Correspondence analysis in a study of the clinical evolution of uncomplicated chronic relapsing alcoholic...

This paper gives an example of the use of correspondence analysis in a study of the risk factors of surgery in uncomplicated chronic relapsing alcoholic pancreatitis (UCRAP) where censored observations are present. Ninety-seven patients were admitted to a long-term follow-up project on the clinical evolution of UCRAP. During follow-up 61 patients underwent surgery, while the observations for the remaining 36 patients were censored. Correspondence analysis was performed on three prognostic variables, selected by a preliminary univariate analysis. A supplementary variable indicating the state of each patient at a particular time (still in follow-up, operated on, lost to follow-up), was projected onto the best factor plane chosen by correspondence analysis. Cox's proportional hazards regression model was also used with the scores of patients on the factor axes as independent variables to evaluate their prognostic importance. Both correspondence analysis and the Cox model showed that the first two axes provide important prognostic information. Furthermore, the former suggested a pattern of censorship for patients with observations censored after 5 years of follow-up. The results of this study confirm that correspondence analysis may be useful in follow-up studies by providing graphic display of important information relative both about the event under study and about censored observations.