Skip to content
Publicly Available Published by De Gruyter July 19, 2022

Identifying predictors of varices grading in patients with cirrhosis using ensemble learning

  • Azadeh Bayani , Azamossadat Hosseini EMAIL logo , Farkhondeh Asadi , Behzad Hatami , Kaveh Kavousi , Mehrdad Aria and Mohammad Reza Zali

Abstract

Objectives

The present study was conducted to improve the performance of predictive methods by introducing the most important factors which have the highest effects on the prediction of esophageal varices (EV) grades among patients with cirrhosis.

Methods

In the present study, the ensemble learning methods, including Catboost and XGB classifier, were used to choose the most potent predictors of EV grades solely based on routine laboratory and clinical data, a dataset of 490 patients with cirrhosis gathered. To increase the validity of the results, a five-fold cross-validation method was applied. The model was conducted using python language, Anaconda open-source platform. TRIPOD checklist for prediction model development was completed.

Results

The Catboost model predicted all the targets correctly with 100% precision. However, the XGB classifier had the best performance for predicting grades 0 and 1, and totally the accuracy was 91.02%. The most significant variables, according to the best performing model, which was CatBoost, were child score, white blood cell (WBC), vitalism K (K), and international normalized ratio (INR).

Conclusions

Using machine learning models, especially ensemble learning models, can remarkably increase the prediction performance. The models allow practitioners to predict EV risk at any clinical visit and decrease unneeded esophagogastroduodenoscopy (EGD) and consequently reduce morbidity, mortality, and cost of the long-term follow-ups for patients with cirrhosis.

Introduction

Among all the mortality reasons, about 3.5% happen due to chronic liver diseases each year [1]. One of the most harmful consequences of chronic liver diseases is liver cirrhosis. It may lead to liver cell necrosis and degeneration [2]. Esophageal varices (EV) are one of the severe consequences of cirrhosis. EV may occur between 60% and 80% in patients with cirrhosis. According to the high mortality rate of EV among patients with cirrhosis, diagnosing EV, especially its grades from moderate-severe EV, namely high bleeding risk EV, needs to be considered [3]. Guidelines recommend frequent screening with esophagogastroduodenoscopy (EGD) [4, 5]. The EGD has been introduced as the “gold standard” for evaluating EV grades, recommending that all patients undergo endoscopy once cirrhosis is confirmed in patients with chronic liver diseases and EV surveillance every 1–3 years [6], [7], [8]. However, many patients undergoing screening do not have EV or have only small EV without high-risk features. In addition, EGD may cause the risks of sedation and procedure-related complications [4] such as bleeding, cardiac tear, and discomfort [3].

One of the recommended methods to rapidly and non-invasively predict EV can be using the laboratory data in the liver tests [9]. Several models have been introduced to identify EV non-invasively [4, 10]. One validated score is the Baveno VI criteria, which involves transient hepatic elastography [4, 11, 12]. However, these methods have not been widely adopted in clinical practice. Thus, a more practical and accurate approach to reliably identify EV and its relevant predictors [1] to reduce the burden of unnecessary EGDs and thus better identify those who stand to benefit from EV screening would be clinically helpful [4]. Machine learning (ML) is a field of computer science that benefits various algorithms to recognize patterns in extensive data [13] and can lead to the optimization of hospital resources by predicting patients with severe risk factors [14]. Benefiting ML solutions enable significantly more comprehensive, accurate, and fast models [13]; also, high-performance predictions can reduce mortality of cirrhosis patients caused by EV [1]. There are several studies [3, 4, 14], [15], [16], [17] that used statistical or ML methods to identify predictive factors in disease related to chronic liver diseases and especially the risks for variceal bleeding or re-bleeding. The final accuracy obtained from these studies ranged from 60 to 90%. However, to the best of our knowledge, no study considered grades of EV and its predictors to improve its prediction in patients with cirrhosis.

Therefore, in the present study, we tried to improve the performance of predictive methods by introducing the most important factors which have the highest effects on the prediction of EV grades among patients with cirrhosis. In the present study, the ensemble learning methods were used to choose the most powerful predictors of EV grades solely based on routine laboratory and clinical data to help clinicians and thereby help avoid the risks and costs of unnecessary EGD and improve the further predictions.

