Tutorials › Pharmacometrics › Exposure-Biomarker-Response Modeling
Pharmacokinetics · PK/PD Foundations

Exposure-Biomarker-Response Modeling

Learn how exposure-biomarker-response models connect drug exposure to pharmacodynamic biomarkers and clinical or biological responses—and how these models help explain drug action, quantify relationships, and support dose selection.

Intermediate PK/PD Modeling Exposure-Response Pharmacometrics
01 · The big picture

1. What Is Exposure-Biomarker-Response Modeling?

Exposure-biomarker-response (EBR) modeling describes how drug exposure relates to one or more pharmacodynamic biomarkers and, ultimately, to a biological or clinical response.

The central idea is that drug concentration or exposure may not be the endpoint of interest. A drug produces pharmacological effects through biological mechanisms, and biomarkers can provide measurable intermediate signals between exposure and the final response.

Exposure C(t), AUC, Cmax or other exposure metric Biomarker target engagement pathway modulation physiologic response Response clinical endpoint or downstream effect Exposure → biological activity → response

An EBR model provides a quantitative framework for connecting drug exposure with biomarkers and downstream responses.

Core idea: an exposure-biomarker-response model can turn a collection of concentration, biomarker, and response observations into a coherent quantitative description of how drug exposure translates into pharmacological activity.
02 · Why model the relationship?

2. Why Is Exposure-Biomarker-Response Modeling Useful?

A concentration measurement tells us about exposure, but exposure alone does not necessarily tell us whether the drug is engaging its target or producing the intended biological effect.

Biomarkers can provide intermediate evidence of pharmacological activity. For example, a biomarker might quantify target occupancy, enzyme inhibition, receptor activation, pathway modulation, or a physiological change.

Modeling these relationships can help answer questions such as:

  • Does increasing exposure produce greater target engagement?
  • What exposure is associated with a specified biomarker response?
  • Is the biomarker response delayed relative to plasma concentration?
  • Does the biomarker mediate part of the relationship between exposure and clinical response?
  • Is there evidence of a plateau or maximum achievable response?
  • How much exposure is required to achieve a desired pharmacological effect?
  • Do patient characteristics modify the exposure-response relationship?

The value of the model is therefore not simply statistical association. The objective is to construct a quantitative representation that is scientifically interpretable and useful for prediction.

03 · The modeling framework

3. The Exposure-Biomarker-Response Framework

A useful conceptual framework separates the problem into linked components.

ComponentTypical variablePurpose
Exposure Concentration, AUC, Cmax, average concentration Quantifies the amount or intensity of drug exposure.
Biomarker Target engagement, enzyme activity, pathway marker Measures a biological consequence of exposure.
Response Clinical endpoint, symptom score, physiologic measurement Represents the downstream effect of treatment.
Covariates Body weight, renal function, disease severity, baseline value Explain systematic differences between individuals or observations.
Random variability Between-subject and residual variability Represents variation not explained by the structural model.

The relationships can be modeled sequentially, jointly, or hierarchically. The appropriate strategy depends on the scientific question, study design, sampling schedule, and amount of information available.

\[ \text{Dose}\rightarrow\text{PK}\rightarrow\text{Exposure}\rightarrow\text{Biomarker}\rightarrow\text{Response} \]
04 · Exposure

4. Defining Drug Exposure

The first component of an EBR analysis is a meaningful measure of drug exposure. Exposure can be represented in several ways depending on the mechanism and time scale of the pharmacological effect.

Exposure measureExampleWhen it may be useful
Concentration at time t \(C(t)\) Rapidly changing effects or direct concentration-response relationships.
Cmax Maximum observed or model-predicted concentration Effects associated with peak exposure.
AUC \(AUC_{0-\tau}\) Effects related to overall exposure over a dosing interval.
Average concentration \(AUC/\tau\) Effects related to typical exposure during repeated dosing.
Time-varying concentration \(C(t)\) Dynamic biomarker or response models.
Effect-compartment concentration \(C_e(t)\) Delayed pharmacodynamic effects or hysteresis.

A major advantage of pharmacometric modeling is that exposure does not necessarily have to be reduced to a single summary statistic. A PK model can provide an estimated concentration-time profile that becomes the input to a downstream biomarker or response model.

