Table 1.

Executive summary

Topic	Recommendations and comments
Exposure metric considerations	• This is a key component of any analysis and may include dose, concentration, time-averaged concentration, time above a threshold, or area-related metrics • Be careful of dose adjustments and dropouts, and their effect on exposure metrics • In choosing a metric consider whether the relationship may be a direct effect (like nausea/vomiting) or a time delay effect like tumor growth
Safety and efficacy endpoints: categorical endpoints
• Examples: presence/absence of nausea/vomiting, presence/absence of grade 3 or higher neutropenia, RECIST	• Primarily logistic modeling (or some modification thereof) is used • 10 Events/predictive variable is recommended for precise estimation of regression coefficients • Consider the confounding effect of drug clearance on the outcome (higher clearance may lead to poorer treatment outcomes); use clearance as a covariate in the model • If time-dependent categorical endpoints are of interest, use Markov models
Safety endpoints
• Time course of myelosuppression	• Empirical approaches (e.g., maximum % decrease from baseline vs. AUC) o Empirical models have limited values extrapolating outside the doses or dosing regimen tested o Need to use some integrated measure of exposure like AUC because of time delay between first dose and peak effect • Semi-mechanistic models, like a Friberg model, allow assessment of time course of myelosuppression
• QTc interval prolongation	• Recommend following Garnett white paper [63] • Oncology trials may not be able to study 2x-above therapeutic exposure for safety reasons • 20 ms is generally accepted as the upper safety threshold compared to 10 ms in healthy volunteers
Efficacy endpoints
• Time to event: survival	• Can use nonparametric (Kaplan–Meier), semiparametric (Cox proportional hazard), or parametric (accelerated lifetime) models • KM curves often assessed by quartile of exposure vs OS or PFS • CPH models should include other prognostic covariates, like baseline tumor size and drug clearance, for controlling these confounding effects • May be subject to inherent selection bias and immortal time effects
• Tumor growth dynamics	• Allows for a better understanding of the entirety of a patient’s tumor burden growth/shrinkage time-course to assess the possible impact of dose or schedule selection on disease response • Many different models to choose from • Secondary parameters may be more intuitively linked with survival outcomes in time-to-event analyses • Informative censoring may affect parameter estimates • Pretreatment tumor growth trajectories may allow better interpretability • Rely on prespecified target lesions which may not be indicative of overall disease burden
Hematologic malignancies
	• All recommendations in the above sections can be applied to hematologic malignancies, where total target tumor size is replaced by the appropriate continuous tumor burden metric for that particular malignancy • May require bounded endpoint models, e.g., minimal residual disease
Tumor biomarker and disease progression
	• Example: PSA kinetics or circulating tumor cells • May require semi- or mechanistic models to explain
Immunogenicity
	• May affect both exposure and response (safety/efficacy) • Need to consider anti-drug antibodies (ADAs) vs neutralizing antibodies • Covariate analysis may include binary grouping (presence or absence of ADAs) or titer in models
Cell therapies
	• CAR-T cells display strange kinetics compared to traditional small molecules or biologics • 4 Phases including distribution, expansion, contraction, and persistence • Cannot use allometric principles for scaling of doses

Each topic in the table is covered in the text. Bulleted recommendations and comments are excerpts from the text. Read text for further details