A multicenter evaluation of variables affecting the predictability of creatinine clearance.

Data from 722 urine collections (627 patients) from 14 medical facilities were used to compare measured creatinine clearance values with clearance estimates calculated using three "urine-free' mathematical formulas. The influence of two patient weight variables (actual weight, lean body weight) and the level of renal function on clearance prediction was assessed. In addition, site parameters (region, facility), time parameters (month, day, day of the week), and patient parameters (age, sex, height, weight, diagnoses) were evaluated for their statistical influence on the relationship between measured clearance and estimates calculated with one of the methods. Strong, statistically significant correlations were observed between clearance values estimated with each prediction method and measured clearances. Although statistical differences were noticed between mean values predicted with each method, these differences were clinically insignificant. Use of lean body weight for calculation of creatinine excretion produced clearance estimates that were significantly lower than those obtained with actual weight and substantially closer to measured clearance values. Among patient variables, age and two specific diagnoses (congestive heart failure and pregnancy), were statistically associated with variability in the relationship between measured and predicted clearance. Interregional and interfacility differences in the measured-predicted clearance relationship, variability associated with the day of the week of collection, and tendency of the urine-free formulas to over-predict clearance support potential inaccuracy of measured clearance determination. The results suggest that inaccuracies in clearance determination, rather than inadequacies of urine-free prediction methods, account for the majority of the observed variability in the relationship.