Important distinction: using AUC or Cmax as a predictor is different from modeling the full time-varying concentration. The choice should reflect the expected mechanism and the temporal structure of the data.
05 · Biomarkers

5. What Role Does the Biomarker Play?

A pharmacodynamic biomarker is a measurable variable that changes in response to drug action and can provide information about pharmacological activity.

Biomarkers can occur at different levels of a biological pathway. A biomarker close to the drug target may provide evidence of target engagement, while a downstream biomarker may reflect pathway modulation or a more integrated physiological effect.

CLOSER TO TARGET Target Proximal biomarker Downstream biomarker Effect Biomarkers can occupy different positions along the pathway from drug exposure to clinical effect. Biological proximity does not automatically imply greater clinical relevance.

Biomarkers may provide intermediate measurements between drug exposure and downstream response. Their interpretation depends on the biology, measurement properties, and intended use.

In EBR modeling, the biomarker may be treated as an outcome, an intermediate variable, a mediator, or an input to a downstream response model.

06 · Direct relationships

6. Direct Exposure-Biomarker Models

The simplest EBR model assumes that the biomarker responds directly to exposure without an important delay.

For an increasing biomarker response, an \(E_{\max}\) model is often useful:

\[ B(C)=B_0+\frac{B_{\max}C}{EC_{50}+C} \]

Here:

  • \(B_0\) is the baseline biomarker value.
  • \(B_{\max}\) is the maximum drug-related increase above baseline.
  • \(EC_{50}\) is the concentration producing half of the maximum drug-related effect.
  • \(C\) is the relevant drug concentration.

For an inhibitory biomarker relationship, an analogous model can be written as:

\[ B(C)=B_0-\frac{I_{\max}C}{IC_{50}+C} \]

The choice between stimulatory and inhibitory forms should reflect the expected direction of the pharmacological effect.

Interpretation: \(EC_{50}\) or \(IC_{50}\) is a model parameter describing potency under the specified model. It should not automatically be interpreted as a universal biological constant.
07 · Nonlinearity

7. Hill Models and Steep Exposure-Response Relationships

Some exposure-biomarker relationships are more or less steep than the basic \(E_{\max}\) model permits. A Hill coefficient can be introduced to provide additional flexibility:

\[ B(C)=B_0+\frac{B_{\max}C^{\gamma}}{EC_{50}^{\gamma}+C^{\gamma}} \]

where \(\gamma\) is the Hill coefficient.

When \(\gamma=1\), the model reduces to the standard \(E_{\max}\) form. Values greater than one produce a steeper transition around \(EC_{50}\), whereas values below one produce a shallower relationship.

The Hill coefficient can improve empirical description of a nonlinear relationship, but its biological interpretation requires care. A fitted Hill coefficient does not necessarily correspond directly to a specific molecular binding mechanism.

08 · Time delay

8. What If Biomarker Response Is Delayed?

Drug concentration and pharmacodynamic response do not always change simultaneously. A delayed biomarker response can occur because of distribution to the site of action, receptor kinetics, signal transduction, turnover of biological components, or other mechanisms.

A simple way to represent distributional delay is an effect-compartment model:

\[ \frac{dC_e(t)}{dt}=k_{e0}\left[C(t)-C_e(t)\right] \]

where \(C_e(t)\) is the effect-site concentration and \(k_{e0}\) controls the equilibration rate between the observed plasma concentration and the effect compartment.

The biomarker can then be modeled as a function of \(C_e(t)\) rather than directly as a function of plasma concentration:

\[ B(t)=B_0+\frac{B_{\max}C_e(t)}{EC_{50}+C_e(t)} \]

This approach can generate a hysteresis relationship, where the biomarker at a given plasma concentration differs depending on whether concentrations are rising or falling.

Key modeling question: if biomarker response appears delayed, first consider whether the delay is biological, measurement-related, or caused by an inappropriate exposure metric.
09 · Biomarker dynamics

9. Turnover Models for Dynamic Biomarkers

Many biomarkers are not simply instantaneous functions of concentration. They are produced and removed continuously. A turnover model can represent this process explicitly.

