Development and Validation of a Model to Predict Rebleeding Within Three Days After Endoscopic Hemostasis for Peptic Ulcer Bleeding

Background: Peptic ulcer bleeding remains a typical medical emergency with signicant morbidity and mortality. Peptic ulcer rebleeding often occurs within three days after emergency endoscopic hemostasis. Our study aims to develop a nomogram to predict rebleeding within three days after emergency endoscopic hemostasis for peptic ulcer bleeding Methods: We retrospectively reviewed the data of 386 patients with bleeding ulcers who underwent emergency endoscopic hemostasis between March 2014 and October 2018. The least absolute shrinkage and selection operator method were used to identied predictors. The model was displayed as a nomogram. Internal validation was carried out using bootstrapping. The model was evaluated using the calibration plot, decision-curve analyses and clinical impact curve. Results: Overall, 386 patients meeting the inclusion criteria were enrolled, with 48 patients developed rebleeding within three days after initial endoscopic hemostasis. Predictors contained in the nomogram included albumin, prothrombin time, shock, haematemesis/melena and Forrest classication. The model showed good discrimination and good calibration with a C-index of 0.854 (C-index: 0.830 via bootstrapping validation). Decision-curve analyses and clinical impact curve also demonstrated that it was clinically valuable. Conclusion: This study presents a nomogram that incorporates clinical, laboratory, and endoscopic features, effectively predicting rebleeding within three days after emergency endoscopic hemostasis and identifying high-risk rebleeding patients with peptic ulcer bleeding.

Although the prognosis of patients with PUB has improved with advances in proton pump inhibitors (PPIs) therapy and endoscopic treatment, it remains a potentially life-threatening gastrointestinal emergency. Moreover, many patients die every year due to peptic ulcer rebleeding, especially rebleeding within three days after endoscopic hemostasis [5][6][7] . Therefore, exploring e cient factors and developing tools for early identifying patients with high risk of rebleeding after emergency endoscopic hemostasis is an adequate precaution to improve the prognosis of PUB.
Previous studies have developed several scoring systems to estimate the prognosis of patients with upper gastrointestinal bleeding (UGIB), including the Glasgow-Blatchford score (GBS), Rockall score (RS), and the AIMS65 score [8][9][10] . However, the RS and the AIMS65 score were mainly developed to estimate the mortality risk of patients [11,12] . As for the GBS, by summarizing much research, the guideline suggested using GBS ≤ 1 to identify patients at low risk for rebleeding or mortality [4,5,13] . Besides, the predictive ability of GBS in patients with high-risk ulcers after endoscopic hemostasis is unsatisfactory [11,[13][14][15] . In addition, these scoring systems' complexity has limited their application in routine clinical situations.
Prior researches have been carried out to determine predictors of rebleeding in patients with PUB, and some have been incorporated into predictive models. Factors found to be predictors of rebleeding include Forrest classi cation, use of omeprazole, liver cirrhosis, recent surgery, systolic blood pressure below 100 mmHg, heart rate above 100 bpm, hematemesis, large ulcer size and ulcer site [16][17][18][19][20] . However, study predicting rebleeding within three days after emergency endoscopic hemostasis is rare. Besides, the predictive ability of models mentioned above is unsatisfactory. Thus, the study aimed to establish a novel prediction model displayed as a nomogram to predict rebleeding after emergency endoscopic hemostasis for PUB. The study outcome was rebleeding within three days of the initially successful therapeutic endoscopy. Rebleeding was de ned as recurrent hematemesis or melena with a decrease in hemoglobin by at least 2 g/dL within three days after the initial endoscopic treatment [19] . Shock was de ned as shock index (pulse rate/systolic blood pressure) >1.0. We classi ed patients with only hematemesis or both hematemesis and melena as hematemesis group, and classi ed patients with melena as melena group.

Endoscopic evaluation and pharmacologic therapy
Experienced endoscopists performed all emergency endoscopies within 12 hours of hospital admission. A single-channel endoscope (GIF-XQ290, Olympus Optical Co., Ltd., Japan) was used during the procedure. The endoscopic hemostasis methods included injection therapy, thermal coagulation, mechanical therapy and combined therapy. After successfully hemostasis, the patients will receive highdose intravenous proton pump inhibitors (80 mg of intravenous injection, then continuous infusion of 8 mg per hour for 72 hours.). Then, the patients will receive 40 mg esomeprazole once daily for 30 days.

