Precision Dosing of Targeted Therapies Is Ready for Prime Time

The treatment of cancer is becoming increasingly personalized and attuned to the molecular characteristics of the tumor. The introduction of imatinib and other molecules has widely been hailed as a breakthrough of precision medicine (1). At the moment, however, these targeted anticancer drugs are still predominantly administered using a one-size-fits-all fixed dose and schedule. Usually, the starting dose is only adjusted in cases of intolerable toxicity. Whilst for some drugs, fixed dosing may be appropriate (e.g., certain monoclonal antibodies; ref. 2), for drugs that exhibit large variability in pharmacokinetic (PK) exposure and have a strong exposure-response or exposure-toxicity relationship, personalized adaptive dosing based on PK measurements, generally known as therapeutic drug monitoring (TDM) or PK-guided dosing, may be superior (3–6).

Reasons why TDM has not been widely implemented yet could include the lack of acknowledgement by regulatory authorities (i.e., drug label change), the lack of access to appropriate PK tests, concerns about the cost-effectiveness, concerns about reimbursement of PK measurements, and treatment costs for higher than approved dosages. Another argument frequently used against the implementation of TDM in routine cancer care is the absence of evidence from randomized controlled trials (RCT). It could be questioned, though, whether these RCTs are really necessary before TDM should be applied in daily clinical practice. In this commentary, we discuss whether randomized studies are needed to confirm the clinical value of precision dosing in oncology.

The current model guiding dose finding in oncology is the maximum tolerated dose (MTD). Fixed doses are often determined by selecting the MTD in early clinical trials. This ancient MTD paradigm sets the dose for all patients, based on only a few, sometimes nonrepresentative patients, and does not take into account their exposure (7, 8). These dose-finding studies generally enroll only a small group of patients (with a median sample size of only 26; ref. 9). Importantly, only a very limited number of patients in the highest dose levels (generally including 6–12 patients) will actually receive and determine the recommended dose for future clinical development. This MTD approach would be pharmacologically justifiable if the concentration-effect curves for toxicity and efficacy would be overlapping, as is the case for classical cytotoxic agents, but which is more often not the case for molecularly targeted agents. For these compounds, the MTD usually holds no particular relation to the optimal effective dose, and may, therefore, very well be suboptimal for this class of drugs (8, 10). Although more innovative designs such as the modified toxicity probability interval (mTPI) and other Bayesian adaptive designs are becoming more widely used, the MTD is still the most commonly used model for dose finding. Puzzlingly, once these small dose-finding trials have been completed, the recommended dose is usually not reconsidered or refined in later trials and subsequent clinical use after approval by the respective authorities. For some molecules, though, phase II studies did continue to examine the optimal dose, of which erdafitinib is a recent example (11). Although there has been an increase in the use of PK–pharmacodynamic (PK–PD) models to guide dose finding, these efforts are still predominantly aimed at identifying a single optimum dose based on population averages, in contrast to an individual level personalized dose.

Many targeted drugs display a solid association between PK exposure and both treatment efficacy and toxicity (5, 6). As PK exposure of anticancer drugs varies greatly between patients receiving the same dose, some patients may be at risk of treatment-related toxicity due to high exposure, while others may experience suboptimal efficacy caused by low exposure. Therefore, using PK exposure to guide dosing decisions in a patient-specific manner (in contrast to using a fixed dose for the whole population) is rational and an important personalized strategy for treatment optimization. Table 1 provides an overview of the extent to which the most frequently prescribed oral targeted anticancer drugs demonstrate exposure-response relationships.

As can be appreciated from Table 1, PK exposure to many oral targeted therapies is related to both efficacy and toxicity. However, not for all oral targeted therapies clinically relevant exposure-response relationships were identified (e.g., enzalutamide and osimertinib). This may be explained by a plateau in the exposure-response curve for these drugs, that are dosed at the flat end of this curve. For some molecular targets (i.e., BRAF and EGFR) the therapeutic window appears to be wider than for others (i.e., ALK, MEK, and VEGF). Also, newer generation kinase inhibitors may have a more robust formulation resulting in reduced variability.

It is thus essential to study for which oral targeted therapies precision dosing holds promise, and for which it might not be worthwhile. Before solid conclusions can be drawn on the absence of an exposure-response relationship, it should be ensured that the study has sufficient power to demonstrate this. Otherwise, absence of evidence is not evidence of absence, as is the case for several underpowered exposure-response analyses for endoxifen (12–14).