A simple baseline turnover model is:

\[ \frac{dB(t)}{dt}=k_{\mathrm{in}}-k_{\mathrm{out}}B(t) \]

At baseline, assuming steady state:

\[ B_0=\frac{k_{\mathrm{in}}}{k_{\mathrm{out}}} \]

Drug exposure can then modify either biomarker production or biomarker loss. For example, if exposure inhibits production:

\[ \frac{dB(t)}{dt} = k_{\mathrm{in}} \left( 1-\frac{I_{\max}C(t)}{IC_{50}+C(t)} \right) -k_{\mathrm{out}}B(t) \]

Alternatively, drug exposure can stimulate production or inhibit loss. These mechanisms produce different time courses even when the observed concentration-response relationship appears similar.

Turnover models are especially useful when the biomarker changes gradually after exposure changes and when the time course itself contains important mechanistic information.

10 · Downstream response

10. Connecting the Biomarker to Clinical Response

An important extension of EBR modeling is to connect the biomarker to a downstream clinical or physiological response.

A conceptual model might be:

\[ C(t)\rightarrow B(t)\rightarrow E(t) \]

For example, a biomarker may quantify pathway suppression, while the final response measures disease activity. The response can be modeled as a function of the biomarker:

\[ E(B)=E_0+\frac{E_{\max}B}{EC_{50,B}+B} \]

Alternatively, the biomarker can enter a linear, logistic, proportional-hazards, count, or other response model depending on the endpoint.

Response typePossible modeling framework
Continuous responseLinear, nonlinear, \(E_{\max}\), Hill, or turnover model
Binary responseLogistic model
Ordinal responseOrdinal logistic or related models
Count responsePoisson or negative-binomial models
Time-to-event responseHazard or survival models
Repeated measurementsLongitudinal mixed-effects or nonlinear mixed-effects models
11 · Joint modeling

11. Sequential Versus Joint Exposure-Biomarker-Response Models

There are several ways to construct an EBR analysis.

Sequential modeling

One approach is to estimate the PK model first, use the resulting exposure predictions in a biomarker model, and then use biomarker predictions in a response model.

\[ \text{PK}\rightarrow\text{Biomarker}\rightarrow\text{Response} \]

This can be practical and computationally convenient, particularly when the different datasets or modeling stages are naturally separated.

Joint modeling

A joint model estimates multiple components simultaneously:

\[ \text{Dose}\rightarrow PK\rightarrow B(t)\rightarrow E(t) \]

Joint modeling can account more directly for uncertainty and shared parameters across linked components. It can also be useful when the biomarker and response contain complementary information about the underlying pharmacology.

The choice depends on the scientific objective, data structure, identifiability, computational requirements, and assumptions about how the different components are connected.

12 · Variability

12. Covariates and Between-Subject Variability

Patients may differ in both exposure and pharmacodynamic sensitivity. An EBR model can therefore incorporate covariates that explain systematic differences between individuals.

For example, a potency parameter could depend on a covariate \(X_i\):

\[ EC_{50,i}=EC_{50,\mathrm{pop}} \left(\frac{X_i}{X_{\mathrm{ref}}}\right)^{\theta} \]

or, for a categorical covariate:

\[ EC_{50,i}=EC_{50,\mathrm{pop}}e^{\theta X_i} \]

Possible covariates include demographic characteristics, disease characteristics, renal or hepatic function, concomitant medications, baseline biomarker values, and genetic or molecular characteristics.

Population modeling can also distinguish typical population parameters from individual variability.

\[ EC_{50,i}=EC_{50,\mathrm{pop}}e^{\eta_i} \]

where \(\eta_i\) represents an individual-specific deviation from the population value.

13 · Baseline

13. Why Baseline Biomarker Values Matter

Biomarkers can vary substantially between individuals before treatment. A model that ignores baseline heterogeneity may incorrectly attribute pre-existing differences to drug exposure.

A simple model can explicitly include an individual baseline:

\[ B_i(t)=B_{0,i}+E_i(t) \]

where \(B_{0,i}\) represents the individual's baseline biomarker level and \(E_i(t)\) represents the drug-related change.

Baseline can also be incorporated into downstream response models as a covariate. This is particularly important when the clinical response depends on both treatment effect and the starting disease state.

