Methods for analysis of the occurrence of abscess in patients with pancreatitis.

ABSTRACT

OBJECTIVE: Standard survival analysis methods are useful for data involving censored cases when cures do not generally occur. If the object is to study, for instance, the development of a complication in the progress of an infectious disease, some people may be cured before complications develop. In this article, we provide methods for the analysis of data when cures do occur. An example is a study of prognostic factors for pancreatic abscess in patients with pancreatitis, some of whom leave the risk set because the pancreatitis clears. DESIGN: We present methods for estimating the survival curves and comparing hazard function for two objectives: (1) the occurrence of an abscess, irrespective of whether the patients are cured or not, and (2) the occurrence of an abscess for patients who, at that stage, have not been cured. PATIENTS We illustrate the applications of the methods using a sample of 50 patients with severe pancreatitis. RESULTS: To study the occurrence of an abscess, regardless of whether the patients are cured or not, we show that the appropriate strategy is to assign to the cured patients an infinite time to the appearance of an abscess. If the cured were considered censored at the moment the pancreatitis cleared, this would result in an overestimation of the hazard of presenting an abscess. On the other hand, if the objective is to compare the occurrence of abscess according to an exposure for patients who have not been cured, one needs to censor the cured patients at the time they are cured. CONCLUSIONS: For the analysis of survival data in the context of infectious diseases when cure is possible, it is important to use a censoring strategy that is pertinent to the specific aims of the study. Considering cures as censored at the time of cure is not always appropriate.