Correlating Drug Exposure with Drug Efficacy and Toxicity

Once it is established that the blood levels of a novel oral anticancer drug vary significantly from one patient to another and cannot be practically estimated by means other than direct measurement, the relationship between systemic drug exposure and drug efficacy and drug toxicity should be investigated (Figure 1, Stage 2). Such studies can provide evidence in favor of TDM and define the target (therapeutic) exposure range by demonstrating increased treatment failures at sub-therapeutic exposures and increased toxicity at supra-therapeutic exposures. After all, if low drug levels cannot predict treatment failure and high drug levels cannot predict adverse drug effects, TDM will be of limited value.

Measuring Efficacy

A drug’s effect can be measured in various ways. In oncology, efficacy endpoints such as response rates, progression free survival, or overall survival are typically used.15 These endpoints are commonly derived from histologic and/or radiologic tumor evaluations, but other assessments such as circulating tumor cells, circulating cell-free tumor DNA, microRNA, or protein markers can also be used. In addition, pharmacodynamics markers (e.g. measurable molecules corresponding to drug target inhibition or downstream pathway activity) may be available for some drugs, which may enable close to real-time PD monitoring.16 Thus, for studying the relationship between drug levels and effect, one or more efficacy endpoints, alone or in combination with PD biomarkers, can be used.1–3Measures of drug effect can be represented by dichotomous variables, such as frequency of occurrence, or by continuous variables such as concentrations of tumor markers.
To help decrease methodological heterogeneity in measuring drug response, an international multidisciplinary working group developed RECIST (Response Evaluation Criteria in Solid Tumours) criteria for the evaluation of tumor burden. 17 These criteria describe standardized approaches of solid tumor size measurement, primarily using imaging techniques, and define the outcomes of complete response (CR), partial response (CR), stable disease (SD), and progressive disease (PD). 17
The duration of time it takes to achieve the chosen efficacy endpoints is also important for study design. The lag in time between when drug exposure is initially assessed and when clinical response can be detected is typically on the order of weeks to months or even years. On these timescales, the initial exposure assessment may no longer accurately represent the total drug exposure over the course of treatment. Thus, in studies with long treatment duration, serial exposure assessments over time may be particularly useful for capturing the overall drug exposure more accurately.
The pre-treatment dynamics of outcome measures is also important to consider. For example, high heterogeneity in the pre-treatment rates of tumor growth and trajectories of biomarker levels between individuals in a study population is likely to result in high inter-individual variability in these measures during treatment. Consequently, the statistical power of the study suffers, requiring increased numbers of participants. As a further example, a small decrease in the rate of tumor growth after treatment initiation may be interpreted as disease progression in a patient with a fast-growing tumor and as stable disease in a patient with a slow-growing tumor. This suggests that several pre-treatment assessments of the patient’s baseline tumor size or biomarker levels, as well as the use of a control group, may help more accurately characterize the effect of the drug.

Measuring Toxicity

The side effects that occur during treatment can be a consequence a drug’s effect, related to a drug’s unwanted but expected off-target effect, or they can be idiopathic. Depending on the mechanism, side effects can manifest relatively quickly, within hours or days, or can take months to develop. Similar to the assessment of drug efficacy discussed above, the prevalence and timing of drug toxicity will impact the study design with respect to the number of participants required, the frequency of toxicity assessments, and the duration of toxicity monitoring.
Drug-related toxicity often correlates with drug dose and typically subsides following dose decrease or interruption. However, the occurrence of adverse drug events may also seem stochastic and they may appear and disappear without temporally related dose adjustments. In this context, variations in drug exposure (at the same prescribed dose) may correlate with toxicity. Thus, serial exposure assessments over time may be particularly helpful for relating fluctuations in drug trough levels to toxicity symptoms, especially in individuals concurrently treated with other drugs prone to interactions or toxicities of their own.
The approach to capturing and quantifying adverse drug effect data must also be considered. Self-administered patient questionnaires (patient reported outcomes or PROs) may be used to supplement clinical assessments.18 Toxicity may be represented as dichotomous (either present or not), categorical (based on severity) or even continuous (e.g. elevation in blood pressure) variables. In addition, the National Cancer Institute (NCI) provides Common Terminology Criteria for Adverse Events (CTCAE) to help standardize the description and grading of adverse events.19,20
Of note, drug toxicity itself can sometimes be used to guide dose optimization. Such “dosing to toxicity” strategies have long been used for chemotherapy but may also have a role in dosing of oral targeted small molecule drugs.16 A relevant review on the susceptibility to adverse drug reactions was recently published in this journal.21 The described susceptibility factors included the type of immunological reaction, genetics, age, sex, physiological changes (such as pregnancy), exogenous factors (such as interacting drugs), and diseases. Notably, the authors highlight that there may be significant inter-patient variability in the dose-response curves not only for drug benefits but also for harm, providing an illustration of how some (hypersusceptible) patients may experience toxicity at drug concentrations insufficient for efficacy.21 Importantly, this is one context in which TDM has a clear advantage: using a dosing to toxicity approach for hypersusceptible patients results in continued treatment with drug doses that are ineffective, while TDM informs a change in therapy.