Practical principle: distinguish the baseline level of a biomarker from the change attributable to treatment. A model should make that distinction explicit when the scientific question requires it.
14 · Observation model

14. Modeling Biomarker and Response Variability

Observed biomarkers are not exact measurements of the underlying biological state. Measurement error, biological variability, timing differences, and other sources of residual variation should be considered.

A simple additive observation model is:

\[ B_{\mathrm{obs},ij}=B_{\mathrm{pred},ij}+\epsilon_{ij} \]

A proportional model is:

\[ B_{\mathrm{obs},ij}=B_{\mathrm{pred},ij}(1+\epsilon_{ij}) \]

Combined additive and proportional error models are also commonly used when neither form alone adequately represents the residual variability.

The observation model is important because estimates of exposure-response relationships can be affected by how measurement variability is represented.

15 · Worked example

15. Worked Example: Exposure and Biomarker Response

Consider a hypothetical drug whose pharmacodynamic biomarker decreases with increasing exposure. Suppose the baseline biomarker is 100 units, the maximum inhibition is 80 units, and the \(IC_{50}\) is 10 mg/L.

Step 1: Specify the inhibitory model

\[ B(C)=B_0-\frac{I_{\max}C}{IC_{50}+C} \]

Substituting the parameters:

\[ B(C)=100-\frac{80C}{10+C} \]

Step 2: Calculate the biomarker at 2 mg/L

\[ B(2)=100-\frac{80(2)}{10+2} =100-13.33 \approx86.67 \]

At an exposure of 2 mg/L, the predicted biomarker is approximately 86.7 units.

Step 3: Calculate the biomarker at 10 mg/L

\[ B(10)=100-\frac{80(10)}{10+10} =100-40 =60 \]

At the \(IC_{50}\), the model predicts half of the maximum drug-related inhibition, producing a biomarker value of 60 units.

Step 4: Calculate the biomarker at 40 mg/L

\[ B(40)=100-\frac{80(40)}{10+40} =100-64 =36 \]

At 40 mg/L, the predicted biomarker is approximately 36 units. Increasing exposure from 10 to 40 mg/L produces a smaller incremental reduction than increasing exposure from 0 to 10 mg/L because the relationship is approaching its maximum effect.

What the example illustrates: an exposure-biomarker model provides a quantitative description of potency, maximum effect, and the degree to which additional exposure is expected to produce additional biomarker modulation.
16 · Choosing exposure

16. Choosing the Right Exposure Metric

The exposure metric used in an EBR model should be biologically and temporally appropriate.

If the effect is primarily related to…A candidate exposure metric might be…
Instantaneous concentration\(C(t)\)
Peak exposure\(C_{\max}\)
Cumulative exposureAUC
Average exposure during a dosing interval\(AUC/\tau\)
Time above a concentration threshold\(T>C_{\mathrm{threshold}}\)
Delayed pharmacological actionEffect-site concentration or dynamic turnover model

Choosing an exposure metric solely because it produces a strong empirical association can be misleading. The metric should also make sense in the context of the drug's pharmacology and expected mechanism of action.

17 · Hysteresis

17. Recognizing Hysteresis in Exposure-Biomarker Relationships

Hysteresis occurs when the biomarker response at a given plasma concentration differs depending on whether the concentration is increasing or decreasing.

A concentration-response plot may therefore form a loop rather than a single curve.

Plasma concentration Biomarker effect Rising exposure Falling exposure

A hysteresis loop indicates that plasma concentration alone may not adequately explain the instantaneous biomarker response.

Hysteresis can arise from effect-site equilibration, signal transduction, indirect mechanisms, or other biological delays. It is therefore a useful diagnostic feature when evaluating an exposure-biomarker model.

18 · Model selection

18. Choosing an Appropriate EBR Model

Several candidate models may describe the same data reasonably well. Model selection should therefore consider both statistical performance and biological plausibility.

ModelUseful whenKey consideration
Linear exposure-responseResponse changes approximately proportionally over the observed exposure rangeMay extrapolate poorly beyond the observed range.
\(E_{\max}\) modelResponse approaches a plateauCan quantify potency and maximum effect.
Hill modelResponse is unusually steep or shallowAdditional parameter can increase flexibility and identifiability demands.
Effect-compartment modelDelayed response relative to plasma concentrationRequires informative temporal sampling.
Turnover modelBiomarker production and loss are importantProvides mechanistic time-course interpretation.
Joint PK/PD modelExposure and response dynamics are tightly linkedMore complex estimation and model evaluation.

