Validation of a multivariate model predicting presence and size of varices

Background: A model developed by our group identified low platelets and advanced Child-Pugh class (CPC) as being associated with large varices. Goals: To validate a defined cut-off of platelets <= 80,000/mu L in CPC-A for large varices and platelets <= 90,000/mu L in CPC-B/C for any varices. Study: Validation cohorts consisted of patients with cirrhosis undergoing screening for varices from Oregon Health and Science University (n = 152), Indiana University (n = 252), and Genoa, Italy (n = 101). Similar clinical and laboratory data were collected as for the original cohort. To assess the ability of these cut-offs to predict presence of large and any varices, sensitivity, specificity, positive and negative predictive values, likelihood ratios, and the c-statistic were measured. Results: The validation cohorts were statistically different from the original cohort with regards to CPC and prevalence of large varices. Combining the original (n = 301) and validation cohorts resulted in a negative predictive value of 92.1% for platelets <= 80,000/mu L in CPC-A for large varices and positive predictive value of 80.1% for platelets <= 90,000/mu L in CPC-B/C for any varices. Combining the 4 cohorts yielded a c-statistic of 0.67 (95% confidence interval: 063-0.72). No other factors such as splenomegaly and Model for End-Stage Liver Disease score were identified as significant. Conclusions: This study confirms the validity of a previous model identifying low platelets and advanced CPC class as predictors of large varices. Despite combining the cohorts, no other risk factors were identified.