Materials and methods

Dataset preparation

To detect prognostic factors for the prediction and grading of EV using predictive ML models, the method was divided into the following phases: 1. Data preparation: which included data cleaning, resolving missing data, and data transformation 2. Feature selection: detecting the best subset of the most relevant features for considering as the outputs of our model; and 3, building the models: using the most predictive features to build the models. To ensure obtaining the best result considering all data, 5-fold cross-validation was applied. With this method, the training and test sets cross over in successive rounds so that each row of data has a chance of being tested against. Figure 1 illustrates the process.

Figure 1: 
Method approach pathway.
Figure 1:

Method approach pathway.

The present study analyzed a dataset of 490 patients with cirrhosis gathered by the Liver and Gastroenterology Research Center of Shahid Beheshti University of Medical Sciences between 2014 and 2021. Two hepatologists labeled the data, and the disagreements were resolved by consulting with a third expert. The dataset consisted of patients with cirrhosis, 254 patients had no EV, and 236 patients had a history of EV, among whom 72 patients had grade 1, 55 patients had grade 2, and 109 patients had grade 3 with degrees of the fundus. The features registered from the patients are represented in Table 1. In the pre-processing phase, the missing values in numerical features of the records that had less than 20% missing in their features were replaced with the mean of that column, and there were two records that had missing values in their categorical features so they were omitted horizontally. Also, since the numerical features were in different scales to avoid the significant ones impacting the model just because of their large magnitude and to bring all features to the same condition, the scales of all numerical features were standardized using a standard scaler library to avoid bias in prediction [18]. This study was performed using Python programming language in Anaconda open-source environment. TRIPOD checklist for prediction model development was completed (Transparent Reporting of Multivariate Prediction Model for Individual Prognosis or Diagnosis Statement (See Supplementary File 1).

Table 1:

The input variables of the study.

No. Variable Mean (SD) Description
1 Age 63.5 (15.76)
2 Sex Categorical
1 – Male
0 – Female
3 Etiology Categorical 1 – Hepatitis B virus
2 – Hepatitis C virus
3 – Alcohol consumption
4 – A form of fatty liver
5 – Same as the medical term
6 – Primary biliary cholangitis
7 – Primary sclerosing cholangitis
8 – Same as the medical term
9 – Cryptogenic liver disease
10 – Wilson’s disease
1 – HBV
2 – HCV
3 – Alcohol
4 – NASH
5 – Autoimmune
6 – PBC
7 – PSC
8 – Metastasis
9 – Cryptogenic
10 – Wilson
11 – Autoimmune+PBC
12 – HBV+NASH
13 – Autoimmune+PSC
14 – Myelofibrosis
15 – Unknown
4 Ascites Categorical The degrees of ascites among patients
0 – No
1 – MILD-Slight
2 – Moderate
3 – Severe
5 SBP Categorical Spontaneous bacterial peritonitis or ascites fluid infection
1 – Yes
0 – No
6 Encephalopathy Categorical Same as the medical term
0 – No
1 – Grade 1–2
2 – Grade 3–4
7 Bleeding Categorical Varices bleeding
1 – Yes
0 – No
8 Band. Ligation Categorical Same as the medical term
1 – Yes
0 – No
9 Child_Pugh score Categorical Rates which are used to the severity of long-term liver diseases
5–14
10 child.class Categorical The classes for the previous item
A
B
C
11 MELD.score Categorical A score used as the model for end-stage liver disease
6–40
12 WBC 435.94 (16.54) White blood cell
13 Hb 2.57 (0.27) Hemoglobin
14 MCV 7.89 (22.7) Mean corpuscular volume
15 MCH 3.36 (6.37) Sign of macrocytic anemia
16 PLT 11.7 (42.8) Platelets in liver disease
17 Bili.T 1.96 (1.23) Conjugated bilirubin
18 Bili.D 1.86 (1.23) Direct bilirubin
19 AST 7.53 (32.9) Aspartate aminotransferase
20 ALP 23.95 (101.82) Alkaline phosphatase
21 PT 2.84 (7) Prothrombin time
22 INR 1.84 (0.67) International normalized ratio
23 ALB 2.17 (6.45) Albumin; protein made by the liver
24 K 2.05 (0.97) Vitamin K
25 Cr 1.08 (0.72) Serum creatinine
26 Na 136.57 (30.32) Serum sodium level

Base model

Ensemble learning is a method that combines the predictions of several learning methods into a single “ensemble” model to obtain better performance. Common ensemble learning methods include bagging, boosting, and stacking, amongst others [19]. In bagging, the model considers different random samples of a dataset, trains the models based on each sample, and then takes the averages or votes on the results. The models usually use decision trees to decrease the overall variance. Boosting, in contrast, starts by creating a single model on the dataset and produces a series of new sub-models. It iteratively trains each model, continuing until a stopping condition is met [9].

