Nutrition Markers in Variceal Bleeding

Evidence

This retrospective observational study examined whether nutritional indicators could help characterize esophagogastric variceal bleeding in patients with cirrhosis treated at the First Hospital of Shanxi Medical University from March 2023 through March 2025. The analytic cohort contained 416 patients: 135 in the esophagogastric variceal bleeding group and 281 in the non-esophagogastric variceal bleeding group. The study therefore compared people classified by bleeding status within a treated cirrhotic population, rather than assigning nutritional exposures or testing an intervention. (pmid:42903201)

The investigators compared eight commonly used clinical nutritional assessment and screening tools. They used univariate logistic regression to assess the association of each tool with esophagogastric variceal bleeding occurrence and receiver operating characteristic analysis to evaluate each tool’s diagnostic performance. In this comparison, the Malnutrition Universal Screening Tool had an odds ratio of 1.444 for esophagogastric variceal bleeding, with a 95% confidence interval of 1.222 to 1.705 and p <0.001***; its area under the curve was 0.698. The report states that the area under the curve values for the other nutritional-screening indicators were all below 0.6. (pmid:42903201)

The study then constructed an eXtreme Gradient Boosting prediction model using permutation-importance-based feature selection. Its reported input variables were hemoglobin, Malnutrition Universal Screening Tool score, aspartate aminotransferase, serum creatinine, adjusted body mass index, alanine aminotransferase, neutrophils, potassium, monocyte count, and mid-arm muscle circumference. This is an integrated clinical prediction approach: nutritional measurements were included alongside laboratory and blood-cell variables, so its reported performance does not isolate the contribution of any individual nutritional measure. (pmid:42903201)

For the final eXtreme Gradient Boosting model, the reported area under the curve was 0.860, accuracy was 84.3%, recall was 66.7%, precision was 81.8%, F1-score was 80.9%, and Brier score was 0.122. These measures describe distinct aspects of model behavior in the analyzed cohort: discrimination, classification results, and calibration-oriented error were each reported. Statistical and predictive performance measures do not, by themselves, establish clinical significance. (pmid:42903201)

Shapley Additive Explanations analysis was used to interpret model predictions. The report identifies Malnutrition Universal Screening Tool score, adjusted body mass index, and mid-arm muscle circumference as important predictors. Within this cohort’s model behavior, higher Malnutrition Universal Screening Tool score, higher adjusted body mass index, and lower mid-arm muscle circumference were associated with increased esophagogastric variceal bleeding risk. These are directions reported from an observational prediction analysis, not demonstrated biological effects of changing any of those measures. (pmid:42903201)

Analysis — Cohort-specific prediction pattern

The central pattern is that a nutritional screening measure showed an observed association with esophagogastric variceal bleeding in the single-tool comparison, while nutritional indicators also appeared among the inputs and important predictors in the multivariable eXtreme Gradient Boosting model. The study design and selection process mean this is a pattern among patients with cirrhosis already classified into bleeding and non-bleeding groups at one hospital. It is consistent with the reported retrospective prediction objective, but it cannot determine whether nutritional status caused bleeding, whether bleeding status influenced measured nutritional indicators, or whether unmeasured clinical differences explain part of the association. (pmid:42903201)

The comparison between the Malnutrition Universal Screening Tool and the integrated model should also be read as a comparison between different analytic constructs. The former was assessed as a single nutritional-screening indicator in univariate logistic regression and receiver operating characteristic analysis. The latter combined nutritional, hematologic, biochemical, and anthropometric variables after permutation-importance-based feature selection. A higher area under the curve for the multivariable model therefore describes its reported cohort-level discrimination, not a causal effect attributable to nutrition measures or proof that the individual screening tool is sufficient on its own. Shapley Additive Explanations provide an account of modeled feature contributions within this dataset; the source explicitly cautions that their directions should not be interpreted as causal. (pmid:42903201)

Limitations

The retrospective design cannot establish causation. The study compared patients selected from a single hospital’s treated cirrhotic population and classified them by esophagogastric variceal bleeding status, creating scope for selection bias and for differences between groups that were not measured or not accounted for in the supplied evidence. The observed associations are therefore limited to the cohort, outcome classification, variables, and analytic methods reported here. (pmid:42903201)

The prediction results also require separation from outcome benefit. The reported accuracy, recall, precision, F1-score, area under the curve, and Brier score characterize performance of the stated model in this study cohort; they do not establish clinical significance. The supplied material identifies the model inputs and its interpretation method, but it does not provide an extraction-supported account of external validation, prospective testing, or performance in a separate setting. An extraction gap is not evidence that those items were absent from the study. (pmid:42903201)

Finally, the nutritional screening comparison and model interpretation measure associations and modeled importance, respectively. They do not establish that higher or lower values of the identified nutritional indicators independently produce esophagogastric variceal bleeding. The source specifically states that Shapley Additive Explanations-derived directions describe model behavior within this cohort and should not be read causally. (pmid:42903201)