Standardization of Assays and Methods

For the vast majority of new drugs there are no FDA-approved quantitative assays. Instead, new drugs are typically quantified by assays developed in individual laboratories, known as laboratory developed tests (LDTs). The required levels of quality assurance for these tests vary widely, in part depending on the laboratory’s local and other regulations (e.g. CLIA, GLP). In addition, LDTs developed in different labs may employ distinct methodologies (e.g.immunoassays, liquid chromatography-tandem mass spectrometry, etc.). Taken together, this can lead to significant inter-laboratory and sometimes even intra-laboratory differences in results. External proficiency testing programs can help minimize such differences but, more often than not, such programs do not exist for new drugs. Therefore, it is important to be aware that lab-to-lab differences in the measurement of drug levels may be a significant contributor of noise in TDM studies. Utilizing the same laboratory with a thoroughly validated method for all drug level measurements for a precision dosing study may be a worthwhile consideration.
The same holds true for methods and approaches for quantifying drug effects. The challenges associated with bioanalytical measurements of pharmacodynamics biomarkers are analogous to those for drug assays. Similarly, there may be significant inter-institution and even intra-institution variability in imaging or anatomical techniques used for tumor assessments. Again, this variability may be a considerable source of noise in TDM studies.
In order to improve experimental reproducibility as well as applicability and translatability of results, attempts should be made to standardize the assays and methods. As mentioned above, RECIST criteria can help standardize solid tumor size measurements and NCI’s CTCAE can help standardize assessments of drug toxicity.17,19Similarly, guidance from the NCI also exists for the development and incorporation of biomarkers studies in drug trials.22The standardization of assays for oral small molecules for cancer is lagging, although some proficiency testing programs have recently become available.23

Study Design Considerations for Exposure-Response Relationships

In contrast to biomarker studies, which can obtain useful data through retrospective analysis of repository samples collected during routine patient care, TDM studies aiming to investigate the correlation of drug levels with effects and toxicity will likely require prospective collection of samples. This is because the relative timing of drug intake and blood sampling is critically important to interpreting the obtained drug level results. In samples without associated data on timing of last drug intake (most repository samples), the drug levels may represent trough, peak, or intermediate time points. In addition, exposure-response relationship studies are typically observational (no dose adjustment based on results) rather than interventional, because dose adjustment after blood level measurement but before response measurement would confound interpretation of results. A large number of examples of such studies for oral small molecule anticancer drugs have been summarized in numerous reviews.1–5
Although the necessity to conduct such studies prospectively presents certain challenges (e.g. obtaining preliminary data for a grant proposal and long accrual times), prospective studies tend to be less prone to certain types of biases such as recall bias and non-recorded confounders. Other types of bias, such as selection bias, can still occur in prospective studies.24,25
As discussed above, numerous choices are available with respect to the frequency and duration of exposure sampling as well as the timing, prevalence, and quantification of clinical endpoints and toxicity. Consequently, study design and power calculations should take into account the temporal relationships between drug levels and efficacy and toxicity as well as the anticipated frequency of measured outcomes and adverse events. Although there are numerous resources to guide power calculations for PK studies, the literature on power calculations for TDM studies seems to be lacking.26,27
Data from the exposure-response relationship studies can be described using various forms of regression analysis or more simply by comparing the outcomes of patients stratified by, for example, Cmin quartiles or deciles.25,28–30Ultimately, the goal of such studies is to define a therapeutic exposure range below which there is increased risk of lack of efficacy and above which there is increased risk of toxicity.1–5
It should be mentioned that in vitro and pre-clinicalin-vivo experiments may also demonstrate concentration-effect relationships and can be used to supplement the results obtained in clinical studies. 4,5 For solid tumors, blood level measurements may be complemented by in vivo studies that also measure drug concentrations in tumor tissue.31