The simplest model that adequately represents the scientific question is often preferable to unnecessary complexity. However, an overly simple model can hide important dynamics or lead to misleading parameter interpretation.

19 · Diagnostics

19. How Do We Evaluate an EBR Model?

An EBR model should be evaluated using multiple complementary diagnostics.

  • Observed versus predicted plots: assess whether the model reproduces the central tendency of the data.
  • Residual plots: identify systematic deviations and inappropriate error assumptions.
  • Time-course plots: reveal whether the model captures onset, peak, delay, and recovery.
  • Individual predictions: assess whether the model represents subject-specific trajectories.
  • Parameter precision: evaluate whether important parameters are estimated with adequate information.
  • Biological plausibility: determine whether parameter estimates make sense in the context of the pharmacology.
  • Simulation-based checks: evaluate whether simulated data resemble the observed data under the fitted model.
  • Sensitivity analysis: examine whether important conclusions depend strongly on uncertain assumptions.
Good fit is not enough: an EBR model should reproduce important features of the observations while also providing parameters and predictions that are scientifically interpretable.
20 · Identifiability

20. Identifiability and Informative Study Design

Complex EBR models require sufficiently informative data. A model containing many parameters may fit the observations while leaving some parameters weakly identified.

For example, a model containing both an effect-compartment rate constant and a turnover process may require rich temporal sampling to distinguish the two mechanisms.

Important design considerations include:

  • Sampling before and after expected onset of effect.
  • Sampling around peak exposure and peak biomarker response.
  • Observations during the declining exposure phase.
  • A sufficient range of exposure levels.
  • Repeated observations within individuals when appropriate.
  • Measurement of baseline biomarker values.
  • Sampling that captures recovery or washout when relevant.

Study design and model development are therefore closely connected. A sophisticated model cannot recover information that the study did not collect.

21 · Dose selection

21. Using EBR Models for Dose Selection

One important application of exposure-biomarker-response modeling is dose selection.

A typical development framework may proceed from dose to exposure, exposure to biomarker effect, and biomarker effect to clinical response:

\[ \text{Dose}\rightarrow\text{Exposure}\rightarrow\text{Biomarker}\rightarrow\text{Clinical response} \]

Suppose a model predicts that the biomarker reaches near-maximal modulation at a particular exposure range. If the downstream clinical response also increases over that range, the model can help characterize the exposure range of interest for subsequent dose selection.

The model can also identify situations in which increasing dose produces substantially greater exposure but little additional biomarker or clinical effect.

Important: model-based dose selection is conditional on the fitted model, parameter uncertainty, observed exposure range, and assumptions about the relationship between biomarker and clinical response.
22 · Simulation

22. Simulation From an EBR Model

Once an EBR model has been established, simulation can be used to explore expected biomarker and response trajectories under alternative dosing conditions.

For example, a model can be used to simulate:

  • Different doses.
  • Different dosing intervals.
  • Different patient characteristics.
  • Alternative exposure levels.
  • Expected biomarker inhibition or stimulation.
  • Expected probability of achieving a response threshold.
  • The effect of parameter uncertainty on predicted outcomes.

Simulation is particularly useful because the full model preserves the time dimension rather than reducing every treatment regimen to a single observed summary.

23 · Population modeling

23. EBR Models in Population Pharmacology

In population PK/PD modeling, exposure-biomarker-response relationships can be estimated across many individuals while accounting for between-subject variability.

A generic population model can be written as:

\[ \theta_i=f(\theta_{\mathrm{pop}},X_i,\eta_i) \]

where \(\theta_i\) is an individual's parameter, \(\theta_{\mathrm{pop}}\) is the population-typical parameter, \(X_i\) represents covariates, and \(\eta_i\) represents unexplained between-subject variability.

This framework allows the model to distinguish several sources of variation:

SourceExample
Exposure variabilityDifferences in clearance or bioavailability
Pharmacodynamic variabilityDifferences in \(EC_{50}\), \(IC_{50}\), or \(E_{\max}\)
Baseline variabilityDifferent starting biomarker values
Residual variabilityMeasurement and unexplained within-subject variation
24 · Interpretation

24. What EBR Models Do Not Tell Us Automatically

Exposure-biomarker-response modeling provides a quantitative framework, but several important limitations should be kept in mind.

