Introduction
Bevacizumab (Avastin
®, Genentech Inc.) is a humanized monoclonal immunoglobulin G (IgG) 1 antibody that specifically binds and neutralizes the biological activity of vascular endothelial growth factor A (VEGF-A), a key isoform of VEGF involved in angiogenesis, and a well-characterized pro-angiogenic factor [
1]. Bevacizumab causes inhibition of tumor angiogenesis by blocking VEGF-A from binding to its receptors and leads to tumor growth inhibition. Bevacizumab in combination with standard therapy has received marketing authorization for use in the treatment of various cancers including metastatic colorectal cancer (CRC) [
2,
3], non-small cell lung cancer (NSCLC) [
4], breast cancer [
5], renal cell carcinoma [
6], cervical cancer [
7] and ovarian cancer [
8].
A population pharmacokinetic (PK) model has been previously developed [
9]. Bevacizumab PK showed dose linearity within the dose range of 1–20 mg/kg, a slow clearance, a volume of distribution consistent with limited extravascular distribution and a terminal half-life of approximately 20 days. Clearance (CL) and central volume of distribution (V1) increased with body weight and were higher in male patients. CL decreased with increasing albumin and decreasing alkaline phosphatase. There has been no evidence for anti-therapeutic antibodies (ATAs) for bevacizumab in metastatic solid tumors based on the large number of historical clinical studies, and ATA was detected in only 0.6 % of the patients with colon cancer (adjuvant setting) [
10].
However, the previous analysis had several limitations. Several important covariates were not evaluated in previous analysis including ethnicity (e.g., Asian vs. non-Asian), indications and baseline VEGF-A. First, bevacizumab has been widely used across ethnic groups (e.g., Asian vs. non-Asian), and supplementary Biologics License Applications have been submitted to health authorities for approval of using bevacizumab for new indications or new combinations based on data from limited ethnic groups while the target population contains much broader ethnic groups. Therefore, it is important to evaluate ethnicity (Asian vs. non-Asian) as a covariate. Second, it has been shown that bevacizumab clearance is 50 % higher and exposure is 50 % lower in gastric cancer as compared to other types of solid tumors [
11], making it important to evaluate indication as a covariate. Finally, several studies have shown the predictive value of baseline VEGF-A for bevacizumab treatment effect on progression-free survival and/or overall survival, meaning that only patients with high VEGF-A levels may benefit from bevacizumab treatment, for example in gastric cancer [
12] and metastatic breast cancer [
13]. Therefore, it is important to evaluate baseline VEGF-A as a covariate. The reason why these important covariates were not evaluated in the previous analysis is likely that these evidences showing the importance of these covariates all appeared after the previous analysis, and therefore, the significance of these covariates may not have been fully realized at that time, and/or the data were unavailable at that time.
Other limitations of the previous analysis include utilization of FO (first-order) instead of FOCE (first-order conditional estimation) algorithm in NONMEM [
14], limited number of studies (
n = 6), patients (
n = 491) and indications (mainly CRC, NSCLC and breast cancer), etc. Therefore, an updated analysis is warranted.
The objectives of the current analysis were to develop a robust population PK model in adult patients with solid tumors and to evaluate the influence of patient variables on bevacizumab PK, which can be used to simulate bevacizumab exposure to optimize bevacizumab dosing strategies.
Discussion
This analysis is a comprehensive PK evaluation of bevacizumab in adult cancer patients in Phases I–IV studies as a single agent or in combination with chemotherapy for both single- and multiple-dose administration with both rich and sparse bevacizumab serum concentration data. A robust population PK model was built based on a large PK population of 1792 patients from 15 studies and then externally validated using data from 146 Japanese patients in three independent studies. This model consolidated all bevacizumab PK data in one model, can timely support simulations and decision making when needed, can help develop consistent pharmacokinetic messages of bevacizumab for investigators and health authorities given that multiple PK models have been developed for bevacizumab and contained inconsistent messages, and can support future studies of bevacizumab in other indications. As mentioned in the “Introduction,” many important covariates that were not evaluated in the previous analysis (e.g., Asian vs. non-Asian, indications, baseline VEGF-A) were evaluated in this analysis.
Typical population PK parameter estimates were similar as previously published [
9]. The low IIV of 29 and 18.3 % observed for CL and V1 was typical for antibody drugs [
25]. The pcVPC demonstrated adequate fit and predictive performance of the final model (Fig.