We and others have previously summarized the available data supporting PK-guided dosing of anticancer drugs (3, 5, 6, 15–17). Moreover, prospective clinical trials have demonstrated the safety and feasibility of PK-guided dosing for several agents (18–25). When possible, cost-neutral strategies to optimize exposure could be applied, i.e., administration of the drug with food or optimized time of intake (20, 26). In fact, from the patient and treating-physician perspectives, PK-guided dosing requires no complicated procedures and consists of simple and convenient interventions. For most compounds, samples could be collected 4, 8, and 12 weeks after start of treatment and every 12 weeks thereafter (except from compounds with intermittent dosing schedules or a long elimination half-life), which could be combined with regular visits to the outpatient clinic and blood sampling for routine safety monitoring. This would thus not be expected to negatively affect quality of life. Alternatively, methods for self-sampling at home could be developed so that results are already available when the patient visits the outpatient clinic. If a dose adjustment is made, the next PK sample could be drawn after 4 weeks for most compounds, so this could again be combined with a regular visit. Although the target exposure is mostly based on a certain trough level (i.e., concentration right before administration of the next dose), it is often sufficient to obtain a single blood sample at a random time point, as trough levels can then be estimated (27, 28). From a financial perspective, these expensive drugs should be supplied for one month at a time, so that the ordered amount is completed by the time a potential dose adjustment would be made. In this way, no medication needs to be thrown away, preventing unnecessary cost implications for the patient or payer.

Several cost-effectiveness analyses have been performed for oral targeted therapies based on retrospective data (i.e., abiraterone, imatinib, and tamoxifen), and have demonstrated TDM to be cost effective (29–31).

Axitinib is an interesting example for which precision dosing is already being performed, with dose titration based on toxicity being included in the drug label (32). In fact, exposure to axitinib is related to both efficacy and toxicity (33). Dose titration based on toxicity may result in getting patients within the right exposure range, without measuring PK but by using toxicity as a surrogate for exposure instead. Similarly, toxicity-adjusted dosing was demonstrated to be feasible for sunitinib as well (34). However, treatment tolerability is only part of what should be aimed for, i.e., we want to maximize efficacy without risking intolerable toxicity by dosing patients within the therapeutic window instead of above it. An intriguing approach is to use PK–PD modeling to simulate an RCT (instead of actually conducting it) to determine whether PK-guided dosing would be superior to toxicity-adjusted dosing for drugs like axitinib and sunitinib. There are many recent precedents for PK–PD modeling leading to a label change, such as the approvals for less frequent dosing of pembrolizumab and nivolumab (35, 36). Advantages of this approach include that answers on clinically relevant questions can be obtained faster, as conducting the actual clinical trial would take several years, whereas the simulations can be performed within a few months, which would also save considerable amounts of money.

Pharmacogenetically-guided dosing offers an additional strategy for treatment individualization, as the initial dose can be adjusted based on polymorphisms in the genes encoding for metabolizing enzymes and drug efflux transporters. Nevertheless, variability in exposure will remain, as genotype is only one of the many factors affecting exposure. In fact, measured drug concentrations are the translation of all of these factors, allowing for better precision dosing. Ideally, these two approaches should be combined by first selecting the right starting dose based on pharmacogenetics, which could then be further optimized by PK-guided dosing. However, if one of the two approaches should be preferred over the other, PK-guided dosing takes into account all factors that introduce variability, including pharmacogenetics.

Despite the strong rationale for PK-guided dosing in oncology, history suggests that confirmatory RCTs are rarely feasible. Only a very limited number of RCTs of fixed versus personalized dosing have ever been conducted, all of them for classical cytotoxics, underscoring the difficulty and impracticality of conducting RCTs for this specific application (37–40). Obstacles that must be faced when conducting such RCTs include the large number of patients required, often with rare tumor types, resulting in difficulties in patient accrual, which could be further compounded by competitive studies. Also, the lack of interest by industry and third-party funding makes it challenging to secure sufficient financial support for these types of trials.

Although it should be acknowledged that RCTs are currently considered the gold standard for most interventions before implementation in routine clinical practice, RCTs are both impractical and unnecessary when considering TDM. Impractical because large sample sizes would be required, particularly because the majority of patients would not need a PK-guided intervention, which will inevitably increase the costs and burden of the trial. Given the limited scientific rationale for fixed dosing based on the MTD, and the compelling nature of the exposure matching and efficacy arguments (further outlined below), it could be argued that there is no real need to conduct large confirmatory RCTs to investigate individualized dosing for compounds with clearly proven exposure-efficacy relationships (17).