CatBoost

This approach is a new ensemble algorithm based on the gradient boosting framework preferred over other classification methods when categorical features are present in data [20]. CatBoost algorithm benefits the bagging approach; for building the trees, it bootstraps feature selection and randomness [21]. CatBoost’s base model uses an asymmetric tree and uses a serial iteration of a group of classifiers, and when a robust classifier is obtained, the iterations stop. CatBoost also uses combined features, which benefit from the connections between features, so remarkably empowers the feature dimension [21].

The estimated output can be described as follows:

Z = H ( x i ) = j = 1 J c j 1 { x ϵ R j }

Where H(x i ) is a decision tree function of the explanatory variables x i , and R j is the disjoint region corresponding to the leaves of the tree [22].

XGBoost

XGB or Extreme Gradient Boosting classifier is one of the popular ML algorithms that perform based on gradient boosting that trains decision trees in sequence such that later ones correct residuals of previous trees and the previous models are corrected and generate a final optimal model [23]. The purpose is to search for the best parameters for a model using an objective function, Obj(θ), that measures the performance using a particular set of parameters. An objective function must always contain two parts: training loss (L) and regularization (Ω) [24]:

Obj ( θ ) = L ( θ ) + Ω ( θ )

The L(θ) measures how predictive the model is on training data. The Ω(θ) controls the complexity of the model, which prevents overfitting [24].

Evaluation of the models

The models were applied to the data with a 5-fold cross-validation technique. The data labels defined the different grades of EV from grade 0 (without the presence of EV) to grade 3 (severe EV with the presence of fundus). The feature importance is automatically calculated in the predictive models by averaging all the decision trees in the model [25], and by using the important_feature function the most important features were extracted. Then for each algorithm, the 15 top obtained features were considered for the model. For evaluating the performance of the models, confusion matrix and performance indexes, including precision, recall, and accuracy, were used. These indexes are defined as follows:

true negative (TN); true positive (TP); false negative (FN), and false positive (FP)

(1) Accuracy = TN + TP TN + TP + FN + FP
(2) Precision = TP Tp + FP
(3) Recall  = TP TP + FN

The hyperparameters for Catboost method were as follows (iterations=75, learning_rate=0.03, depth=2,loss_function=‘RMSE’, l2_leaf_reg=0.2), and for XGB the following hyperparameters were applied (objective =‘reg:linear’, colsample_bytree=0.3, learning_rate=0.3, max_depth=5, alpha=10, n_estimators=10).

Results

In the present study, CatBoost and XGB Classifiers have been used based on a total of 26 laboratory and clinical features. The important features of the CatBoost and XGB Classifier have been demonstrated in Figures 2 and 3, respectively. To create test and train datasets, 5-fold cross-validation was applied. To choose the most important features, it is better to compare the performance of each model and their prediction power. In Table 2, the comparison of the average performances for both models has been demonstrated, including the average precision, recall, accuracy, and mean squared error.

Figure 2: 
The most important features using the CatBoost classification model for predicting EV grades.
WBC, white blood cell; K, vitamin K; INR, international normalized ratio; MCV, mean corpuscular volume; PT, prothrombin time; Cr, serum creatinine; Bili.T, conjugated bilirubin; PLT, platelets in liver disease; MCH, sign of macrocytic anemia; Na, serum sodium level; Hb, hemoglobin; ALT, alanine transaminase; ALP, alkaline phosphatase; AST, aspartate aminotransferase; Bili.D, direct bilirubin; SBP, spontaneous bacterial peritonitis; Alb, albumin.
Figure 2:

The most important features using the CatBoost classification model for predicting EV grades.

WBC, white blood cell; K, vitamin K; INR, international normalized ratio; MCV, mean corpuscular volume; PT, prothrombin time; Cr, serum creatinine; Bili.T, conjugated bilirubin; PLT, platelets in liver disease; MCH, sign of macrocytic anemia; Na, serum sodium level; Hb, hemoglobin; ALT, alanine transaminase; ALP, alkaline phosphatase; AST, aspartate aminotransferase; Bili.D, direct bilirubin; SBP, spontaneous bacterial peritonitis; Alb, albumin.

