From a Biological Question to a Phase I Decision
A Phase I biomarker result is useful only when it answers a question that can change what the development team does next. “Did the dose affect biology?” is too broad to guide a decision: the intended biological step, the measurement, the comparison, and the decision criterion all need to be specified. A target-engagement question, for example, asks whether the administered regimen measurably engages its intended molecular target. A pathway-modulation question asks whether a downstream pathway readout changes. An early-pharmacology question asks whether a functional biological response is observed. These are different questions and can require different biomarker contrasts.
The aim here is to map each question to an estimand: the precise quantity the study is designed to estimate. The estimand is not the observed result, and it is not the rule used to decide whether to progress. Keeping those three elements separate makes a Phase I decision interpretable even when the sample is small and uncertainty remains.
For a typical Phase I question, the population might be enrolled healthy participants or participants with the disease of interest; the treatment condition includes dose and regimen; and the variable includes the biomarker and its measurement time. A summary measure could be a mean change from baseline, a treatment-placebo contrast in mean change, or a difference in response proportions. The event-handling strategy matters: for example, the protocol may define the target as the effect while participants remain on assigned treatment, or may specify what is done when treatment is stopped before the scheduled measurement. A measurement that is simply unavailable is not automatically equivalent to no biological response.
Three Questions, Three Estimand Targets
The biological chain can be represented in broad terms as drug exposure, target engagement, pathway modulation, and a functional pharmacological response. A Phase I study may measure one or more points along that chain, but the question must say which point is intended. The terms below identify the purpose of the estimand, not a claim that one result establishes every later biological step.
| Phase I question | Example biomarker variable | Useful contrast | Illustrative decision |
|---|---|---|---|
| Target engagement | Target occupancy, in percentage points | Mean change from baseline with drug minus mean change with placebo at a specified time | Is the estimated engagement large enough to justify testing a higher or repeated regimen? |
| Pathway modulation | Concentration of a downstream pathway marker, in specified assay units | Difference between treatment groups in mean change at a prespecified time | Is the pathway readout changing in the intended direction by a prespecified amount? |
| Early pharmacology | Presence or absence of a defined functional response | Difference between treatment groups in the proportion meeting the response definition | Is the observed response frequent enough to support further pharmacology evaluation? |
The contrast should follow from the question. An absolute post-dose measurement can mix treatment effects with baseline differences. A within-participant change uses each participant’s baseline, but by itself does not separate a drug-related change from a time trend or other shared influence. A placebo-adjusted change can address that shared trend when the study design and data support the comparison. For any contrast, specify the direction that represents the intended effect and retain the measurement units.
Why Compare Changes Between Groups?
Suppose participants are assigned to a drug regimen or placebo, and a continuous biomarker \(Y\) is measured before treatment and at a prespecified post-dose time. Let \(Y_0\) denote the baseline measurement and \(Y_t\) the measurement at that time. The treatment estimand can be the difference between groups in mean change:
Here \(A=d\) denotes assignment to the specified drug regimen and \(A=0\) denotes assignment to placebo. The parameter \(\tau\) is measured in the units of \(Y\). The corresponding sample estimator is the observed mean change in the drug group minus the observed mean change in the placebo group.
The reason for this contrast can be seen by writing each group’s mean change as a treatment-related component plus a common time-related component. If the drug causes a mean change \(\delta\), while both groups share a mean time trend \(c\), then the expected drug-group change is \(\delta+c\) and the expected placebo-group change is \(c\). Subtracting gives:
This cancellation is a justification for the comparison, not a guarantee that every bias disappears. It relies on an appropriate comparator and design, consistent measurement, and adequate handling of missing data and events after assignment. In a randomized study, random assignment supports a causal comparison of the assigned regimens in expectation, subject to those conditions. Without a suitable comparator, a change from baseline alone should not be described as the drug’s effect.
Worked Examples: Specify the Contrast and the Decision
Worked Example: Target-Engagement Estimand
In a hypothetical randomized Phase I study, 8 participants receive a single drug dose and 8 receive placebo. Target occupancy is measured at baseline and 4 hours after dosing, in percentage points. The mean change is 54 percentage points in the drug group and 6 percentage points in the placebo group. The protocol’s illustrative screening rule is to proceed to a repeated-dose engagement assessment if the placebo-adjusted mean change is at least 40 percentage points.
Estimand. The target is the difference between assigned groups in mean change in target occupancy from baseline to 4 hours after the specified dose, among the study population, in percentage points.
Estimate. The observed contrast is:
The estimate is 48 percentage points, which exceeds the illustrative 40-percentage-point screening criterion. Under that prespecified rule, the result supports proceeding to the repeated-dose assessment. It does not establish that the target is engaged at every dose, time, or in every future population. With only 8 participants per group, the estimate is also imprecise; the decision should be accompanied by an uncertainty assessment rather than presented as a precise population truth.
