Published online Mar 21, 2025. doi: 10.3748/wjg.v31.i11.101804
Revised: January 22, 2025
Accepted: February 20, 2025
Published online: March 21, 2025
Processing time: 167 Days and 15.7 Hours
A recent study by Peng et al developed a predictive model for first-instance secondary esophageal variceal bleeding in cirrhotic patients by integrating clinical and multi-organ radiomic features. The combined radiomic-clinical model demonstrated strong predictive capabilities, achieving an area under the curve of 0.951 in the training cohort and 0.930 in the validation cohort. The results highlight the potential of noninvasive prediction models in assessing esophageal variceal bleeding risk, aiding in timely clinical decision-making. Additionally, manual delineation of regions of interest raises the risk of observer bias despite efforts to minimize it. The study adjusted for clinical covariates, while some potential confounders, such as socioeconomic status, alcohol use, and liver function scores, were not included. Additionally, an imbalance in cohort sizes between the training and validation groups may reduce the statistical power of validation. Expanding the validation cohort and incorporating multi-center external validation would improve generalizability. Future studies should focus on incorporating long-term patient outcomes, exploring additional imaging modalities, and integrating automated segmentation techniques to refine the predictive model.
Core Tip: A study by Peng et al developed a combined multi-organ radiomics and clinical model to predict the risk of first-instance secondary esophageal variceal bleeding in cirrhotic patients. The model demonstrated high predictive accuracy, with an area under the curve of 0.951 in the training cohort, highlighting its potential for noninvasive risk stratification. By integrating radiomic features and clinical data, the model offers a more detailed approach to predicting esophageal variceal bleeding, helping clinicians make timely decisions. While the findings are promising, validating the model further using diverse and larger patient cohorts and considering including additional factors to maximize its clinical relevance and applicability is recommended.
- Citation: Krishnan A. Improving radiomics-based models for esophagogastric variceal bleeding risk prediction in cirrhotic patients. World J Gastroenterol 2025; 31(11): 101804
- URL: https://www.wjgnet.com/1007-9327/full/v31/i11/101804.htm
- DOI: https://dx.doi.org/10.3748/wjg.v31.i11.101804
We read with great interest the recently reported work by Peng et al[1], which aimed to create a predictive model utilizing clinical data and multi-organ radiomic features to estimate a non-invasive prediction of risk for first-instance secondary esophageal variceal bleeding (EVB) in cirrhotic patients[1]. Esophageal varices (EV) are present at diagnosis in almost half of patients with cirrhosis, and EVB is a life-threatening complication of portal hypertension, with mortality rates remaining high despite advancements in medical care[2,3]. Assessing the hepatic venous pressure gradient is considered the gold standard for diagnosing and evaluating the severity of portal hypertension[3,4]. Endoscopic examination is the primary technique for identifying EV. However, screening for EV in clinical settings may not be cost-effective due to the procedure’s invasive nature[3]. Consequently, noninvasive techniques for measuring liver fibrosis have become increasingly important. Biomarkers such as type III procollagen, type IV collagen, and hyaluronic acid are frequently used[5]. However, their effectiveness in predicting EVB remains uncertain.
On the other hand, the advancement of liver fibrosis, resulting from increased intrahepatic resistance, plays a significant role in the development of portal hypertension in liver cirrhosis. Consequently, recent studies have investigated how additional noninvasive markers and scoring systems can help predict the presence of EV and the risk of EVB. Specifically, the aspartate aminotransferase to platelet ratio index, fibrosis-4 index, fibrosis index, model for end-stage liver disease (MELD), aspartate aminotransferase to alanine aminotransferase ratio, and King score have been evaluated as a simple, noninvasive alternative to predict the presence of EV in cirrhotic patients[6-8]. In this context, the present study conducted by Peng et al[1] represents a significant advancement in predictive analysis, emphasizing the potential of radiomics as an effective predictive tool. Integrating radiomics into clinical practice offers a hopeful, noninvasive strategy for the early identification of high-risk patients. This method can enable prompt interventions, which may improve patient outcomes. I acknowledge the authors’ rigorous efforts and valuable contributions as the study addresses a critical need to predict EVB in patients with cirrhosis using radiomic features and clinical data. While the study represents a considerable advancement, I want to share a few constructive suggestions to improve its robustness and clinical applicability.
The study offered a comprehensive analysis of six distinct prediction models, including a detailed assessment of the radiomic score, clinical score, and combined radiomic-clinical models. The radiomics-clinical model demonstrated strong predictive performance, with an area under the curve of 0.951 in the training cohort and 0.930 in the validation cohort, which highlights the practical value of integrating radiomic and clinical data in predictive modeling. However, the researchers used a systematic approach to select features using the least absolute shrinkage and selection operator regression, including 3948 initial radiomic features, creating a risk of overfitting. Although the selection process was statistically rigorous, it may not contain all clinically relevant variables. To address this issue, using dimensionality reduction methods such as principal component analysis or t-distributed stochastic neighbor embedding proves advantageous[9]. These approaches reduce data redundancy while retaining the most significant predictive features, improving the model’s robustness and interpretability.