Interestingly, the paradigm of dose adaptation aiming for a PK exposure is already applied routinely in special patient populations, if not at the individual patient level. Dose recommendations for patients with renal or hepatic impairment, pediatric patients, and to guide drug–drug and drug–food interaction management strategies are based on matching the PK exposure of special populations to that of the reference population at the approved dose (41–49). This model for dose adjustment based on exposure matching is generally accepted and part of multiple guidelines of regulatory authorities (41–48), and no or very limited follow-up studies on efficacy endpoints are required to support these dosing recommendations (in acknowledgement of their impracticality).

Yet, the logical extrapolation of PK exposure matching to the level of the individual patient by optimizing exposure in patients with very high or very low concentrations, as is done in special populations, is currently considered unconventional in oncology. This is all the more remarkable because interindividual differences in exposure are often much greater than the average impact of organ dysfunction, drug interaction, or food on exposure (3, 50–52).

If it is accepted that drugs should reach a certain target exposure to be effective in special patient populations, it should similarly also be considered beneficial to adapt the dose to target this exposure for any individual patient, which is the core principal of PK-guided dosing.

An additional argument why confirmatory RCTs should not be required in some cases is that retrospective evidence showing that a certain treatment is ineffective in a subgroup of patients is regarded sufficient to formally recommend exclusion of these patients in treatment guidelines and health authority–endorsed drug labels. For example, retrospective analyses for antibodies against the EGFR (i.e., panitumumab and cetuximab) demonstrated that efficacy is limited to patients with wild-type KRAS wildtype (53, 54), resulting in subsequent updates of the drug labels and treatment guidelines and these drugs no longer being prescribed to patients with KRAS-mutated tumors. Analogous to that, EGFR tyrosine kinase inhibitors erlotinib and gefitinib were initially approved for the treatment of all patients with non–small cell lung cancer (NSCLC), but when retrospective analyses showed that efficacy was confined to patients with activating EGFR mutations (55–57), further development was restricted by targeting these patients. While these two examples concern efficacy, the same logic has been applied regarding toxicity. Recently, the drug label of 5-fluorouracil and capecitabine has been updated to include dose adjustments in patients harboring polymorphisms in the DPYD gene that are associated with an increased risk of severe toxicity (58). In all three examples, no randomized confirmatory trials have ever been performed apparently and these were not considered essential to persuade the field and to change clinical practice, and could well be considered unethical due to the lack of equipoise.

Indeed, retrospective studies have shown that efficacy of several anticancer drugs is restricted to the subset of patients with a PK exposure above certain efficacy thresholds. For example, progression-free survival (PFS) in patients treated with pazopanib with an exposure below the minimum plasma concentration (C_min) target of 20.5 mg/L is similar to the placebo arm of the pivotal trial (59, 60). Analogous analyses can be made for several other anticancer drugs, e.g., abiraterone, imatinib, and sunitinib (18, 20, 21).

Tellingly, TDM has been implemented as routine care based on similar retrospective data and without confirmatory trials having been performed for many other drugs with a narrow therapeutic window (e.g., antiepileptics, anti-infectives, immunosuppressants, and digoxin).

To advance precision dosing from exploratory studies into standard of care, we propose a development pathway as visualized in Fig. 1 (61). First, as part of the learning phase, sufficient information should be obtained to assess whether a compound has a suitable pharmacological profile (i.e., high interindividual variability and relatively lower intraindividual variability). This could be achieved by characterization of the variability in PK exposure within and between patients in the pivotal trials. Also, sound technical support should be in place (i.e., validated bioanalytical methods, convenient blood sampling, and logistics). It is important that bioanalytical methods are publicly available so that they can be reproduced by other centers, and that (inter)national cross-validation programs are initiated. As samples of most compounds are stable under ambient conditions, regional collaborations can be a viable option in clinical practice. Most available bioanalytical assays measure the total drug concentration (i.e., protein-bound plus free drug). As exposure-response analyses are also performed using total drug concentrations, the total drug concentration could be used as a surrogate for the pharmacologically-active free drug concentration. It should be kept in mind, though, that under certain circumstances the unbound fraction can change (e.g., organ dysfunction), in which case the total drug concentration may no longer be a representative measure of exposure. Most importantly, well-established exposure-therapeutic response and exposure-toxicity relationships should have been demonstrated (4). These exposure-response analyses could take multiple forms. For some drugs, elaborate model-based analyses are available (62), while only simple analyses (e.g., using exposure quartiles) have been used for others (63). These more straightforward analyses could be incorporated as prespecified analyses as an endpoint in pivotal phase III trials, as only limited additional sampling is needed (e.g., C_min), and patients will already need blood sampling for routine safety monitoring. This could be further incentivized by regulatory authorities. In the case of combination therapies, exposure-response relationships would preferably be determined in patients treated with the combination regimen, as it could be imagined that different target exposures would apply for combination therapy compared with monotherapy. Then, based on what has been learned, PK targets and dosing algorithms should be defined. Dosing algorithms should take into account the possibility of cost-neutral interventions (i.e., concomitant intake with food or optimizing the dosing schedule; refs. 20, 26, 64), the MTD or maximum administered dose in phase I studies, and the available capsule or tablet sizes.