Figure 3: 
The most important features using the XGB classifier classification model for predicting EV grades.
Cr, serum creatinine; INR, international normalized ratio; Bili.T, conjugated bilirubin; MCV, mean corpuscular volume; AST, aspartate aminotransferase; PT, prothrombin time; WBC, white blood cell; PLT, platelets in liver disease; K, vitamin K; Na, serum sodium level; ALT, alanine transaminase; MCH, sign of macrocytic anemia; SBP, spontaneous bacterial peritonitis; ALP, alkaline phosphatase; Bili.D, direct bilirubin; Hb, hemoglobin; Alb, albumin.
Figure 3:

The most important features using the XGB classifier classification model for predicting EV grades.

Cr, serum creatinine; INR, international normalized ratio; Bili.T, conjugated bilirubin; MCV, mean corpuscular volume; AST, aspartate aminotransferase; PT, prothrombin time; WBC, white blood cell; PLT, platelets in liver disease; K, vitamin K; Na, serum sodium level; ALT, alanine transaminase; MCH, sign of macrocytic anemia; SBP, spontaneous bacterial peritonitis; ALP, alkaline phosphatase; Bili.D, direct bilirubin; Hb, hemoglobin; Alb, albumin.

Table 2:

Comparison of the performance for the two models.

Type of model Precision Recall Accuracy Mean squared error
CatBoost model results 1.0 1.0 1.0 0.0314
XGB classifier results 0.8472 0.8626 0.9202 0.0957

As shown in Figure 2, child_score, white blood cell (WBC) and vitamin K (K) are the top three features that have the most impact in classifying EV grades. International normalized ratio (INR), mean corpuscular volume (MCV), prothrombin time (PT), and age were the other essential features on the EV prediction.

As shown in Figure 3, Ascite_score, age, and Cr are the top three features that have the most impact in classifying EV grades. After them, MELD_score, INR and Bilirubin were the most important features for EV prediction.

According to Table 2, the CatBoost model had the best performance, with 100% accuracy in predicting EV grades. The XGB Classifier had 92% average accuracy and the average precision of 84%. Also, the amount of mean squared error for the Catboost algorithm was 0.03, which was very low comparing the XGB classifier.

Also, the confusion matrix for each model has been shown in Figures 4 and 5.

Figure 4: 
Confusion matrix for CatBoost model.
Figure 4:

Confusion matrix for CatBoost model.

Figure 5: 
Confusion matrix for XGB classifier.
Figure 5:

Confusion matrix for XGB classifier.

According to the confusion matrixes, the Catboost model predicts all the variables correctly with 100% precision. However, Figure 5 shows the XGB classifier had a better performance for grades 0 and 1, and totally the accuracy was 91.02%.

Discussion

A large number of patients who have cirrhosis undergo unnecessary and costly EV screening [4]. In the present study we developed an ML-based method to choose the most important features to predict EV grades that can contribute to finding a subset of variables that increase the performance of prediction algorithms by reducing the dimensionality of the problem [1]. In the present study 26 variables were considered as independent variables. The best feature selection method should reach a high precision after applying the variables in the prediction model [16]. In the present study, the Catboost model had the best performance with an accuracy equal to one; therefore, the most important variables according to the output of this algorithm can be selected as the predicting variables of EV grading. Several studies considered acute EV bleeding problems in chronic liver diseases. Most of them predicted EV bleeding using ML [1, 3, 26, 27], while others used statistical approaches to predict EV bleeding as a non-invasive method [4, 14, 17]. Using two ensemble learning algorithms of CatBoost and XGB Classifier, we demonstrated the important features in our study. For the CatBoost model, the Child-Pugh score, WBC, K, INR, and MVC were top-five significant features. Other studies [15, 28, 29] also indicated Child-Pugh score as a significant predictor, especially in bleeding EV, which is consistent with the finding of this study. Previous studies [4, 30] showed that the INR has a significant association with EV re-bleeding, obtained as a key predictor in the present study.

