The study was approved by the institutional review board (IRB) of the University of Texas (UT) Southwestern Medical Center (IRB102005-009).
Patients and setting.
The study was performed at UT Southwestern Medical Center in Dallas, TX. The first patient was enrolled 1 March 2004, and the last patient was enrolled 25 January 2010. All subjects gave written informed consent after the study was explained to the participant in detail.
Inclusion criteria was HCV infection by serum anti-HCV enzyme immunoassay, an HCV viral load of >1,000 IU/ml, documented HCV genotype by a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory, age of >18 to 65 years, and willingness to use two or more methods of birth control in women of childbearing age. Subjects were excluded if they had uncontrolled diabetes mellitus, psychiatric illness, autoimmune disease, decompensated liver disease, or prior interferon therapy. Patients with other liver diseases, such as alcoholism, hepatitis B virus infection, Wilson's disease, hemochromatosis, or alpha-1 antitrypsin, were excluded. Patients were excluded from the study if they were unwilling to be admitted for a 48-h period for serial blood draws. Patients who did not have serum samples at the needed time points were excluded. HIV-infected patients either had to be antiretroviral therapy (ART) naïve or, if on ART, they had to be on a stable ART regimen for at least 12 weeks. The ART regimen could not include didanosine due to the risk of fatal hyperlactatemia. Patients had to have a CD4 T-cell count of >200 cells/mm3 within the 12 weeks prior to enrollment. Subjects were excluded if they had symptomatic HIV disease, including an AIDS-defining illness.
This was an open-label observational study of HCV-infected patients who were otherwise candidates for treatment with PEGIFN/RBV. Patients that met inclusion criteria and consented to participate were admitted for 48 h to initiate treatment. All patients received subcutaneous pegylated alpha interferon 2a (IFN-α2a; 180 μg/week) plus ribavirin (termed PEGIFN/RBV) daily for 48 weeks, which was dosed based on weight: if the patient was ≤75 kg, a dose of 1,000 mg daily was used; if the patient was >75 kg, the dose was 1,200 mg. Ribavirin doses were administered as a twice-daily regimen. Blood was drawn at 0, 4, 24, 96, and 168 h after the initial subcutaneous injection and then spun at room temperature at 1,500 rpm for 10 min within 4 h of collection. The plasma was decanted into freezer tubes and stored in a −80°C freezer for subsequent HCV RNA quantification and for drug assays. Blood was also collected on days 3, 5, 9, 11, 14, 21, 28, 42, and 56 for additional HCV RNA quantification, which was performed using the VERSANT 3.0 branched DNA technology (Siemens Medical Solutions Diagnostics, Tarrytown, NY) with a dynamic range of 615 to 7,692,310 IU/ml.
Standard safety assessments and urine pregnancy tests in women were conducted at each study visit. As PEGIFN/RBV was the standard of care, adverse events that were known side effects of the medications were monitored and managed by study investigators as dictated by best practice.
Human genomic DNA was extracted from whole blood or serum samples with a QIAamp DNA blood minikit (Qiagen). The targeted region, including SNP rs12979860, was amplified by PCR, and PCR products were purified and sequenced (14
Serum samples were kept frozen at −80°C after processing and then sent to Hoffman-La Roche, Inc., laboratories in Nutley, NJ, on dry ice. PEGIFN concentrations were determined using a sandwich enzyme-linked immunosorbent assay (ELISA), in which capture antibody (affinity-purified rabbit polyclonal anti-PEGIFN) was coated onto microtiter plates. Serum samples were added to the plate and incubated. The plate was washed, detection antibody (mouse monoclonal anti-PEG) was added, and then the plate was further incubated. The plate was washed again, and peroxidase-conjugated goat anti-mouse IgM was added and incubated. After a final wash, a substrate solution was added. Color developed in proportion to the amount of PEGIFN present in the sample. This assay has a range of 250 to 5,000 pg/ml. The intrabatch precision was 2.5 to 8.4%, and the accuracy was −5.5 to 13.2%. The interbatch precision was 11.7 to 16.6%, and the accuracy was 0.4 to 9.0%.
Population pharmacokinetic analyses.
PEGIFN concentrations were modeled using the ADAPT 5 program (15
). A one-compartment model, a one-compartment model with lag, a two-compartment model, a two-compartment model with lag, and a three-compartment model were examined. First, initial guesses of the population pharmacokinetic parameter estimates for each model were identified using the standard two-stage approach. The estimates were then used in further pharmacokinetic analysis using the maximum likelihood solution via the expectation-maximization (MLEM) algorithm. Four criteria were used to choose the best compartment model: Akaike's information criterion, Bayesian information criterion, negative log likelihoods, and the law of parsimony (16
). The relationships between pharmacokinetic parameter and the following covariates were examined in log-log scatter plots for the selected model: self-identified race (African-American versus other, given the known poor treatment response in African-Americans), IL-28B SNPs, HIV status, weight, CD4 count, gender, and age. Fractal relationships are easiest to spot with log-transformed data, i.e., in log-log plots (11
). When a relationship was identified, the slope of the relationship between covariate and pharmacokinetic parameter was calculated. Those covariates and initial estimates of slope were then added in the subroutine COVMOD of ADAPT. The relationship between covariate and pharmacokinetic parameter was then determined using MLEM to yield the final model estimates.
CART and survival analysis.
Classification and regression tree (CART) analysis was used to identify the best predictors of virologic outcome and to identify the threshold cutoff values for such predictors. CART analysis models are very accurate at identifying and estimating complex high-order nonlinear interactions. The outcomes of this nonparametric and recursive partitioning analysis is predictive accuracy, as opposed to statistical association with the standard statistical approaches. The outcomes examined were SVR, defined as HCV RNA of <615 IU/ml at least 24 weeks after termination of antiviral therapy, as well as rapid virologic response (RVR), defined as HCV RNA of <615 IU/ml at week 4. Variables examined for prediction of RVR and SVR included PEGIFN peak concentration, trough concentration, area under the concentration-time curve (AUC) at 168 h (1 week), patient weight, age, IL-28B SNPs, CD4 count, initial HCV viral load, gender, race, and HIV status. All outcomes were examined so that they could be ranked by the most important determinant of outcome. First, CART analysis was used for rank variables. Significant variables were chosen if the CART score was ≥20%. The analysis also identified clinically meaningful cutoff points for the selected continuous variables. The splitting criterion was based on the Gini index (19
). Several trees were constructed, pruned, and subjected to 10-fold cross-validation (20
). The optimum tree was selected based on relative misclassification costs, PK/PD considerations, and biological plausibility. CART was performed using the Salford Predictive Miner System software (San Diego, CA) (20
We validated the CART analysis findings using standard statistical approaches familiar to most clinicians. First, the main predictor of SVR (highest decision node) in CART was examined in a survivor analysis to determine time to undetectable viral load in patients above the cutoff value versus those below. Since patients who failed therapy were taken off PEGIFN/RBV after week 12 and had no viral loads done after that, data for these patients was censored at the 12-week time point. In addition, since the ratio of hazard functions was not the same at all time points, the Gehan-Breslow-Wilcoxon method was used to compare the survival curves. The CART outputs, including data obtained by examination of surrogate and competitor variables, were used as inputs for parametric univariate analysis and multivariate logistic stepwise analyses. Modifiable and clinically important variables were initially added in the several models examined and then sequentially (stepwise) removed using backward regression if P was <0.1 based on the Wald statistic. The survival analysis and the multivariate analyses were performed in SPSS version 12.