  • Association does not automatically establish mechanism. A strong exposure-biomarker relationship may be consistent with several biological explanations.
  • A biomarker is not automatically a surrogate endpoint. Predictive or mechanistic relevance requires additional evidence.
  • Exposure metrics can obscure time. AUC may hide important differences in concentration-time profiles.
  • Model parameters depend on model structure. \(EC_{50}\), \(E_{\max}\), turnover parameters, and other quantities are conditional on the specified model.
  • Extrapolation can be uncertain. Relationships outside the observed exposure range may be strongly dependent on assumptions.
  • Biomarker measurement error matters. Noisy measurements can affect estimates of exposure-response relationships.
  • Confounding can occur. Disease progression, baseline differences, concomitant treatments, and other factors can influence both biomarkers and responses.
Modeling principle: an EBR model should be interpreted as a quantitative representation of the available evidence, not as proof that the proposed biological pathway is the only possible mechanism.
25 · Practical workflow

25. A Practical Exposure-Biomarker-Response Modeling Workflow

  1. Define the scientific question. Decide whether the goal is target engagement, biomarker characterization, exposure-response characterization, dose selection, or prediction.
  2. Characterize exposure. Develop or obtain an appropriate PK model and determine whether concentration, AUC, Cmax, or a dynamic exposure measure is most appropriate.
  3. Explore biomarker data. Examine baseline values, time courses, exposure ranges, and response patterns.
  4. Identify temporal relationships. Look for delays, hysteresis, turnover, or other dynamic features.
  5. Specify candidate models. Consider linear, \(E_{\max}\), Hill, effect-compartment, turnover, or other mechanistic models as appropriate.
  6. Define the observation model. Account for measurement and residual variability.
  7. Estimate parameters. Use an estimation approach appropriate for the data and model.
  8. Evaluate diagnostics. Examine predictions, residuals, time courses, parameter precision, and biological plausibility.
  9. Assess variability and covariates. Determine whether patient characteristics explain systematic differences in exposure or response.
  10. Validate the model where possible. Use external data, internal validation, simulation, or other appropriate methods.
  11. Simulate relevant scenarios. Explore alternative doses, exposure levels, and patient characteristics.
  12. Interpret within the model's scope. Clearly distinguish observed data from model-based predictions and assumptions.

26. Key Takeaways

  • Exposure-biomarker-response modeling connects drug exposure to biological activity and downstream response.
  • Drug concentration is often the starting point, but the ultimate scientific question may concern target engagement, pathway modulation, or clinical effect.
  • Biomarkers can provide measurable intermediate information between exposure and clinical response.
  • Simple \(E_{\max}\), inhibitory \(E_{\max}\), and Hill models can describe nonlinear exposure-biomarker relationships.
  • Effect-compartment and turnover models can represent delayed or dynamic biomarker responses.
  • The choice of exposure metric should reflect the expected pharmacology and temporal behavior of the effect.
  • Baseline biomarker values, between-subject variability, residual variability, and covariates can materially influence EBR model interpretation.
  • Sequential and joint modeling approaches provide different ways to connect PK, biomarker, and response components.
  • Model parameters such as \(EC_{50}\), \(IC_{50}\), and \(E_{\max}\) are conditional on the model structure and data.
  • Good model evaluation requires more than a visually good fit; diagnostics, parameter precision, biological plausibility, and predictive performance should all be considered.
  • Informative study design is essential because complex EBR models require adequate exposure range and temporal sampling.
  • EBR models can support dose selection and simulation, but model-based predictions remain conditional on assumptions and uncertainty.
Next step

Where to Go Next

A natural progression is to study PK/PD models in greater detail, followed by direct and indirect response models, turnover models, effect-compartment models, \(E_{\max}\) and Hill models, population PK/PD, and covariate modeling.

The next step is to examine how a time-varying PK model can be connected to a pharmacodynamic model so that exposure, biomarker dynamics, and clinical response are represented within one quantitative framework.

← Back to Pharmacokinetics Tutorials