The second model chose ascites grades, age, Cr, MELD score, and INR as the top five important features. The Child_Pugh score is not one of the top important features associated with the EV grades prediction. According to refs. [1, 31], the MELD score has been proven as a critical factor in predicting cirrhosis mortality; however, ref. [4], did not find the MELD score as a significant factor in EV bleeding prediction and also did not find Cr significant (p-value=0.95). Also, a study [17] on mortality prediction of EV bleeding in cirrhotic patients showed that the Child-Pugh score is a more significant factor than MELD scores. The reason for these differences may be due to the fact that all the previously mentioned studies considered EV bleeding or re-bleeding as their target groups, but in the present study, our purpose was to predict EV grading for better management of patients with cirrhosis; therefore, the risk factors to predict these grades obtained differently.

To confirm the power of each obtained feature, the prediction models were performed. The two models showed a good performance for grade 0, the group of patients without EV. The Catboost model predicated all the grades with 100% accuracy. However, the XGB classifier had an average accuracy of 92% and had the best results for grades 0 and 1, and for grades 2 and 3, the correct predictions were 69 and 83.4%, respectively. Therefore, the features reported by the Catboost model may be selected as the most significant features for EV prediction studies.

Previous studies considered EV bleeding in hepatitis patients, which used ML models reached 0.82 [4], and 79% [32], 68.9% [16] for accuracy. A study [33] developed a model to predict the bleeding source for patients with acute gastrointestinal bleeding using ML models. They reported that the Random Forest model best predicted the target with accuracies of approximately 80%, and the area under the ROC curve for RF was more significant than 0.85. Another study [34] developed models to predict risk assessment in acute lower-gastrointestinal hemorrhage. They indicated that ANN performed well in predicting death (97%), recurrent bleeding (93%), and need for intervention (94%). Although the previous studies did not consider EV for prediction and mainly focused on EV bleeding, the present study can claim to have improved accuracy in predicting EV grades significantly since the findings had a high performance, which may be due to the strength of ensemble learning. Also, several studies considered re-bleeding prediction, and they reported that ANN performed significantly better in predicting the composite outcome (accuracy 0.76, 95% CI 0.70–0.83) compared with bleed (0.49, 95% CI 0.42–0.57) [35], the accuracy of LR model was 0.81 [36], and finally, another study reported ANN accuracy for re-bleeding prediction as 0.66% [37] which did not represent high accuracies.

The present study also had limitations to consider. The first was that the data was not gathered from multi centers, so there may be some limitations on the model representativeness and generalization. Another limitation was the low number of datasets that limited the number of samples in each class. By applying a cross-validation approach, this limitation was most reduced. Future studies considering more datasets should be considered to gain a more reliable and validated model. Of the strengths of this study, it can be mentioned that the most important predictive factors were introduced to improve the performance of prediction models in grading EV based on clinical and laboratory data. The conducted model may reduce routine EGD’s unnecessary costs and the consequences for patients with cirrhosis.

Conclusions

The significant predictors of EV prediction were introduced. Applying ML models, especially ensemble learning models, can remarkably increase the prediction performance. Using a model solely based on routine physical examination and laboratory tests allows practitioners to predict EV grades at any clinical visit and decrease unneeded EGD and consequently reduce morbidity, mortality, and cost of the long-term follow-ups for patients with cirrhosis.


Corresponding authors: Azamossadat Hosseini and Farkhondeh Asadi, PhD, Department of Health Information Technology and Management, School of Allied Medical Sciences, Shahid Beheshti University of Medical Sciences, Darband St, Ghods Square, Tehran, Iran, Phone: +98 9123187253 (A. Hosseini), E-mail: (A. Hosseini), (F. Asadi)

  1. Research funding: None declared.

  2. Author contributions: Study design: Farkhondeh Asadi and Azadeh Bayani. Database synthesis: Azadeh Bayani, Behzad Hatami, and Azamassadat Hosseini. Initial manuscript draft: Azadeh Bayani and Mehrdad Aria. Critical revision of the manuscript: Azadeh Bayani, Mohammad Reza Zali and Kaveh Kavousi. All authors have accepted responsibility for the entire content of this manuscript and approved its submission.

  3. Competing interests: Authors state no conflict of interest.

  4. Informed consent: An informed consent was obtained from all participants. The raw data did not contain any personal identifying information that can be linked to particular individuals, and was anonymized before its use.

  5. Ethical approval: The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Medical Ethics Committee of Shahid Beheshti University of Medical Sciences due to the retrospective nature of the study and that only anonymous data were used (ethics code: IR.SBMU.RETECH.REC.1400.588).