Statistical analysis
For normally distributed data, continuous variables were presented as the mean ± standard deviation (SD) and the differences between the rebleeding and no-rebleeding groups were compared using Student's t-test. For non-normal distributed data, continuous variables were presented as the median and interquartile range (IQR) and the Mann-Whitney rank-sum test was used to analyze the difference between the two groups. Categorical variables are presented as proportions, and the chi-square test or Fisher's exact test was used accordingly.
The least absolute shrinkage and selection operator (LASSO) method, which is suitable for the regression of high-dimensional data [21] , was used to select the most useful predictive features from the primary data set. And we used the 1 standard error of the minimum criteria (the 1-SE criteria) value as cutoff. Albumin, prothrombin time (PT), shock, haematemesis/melena and Forrest classi cation were used to construct a nomogram.
Next, calibration curves and the concordance index (C-index) were calculated to evaluate the performance of the model in predicting prognosis. The values of C index of 0.5 and 1.0 respectively represent the random chance and good ability of the model to predict rebleeding. Besides, decision-curve analysis (DCA) and clinical impact curve were also used to determine the clinical net bene t associated with the use of the model [22] . Finally, the model was internally validated via bootstrapping resampling of the construction data set (with 1000 bootstrap samples per model) to obtain optimism corrected discrimination via the C-index for rebleeding [23] . What's more, we also compared the Area Under Curve (AUC) for the models vs. three clinical risk scores (GBS, RS, and the AIMS65). P < 0.05 were considered to be statistically signi cant. All the statistical analyses were performed by R statistical software 4.1.0 (www.r-project.org)

Results
Clinical characteristics A total of 386 patients with PUB who underwent emergency endoscopic hemostasis during the study period at our centre were enrolled ( Figure.1). Among these patients, 48 had rebleeding within three days after initial endoscopic hemostasis. The included patients' median age (IQR) was 56 (43-65) years old, and 313 (81.1%) of these patients were male. The enrolled patients' baseline characteristics are shown in Table 1-3. Compared to patients who did not rebleed, patients who rebled were more likely to present haematemesis and shock at the time of admission. What is more, patients in the rebleeding group seemed to have a faster heart rate, higher AIMS65 score, higher white cell count, lower platelet, lower albumin, prolonged PT, prolonged activated partial thromboplastin time (APPT) and international normalized ratio (INR) (all P 0.05). While no differences were observed between the rebleeding group and non-rebleeding group concerning age, sex, alcohol use, smoking, medication use, PU bleeding history, hypertension, diabetes mellitus, systolic blood pressure, diastolic blood pressure, GBS, Rockall score, haemoglobin level on admission, blood urea nitrogen, creatinine, ulcer location, ulcer size ≥ 2cm and methods of endoscopic hemostasis.

Feature selection based on LASSO method
Rebleeding within three days after emergency endoscopic hemostasis was chosen as the study outcome.
In order to reduce the dimensionality and screen out the most representative risk factors for PU rebleeding within three days after endoscopic hemostasis, LASSO regression analysis was performed on the 48 collected variables using the 1-SE criteria value as the cutoff. And as a result, ve variables that predict rebleeding within three days after the initially successful therapeutic endoscopy for PU screened out, including albumin, PT, shock, haematemesis/melena, Forrest classi cation ( Figure.2A and 2B).

Development and assessment of the nomogram
To predict rebleeding within three days after emergency endoscopic hemostasis for PUB, we conducted multivariable logistic regression analysis using the ve predictors selected by the LASSO method. And construct an accurate and stable nomogram ( Figure.3.). The equation built for model was LogitP = -0.148-0.21 * albumin +1.763 * PT +1.873 * shock-1.363 * haematemesis/melena -0.281 * Forrest classi cation (Table 4).  The DCA demonstrated that this model improved patient outcomes compared with either treat-all or treatnone strategies by helping assess the risk of rebleeding in patients and informing interventions ( Figure.5A). The DCA showed more bene t in the current study with a threshold probability > 0.0% using the nomogram. Besides, the clinical impact curve for the model was also visually indicated that nomogram conferred high clinical net bene t and con rmed the clinical value of this model ( Figure.5B).
And the model also performed better than GBS, RS, and AIMS65 ( Figure.6).

Internal validation
Finally, this model was internally validated using bootstrapping resampling of the construction data set (with 1000 bootstrap samples per model). Moreover, the C-index for this nomogram was 0.830, which suggested high accuracy.

