Why Link Dose, Exposure, and a Pharmacodynamic Biomarker?
In phase I, dose is assigned, exposure is measured, and a pharmacodynamic (PD) biomarker may show whether the drug is producing a biological response. These quantities are related but not interchangeable. The same dose can produce different exposures across participants, and a measured biomarker change is evidence of biological activity only if the biomarker, assay, and sampling time are appropriate for the question.
A useful decision process therefore asks three linked questions: does exposure increase as dose increases; does the biomarker respond as exposure increases; and is the emerging response useful for choosing the next dose, given the available safety information? The biomarker can help guide escalation or identify a dose for further evaluation. It does not, by itself, establish clinical benefit, an acceptable safety profile, or a recommended dose.
The direction of response must be specified before interpreting the data. If the biomarker is expected to increase, define the response in the opposite direction. The baseline and post-dose measurements should use the same assay and units, and the post-dose time should be selected to capture the intended pharmacology. A prespecified biomarker target is a decision benchmark, not automatically a validated threshold for clinical efficacy.
A Hypothetical Phase I Case
Suppose a hypothetical first-in-human study evaluates four dose levels in sequential cohorts of six participants. The drug is expected to reduce a circulating biomarker. At a specified post-dose time, the study measures the biomarker and the 0–24-hour area under the concentration-time curve, denoted AUC\(_{0-24}\), in mg·h/L. The study team has prespecified 50% inhibition as a useful level of biological activity for this case. That benchmark is a development decision criterion, not a claim that 50% inhibition produces clinical benefit.
For this example, the mean AUC\(_{0-24}\) values and mean individual percentage inhibitions are:
| Dose (mg) | Mean AUC0-24 (mg·h/L) | Mean inhibition (%) |
|---|---|---|
| 1 | 10 | 10.3 |
| 3 | 30 | 26.0 |
| 10 | 100 | 48.0 |
| 20 | 200 | 60.0 |
The progression is consistent with a dose-related increase in exposure and a larger average biomarker response at higher exposure. These cohort summaries are useful for an initial decision, but they hide participant-level variation. The next example shows how to quantify that variation for one cohort.
Worked Example: Calculating Biomarker Inhibition and Its Uncertainty
At 20 mg, six hypothetical participants have baseline biomarker values of 100, 110, 90, 120, 80, and 100 assay units. Their corresponding post-dose values are 49, 49.5, 38.7, 48, 28, and 28 assay units. Calculate each percentage inhibition and summarize the cohort response.
For each participant, use \(100(B_i-Y_i)/B_i\). The six values are \(100(100-49)/100=51\%\), \(100(110-49.5)/110=55\%\), \(100(90-38.7)/90=57\%\), \(100(120-48)/120=60\%\), \(100(80-28)/80=65\%\), and \(100(100-28)/100=72\%\).
The sample mean inhibition is \(\bar{x}=(51+55+57+60+65+72)/6=60.0\%\). The sample standard deviation is \(s=\sqrt{284/5}=7.54\) percentage points, so \(SE=s/\sqrt{n}=7.54/\sqrt{6}=3.08\) percentage points.
For a two-sided 95% confidence interval using a t distribution with \(df=n-1=5\), the critical value is 2.571. The interval is \(60.0\%\pm2.571(3.08\%)=60.0\%\pm7.92\%\), or 52.1% to 67.9%. Under the assumptions of independent participants and an approximately normal distribution of individual responses, this interval is a plausible range for the population mean inhibition at this dose and sampling time. It is not a 95% probability statement about the fixed population mean.
In this small hypothetical cohort, the interval is above the 50% benchmark. That supports the view that the dose reached the prespecified level of biological activity in the population, subject to the assumptions and limitations of the interval. It does not establish that the biomarker is a surrogate for clinical benefit.
Relating Biomarker Response to Exposure
A dose-response summary relates average response to assigned dose. An exposure-response model instead relates response to measured drug exposure, which can be informative when participants receiving the same dose have different concentrations. For this case, consider the simple Emax model:
Here, \(E(C)\) is predicted percentage inhibition at AUC \(C\), \(E_{\max}\) is the maximum response represented by the model, and \(EC_{50}\) is the exposure producing half of that maximum. The model assumes a monotone, saturating relationship over the relevant exposure range and a suitable correspondence between the exposure summary and the biomarker measurement time. It does not account automatically for delayed response, baseline differences, measurement error, or other causes of biomarker change.
Worked Example: Interpreting an Illustrative Emax Fit
Suppose an illustrative fit to the cohort means gives \(E_{\max}=80\%\) and \(EC_{50}=66.7\) mg·h/L. Calculate the predicted response at the four mean exposures and interpret the result.
At AUC 10 mg·h/L, \(E(10)=80(10)/(66.7+10)=800/76.7=10.4\%\). At AUC 30 mg·h/L, \(E(30)=2400/96.7=24.8\%\). At AUC 100 mg·h/L, \(E(100)=8000/166.7=48.0\%\). At AUC 200 mg·h/L, \(E(200)=16000/266.7=60.0\%\).
These predictions are close to the observed cohort means of 10.3%, 26.0%, 48.0%, and 60.0%. In this hypothetical case, the model provides a compact description of the observed pattern. Because it is applied to mean exposures and mean responses, it is illustrative: fitting participant-level data while accounting for measurement and between-participant variability would be needed for a fuller analysis.
