Objectives: We aimed to develop a model based on noninvasive variables for the prediction of clinically significant portal hypertension (CSPH) and of esophageal varices (EV) in patients with compensated liver disease.
Methods: Sixty patients with compensated liver cirrhosis diagnosed by histology were included in the training set. All patients had physical examination, laboratory tests, abdominal color-Doppler ultrasound, upper digestive tract endoscopy, and measurement of hepatic venous pressure gradient. Predictive models for the presence of CSPH and of EV were calculated. The models were validated in an independent series of 74 patients with compensated liver disease.
Results: Clinical and laboratory variables were selected in the final models, while ultrasonography did not add statistical power for the prediction of CSPH and EV. The model for prediction of CSPH included albumin, INR, and ALT. The best cutoff had 93% sensitivity and 61% specificity in the training set, and correctly classified 77% of patients in the validation set. Spider angiomas, ALT, and albumin predicted EV. The best cutoff of the model in the training set had a sensitivity of 93% and a specificity of 37% and correctly classified 72% of cases in the validation set.
Conclusions: Noninvasive prediction of EV in well-compensated cirrhotic patients is not accurate. However, a model obtained by combining simple laboratory variables has a high sensitivity to predict CSPH in this population and may be useful to select the subset of patients requiring screening endoscopy. By this method, endoscopic screening could be obviated in about 40% of patients.