BACKGROUND: We analyzed a large number of demographic and biochemical variables to identify predictors of hospitalization in subjects on peritoneal dialysis (PD). METHODS: All patients initiated on PD at our center from January 1990 through December 1999 were included. The following variables at the initiation of PD were included: demographics, clinical data, nutritional and adequacy parameters, transport characteristics, and various co-morbidities. Co-morbidities were graded for severity using a modified version of the Index of Coexistent Disease. Variables included during the course of PD consisted of weighted time average of a number of laboratory, adequacy, and nutritional parameters along with the number of peritonitis episodes per year. Stepwise linear regression was used following a univariate screening procedure to identify independent predictors of the outcome of hospitalization days per month on PD. RESULTS: The subject population consisted of 191 subjects (105 men, 86 women; 180 Caucasians, 10 African-American, 1 Asian). The mean age was 61 +/- 13 (SD) years and mean duration of follow-up was 21 +/- 18 months. The baseline variable analysis revealed that the presence of partner to perform PD predicted increased hospitalization (p < 0.0001). Additionally, the presence and severity of peripheral vascular disease and residual renal Kt/V at baseline (negative association) predicted increased hospitalization. In the analyses of ongoing variables, stepwise linear regression solely identified weighted time average albumin as a strong negative predictor of hospitalization (p < 0.0001). CONCLUSION: A comprehensive analysis of a large number of variables revealed that serum albumin during the course of PD (negative association) and the need for partner to perform PD strongly predicted increased hospitalization in PD subjects.