Four methods of predicting vancomycin maintenance doses on the basis of the relationship between total-body or renal clearance of the drug and creatinine clearance were evaluated retrospectively using data from 24 patients who received vancomycin hydrochloride. Data for 18 men and 6 women with creatinine clearances of 10-130 mL/min were used; the mean (+/- S.D.) vancomycin maintenance dose was 22.0 +/- 11.2 mg/kg/day. The actual vancomycin maintenance dose needed to sustain an average steady-state vancomycin concentration of 15 mg/L was determined based on individual pharmacokinetic values. Predicted vancomycin maintenance doses were generated using the vancomycin-clearance:creatinine-clearance relationships derived separately by Nielsen, Moellering, Rotschafer, and Matzke. Orthogonal regression analysis was used to determine relationships between actual and predicted maintenance doses. Predictive ability of each method was assessed for bias (mean error) and precision (root mean squared error). Mean error and root mean squared error (+/- S.D.), respectively, for the four methods were as follows: Nielsen -262 +/- 451, 522 +/- 718; Moellering -91 +/- 439, 448 +/- 547; Rotschafer 192 +/- 602, 632 +/- 682; and Matzke -76 +/- 101, 408 +/- 517. Doses predicted by the Nielsen relationship were significantly lower than the actual vancomycin maintenance doses. Use of the Rotschafer relationship resulted in substantial overprediction, which increased as creatinine clearance declined. Doses predicted by the Moellering and Matzke relationships were slightly less than but not significantly different from the actual dose. The Matzke method demonstrated the least bias and was the most precise.(ABSTRACT TRUNCATED AT 250 WORDS)