The manual delineation of regions of interest introduces the potential for observer bias despite attempts to mitigate subjectivity through inter- and intra-observer correlation coefficient values[10]. To address these challenges, automated or semi-automated segmentation algorithms could help reduce observer bias related to the manual delineation of regions of interest. Machine learning or artificial intelligence-based segmentation can provide more consistent and reproducible results, thereby minimizing the variability introduced by human observers. These methods provide consistent and objective delineation, improving replicability and scalability in clinical practice.
A detailed list of potential confounding factors should be considered. The authors have commendably accounted for important covariates, including age, sex, cirrhosis etiology, hypertension, cirrhosis complications (ascites, portal vein thrombosis), Child-Pugh grade, laboratory parameters (transaminases, bilirubin, glutamyl transpeptidase, platelet, prothrombin time, serum albumin, and alpha-fetoprotein) which are known predictors of cirrhosis-related complications. We noticed that other potential confounders, such as socioeconomic status, alcohol use, smoking status, body mass index, and liver function scores (e.g., MELD, MELD-Na) or comorbidities were not included as covariates in the regression analysis[11-13]. These confounders could impact liver function and variceal bleeding risk, potentially skewing the model’s predictive performance. Therefore, incorporating a more comprehensive set of potential confounders, including these covariates, could strengthen the robustness of the findings, particularly after excluding multicollinearity. Stratified analyses or sensitivity testing could also assess how these factors affect the model’s performance[14].
The training group consisted of 145 individuals, while the validation group comprised 63, indicating an imbalance in cohort sizes. This imbalance could potentially lower the statistical power during the validation phase. It is recommended to consider augmenting the validation cohort to address this issue. By increasing the size of the validation cohort, the study’s statistical power would be strengthened, leading to a more comprehensive evaluation of the model’s generalizability[15,16]. Moreover, a larger, independent external validation cohort from multiple centers could further validate the reproducibility of the findings across diverse clinical settings. Additionally, prospective validation should be considered to verify that the findings are relevant in real-world contexts.
In the context of future research, I want to highlight a few key points for consideration. Firstly, long-term patient outcome data, including survival rates and rebleeding events, can be integrated into the model to predict EVB and serve as a tool for predicting long-term prognosis, improving the model’s usefulness in guiding clinical decisions. Secondly, future studies could be explored using additional imaging techniques (such as magnetic resonance imaging or ultrasound) along with radiomics and clinical data to improve the accuracy of EVB prediction. Using multiple modalities may provide additional data, potentially leading to more robust models for assessing EVB risk. Lastly, extending study populations and incorporating automated segmentation techniques can further improve the reliability and generalizability of the findings.
In conclusion, the study by Peng et al[1] demonstrated the potential of using radiomics analysis combined with clinical data to predict EVB in cirrhotic patients. The results suggested that radiomics features from different imaging techniques and clinical parameters can improve the precision of predicting EVB risk, which can be valuable in creating fast and well-informed clinical decisions. The suggestions offered are intended to address these challenges. Future studies can build on these insights to refine predictive models, ultimately contributing to better patient outcomes in managing cirrhosis-related complications.
