Work

Performance of the Padua Prediction Score in Critically Ill Patients With Gastrointestinal Bleeding: A 24-Hour Landmark Cohort Study Using MIMIC-IV

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background Critically ill patients with gastrointestinal bleeding may have substantial venous thromboembolism risk, but pharmacologic prophylaxis can complicate hemostatic management and raise concern for recurrent bleeding. The usefulness of the Padua Prediction Score in this setting remains uncertain. Methods We performed a retrospective 24-hour landmark cohort study using MIMIC-IV version 3.1. Adults with gastrointestinal bleeding among the first three discharge diagnoses, an intensive care unit admission, and a hospital stay of at least 72 hours were included. The primary endpoint was a non-primary-position venous thromboembolism diagnosis code during the index hospitalization. A stricter sensitivity endpoint additionally required treatment-dose anticoagulation initiated after the landmark. We evaluated discrimination, threshold performance, apparent calibration, multivariable associations, and exploratory overlap-weighted associations of early prophylaxis with venous thromboembolism and severe rebleeding proxies. Results Among 2,758 patients, 101 (3.7%) met the primary endpoint. The median age was 69 years, and 60.2% were male. The Padua score had an area under the receiver operating characteristic curve of 0.610 (95% confidence interval, 0.555–0.665). At the conventional threshold of ≥ 4, sensitivity was 86.1%, specificity 25.1%, positive predictive value 4.2%, and negative predictive value 97.9%. The maximum-Youden threshold was ≥ 6, with sensitivity of 57.4% and specificity of 62.5%. Results were similar for the strict endpoint, with 52 events and an area under the curve of 0.641 (95% confidence interval, 0.569–0.713), and among patients with complete core Padua data, with an area under the curve of 0.644 (95% confidence interval, 0.575–0.713). Exploratory weighted associations between prophylaxis and thrombotic or rebleeding proxies were imprecise and not statistically significant. Conclusions In critically ill patients with gastrointestinal bleeding, the conventional Padua threshold captured most venous thromboembolism-coded cases but classified three quarters of the cohort as high risk and had poor specificity. The score should not be used alone to guide prophylaxis; decisions should incorporate bleeding activity, hemostatic stability, and repeated clinical reassessment.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7225872279
Cites: 17
Cited by:
Related to: 10