Discussion
PUB is the most common cause of acute non-variceal upper gastrointestinal bleeding. Although the development of endoscopic technology has effectively improved the prognosis of PUB, rebleeding is still one of the common complications [4] . Therefore, it is imperative for clinicians to identify high-risk rebleeding patients after emergency endoscopic therapy, for which clinicians can give more powerful measures. According to our clinical observation, peptic ulcer rebleeding often occurs within three days after emergency endoscopic hemostasis. However, to our knowledge, there were few studies focused on rebleeding within three days. Anne C Travis et al. developed a model to predict rebleeding for NVUGIB. However, this model's study outcome was rebleeding within 30 days of the initially successful therapeutic endoscopy, and the predictive ability of this model was barely satisfactory (AUC = 0.752) [24] . Zhiyu Dong et al. established a new scoring system to predict poor clinical outcomes for NVUGIB, while this study had a small sample size and the predictive ability is still low [25] . There were also other clinical trials carried out to build models to predict rebleeding for PUB. However, these studies either did not focused on rebleeding within three days after emergency endoscopic hemostasis or had the poor predictive ability [18,[26][27][28] .
Hence, in the present study, we developed a novel model to predict rebleeding within three days after emergency endoscopic hemostasis for peptic ulcer bleeding, and the model showed an excellent discriminatory ability (C-index: 0.854).
In the present study, ve predictors identi ed by the LASSO method were incorporated into the nomogram, and the model was proven to be of excellent performance in internal validation. The Forrest classi cation is mainly used to stratify ulcer bleeding patients and guide management decisions, including endoscopic and pharmacological therapy. Moreover, many studies proved that the Forrest classi cation had excellent predictive value for rebleeding peptic ulcers [4,18,26] . Thus the Forrest classi cation is a stronger predictor for rebleeding in PUB. Hypoalbuminemia is a risk factor of mortality in certain diseases, and the correlation between hypoalbuminemia and the prognosis of PUB has been reported. Hsiu-Chi Cheng et al. indicated that hypoalbuminemia in patients with peptic ulcer bleeding could be an alarm indicator of recurrent bleeding [29] . In our study, albumin≤30 g/L scored more in the nomogram than albumin 30 g/L, which is similar to the previous study. Patients admitted to the hospital with hematemesis and shock heralded more dangerous gastrointestinal bleeding and heralded a worse prognosis [4] . Hence, hematemesis and shock scored more in the nomogram. Prolonged PT indicates deranged coagulation function and was another valuable predictor for rebleeding in patients presenting with PUB [30] . For patients with prolonged PT, the doctor should pay more attention to or take an extra intervention.
There were several advantages in the present study. First, this is the rst study to construct a model that incorporated variables from clinical, laboratory, endoscopic features for predicting rebleeding within three days after initial endoscopic therapy for PUB. Second, informative variables were identi ed using the LASSO method, which can avoid the statistical defects of over tting compared with using univariate analysis. Third, our model had an excellent predictive ability. What is more, the model was displayed as a nomogram which was intuitive and easy to use in clinical practice.
However, there were some limitations in the present study. First, the present study was a single-centre retrospective study. Second, our study only had internal validation. However, the model had a good performance in predictive ability (C-index: 0.854), and the internal veri cation performance was also good (C-index via bootstrapping validation: 0.830). This model needs to be prospectively validated on a distinct group of patients in the future.

Conclusion
In conclusion, we established and internally validated a nomogram to predict rebleeding within three days after emergency endoscopic hemostasis. This nomogram incorporated variables from clinical, laboratory and endoscopic features and can be conveniently used to identify high-risk patients after emergency endoscopic hemostasis, which can help doctors pay more attention to or give extra intervention. Availability of data and materials: The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Competing interests: All authors declare that they have no competing interests.
Funding: Not applicable.
Author contributions: YK L and YL X collected the data, analyzed relevant information, and wrote the manuscript; ZH Z, XL P, SH L, WD L, BiM L, Y Z and YX C clinically managed the patient. S X designed the Figures Figure 1 The owchart of patients included in the present study  Nomogram predicting the probability of rebleeding within three days following endoscopic therapy for peptic ulcer bleeding.
Page 17/18 Figure 4 Calibration curves of the nomogram in the primary cohort.  Comparison of ROC among the nomogram, Glasgow-Blatchford score system, Rockall score system and AIMS65 score system. ROC: receiver operating characteristic; GBS: Glasgow-Blatchford score.