Photo Credit: Mohammed Haneefa Nizamudeen

The following is a summary of “Prediction of gastrointestinal bleeding hospitalization risk in hemodialysis using machine learning,” published in the October 2024 issue of Nephrology by Larkin et al. 

Researchers conducted a retrospective study to assess the ability of machine learning to predict the 180-day hospitalization risk for gastrointestinal bleeding (GIB) in patients undergoing hemodialysis (HD). 

They developed an eXtreme Gradient Boosting (XGBoost) and logistic regression model using an HD dataset from the United States (2017–2020). Patient data was randomly divided into training (50%), validation (30%), and testing (20%) sets. Treatments within 180 days prior to GIB hospitalization were classified as positive observations, while others were negative. The models accounted for 1,303 exposures and covariates, and performance was evaluated using unseen testing data. 

The results showed that the incidence of 180-day GIB hospitalization was 1.18% in the HD population (n = 451,579) and 1.12% in the testing dataset (n = 38,853). The eXtreme Gradient Boosting (XGBoost) model achieved an area under the receiver operating curve (AUROC) of 0.74 (95% confidence interval (CI) 0.72, 0.76), while logistic regression had an AUROC of 0.68 (95% CI 0.66, 0.71). XGBoost’s sensitivity and specificity were 65.3% (60.9, 69.7) and 68.0% (67.6, 68.5), respectively, compared to 68.9% (64.7, 73.0) and 57.0% (56.5, 57.5) for logistic regression. Both models linked GIB hospitalization risk to older age, anemia and iron disturbances, recent hospitalizations, and bone mineral metabolism markers. XGBoost emphasized the importance of serum 25 hydroxy (25OH) vitamin D levels, while logistic regression highlighted parathyroid hormone (PTH) levels. 

The study concluded that machine learning, particularly XGBoost, is effective for early detection of GIB risk in patients with HD, though further external and prospective validation is necessary, especially regarding the unexpected association with bone mineral metabolism markers. 

Source: bmcnephrol.biomedcentral.com/articles/10.1186/s12882-024-03809-2#Abs1