PK sampling design for first-in-human (FIH) Phase I oncology studies is frequently treated as a protocol finalization task rather than a protocol design task. The PK section of the protocol is written by the clinical pharmacologist in the last two to four weeks before IND submission. By that point, critical decisions that constrain PK sampling design - the visit schedule, the blood volume policy, the number of PK-intensive cohorts, the maximum clinic hours per day - have already been locked by the study team in the main protocol body. The PK sampling plan is retrofitted to work within those constraints rather than the constraints being designed to support an optimal sampling strategy.
This matters because the quality of the PK data collected in a Phase I study determines whether the population PK model built from that data will be useful for subsequent dose decision support. This article identifies the five decisions that most affect PK sampling scheme quality in Phase I oncology studies - and the ones that need to be made before, not after, the visit schedule is locked.
Decision 1: Full Intensive vs. Sparse Sampling Strategy
The first decision is whether cycle 1 (and possibly cycle 2) will use intensive PK sampling (8+ samples per dose) or a sparse sampling approach (2-4 samples per dose). This decision drives everything else in the PK sampling design and should be made jointly by the clinical pharmacologist, sponsor, and principal investigator before the visit schedule is drafted.
The argument for intensive sampling in Phase I is strong: you have a new drug with unknown pharmacokinetics, and collecting a full concentration-time profile from at least the first cohort allows model-independent (noncompartmental analysis) AUC and half-life estimation without depending on a model that has not yet been built. The argument for sparse sampling even in Phase I is that intensive sampling is logistically demanding, increases patient burden, and in multicenter trials is difficult to standardize across sites. The hybrid approach - intensive sampling in sentinel subjects (first two patients per cohort at the lowest dose levels) and sparse sampling for all subsequent patients - provides model-building data from sentinel subjects while reducing burden in the rest of the cohort.
Decision 2: Sampling Window Relative to the Elimination Half-Life
Sampling window design for a new drug requires an estimate of the expected half-life. For small molecule drugs, preclinical PK data in at least two species (rat and dog or monkey) is available before FIH and provides a rough half-life estimate. The human half-life will differ from preclinical estimates, sometimes substantially, but the preclinical data constrains the expected range enough for practical sampling window design.
The minimum sampling window requirement is typically 3-4 half-lives post-dose to characterize the elimination phase. For drugs with preclinically estimated human half-lives of 4-8 hours, this means samples through 24-32 hours post-dose - achievable in a one-day clinic visit with an overnight collection from a pre-connected catheter or via patient-reported home collection. For drugs with half-lives in the 24-72 hour range, the window extends to day 3-5, requiring multiple sampling days or patient diaries for home collections.
The visit schedule in the main protocol determines when patients can be scheduled for sample collection. If the protocol specifies a single clinic visit for cycle-1, day-1 PK collection and the drug's half-life is 48 hours, the sampling window is by design inadequate for terminal elimination characterization. This constraint, once locked into the visit schedule, cannot be corrected without a protocol amendment. It must be identified before the visit schedule is finalized.
Decision 3: Number of Cycles with PK Sampling
For most agents, at least two cycles of PK data per patient are valuable for population PK model building. Cycle 1 data characterizes distribution and elimination from a clean baseline. Cycle 2 data (typically on day 1 of cycle 2, using trough or limited samples rather than a full profile) provides individual steady-state AUC information and allows assessment of whether clearance changes from cycle 1 to cycle 2 - which it does, meaningfully, for agents that induce or inhibit their own metabolism (CYP autoinduction is not rare in oncology drugs) or for agents that alter renal function cumulatively.
If the protocol specifies PK collection only in cycle 1, multi-cycle PK data will not be available for the population model. The model built from cycle 1 data only will not capture time-varying covariates or cycle-to-cycle clearance changes, limiting its utility for dose individualization in subsequent cycles. The decision about multi-cycle PK sampling must be made at protocol design time because it affects visit schedules and informed consent language across all cycles, not just cycle 1.
