","lastPublishedDoi":"10.21203/rs.3.rs-6194668/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6194668/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eObjective\u003c/b\u003e\u003c/p\u003e \u003cp\u003eTo construct a clinical assessment prediction model for the safe discharge of older patients with upper gastrointestinal bleeding via machine learning algorithms.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e \u003cp\u003eA retrospective analysis was conducted on 1,032 older patients with upper gastrointestinal bleeding who were admitted through the emergency departments of Zhongda Hospital, Shanghai Changhai Hospital, Lishui Central Hospital, and Zhengzhou First Hospital from January 2018 to December 2020 to obtain relevant epidemiological, treatment, and prognostic data on Chinese older patients with upper gastrointestinal bleeding. This study aimed to clarify the comorbidities and medication history, onset conditions, causes, treatment interventions, and prognostic status of older patients with upper gastrointestinal bleeding, as well as the high-risk factors associated with death. A total of 218 older patients with upper gastrointestinal bleeding admitted to Zhongda Hospital from January 2021 to December 2022 who met the same criteria were subsequently selected as the validation group. A clinical assessment prediction model suitable for the older population was constructed via machine learning to predict safe discharge from upper gastrointestinal bleeding.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe results revealed that the nine diagnostic variables of HB, SBP, INR, BUN, Alb, CR, HR, age, and CCI were used for algorithm modeling. The importance of the elements was consistent with the ranking in the machine learning algorithm. Among the machine learning algorithms, the random forest algorithm was the best clinical prediction model. For the validation group, the AUC of the RF model for the prediction of safe discharge was 0.889. The model prediction accuracy was 0.830 (0.781\u0026ndash;0.880), the sensitivity was 0.868 (0.806\u0026ndash;0.928), the specificity was 0.786 (0.705\u0026ndash;0.867), the positive predictive value was 0.832 (0.765\u0026ndash;0.898), and the negative predictive value was 0.828 (0.751\u0026ndash;0.905), all of which were better than those of traditional scoring systems (CANUKA, AIMS65, MAP, ABC, and GBS).\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e\u003c/p\u003e \u003cp\u003eMachine learning algorithms can form more accurate prediction models than traditional scoring systems.\u003c/p\u003e","manuscriptTitle":"Clinical assessment and prediction model construction for older patients with upper gastrointestinal bleeding","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-07 17:20:40","doi":"10.21203/rs.3.rs-6194668/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-05-05T03:40:01+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-29T06:23:15+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-10T09:07:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-09T12:12:25+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Geriatrics","date":"2025-04-09T12:07:17+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-geriatrics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bgtc","sideBox":"Learn more about [BMC Geriatrics](http://bmcgeriatr.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bgtc/default.aspx","title":"BMC Geriatrics","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"af860267-3394-4d4d-87fc-2c1e23dcccc2","owner":[],"postedDate":"May 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-05-07T17:20:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-07 17:20:40","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6194668","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6194668","identity":"rs-6194668","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}