Derivation and Internal Validation of a Clinical Prediction Tool for 30-Day Mortality in Lower Gastrointestinal Bleeding.
Am J Med. 2017 Jan 05;:
Authors: Sengupta N, Tapper EB
BACKGROUND AND AIMS: There are limited data to predict which patients with lower gastrointestinal bleeding are at risk for adverse outcomes. We aimed to develop a clinical tool based on admission variables to predict 30-day mortality in lower gastrointestinal bleeding.
METHODS: We used a validated machine learning algorithm to identify adult patients hospitalized with lower gastrointestinal bleeding at an academic medical center between 2008 and 2015. The cohort was split randomly into a derivation and validation cohort. In the derivation cohort, we used multiple logistic regression on all candidate admission variables to create a prediction model for 30-day mortality, using area under the receiving operator characteristics curve (AUC) and misclassification rate to estimate prediction accuracy. Regression coefficients were used to derive an integer score, and mortality risk associated with point totals were assessed.
RESULTS: In the derivation cohort (n=4044), 8 variables were most associated with 30-day mortality: age, dementia, metastatic cancer, chronic kidney disease, chronic pulmonary disease, anticoagulant use, admission hematocrit and albumin. The model yielded a misclassification rate of 0.06, and AUC of 0.81.The integer score ranged from -10 to 26 in the derivation cohort, with a misclassification rate of 0.11 and AUC of 0.74. In the validation cohort (n=2060), the score had an AUC of 0.72 with misclassification rate of 0.12. After dividing the score into 4 quartiles of risk, 30-day mortality in the derivation and validation set was, respectively, 3.6% and 4.4% in Quartile 1, 4.9% and 7.3% in Quartile 2, 9.9% and 9.1% in Quartile 3, and 24% and 26% in Quartile 4.
CONCLUSION: A clinical tool can be used to predict 30-day mortality in patients hospitalized with lower gastrointestinal bleeding.
PMID: 28065767 [PubMed - as supplied by publisher]