Decision 4: Bioanalytical Method Validation Timeline
PK sampling is only useful if the bioanalytical method (typically LC-MS/MS for most small molecules) is validated before the first samples are collected. Full validation per FDA Bioanalytical Method Validation guidance takes 8-12 weeks for a new analyte. This timeline is often not integrated into Phase I startup planning, leading to one of two common problems: samples collected before method validation is complete (stored in the correct conditions but not analyzed, creating a risk of sample degradation if storage is prolonged), or first-patient dosing delayed while bioanalytical validation catches up.
The bioanalytical method development timeline should appear on the critical path for Phase I startup. It is a prerequisite for PK data collection, and it takes long enough that a two-week slip in the preclinical reference standard synthesis (the compound used for the calibration curve) can delay the entire PK program by six weeks. Clinical pharmacologists who are not engaged in the startup timeline planning process often do not catch this dependency until it becomes a crisis.
Decision 5: Handling of PK Data in Real Time vs. Post-Trial
The fifth decision is whether individual PK data will be reviewed in real time during the trial to inform dose decisions, or whether PK data will be collected and analyzed post-trial for population PK modeling only. This is not a binary choice - most Phase I designs use a blend - but the default position should be specified at protocol design time.
Real-time review of PK data adds operational complexity: results must be reported from the bioanalytical laboratory quickly enough to influence dosing decisions, and there must be a defined process for who reviews the data, in what format, with what decision authority. But real-time review provides clinical information that cannot be recovered post-trial: a patient with an unusually high Cmax on cycle 1 whose toxicity is caught early because the PK flagged the outlier exposure is a patient whose outcome is better because of the real-time review. Post-trial analysis of that same data, however thorough, cannot retroactively change what happened in cycle 1.
As we describe in our article on sparse sampling and MAP Bayesian estimation, the infrastructure for real-time PK review - a TDM platform connected to the bioanalytical laboratory, with MAP Bayesian updating and cycle-to-cycle AUC tracking - is the same infrastructure that supports dose individualization in Phase II. Building that infrastructure during Phase I, even if only used for safety monitoring rather than active dose adjustment, provides a validated system that is immediately available when Phase II begins.
The Integration Question: When Does PK Connect to Dose Decisions?
In most Phase I oncology protocols, PK data does not drive dose decisions in the dose-escalation phase - BOIN or 3+3 decisions are made based on DLT observations. But PK data is increasingly used as supporting information for dose-escalation decisions, particularly in exposure-response designs where a target Cmax or AUC is used alongside DLT observations to define an optimal biological dose range.
If the protocol is designed for exposure-informed escalation, the PK sampling scheme must produce timely enough data to be available at the dose-escalation review. If the bioanalytical laboratory's turnaround time is 5 business days, the dose-escalation committee meeting must be scheduled at least 5 days after the last PK sample in the cohort is drawn. These timing dependencies are protocol design questions that cannot be resolved by the clinical pharmacologist alone - they require coordination with the bioanalytical laboratory, the data management team, and the sponsor's operations team before the protocol is finalized.
Conclusion: PK Sampling Design Is Protocol Architecture, Not a Footnote
The five decisions above - sampling strategy intensity, sampling window versus half-life, multi-cycle sampling, bioanalytical timeline, and real-time versus post-trial review - are not details to be sorted out at protocol finalization. They are architectural decisions that determine whether the Phase I study will produce PK data of sufficient quality for population model building, real-time safety monitoring, and eventual Phase II dose optimization.
Clinical pharmacologists need to be involved in Phase I startup planning from the protocol concept stage, not called in to write the PK section after the visit schedule is locked. The cost of retrofitting an adequate PK sampling plan around a visit schedule that was not designed to support it is paid in data quality for the entire drug development program.
DoseMind provides pre-protocol PK sampling design consultation as part of our Phase I integration package. Contact the team at hello@dosemind.com to discuss your study's requirements before the visit schedule is finalized.