Computer analysis of fibrinogen tracer data

Abstract

Biological fibrinogen tracer data were simulated on a high-speed digital computer by superposing normally distributed random errors of 2 and 5% on values generated from sums of two, three, and four exponential terms in which the slowest decaying term was given the value 0.7 e −0.15 t . Thirty sets of 10 or 15 data points ( x i , t i ) with equidistant or initially concentrated distribution and with weights 1 or x i −2, only differing with respect to the random numbers to assign the error were analyzed by a damped Gauss-Newton iterative technique. From the fitted sums of exponentials, metabolic parameters (fractional catabolic rate constant, radioactivity distribution ratio, and transcapillary transfer rate constant) were calculated by compartmental analysis. The criterium of the sum of squares of deviates did not allow to determine the number of terms from which the data were generated, because its value was mainly determined by data error and much less by systematic error, introduced by underestimation of the number of terms. Reliable estimates could be obtained for only the fractional catabolic rate and to a lesser extent for the radioactivity distribution ratio, while the estimation of the fractional transcapillary efflux rate constant was erratic. Those estimates were most reliable when fitting the data with a sum of two exponential terms, regardless of the number of terms in the generating function. When fitting data generated from a sum of three exponential terms with a sum of three, convergence was obtained in only half to two thirds of the simulations, while convergence was always obtained when fitting these data with a sum of two exponential terms.