Upper digestive tract hemorrhage in cirrhotic patients. Contribution of multidimensional analysis to the study of variables related to the cause of the hemorrhage.

Sixteen clinical and biological variables were recorded in 116 episodes of upper digestive tract hemorrhage of known cause in cirrhotic patients. One-dimensional analysis reveals a significant correlation between six variables and the rupture of esophagogastric varices, whereas multiple linear regression and partial correlation analysis reduces the significant variables to two: a history of digestive hemorrhage and the nonalcoholic etiology of the cirrhosis. A value of the discriminant function exists for which the specificity and, consequently, the positive predictive value are equal to 100%, but with a sensitivity of 39%. These results mean that, in an explanatory approach, partial correlation analysis seems to constitute an indispensable complement to analysis of clinical and biological variables, since it reduces the chances of unwarranted explanatory interpretation. However, in a pragmatic approach, the recording of 16 variables does not permit a clear discrimination between ruptured varices and nonruptured varices; this suggests that either other factor(s) remain to be discovered or else that those related to ruptured varices and to acute ulcerations are the same.