ABSTRACT
A computer program based on a Bayesian statistical model has been developed for calculating tracer clearance from any number of plasma samples drawn at arbitrary time intervals. Bayesian prior parameters were calculated from clinical data for Tc99m-MAG3, Tc99m-EC, I131-OIH, Tc99m-DTPA, and Yb169-DTPA and then used to calculate clearance from prospective data. Clearance estimates using only one or two plasma samples were found to closely approximate the results of using multiple samples. When only one or a few samples are available, the program supplements the observed data by a Bayesian prior probability distribution (based on prior clinical measurements) to achieve agreement with multisample clearance. When many points are available, the observed data overwhelm the prior probability, and results approach those of conventional curve fitting but with less sensitivity to bad data points and less risk of fitting failure.
Subject(s)
Humans , Radiopharmaceuticals/pharmacokinetics , Kidney/physiology , Computer Simulation , Bayes Theorem , Kidney Glomerulus/physiology , Ytterbium/pharmacokinetics , Models, Statistical , Probability , Kidney Function Tests , Radiopharmaceuticals/blood , Iodine Radioisotopes/pharmacokinetics , Metabolic Clearance Rate , Technetium/pharmacokinetics , Kidney Tubules/physiologyABSTRACT
UNLABELLED: Because recent reports have questioned the traditional 2-compartment model for calculating tracer clearance after a single intravenous injection, a 3-compartment model was evaluated in this study. METHODS: Bayesian statistics were used, which facilitated curve fitting by treating all subjects simultaneously. (99m)Tc-Mercaptoacetyltriglycine clearance data from 154 adults and 109 children were measured at several centers, typically 6-9 plasma samples spanning 5-90 min, and fitted by 2- and 3-compartment Bayesian models. RESULTS: Clearance estimates were found to be systematically lower for the 3-compartment model than for the 2-compartment model. A single-sample procedure based on the 3-compartment model was found to eliminate most of the known discrepancy between formulas based on single-injection and continuous-infusion reference methods. CONCLUSION: A 3-compartment model led to lower and probably more accurate clearance estimates than the conventional 2-compartment model. A new single-sample method is presented, based on the 3-compartment model as reference standard.
Subject(s)
Kidney/metabolism , Models, Biological , Technetium Tc 99m Mertiatide/pharmacokinetics , Adolescent , Adult , Bayes Theorem , Child , Child, Preschool , Computer Simulation , Humans , Infant , Kidney/diagnostic imaging , Metabolic Clearance Rate , Quality Control , Radioisotope Renography/methods , Radiopharmaceuticals/blood , Radiopharmaceuticals/pharmacokinetics , Reproducibility of Results , Sensitivity and Specificity , Technetium Tc 99m Mertiatide/bloodABSTRACT
UNLABELLED: Nonlinear Bayesian regression permits curve fitting to a group of subjects simultaneously rather than individually. We evaluated this approach for interpreting plasma clearance curves with the goal of reducing curve-fitting failures and dealing objectively with problem datasets that may arise in clinical settings. METHODS: (99m)Tc-Diethylenetriaminepentaacetic acid plasma clearance curves from 79 subjects were analyzed. The data typically comprised 7-9 samples obtained from 5-10 to 180-240 min after injection. A 2-compartment model was fitted by Bayesian regression to yield compartmental hyperparameters V1, L21, and L12 corresponding to the volume of the compartment into which tracer was injected and the transfer rates from compartment 1 to compartment 2 and from compartment 2 to compartment 1, respectively. This also yielded a clearance estimate for each subject. RESULTS: Estimated hyperparameters were V1 = 8.9 L, L21 = 0.026 min(-1), and L12 = 0.040 min(-1). Conventional methods led to fitting failures in 2 of the 79 subjects but there were no failures with the Bayesian method. The hyperparameters were used to calculate the glomerular filtration rate for each subject from a single plasma sample with a root-mean-square error of 7.3 mL/min, which was not significantly different from the widely used Christensen-Groth formula. CONCLUSION: Fewer fitting failures were encountered than with conventional methods, offering an objective means of dealing with problem data. This conceptually simple model can be used directly to calculate clearance from a single plasma sample. It requires only the 3 parameters described above, whereas the Christensen-Groth method requires 6 parameters.