1). The median prediction (blue band) may appear to be slightly below the median observation (blue line) beyond day 112, suggesting a possible tendency of under-prediction. However, this tendency is likely irrelevant given (1) the small degree of under-prediction, (2) the sparseness of data beyond day 112, (3) the good predictive performance for 2.5th and 97.5th percentiles (red) across all time points, as well as (4) the complexity and heterogeneity of the data. The external validation (Figs.
3,
4) demonstrated good predictive performance of the final model with no apparent systemic bias and the similarity in bevacizumab PK between Asian and non-Asian adult cancer patients. Although there may appear to be a tendency of over-prediction of the variability (Fig.
3), this tendency is likely irrelevant because the model was built based on more heterogeneous data (15 studies over a decade across various ethnic groups) while the validation data were more homogeneous (three studies over a few years in Asian patients only).
Factors significantly associated with bevacizumab PK were similar as previously published [
9]: CL and V1 increased with BWT and were higher in males, and CL decreased with increasing albumin and decreasing BALP. It is well known that CL of other IgG antibodies is faster in patients with lower serum albumin levels [
25], likely due to two reasons. First, the level of albumin correlates with disease status. Second, the recycling of albumin and IgG is both mediated by FcRn (neonatal Fc receptor) [
25], and therefore, albumin levels may reflect the abundance and efficiency of FcRn. The effect of BALP on bevacizumab CL is likely because BALP is an indicator of disease burden, such as liver or bone metastases. CL was found to be 15.6 % lower in patients treated with interferon alpha. However, this effect was within the overall PK variability and therefore may be clinically irrelevant.
Similar to the previous analysis [
9], tumor burden was not included in the final model in this analysis. Among solid tumors, tumor burden is usually an indicator of disease severity and health status. It is usually defined as the sum of longest diameters of target lesions under RECIST (Response Evaluation Criteria in Solid Tumors) criteria for systematic tumors and under other criteria for other tumors (e.g., brain tumors). Inclusion of tumor burden in bevacizumab PK model may not be crucial. First, tumor burden as an indicator of disease burden and health status could already be represented by albumin and BALP in the model. Second, tumor burden as a source of VEGF-A (target of bevacizumab) is irrelevant for bevacizumab PK because bevacizumab molar concentration is thousands of times higher than that of VEGF-A [
10], and there has been no evidence of target-mediated drug disposition (TMDD) for bevacizumab [
10]. Third, in previous analyses, tumor burden alone showed relatively low impact on bevacizumab exposure in the sensitivity analysis (similar to Fig.
2, data not published). Finally, the final model demonstrated adequate fitting and superior predictive performance without incorporating tumor burden.
On the other hand, three factors made it impossible to test baseline tumor burden as a covariate in this analysis. First, tumor response criteria were inconsistent across these 15 studies that were conducted across a time span of over a decade. Several different versions of RECIST and other criteria (e.g., Macdonald criteria for glioblastoma in BO21990) were used. Second, the methods used to measure tumor burden were inconsistent across studies, such as CT (computerized tomography) scans and MRI (magnetic resonance imaging). Finally, unit of length (mm) and area (mm2) both exist in tumor burden data, which cannot be converted to each other. In fact, inclusion of tumor burden in the model would greatly reduce the applicability of the model due to the continuous advancement in tumor response criteria and measurement methods, and due to different tumor response criteria and measurement methods across cancer types, for example RECIST version 1.0 versus version 1.1, RECIST criteria versus Macdonald criteria or RANO (Response Assessment in Neuro-Oncology) criteria, CT scans versus MRI.
In conclusion, a robust population PK model for bevacizumab in adult cancer patients was built and externally validated, which may be used to simulate concentration-time profile in adult cancer patients in future studies [
11,
26]. Baseline body weight, albumin, alkaline phosphatase and gender were the covariates with the greatest influence on bevacizumab CL and V1, supporting body weight-based dosing of bevacizumab. No difference in bevacizumab PK was observed between Asian and non-Asian patients. Given the similarity in PK among many monoclonal antibodies, this may inform PK studies in different ethnic groups (e.g., Asian vs. non-Asian) for other therapeutic antibodies without TMDD and significant race-dependent target expression.