Developing TDM with oral targeted therapies has been an appealing strategy to reduce inter-individual variability in PK due to multiple causes, such as DDI, genetic polymorphism (PGx), poor adherence or comorbidities.
Bertolaso et al. have demonstrated that even if TDM is not routinely used for sunitinib in every mRCC patient, pharmacokinetically guided dosing could be useful for frail individuals such as patients with cardiac transplant [13
]. Several PK/PD models have been further proposed as a means to predict the efficacy of sunitinib. For instance, Diekstra et al. have proposed a comprehensive PK/PD/PGx model for sunitinib in both mRCC and metastatic colorectal cancer (mCRC) patients. They found that drug exposure was related to efficacy in mCRC patients but, surprisingly, in mRCC patients, monitoring basal levels of sVEGFR2 was more useful than PK to forecast sunitinib efficacy [14
]. Similarly in GIST patients treated with sunitinib, modeling early changes in standardized uptake value upon PET scan imaging helped to predict survival [15
]. In another study, Narjoz et al. found that lean body weight and genetic polymorphisms on the ABCG2 transporter contributed to the PK variability of sunitinib, and that AUC above 1950 ng/mL.h was associated with prolonged survival in mRCC patients [16
]. Here, inter-patient variability in exposure levels ranged from 42% to 45% for AUC and Cmin, respectively. Ideally, TDM should help in customizing dosing in a time-effective manner when patients are not in the right therapeutic window. Indeed, TDM can provide early and relevant information on inadequate exposure levels before clinical signs (e.g., side effects or lack of efficacy) actually show in patients. Implementing TDM with adaptive dosing in routine practice in oncology remains difficult and most oncologists prefer to rely on their clinical judgment, rather than using model-based dosing recommendations [17
]. Logistic considerations such as strict respect for sampling time (i.e., T24 h) to measure trough concentrations have long been hard requirements to meet in routine clinical practice. Vagueness of subsequent dosing recommendations or the difficulties in establishing appropriate target concentrations could also be limitations. However, the rise in oral targeted therapies has shown that drug exposure is correlated with clinical outcome in several settings [18
] In this respect, PK/PD modeling offers valuable help. First, the Bayesian estimate of individual PK parameters from sparse samples can help in simulating trough concentrations, regardless of the exact time of the sampling. Indeed, once individual PK parameters have been calculated with a good-quality estimates, it is possible to determine in silico the trough concentrations and the resulting AUC. Second, provided that the therapeutic window has been identified, the model can then calculate the exact dosing to achieve appropriate exposure [20
]. Here, we have tested such a PK/PD model implemented in Monolix®
to monitor sunitinib and N-desethyl sunitinib levels in mRCC patients and retrospectively propose dosage adjustment if required, based upon two distinct metrics: trough concentrations and plasma AUC. The data we have collected here show how using a PK model helps in simulating trough concentrations when drawing blood samples is not feasible at 24 h. As in routine oncology, sampling the patients precisely at the required time can be difficult, especially with ambulatory patients treated with oral targeted therapies, so using a PK model is a valuable strategy, giving much flexibility to perform TDM. For instance, here only 19.4% of the samples were withdrawn precisely at T24 h, but the model was still able to calculate individual PK parameters and simulate virtual trough concentrations. Overall, 54.8% of patients were found to be inadequately exposed with respect to the target trough concentrations of sunitinib and 74.2% of patients were inadequately exposed with respect to the target AUC. Although not statistically significant because of the small numbers of patients, a trend was observed between both Cmin and AUC and early-onset toxicities, i.e., the higher the exposure, the greater the side effects. This observation calls for the implementation of TDM with sunitinib since most of the patients treated with standard dosing failed to be within the target exposure levels. Our observation is in line with clinical observations by Noda et al., who showed that severe toxicities in mRCC patients were associated with elevated residual concentrations of sunitinib and N-desethyl sunitinib [21
]. Regarding patients who were not in the therapeutic window, model-based recommendations for tailored dosing ranged from 12.5 to 100 mg, i.e., a −75% to + 100% change as compared with initial standard dosing. The discrepancy between Cmin-based or AUC-based recommendations, as explained in the “model simulation” section below, were mainly due to the fact that simulated AUCs were always slightly higher than simulated trough concentration values. Of note, at the bedside, a change in dosing was primarily left to the clinician’s choice, i.e., based upon clinical observations, and not upon exposure levels or subsequent model’s proposals for customized dosing. Interestingly, and although numbers were small, patients with correct baseline exposure had an 80% response rate eventually—whereas patients who were not correctly exposed at baseline had response rate of only about 22.2%, i.e., 3.5 times lower. Very interestingly, the subset of patients with inadequate sunitinib exposure at baseline for whom doses were changed in an empirical manner showed a 72.7% response rate, i.e., close to the values of the patients with the right exposure levels upon standard dosing. This suggests that exposure could be a critical factor for sunitinib efficacy, and that increasing dosing could be an actionable item to improve response rates in patients with low drug levels. Of note, the fact that in this study, globally, no relationship between baseline exposure (i.e., Cmin or AUC) and clinical efficacy was found can be partly explained by the fact that, here, drug exposure was solely measured at baseline, and not after subsequent empirical changes in dosing. As efficacy was evaluated 3 months after that treatment had started, the subsequent changes in dosing in 53.8% of patients were thus a major confounding factor when trying to find a relationship between initial drug exposure and clinical outcome 3 months later. In this respect, here, the collected data on efficacy and the possible usefulness of TDM plus adaptive dosing to improve efficacy cannot be conclusive, an observation already reported by others for mRCC patients [14
]. This calls for implementing longitudinal monitoring of sunitinib concentrations, especially when patients are likely to have their dosing changed over time. In this real-world study with sunitinib, we observed that actual changes in dosing were not as drastic and as frequent as compared with what the PK/PD model would have recommended at baseline. This observation is fully in line with clinical reports on routine TDM with other oral targeted therapies published by others [13
]. Despite several flaws related to its single-institute nature and the limited number of patients, plus the fact that no predictive performance validation step could be run, this study presents several findings. First, although not significant, a trend between exposure levels and severe early-onset toxicities was observed. Importantly, the vast majority (82–88%) of the patients who experienced those severe toxicities would have had their dosing reduced following model recommendations, based upon measurement of baseline exposure levels of sunitinib and N-desethyl sunitinib. This suggests that the incidence of treatment-related toxicities could be reduced by implementing TDM-based, model-driven adaptive dosing with sunitinib.