References

1. Aleksić, A, Nedeljković, S, Jovanović, M, Ranđelović, M, Vuković, M, Stojanović, V, et al.. Prediction of important factors for bleeding in liver cirrhosis disease using ensemble data mining approach. Mathematics 2020;8:1887.10.3390/math8111887Search in Google Scholar

2. Yeom, SK, Lee, CH, Cha, SH, Park, CM. Prediction of liver cirrhosis, using diagnostic imaging tools. World J Hepatol 2015;7:2069. https://doi.org/10.4254/wjh.v7.i17.2069.Search in Google Scholar PubMed PubMed Central

3. Yan, Y, Li, Y, Fan, C, Zhang, Y, Zhang, S, Wang, Z, et al.. A novel machine learning-based radiomic model for diagnosing high bleeding risk esophageal varices in cirrhotic patients. Hepatol Int 2021;16:423–32.10.1007/s12072-021-10292-6Search in Google Scholar PubMed

4. Dong, TS, Kalani, A, Aby, ES, Le, L, Luu, K, Hauer, M, et al.. Machine learning-based development and validation of a scoring system for screening high-risk esophageal varices. Clin Gastroenterol Hepatol 2019;17:1894–901. e1891. https://doi.org/10.1016/j.cgh.2019.01.025.Search in Google Scholar PubMed

5. Garcia‐Tsao, G, Abraldes, JG, Berzigotti, A, Bosch, J. Portal hypertensive bleeding in cirrhosis: risk stratification, diagnosis, and management: 2016 practice guidance by the American Association for the study of liver diseases. Hepatology 2017;65:310–35. https://doi.org/10.1002/hep.28906.Search in Google Scholar PubMed

6. Baltes, A, Akhtar, W, Birstler, J, Olson-Streed, H, Eagen, K, Seal, D, et al.. Predictors of skin and soft tissue infections among sample of rural residents who inject drugs. Harm Reduct J 2020;17:96. https://doi.org/10.1186/s12954-020-00447-3.Search in Google Scholar PubMed PubMed Central

7. Pateu, E, Oberti, F, Calès, P. The noninvasive diagnosis of esophageal varices and its application in clinical practice. Clin Res Hepatol Gastroenterol 2018;42:6–16. https://doi.org/10.1016/j.clinre.2017.07.006.Search in Google Scholar PubMed

8. Haq, I, Tripathi, D. Recent advances in the management of variceal bleeding. Gastroenterol Rep 2017;5:113–26. https://doi.org/10.1093/gastro/gox007.Search in Google Scholar PubMed PubMed Central

9. Chicco, D, Jurman, G. An ensemble learning approach for enhanced classification of patients with hepatitis and cirrhosis. IEEE Access 2021;9:24485–98. https://doi.org/10.1109/access.2021.3057196.Search in Google Scholar

10. Colli, A, Gana, JC, Yap, J, Adams‐Webber, T, Rashkovan, N, Ling, SC, et al.. Platelet count, spleen length, and platelet count-to-spleen length ratio for the diagnosis of oesophageal varices in people with chronic liver disease or portal vein thrombosis. Cochrane Database Syst Rev 2017;26:CD008759. https://doi.org/10.1002/14651858.CD008759.pub2.Search in Google Scholar PubMed PubMed Central

11. De Franchis, R. Expanding consensus in portal hypertension: report of the Baveno VI consensus workshop: stratifying risk and individualizing care for portal hypertension. J Hepatol 2015;63:743–52. https://doi.org/10.1016/j.jhep.2015.05.022.Search in Google Scholar PubMed

12. Sousa, M, Sousa Fernandes, S, Proença, L, Silva, AP, Leite, S, Silva, J, et al.. The Baveno VI criteria for predicting esophageal varices: validation in real life practice. Rev Esp Enferm Dig 2017;109:704–7. https://doi.org/10.17235/reed.2017.5052/2017.Search in Google Scholar PubMed

13. Jordan, MI, Mitchell, TM. Machine learning: trends, perspectives, and prospects. Science 2015;349:255–60. https://doi.org/10.1126/science.aaa8415.Search in Google Scholar PubMed

