Mortality Prediction Scores Vary in Sensitivity and Specificity; Learn One and Know Its Limitations

Use of a validated prediction score is recommended for risk stratification of emergency department patients with upper gastrointestinal (GI) bleeding. Investigators in Shanghai performed a retrospective cohort study to compare the accuracy of the four principal scores in predicting 30-day mortality.

Among 395 consecutive patients with upper GI bleeding, 30-day mortality was 10.4%. The sensitivities of the scores were: AIM65, 71%; Glasgow-Blatchford, 88%; pre-endoscopic Rockall, 98%; pre-endoscopic Baylor, 49%; and for a combination of the AIMS65 and Glasgow-Blatchford scores, 90%.

Overall predictive ability of the scores, as measured by the area under the receiver-operating characteristic curve, was highest for AIMS65 (0.91), followed by Glas…