# Weighting Example

The selection of appropriate weighting schemes becomes even more important when more than one data set is fit simultaneously. As example consider fitting drug concentrations in plasma and drug amount in urine. The magnitude of the data may be quite different. Also, the variance function may be different. Drug concentration data points are essentially independent. That is the uncertainty in one sample is independent of the uncertainty or error in the other samples. The actual values may be dependent on the underlying processes controlling the absorption, distribution, metabolism and excretion but that is the subject of pharmacokinetics. Cumulative amounts excreted into urine values, however are dependent on all the preceding values. A variance or weighting scheme proportional to the square of the value may be quite suitable for plasma concentration data measured by HPLC. For cumulative amount of drug excreted into urine a constant variance or standard deviation scheme may be more useful.

**Figure 13.8.1 A Suitable Weighting Scheme for these Plasma Data may be Constant CV**

As shown in Figure 13.8.1 a constant coefficient of variation (CV) for a series of data points translates into a weighting scheme described as the reciprocal of the observed value squared. Since there are two data sets fitted simultaneously it is important to scale the weight for each line appropriately. Thus, the **b** value of power is 2 and the **a** value is the square of the CV.

**Figure 13.8.2 A Suitable Weighting Scheme for these Urine Data may be Constant SD**

For cumulative amount of drug excreted into urine a constant standard deviation (SD) weighting scheme may be useful. Thus, variance also is constant for each of these data values. Thus, a standard deviation of 5 mg translates into a variance of 25. The **a** value is 25 and the **b** values is 0 since the weight is independent of the observed data value.
If these estimates of CV (for the plasma data) and SD (for the urine data) are reasonable then a good optimization should be possible with reasonable scatter in the weighted residual plots, Chapter 9.

## Boomer Output

One example using Boomer produced:
** FINAL OUTPUT FROM Boomer (v3.0.8) ** 09-Feb-2003 --- 11:38:29 am
Title: Fit to two lines simultaneously
Input: From Ch1308.BAT
Output: To Ch1308.OUT
Data for [Drug] came from Ch1308p.DAT
Data for Drug in Urine came from Ch1308u.DAT
Fitting algorithm: DAMPING-GAUSS/SIMPLEX
Weighting for [Drug] by 1/a*Cp(Obs )^b
With a = 0.2500E-02 and b = 2.000
Weighting for Drug in Urine by 1/a*Cp(Obs )^b
With a = 25.00 and b = 0.0000
Numerical integration method: 2) Fehlberg RKF45
with 2 de(s)
With relative error 0.1000E-03
With absolute error 0.1000E-03
DT = 0.1000E-02 PC = 0.1000E-04 Loops = 2
Damping = 1
** FINAL PARAMETER VALUES ***
# Name Value S.D. C.V. % Lower <-Limit-> Upper
1) ke 0.14030 0.465E-03 0.33 0.00 10.
2) km 0.68668E-01 0.627E-03 0.91 0.00 10.
3) V 25.045 0.159 0.63 0.10 0.10E+03
AIC = -1.28240 Log likelihood = -12.3 Schwartz factor = -16.2293
Final WSS = 0.594419 R-squared = 1.000 Correlation Coeff = 1.000

Try it out yourself with the Boomer .BAT and .DAT files
(Macintosh (OS 9),
Macintosh (OS X),
and DOS versions as zip Archives

return to the Course index

This page was last modified: Tuesday, 24th Oct 2017 at 6:06 pm

Privacy Statement - 25 May 2018

Material on this website should be used for Educational or Self-Study Purposes Only

iBook and pdf versions of this material and other PK material is available

Copyright © 2001-2018 David W. A. Bourne (david@boomer.org)