The model’s implication for the 50% benchmark follows by solving \(50=80C/(66.7+C)\). Multiplying both sides by \(66.7+C\) gives \(3335+50C=80C\), so \(C=3335/30=111.2\) mg·h/L. Thus the model predicts that an AUC of about 111 mg·h/L corresponds to 50% inhibition. This is a model-based estimate, not a directly observed threshold, and its uncertainty matters when applying it to individuals or future cohorts.
Turning the Evidence Into an Escalation Decision
Biomarker-informed escalation is a decision process, not a rule to increase dose until a target is reached. A protocol should define the dose steps, sampling times, biomarker benchmark, review of emerging safety data, and the conditions under which escalation pauses or stops. At each step, the team reviews safety and tolerability independently of the biomarker, then considers exposure and PD evidence together. Escalation proceeds only if the prespecified safety review permits it.
Confirm assay performance, baseline availability, sample timing, and exposure data quality. Determine whether the biomarker was measured at a time relevant to the intended response.
Assess the available tolerability and safety information against the protocol. A favorable biomarker response does not override a safety concern or authorize escalation by itself.
Compare dose, observed exposure, and baseline-adjusted biomarker response. Consider both the cohort summary and participant-level variability; do not assume dose alone predicts response equally well for everyone.
Escalate, hold, or stop according to the protocol and total evidence. If a useful biomarker benchmark is reached with acceptable observed tolerability, further escalation may add little decision value; select a dose for further evaluation while retaining relevant alternatives if uncertainty remains.
Worked Example: Selecting a Dose for Further Evaluation
In the hypothetical case, assume that protocol-defined safety review permits each step through 20 mg, and the 20 mg cohort has the responses calculated above. The 10 mg cohort mean is 48.0%, just below the 50% benchmark, while the 20 mg mean is 60.0% with a 95% confidence interval of 52.1% to 67.9%. How should the team use this evidence?
The 20 mg dose is the first tested dose in this sequence with an interval above the prespecified 50% benchmark. The corresponding mean exposure is 200 mg·h/L, and the illustrative exposure-response model predicts 60.0% inhibition at that exposure. These mutually consistent summaries support selecting 20 mg for further evaluation, provided the separate safety review continues to permit it.
This choice does not prove that 20 mg is the minimum effective dose, that every participant will exceed the benchmark, or that the biomarker response predicts patient benefit. The 10 mg mean is close to the target, and its uncertainty has not been summarized here. Additional participants or a planned comparison of doses could change the assessment. The decision is therefore a reasoned selection for further study, not a final dose recommendation.
Assumptions, Diagnostics, and Alternatives
The calculations and model are most interpretable when baseline and post-dose measurements are reliable, response direction is prespecified, sampling is timed appropriately, and participants within a cohort are analyzed consistently. For a confidence interval based on a t distribution, the participant responses should be independent and sufficiently compatible with an approximately normal sampling distribution; with six participants, substantial skew or outliers can make the interval unreliable. A change from baseline should also be considered alongside any expected time trends or assay drift, especially when there is no untreated or placebo reference.
Diagnostics should include plots of individual response against exposure, not only dose-level means. Check whether increasing exposure generally corresponds to increasing inhibition, whether a few participants dominate the pattern, and whether the fitted curve leaves systematic residual patterns. Examine assay precision and values near the assay’s quantification limits. If higher dose increases exposure but the biomarker does not change, possible explanations include an uninformative biomarker, an unsuitable sampling time, insufficient exposure range, or a mistaken response model. Such a pattern is a reason to investigate, not a reason to force a fitted curve.
The main alternatives answer different questions:
| Approach | Useful for | Main limitation |
|---|---|---|
| Dose-based cohort summaries | Displaying response at the tested dose levels and supporting protocol decisions | Does not describe how differences in individual exposure relate to response |
| Exposure-response model | Summarizing a response trend across measured exposures and estimating a model-based benchmark | Depends on model form, data coverage, and uncertainty; a smooth curve can conceal poor fit |
| Biomarker-guided escalation criterion | Using a prespecified level of biological activity as one input to escalation or dose selection | Cannot establish clinical benefit or replace the protocol’s safety review |
A continuous PD biomarker is most useful when it is analytically reliable, biologically relevant, measured at a justified time, and interpreted within a prespecified decision framework. Its limitations include measurement error, natural within-person variation, small early-phase cohorts, and uncertainty about how a biomarker change relates to outcomes that matter to patients. A dose selected for further evaluation is consequently a development decision based on available evidence, not proof that the dose is optimal.
Check Your Understanding
Use the definitions and hypothetical values in this tutorial. Brief solutions follow the questions.
- If baseline is 80 assay units and the post-dose value is 60 assay units, what is the percentage inhibition?
- For \(E_{\max}=80\%\), \(EC_{50}=66.7\) mg·h/L, and AUC \(=66.7\) mg·h/L, what response does the model predict?
- Why should a 50% biomarker response not automatically be described as a clinically effective response?
- If exposure rises with dose but the biomarker response does not, name two checks to make before using the model to guide another escalation.
Solutions. (1) \(100(80-60)/80=25\%\) inhibition. (2) \(E=80(66.7)/(66.7+66.7)=40\%\), half of \(E_{\max}\). (3) The benchmark describes a prespecified level of biological activity; clinical benefit requires evidence linking the biomarker response to a patient-relevant outcome. (4) Check assay performance and sample timing, and inspect individual exposure-response data for variability, outliers, or a pattern inconsistent with the assumed model. Safety review remains necessary before any escalation.