1. | Peng YJ, Liu X, Liu Y, Tang X, Zhao QP, Du Y. Computed tomography-based multi-organ radiomics nomogram model for predicting the risk of esophagogastric variceal bleeding in cirrhosis. World J Gastroenterol. 2024;30:4044-4056. [PubMed] [DOI] [Cited in This Article: ] [Cited by in RCA: 4] [Reference Citation Analysis (1)] |
2. | Krishnan A, Srinivasan V, Venkataraman J. Variceal recurrence, rebleeding rates and alterations in clinical and laboratory parameters following post-variceal obliteration using endoscopic sclerotherapy. J Dig Dis. 2012;13:596-600. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 2] [Cited by in RCA: 3] [Article Influence: 0.2] [Reference Citation Analysis (0)] |
3. | 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;65:310-335. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 1108] [Cited by in RCA: 1387] [Article Influence: 173.4] [Reference Citation Analysis (3)] |
4. | Gracia-Sancho J, Marrone G, Fernández-Iglesias A. Hepatic microcirculation and mechanisms of portal hypertension. Nat Rev Gastroenterol Hepatol. 2019;16:221-234. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 85] [Cited by in RCA: 158] [Article Influence: 26.3] [Reference Citation Analysis (0)] |
5. | Wang L, Feng Y, Ma X, Wang G, Wu H, Xie X, Zhang C, Zhu Q. Diagnostic efficacy of noninvasive liver fibrosis indexes in predicting portal hypertension in patients with cirrhosis. PLoS One. 2017;12:e0182969. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 23] [Cited by in RCA: 28] [Article Influence: 3.5] [Reference Citation Analysis (0)] |
6. | Giannini EG, Zaman A, Kreil A, Floreani A, Dulbecco P, Testa E, Sohaey R, Verhey P, Peck-Radosavljevic M, Mansi C, Savarino V, Testa R. Platelet count/spleen diameter ratio for the noninvasive diagnosis of esophageal varices: results of a multicenter, prospective, validation study. Am J Gastroenterol. 2006;101:2511-2519. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 164] [Cited by in RCA: 167] [Article Influence: 8.8] [Reference Citation Analysis (0)] |
7. | Sebastiani G, Tempesta D, Fattovich G, Castera L, Halfon P, Bourliere M, Noventa F, Angeli P, Saggioro A, Alberti A. Prediction of oesophageal varices in hepatic cirrhosis by simple serum non-invasive markers: Results of a multicenter, large-scale study. J Hepatol. 2010;53:630-638. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 94] [Cited by in RCA: 88] [Article Influence: 5.9] [Reference Citation Analysis (0)] |
8. | Cifci S, Ekmen N. Evaluation of Non-invasive Fibrosis Markers in Predicting Esophageal Variceal Bleeding. Clin Endosc. 2021;54:857-863. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 3] [Cited by in RCA: 3] [Article Influence: 0.8] [Reference Citation Analysis (0)] |
9. | Zhou H, Wang F, Tao P. t-Distributed Stochastic Neighbor Embedding Method with the Least Information Loss for Macromolecular Simulations. J Chem Theory Comput. 2018;14:5499-5510. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 56] [Cited by in RCA: 54] [Article Influence: 7.7] [Reference Citation Analysis (0)] |
10. | Baeßler B, Weiss K, Pinto Dos Santos D. Robustness and Reproducibility of Radiomics in Magnetic Resonance Imaging: A Phantom Study. Invest Radiol. 2019;54:221-228. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 109] [Cited by in RCA: 155] [Article Influence: 25.8] [Reference Citation Analysis (0)] |
11. | Shahedi M, Halicek M, Li Q, Liu L, Zhang Z, Verma S, Schuster DM, Fei B. A semiautomatic approach for prostate segmentation in MR images using local texture classification and statistical shape modeling. Proc SPIE Int Soc Opt Eng. 2019;10951:109512I. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 1] [Cited by in RCA: 1] [Article Influence: 0.2] [Reference Citation Analysis (0)] |
12. | Carriero A, Groenhoff L, Vologina E, Basile P, Albera M. Deep Learning in Breast Cancer Imaging: State of the Art and Recent Advancements in Early 2024. Diagnostics (Basel). 2024;14:848. [PubMed] [DOI] [Cited in This Article: ] [Reference Citation Analysis (0)] |
13. | Roerecke M, Vafaei A, Hasan OSM, Chrystoja BR, Cruz M, Lee R, Neuman MG, Rehm J. Alcohol Consumption and Risk of Liver Cirrhosis: A Systematic Review and Meta-Analysis. Am J Gastroenterol. 2019;114:1574-1586. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 192] [Cited by in RCA: 214] [Article Influence: 35.7] [Reference Citation Analysis (0)] |
14. | Rezvan PH, White IR, Lee KJ, Carlin JB, Simpson JA. Evaluation of a weighting approach for performing sensitivity analysis after multiple imputation. BMC Med Res Methodol. 2015;15:83. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 11] [Cited by in RCA: 11] [Article Influence: 1.1] [Reference Citation Analysis (0)] |
15. | Weng WZ, Chen JF, Peng XH, Huang M, Zhang J, Xiong J, Cao HJ, Lin BL. Risk factors for underlying comorbidities and complications in patients with hepatitis B virus-related acute-on-chronic liver failure. Epidemiol Infect. 2022;150:e147. [PubMed] [DOI] [Cited in This Article: ] [Cited by in RCA: 2] [Reference Citation Analysis (0)] |
16. | Snell KIE, Archer L, Ensor J, Bonnett LJ, Debray TPA, Phillips B, Collins GS, Riley RD. External validation of clinical prediction models: simulation-based sample size calculations were more reliable than rules-of-thumb. J Clin Epidemiol. 2021;135:79-89. [PubMed] [DOI] [Cited in This Article: ] [Cited by in Crossref: 46] [Cited by in RCA: 55] [Article Influence: 13.8] [Reference Citation Analysis (1)] |