Worked Example: Pathway-Modulation Estimand
In a separate hypothetical study, a pathway marker is measured in normalized assay units at baseline and at a prespecified post-dose time. Ten participants receive drug and ten receive placebo. The mean changes are \(-2.1\) units with drug and \(-0.4\) units with placebo. Lower values represent the intended pathway modulation. The illustrative decision rule is to consider the pathway result supportive if the placebo-adjusted change is at most \(-1.0\) unit.
Estimand. The target is the difference between groups in mean change in the pathway marker at the specified time, measured in normalized assay units.
Estimate. Subtract the placebo mean change from the drug mean change:
The negative estimate is in the intended direction, and its magnitude exceeds 1.0 unit. The result meets the illustrative screening rule for supportive pathway modulation. The rule is specific to this hypothetical development context; it is not a general threshold for pathway biomarkers. The interpretation also depends on the assay measuring the intended marker consistently and on the prespecified sampling time being suitable for the question.
Worked Example: Early-Pharmacology Estimand
In a hypothetical short Phase I study, a functional response is defined in advance as meeting a specified laboratory response criterion at the scheduled assessment. Fifteen participants receive drug and fifteen receive placebo. Nine drug participants and three placebo participants meet the criterion. The illustrative screening rule is to consider a further pharmacology cohort if the estimated difference in response proportions is at least 0.25.
Estimand. The target is the difference in the population proportions meeting the defined response criterion under the drug regimen and under placebo, at the specified assessment.
Estimate. The observed proportions are \(9/15=0.60\) and \(3/15=0.20\). Their difference is:
The estimated difference is 0.40, or 40 percentage points, which exceeds the illustrative 25-percentage-point screening criterion. The rule therefore supports considering a further pharmacology cohort. The estimate is based on only 30 participants, so its uncertainty may be substantial; the decision is a screening decision, not proof that the response rate difference is precisely 40 percentage points.
Diagnostics, Alternatives, and Limits
Before interpreting a result, check that the estimand and the observed data still line up. Confirm that the biomarker, units, time point, treatment regimen, comparator, and direction match the protocol. Review whether baseline and post-dose measurements are available in both groups, whether missing measurements or treatment discontinuations are concentrated in one group, and whether the result depends heavily on a few participants. A decision rule should also specify what happens when data quality is inadequate or uncertainty is too large; otherwise, a numerical threshold can create an appearance of certainty that the design does not support.
The main alternatives answer different questions. An absolute post-dose group difference compares the groups at the assessment time and may be appropriate when baseline values are balanced by design and the analysis is prespecified accordingly. A within-group change from baseline describes change in that group, but without a comparator it cannot distinguish treatment effects from common time trends. A placebo-adjusted change targets the between-group difference in change. For a binary functional response, a difference in proportions directly describes the difference in response frequency; a mean contrast is not suitable for that binary variable unless it is represented through a defined numerical summary.
These contrasts do not make the biological interpretation automatic. A measured change may reflect variability in the biomarker, and an effect on one biological step does not establish an effect on a downstream step. The estimate is also tied to its population, regimen, comparator, and time point. Results from a single dose in healthy participants, for example, should not be silently generalized to a different regimen or patient population. The earlier material on reporting uncertainty in preclinical-to-human predictions is relevant to the broader principle: decisions should acknowledge uncertainty, not hide it behind a point estimate.
Decide whether the Phase I decision concerns target engagement, pathway modulation, or a functional pharmacological response.
Set the population, regimen and comparator, biomarker variable and time, summary contrast, and handling of relevant events or unavailable measurements.
State how the observed data estimate the target quantity, what direction and magnitude would matter, and how uncertainty or data-quality problems affect the action.
Check Your Understanding
Answer each question using the estimand, contrast, and decision concepts above. Brief solutions follow the questions.
- A drug group’s biomarker decreases by 2 units, while the placebo group’s biomarker decreases by 1 unit. What is the placebo-adjusted difference in change, and what can it mean if a decrease is the intended direction?
- Why is the statement “the biomarker changed from baseline, so treatment caused the change” not justified by a single-group study?
- A Phase I team wants to evaluate the proportion of participants who meet a functional response definition. What summary contrast could estimate the treatment effect on that endpoint?
- Name four elements that should be specified before a biomarker estimate is used for a development decision.
- If an estimate meets a prespecified screening threshold but is based on a small cohort, what should be reported alongside the decision?
Solutions. (1) The contrast is \((-2)-(-1)=-1\) unit. It is a one-unit greater decrease with drug than placebo, in the intended direction; it does not by itself establish downstream benefit. (2) A single-group change cannot separate a treatment effect from a time trend or other influences. (3) Use the difference between the drug and comparator proportions meeting the response definition at the specified assessment. (4) Any four of population, treatment regimen, comparator, biomarker variable, time point, summary contrast, and handling of relevant events or unavailable data. (5) Report the estimate with an appropriate assessment of its uncertainty and any limitations in data quality or generalizability; meeting a screening rule is not proof of a precise effect.