14. Wu, CC, Yeh, WC, Hsu, WD, Islam, MM, Nguyen, PAA, Poly, TN, et al.. Prediction of fatty liver disease using machine learning algorithms. Comput Methods Progr Biomed 2019;170:23–9. https://doi.org/10.1016/j.cmpb.2018.12.032.Search in Google Scholar PubMed

15. Şimşek, C, Tekin, E, Sahin, H, Sahin, TK, Balaban, YH. Artificial intelligence to predict esophageal varices in patients with cirrhosis. Acıbadem Üniversitesi Sağlık Bilimleri Dergisi 2021;12:625–9.10.31067/acusaglik.928498Search in Google Scholar

16. Abd El-Salam, SM, Ezz, MM, Hashem, S, Elakel, W, Salama, R, ElMakhzangy, H, et al.. Performance of machine learning approaches on prediction of esophageal varices for Egyptian chronic hepatitis C patients. Inform Med Unlocked 2019;17:100267. https://doi.org/10.1016/j.imu.2019.100267.Search in Google Scholar

17. Krige, J, Spence, RT, Jonas, E, Hoogerboord, M, Ellsmere, J. A new recalibrated four-category child–pugh score performs better than the original child–pugh and MELD scores in predicting in-hospital mortality in decompensated alcoholic cirrhotic patients with acute variceal bleeding: a real-world cohort analysis. World J Surg 2020;44:241–6. https://doi.org/10.1007/s00268-019-05211-8.Search in Google Scholar PubMed

18. Ghosh, P, Azam, S, Karim, A, Hassan, M, Roy, K, Jonkman, M. A comparative study of different machine learning tools in detecting diabetes. Procedia Comput Sci 2021;192:467–77. https://doi.org/10.1016/j.procs.2021.08.048.Search in Google Scholar

19. Latha, CBC, Jeeva, SC. Improving the accuracy of prediction of heart disease risk based on ensemble classification techniques. Inform Med Unlocked 2019;16:100203. https://doi.org/10.1016/j.imu.2019.100203.Search in Google Scholar

20. Hancock, JT, Khoshgoftaar, TM. CatBoost for big data: an interdisciplinary review. J Big Data 2020;7:94. https://doi.org/10.1186/s40537-020-00369-8.Search in Google Scholar PubMed PubMed Central

21. Luo, M, Wang, Y, Xie, Y, Zhou, L, Qiao, J, Qiu, S, et al.. Combination of feature selection and CatBoost for prediction: the first application to the estimation of aboveground biomass. Forests 2021;12:216. https://doi.org/10.3390/f12020216.Search in Google Scholar

22. Jabeur, SB, Gharib, C, Mefteh-Wali, S, Arfi, WB. CatBoost model and artificial intelligence techniques for corporate failure prediction. Technol Forecast Soc Change 2021;166:120658. https://doi.org/10.1016/j.techfore.2021.120658.Search in Google Scholar

23. Jo, YY, Han, J, Park, HW, Jung, H, Lee, JD, Jung, J, et al.. Prediction of prolonged length of hospital stay after cancer surgery using machine learning on electronic health records: retrospective cross-sectional study. JMIR Med Inform 2021;9:e23147. https://doi.org/10.2196/23147.Search in Google Scholar PubMed PubMed Central

24. Kropf, M, Hayn, D, Morris, D, Radhakrishnan, AK, Belyavskiy, E, Frydas, A, et al.. Cardiac anomaly detection based on time and frequency domain features using tree-based classifiers. Physiol Meas 2018;39:114001. https://doi.org/10.1088/1361-6579/aae13e.Search in Google Scholar PubMed

25. Queipo, NV, Nava, E. A gradient boosting approach with diversity promoting measures for the ensemble of surrogates in engineering. Struct Multidiscip Optim 2019;60:1289–311. https://doi.org/10.1007/s00158-019-02325-4.Search in Google Scholar

26. Agarwal, S, Sharma, S, Kumar, M, Venishetty, S, Bhardwaj, A, Kaushal, K, et al.. Development of a machine learning model to predict bleed in esophageal varices in compensated advanced chronic liver disease: a proof of concept. J Gastroenterol Hepatol 2021;36:2935–42. https://doi.org/10.1111/jgh.15560.Search in Google Scholar PubMed

27. Shung, DL, Au, B, Taylor, RA, Tay, JK, Laursen, SB, Stanley, AJ, et al.. Validation of a machine learning model that outperforms clinical risk scoring systems for upper gastrointestinal bleeding. Gastroenterology 2020;158:160–7. https://doi.org/10.1053/j.gastro.2019.09.009.Search in Google Scholar PubMed PubMed Central