Subsequently, as part of the confirming phase, the safety and feasibility of PK-guided dosing strategies should be demonstrated in clinical trials, preferably in a real-life setting, in which TDM is applied in clinical practice, but in which no control group needs to be used. Several trials in this category have already been performed, for example, for abiraterone, imatinib, pazopanib, sunitinib, and tamoxifen (18–22), and this approach is currently also being applied in a prospective study on TDM by the Dutch Pharmacology Oncology Group (DPOG; ref. 65). In these studies, individualized dosing may be demonstrated to be logistically feasible, well tolerated, and to increase the proportion of patients within the target exposure range (i.e., superiority of TDM compared with fixed dosing). The standard for TDM-guided dosing that needs to be applied in these studies is superiority to fixed dosing with regard to the proportion of patients reaching the target exposure, either compared with a control group or to a historical fixed-dosing cohort with available exposure data. As explained in more detail in the sections above, demonstrating superior efficacy might not be feasible (i.e., large sample size) and unnecessary (i.e., efficacy at the target exposure has already been demonstrated to be superior).

At this point, it has been demonstrated that exposure is related to a relevant clinical outcome including efficacy or toxicity, and that exposure can be improved by PK-guided dosing. Hence, it could be logically assumed that individualized dosing would result in better treatment outcomes, at the same level of evidence as FDA and European Medicines Agency (EMA)–endorsed recommendations for dose adjustments in special patient populations. Thus, sufficient evidence for individualized dosing has now been obtained to apply PK-guided dosing in clinical care.

Clinical outcomes and PK data of these patients should then be collected in clinical practice and should be used to further optimize the TDM approach, in order to ensure that with more patients the TDM strategy keeps improving, comparable with a Bayesian algorithm. To facilitate the implementation of individualized dosing, efforts should be made to remove practical barriers as much as possible. This could be done by ensuring that a sound infrastructure is in place for sample collection, shipment, measurement, interpretation, and reporting of the results, with a short turn-around time. As PK-guided dosing recommendations should ideally also be included in the drug label, it is important to work together with the regulatory authorities from an early stage in this pathway.

Currently, there appears to be a missing link between the performed feasibility studies and the widespread implementation of PK-guided dosing (66). Many factors play a role here. First, bioanalytical assays to measure drug concentrations might not be available or a solid infrastructure might be lacking. Second, knowledge about PK-guided dosing and skills to calculate or estimate exposure may be insufficient due to the absence of education and training. Close collaboration between medical oncologists, clinical pharmacologists, and pharmacists is essential here. Third, costs might be a concern. Obviously, dose increases will result in higher treatment costs. However, the costs of drug measurements itself is negligible compared with the total treatment costs. For example, in the Netherlands, measurement of one PK sample costs $90 (i.e., $540 per year), whereas treatment costs of many oral targeted therapies exceed $120 per day (i.e., $43,800 per year). Cost-effectiveness analyses need to be performed to demonstrate the additional value of PK-guided dosing. These are already performed for abiraterone, imatinib, and tamoxifen, for all of which PK-guided dosing was cost-effective (29–31). To limit the financial burden of these expensive treatments, it is essential to perform cost-neutral interventions (i.e., concomitant intake with food or optimization of the dosing schedule) when possible. The above-mentioned factors should be addressed to stimulate the implementation of PK-guided dosing in clinical practice. Furthermore, acknowledgement by regulatory authorities (i.e., change in drug label) would support implementation in routine clinical practice, as inclusion in the label would likely produce an incentive for companies to develop commercially available PK tests as a companion diagnostic.

