Predicting 30-day mortality in upper gastrointestinal bleeding using APUA components and lactate: Development and internal validation.
TL;DR
While the APUA-component model demonstrated modest discrimination, incorporating lactate significantly improved its predictive performance, and decision curve analysis showed a higher net benefit across most evaluated thresholds.
Abstract
Early mortality risk assessment is crucial for acute upper gastrointestinal bleeding (UGIB). This study sought to develop and internally validate a prediction model using age, pulse, urea, and albumin (APUA) components, and to assess the added value of admission lactate levels. This retrospective, single-center study included 700 adults presenting to the emergency department with suspected acute UGIB between 2018 and 2025, excluding confirmed variceal hemorrhage. We developed multivariable logistic regression models, both with and without lactate, using continuous APUA components to predict 30-day all-cause mortality. Complete-case analyses evaluated discrimination, calibration, and net benefit, with internal validation performed using 2,000 bootstrap resamples. Fifty-two patients (7.4%) died within 30 days. Lactate was independently associated with mortality (adjusted odds ratio per 1-mmol/L increase, 1.455; 95% confidence interval, 1.270-1.667). Adding lactate significantly increased the apparent area under the receiver operating characteristic curve (AUC) from 0.693 to 0.777 (AUC difference, 0.084; p = 0.003) and improved the optimism-corrected concordance statistic from 0.670 to 0.756. Calibration and overall predictive accuracy also improved, and decision curve analysis showed a higher net benefit across most evaluated thresholds. Furthermore, lactate enhanced the apparent discrimination of three established comparator scores. While the APUA-component model demonstrated modest discrimination, incorporating lactate significantly improved its predictive performance. Independent external validation is necessary before clinical implementation.