28. Zoli, M, Merkel, C, Magalotti, D, Gueli, C, Grimaldi, M, Gatta, A, et al.. Natural history of cirrhotic patients with small esophageal varices: a prospective study. Am J Gastroenterol 2000;95:503–8. https://doi.org/10.1111/j.1572-0241.2000.01775.x.Search in Google Scholar PubMed

29. Hong, Wd, Zhu, Qh, Huang, Zm, Chen, Xr, Jiang, Zc, Xu, Sh, et al.. Predictors of esophageal varices in patients with HBV-related cirrhosis: a retrospective study. BMC Gastroenterol 2009;9:1–7. https://doi.org/10.1186/1471-230X-9-11.Search in Google Scholar PubMed PubMed Central

30. Trebicka, J, Gu, W, Ibáñez-Samaniego, L, Hernández-Gea, V, Pitarch, C, Garcia, E, et al.. Rebleeding and mortality risk are increased by ACLF but reduced by pre-emptive tips. J Hepatol 2020;73:1082–91. https://doi.org/10.1016/j.jhep.2020.04.024.Search in Google Scholar PubMed

31. Amitrano, L, Guardascione, MA, Bennato, R, Manguso, F, Balzano, A. MELD score and hepatocellular carcinoma identify patients at different risk of short-term mortality among cirrhotics bleeding from esophageal varices. J Hepatol 2005;42:820–5. https://doi.org/10.1016/j.jhep.2005.01.021.Search in Google Scholar PubMed

32. Abd-Elsalam, SM, Ezz, MM, Gamalel-Din, S, Esmat, G, Salama, A, ElHefnawi, M. Early diagnosis of esophageal varices using Boosted-Naïve Bayes Tree: a multicenter cross-sectional study on chronic hepatitis C patients. Inform Med Unlocked 2020;20:100421. https://doi.org/10.1016/j.imu.2020.100421.Search in Google Scholar

33. Chu, A, Ahn, H, Halwan, B, Kalmin, B, Artifon, EL, Barkun, A, et al.. A decision support system to facilitate management of patients with acute gastrointestinal bleeding. Artif Intell Med 2008;42:247–59. https://doi.org/10.1016/j.artmed.2007.10.003.Search in Google Scholar PubMed

34. Das, A, Ben-Menachem, T, Cooper, GS, Chak, A, Sivak, MVJr, Gonet, JA, et al.. Prediction of outcome in acute lower-gastrointestinal haemorrhage based on an artificial neural network: internal and external validation of a predictive model. Lancet 2003;362:1261–6. https://doi.org/10.1016/s0140-6736(03)14568-0.Search in Google Scholar

35. Choi, C, Swingland, J, Ali, A, Bose, S, Ayaru, L. PMO-204 Assessing risk of adverse outcome in acute lower gastrointestinal bleeding: artificial neural network vs sign guidelines and bleed score. Gut 2012;61:A156–7. https://doi.org/10.1136/gutjnl-2012-302514b.204.Search in Google Scholar

36. Augustin, S, Muntaner, L, Altamirano, JT, González, A, Saperas, E, Dot, J, et al.. Predicting early mortality after acute variceal hemorrhage based on classification and regression tree analysis. Clin Gastroenterol Hepatol 2009;7:1347–54. https://doi.org/10.1016/j.cgh.2009.08.011.Search in Google Scholar PubMed

37. Loftus, TJ, Brakenridge, SC, Croft, CA, Smith, RS, Efron, PA, Moore, FA, et al.. Neural network prediction of severe lower intestinal bleeding and the need for surgical intervention. J Surg Res 2017;212:42–7. https://doi.org/10.1016/j.jss.2016.12.032.Search in Google Scholar PubMed PubMed Central


Supplementary Material

The online version of this article offers supplementary material (https://doi.org/10.1515/cclm-2022-0508).


Received: 2022-05-24
Accepted: 2022-06-28
Published Online: 2022-07-19
Published in Print: 2022-11-25

© 2022 Walter de Gruyter GmbH, Berlin/Boston

Downloaded on 28.3.2024 from https://www.degruyter.com/document/doi/10.1515/cclm-2022-0508/html
Scroll to top button