According to the label, abiraterone acetate should be administered at a fixed dose of 1,000 mg once daily under modified fasting conditions. However, it has been demonstrated that patients with a C_min ≥ 8.4 ng/mL have a significantly better PFS (i.e., 12.2 vs. 7.4 months; ref. 67). This has later been confirmed in an independent patient cohort (68). In clinical practice, 35% to 42% of patients do not reach this efficacy threshold and might thus benefit from PK-guided dosing (67, 68). As it was known that concomitant intake with food resulted in a relevant increase in abiraterone exposure, the DPOG-TDM study investigated whether PK-guided dosing using a food intervention as a first step in case of low exposure was feasible in clinical practice. In this study, 20 out of 32 patients had an abiraterone C_min < 8.4 ng/mL at a certain timepoint during treatment. These patients were recommended to take abiraterone acetate concomitant with a light meal or a snack, which resulted in an increase in C_min from 6.9 ng/mL to 27 ng/mL without additional toxicities. This intervention led to adequate exposure in the majority of patients (i.e., 87.5%; ref. 20).

For the ALK inhibitor alectinib, PFS was significantly longer in patients with C_min ≥ 435 ng/mL compared with patients with an exposure below this threshold (69). At the approved dose of 450 mg twice daily, 37% of patients are underexposed and treatment outcomes for this subgroup may be improved by PK-guided dosing. An RCT is planned comparing fixed dosing versus TDM-guided dosing of alectinib (the Adapt Alec Trial), in which a total of 220 patients need to be enrolled, which is almost comparable with the phase III trial (i.e., ALEX study, in which 303 patients were included) and accrual is planned to take 4 years (70, 71).

Exposure to imatinib has been related to efficacy for both chronic myeloid leukemia and gastrointestinal stromal tumors with C_min thresholds of 1,000 ng/mL and 1,100 ng/mL, respectively (72–74). An RCT comparing fixed dosing and TDM-guided dosing failed to demonstrate the additional value of TDM-guided dosing because treating physicians did not implement the recommended dose increases in patients with low exposure (18). However, several studies have since then shown that PK-guided dosing of imatinib is feasible in clinical practice (23, 25).

For pazopanib, trough levels ≥20 mg/L have been linked to prolonged PFS in different treatment settings (59, 75, 76). In clinical practice, 16% to 30% of patients do not reach this threshold (59, 76). Furthermore, PK-guided dosing of pazopanib was demonstrated to be feasible (19). Apart from dose increases, cost-neutral interventions can be applied as first steps in case of low exposure (i.e., splitting intake moments from 800 mg once daily to 400 mg twice daily and concomitant intake with food; refs. 26, 64, 77).

Fixed dosing strategies are suboptimal in the era of precision medicine. PK-guided dosing is a promising tool to lead to more rational, personalized, and optimal treatment with targeted anticancer drugs. With the available data, evidence for TDM is already quite robust. Together with continuous optimization of the TDM approach, it is justified to implement individualized dosing based on evidence deduced from exposure-response analyses and feasibility trials, without the need for a large confirmatory RCT. We propose a development pathway to demonstrate the clinical relevance of precision dosing in oncology.

R.B. Verheijen reports other support from Johnson & Johnson, AstraZeneca, Galapagos, and Chinook Tx outside the submitted work. R.H.J. Mathijssen reports grants from Astellas, Bayer, Boehringer Ingelheim, Cristal Therapeutics, Novartis, Pamgene, Pfizer, Roche, Sanofi, and Servier outside the submitted work. N. Steeghs reports grants from AB Science, AbbVie, Actuate Therapeutics, AIMM Therapeutics, Amgen, Array, AstraZeneca/MedImmune, Bayer, Blueprint Medicines, Boehringer Ingelheim, Bristol Myers Squibb, Cantargia, CellCentric, Cytovation, Deciphera, Ellipses Pharma, Genentech/Roche, GlaxoSmithKline, Incyte, Lilly, Merck Sharp & Dohme, Merus, Molecular Partners, Novartis, Pfizer, Pierre Fabre, Roche, Sanofi, Taiho, and Takeda outside the submitted work. A.D.R. Huitema reports grants from Roche, Novartis, and Pfizer during the conduct of the study. No disclosures were reported by the other authors.

J.H. Beumer was supported by NCI grants UM1CA186690 and P30CA47904.