| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Briefs |
1
Div. of Renal Dis. and Hypertension, Dept. of Internal Med., and
2
Div. of Immunol. and Organ Transplant., Dept. of Surgery, The University of Texas Medical School at Houston, 6431 Fannin, Suite 4.163, Houston, TX 77030;
a author for correspondence and reprint requests: fax 713-794-1197
Sirolimus (SRL), a promising new immunosuppressive macrolide (1)(2)(3), displays unpredictable pharmacokinetic properties in humans. Measurement of a single trough concentration may reflect total drug exposure for drugs with consistent bioavailability and elimination properties, but not for drugs such as SRL that have variable bioavailability and unpredictable elimination characteristics. For such agents, area-under-the-concentration–time-curve (AUC) measurements have been used to estimate total drug exposure because they often correlate with pharmacodynamic effects (4)(5)(6). Although the relation between SRL AUC, the therapeutic effects of SRL, and toxicity has not been investigated, such a study might help elucidate the optimal therapeutic regimen for the drug.
Sampling regimens for full pharmacokinetic profiles are frequently not clinically feasible because of both time and cost restraints. Therefore, limited sampling strategies have been devised to obtain reasonable estimates of drug exposure. Limited sampling AUC strategies have been used to monitor the exposure of patients to oil-based (Sandimmune®; Sandoz, Basel, Switzerland) (7)(8) or microemulsion (Neoral®, Sandoz) (9) formulations of cyclosporine (CsA) and to chemotherapeutic agents such as cyclophosphamide (10). The present report describes a clinically reliable sampling strategy to predict the AUC values for SRL to more efficiently evaluate the clinical significance of total exposure.
In our study, 27 renal transplant recipients underwent a total of 77 full SRL pharmacokinetic profiles during the first year after the transplant procedure; 7 patients underwent pharmacokinetic profiling before the transplant procedure. SRL was administered once daily in a nonaqueous mixture (Rapamune®) provided by Wyeth–Ayerst Research, Princeton, NJ. The clinical protocol for pharmacokinetic monitoring was approved by our institutional Committee for the Protection of Human Subjects. The whole-blood samples were assayed with a validated HPLC method to quantify SRL concentrations (11).
The linear trapezoidal rule (12) was used to calculate the AUC values for each patient profile from the concentrations in the full set of blood samples drawn before dosing as well as 1, 2, 4, 6, 10, 14, and 24 h thereafter. Simple linear regression models were applied to assess the correlation between the AUC value and single sample concentrations. A nonlinear relation between the variables was excluded by using multiple curve estimation procedures (quadratic, cubic, power, inverse, "s," and log). The relative ability of single vs multiple sample concentrations to predict the AUC value was assessed with a stepwise forward selection multiple regression technique (13). This method called for the independent variables to be added to the regression equation sequentially in order of diminishing importance, and for the coefficient of determination (r2) and the regression coefficient to be calculated at each step. We developed multiple linear regression models in which two, three, or four concentration time points served as the independent variables. The relative fit of each model was determined on the basis of the r2 value. The models were then validated by comparing the predicted AUC with the value calculated from the full pharmacokinetic profile. We used the concentration data from several combinations of samples to generate model linear regression equations. The prediction error for each set of values for a given patient was calculated as {[(predicted AUC - full AUC)/full AUC] x 100}. In addition, the mean prediction error and the standard deviation values among all patient data were displayed in frequency histograms that were evaluated for the presence of a normal distribution.
Simple and multiple linear regression and ANOVA analyses were performed to investigate the importance of other factors that might influence the prediction error, such as time after transplantation, absolute AUC values, demographic characteristics, and CsA pharmacokinetic parameters. A power analysis was performed to investigate the likelihood of a type-II error. The statistical analyses were performed with SPSS software (Version 7.0 for Windows 95; SPSS, Chicago, IL) (13) and NCSS software (for the power analysis; PASS, Version 1.0 for DOS) on an IBM-compatible personal computer with a Pentium processor.
Table 1
shows the correlation between the SRL AUC value calculated from
the full pharmacokinetic profile and concentrations obtained at several
time points. The concentration values for the samples obtained at
24 h (C24) correlated better with the full AUC
values and showed lower prediction error values than C0
values, presumably because the C24 sample was obtained at
the precise time by hospital personnel, whereas the C0
value was dependent on patient recollection. Although the SRL AUC value
was well predicted on the basis of data from one early (C2)
and one late (C14) sample (r2 =
0.99), a strategy that involves a late sample compromises the clinical
utility of the limited sampling approach. Table 1
focuses on the
correlation coefficients and regression equations for various
combinations of early time points. Correlation coefficients >0.90 were
observed for samples obtained at 2 and 6 h; 2, 4, and 6 h; 0,
2, 4, and 6 h; and at 24 h after dosing.
|
We assessed the prediction error of these sets of samples (Table 1
).
Although good coefficients of determination were observed for the
samples collected at all time points except for the C0
samples, only two sets of samples, namely those obtained at 0, 2, and
6 h and those obtained at 0, 2, 4, and 6 h, were clinically
acceptable, meaning that >85% of the AUC estimates fell within the
±15% prediction error range. The standard deviation of the mean
prediction error for each of these two sets of data was less than half
that of the 24-h estimate alone. The distribution results were narrower
around the mean and reflected a more normal distribution with the 0-,
2-, 6-h and 0-, 2-, 4-, 6-h models than with other models.
The factors of time after transplantation and absolute AUC values; demographic characteristics of patient weight, age, gender, and race; and CsA AUC values did not affect the prediction error values of the various models (data not shown). However, the power of the multiple regression and ANOVA models was not sufficient to exclude a type-II error.
The therapeutic window for SRL concentrations has yet to be defined. Our previous observations (unpublished) suggest that the hyperlipidemic and pancytopenic toxic effects of SRL are concentration-dependent and correlate with C0 values. Monitoring SRL AUC as a measure of total drug exposure may allow us to refine further the relationship between these toxic effects and blood concentrations and (or) to elucidate the relation between concentration and the immunosuppressive efficacy of SRL.
Because a patient would have to be hospitalized for a full day to
obtain a full pharmacokinetic profile of SRL, the cost of
hospitalization and the discomfort of the patient may make it not
clinically feasible to obtain a complete set of samples routinely.
Therefore, a limited sampling strategy in which samples are obtained at
0, 2, 4, and 6 h and at 0, 2, and 6 h provides a reasonable
prediction of the AUC value calculated from a full set of samples.
Because the mean prediction error and standard deviations for both
models are almost identical, the latter model with fewer time points is
preferable. When this limited sampling strategy (0, 2, and 6 h)
was used to predict AUC values, 66 of the 77 calculated concentrations
displayed a <15% difference from the corresponding full AUC value
(Table 1
). Although further studies are warranted to identify
situations in which a full AUC profile is required because of a large
prediction error, we believe that the limited sampling strategy (0, 2,
and 6 h) described herein represents an efficient approach to
assess total exposure to SRL.
Acknowledgments
This work was supported by a grant from the National Institutes of Diabetes and Digestive and Kidney Diseases (NIDDK 38016–09).
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
