Rockall scores were calculated in 247 patients with major peptic ulcer bleeding entered into a randomized trial of endoscopic therapy. The observed rates of recurrent bleeding and mortality after endoscopic therapy were compared with predicted rates derived from Rockall's study group. The validity of the Rockall system was assessed in terms of calibration and discrimination.
Rates of recurrent bleeding and mortality after endoscopic therapy increased with an increasing Rockall score. Observed rates of recurrent bleeding and mortality were below predicted rates, and calibration of the Rockall system was poor (Mantel-Haenszel chi square = 25.8, p < 0.0001 for recurrent bleeding; Mantel-Haenszel chi square = 15.1, p < 0.0001 for death). For the prediction of recurrent bleeding, the area under the receiver operating characteristic curve was low (63.4 % ), but the system was satisfactory when predicting mortality (area under the resulting curve, 84.3 % ).
After endoscopic therapy for a bleeding peptic ulcer, the Rockall scoring system can identify patients at high risk of death, but it is inadequate for the prediction of recurrent bleeding.