1. What Is an Indirect-Response PK/PD Model?
An indirect-response PK/PD model describes drug effects that arise because a drug changes the rate of production or loss of an endogenous response, rather than directly changing the response itself.
This distinction is important because many pharmacologic effects do not disappear as soon as plasma drug concentration falls. A drug may inhibit the production of a biomarker, stimulate its production, inhibit its loss, or stimulate its loss. The measured response then evolves according to its own turnover dynamics.
In a simple turnover model, the response is represented by a balance between an input or production rate and an output or loss rate:
At baseline, these rates are often assumed to balance:
The drug then changes one of these rates as its concentration changes.
An indirect-response model links drug concentration to the production or loss of a response. The response itself has turnover dynamics that create temporal delay.
2. Why Do We Need Indirect-Response Models?
A direct concentration-effect model can be useful when effect closely tracks drug concentration. For example, an immediate pharmacologic response may be represented with an Emax relationship:
But some responses do not track concentration instantaneously. A biomarker may have a natural production and degradation process, and changing one of those processes takes time to alter the measured level.
This creates several characteristic observations:
- The maximum response may occur after Cmax.
- The response may remain altered after plasma concentration has substantially declined.
- Repeated dosing can produce a response profile that differs substantially from the concentration profile.
- The magnitude and duration of the response can depend on the turnover rate of the measured endpoint.
Indirect-response models provide a mechanistic way to represent these patterns without requiring the response to be an instantaneous function of concentration.
3. The Turnover Model
The foundation of indirect-response modeling is the turnover model. Let R(t) denote the response or biomarker level. A basic turnover equation is:
Here:
| Parameter | Meaning | Interpretation |
|---|---|---|
| R | Response or biomarker amount/concentration | The quantity being modeled dynamically |
| Rin | Zero-order production/input rate | Rate at which the response is generated |
| kout | First-order loss rate constant | Fractional rate at which the response is removed |
| R0 | Baseline response | Response before drug effect |
At baseline, before drug exposure alters the system, the response is at steady state. Therefore:
which gives:
This relationship is extremely useful because it allows the production rate to be expressed in terms of the baseline response and turnover rate.
4. The Four Classical Indirect-Response Models
The classical indirect-response framework contains four basic model types. They differ according to whether the drug inhibits or stimulates response production or response loss.
| Model | Drug action | Typical effect |
|---|---|---|
| Type I | Inhibits response production | Response decreases |
| Type II | Stimulates response loss | Response decreases |
| Type III | Stimulates response production | Response increases |
| Type IV | Inhibits response loss | Response increases |
These four structures are often introduced using an Imax or Emax function to connect drug concentration to the affected rate.
The four classical indirect-response structures are defined by which turnover process the drug modifies.
5. Type I: Inhibition of Response Production
In a Type I model, drug concentration inhibits the production rate. An Imax function is commonly used:
where:
- Imax is the maximum fractional inhibition of production.
- IC50 is the concentration producing 50% of the maximum inhibitory effect.
- C is the drug concentration driving the PD effect.
For a fully inhibitory drug effect with Imax = 1, the production rate approaches zero as concentration becomes very large.
Because production is reduced while loss continues, the response falls. Importantly, it does not necessarily fall instantaneously. The response must turn over before the measured level changes substantially.
6. Type II: Stimulation of Response Loss
In a Type II model, the drug increases the loss rate of the response:
Here, drug concentration increases the effective loss rate. The response therefore decreases because the system is removing the response faster than it did at baseline.
The distinction between Type I and Type II can be subtle when looking only at the observed response. Both can produce a decrease in R. Their mechanistic interpretation is different:
| Feature | Type I | Type II |
|---|---|---|
| Drug target in model | Production | Loss |
| Production rate | Decreases | Unchanged |
| Loss rate | Unchanged | Increases |
| Response direction | Decreases | Decreases |
Choosing between these structures should therefore be based on the scientific mechanism, experimental evidence, and the information contained in the data—not solely on which equation gives the closest visual fit.
7. Type III: Stimulation of Response Production
In a Type III model, drug concentration increases the production rate:
The drug therefore adds to the normal production process. As concentration rises, the production rate increases, causing the response to increase over time.
The Emax parameter describes the maximum fractional increase in production, while EC50 determines the concentration associated with half of that maximum stimulatory effect.
At high concentrations, the drug-driven production approaches:
The eventual magnitude of the response also depends on the turnover process represented by kout.
8. Type IV: Inhibition of Response Loss
In a Type IV model, the drug decreases the loss rate of the response:
Because less response is removed, the response increases.
Type III and Type IV therefore produce the same broad direction of response change but act on different turnover processes:
| Feature | Type III | Type IV |
|---|---|---|
| Drug target in model | Production | Loss |
| Production rate | Increases | Unchanged |
| Loss rate | Unchanged | Decreases |
| Response direction | Increases | Increases |
9. Why Does the Response Lag Behind Concentration?
One of the most important features of indirect-response models is the ability to produce a delay between drug concentration and observed response.
Suppose concentration changes rapidly after a dose. The drug immediately changes a production or loss rate, but the response itself changes only as the differential equation accumulates that rate imbalance.
The response turnover rate determines how quickly the system can react. For a first-order loss process, a characteristic response turnover half-life is:
A smaller kout corresponds to slower turnover and therefore a longer response time scale. A larger kout corresponds to faster turnover.
An indirect-response system can produce a delayed response because the drug modifies a turnover rate rather than instantaneously setting the response level.
This delay is mechanistic rather than simply a statistical lag term. The model explains the delay through the dynamics of response production and loss.
10. Baseline and Drug-Modified Steady State
At baseline, the turnover model satisfies:
After a drug effect is introduced, the system may move toward a different equilibrium, depending on whether production or loss has been altered.
For example, consider a Type I model:
At a constant concentration C, the drug-modified steady-state response is:
Using Rin = koutR0 gives:
This result shows that a sustained concentration can shift the response toward a new equilibrium. However, the system does not generally reach that equilibrium instantaneously; the response approaches it according to its turnover dynamics.
11. The Main Parameters in an Indirect-Response Model
Indirect-response models contain both PK parameters and PD/turnover parameters. Keeping these roles conceptually separate is essential.
| Parameter | Role | What it controls |
|---|---|---|
| CL | PK | Drug elimination and therefore the concentration-time profile |
| V | PK | Relationship between drug amount and concentration |
| R0 | PD/turnover | Baseline response |
| kout | PD/turnover | Response turnover time scale |
| Rin | PD/turnover | Baseline production/input rate |
| IC50 | PD | Concentration associated with half-maximal inhibition |
| EC50 | PD | Concentration associated with half-maximal stimulation |
| Imax | PD | Maximum fractional inhibition |
| Emax | PD | Maximum fractional stimulation |
A useful modeling decomposition is therefore:
The PK model determines the concentration driving the PD model. The indirect-response model then determines how that concentration changes the response trajectory.
12. How Do You Choose Among the Four Model Types?
The choice of indirect-response structure should begin with the scientific mechanism. The observed direction of the response alone is not sufficient to distinguish production effects from loss effects.
| Observed response | Possible mechanism | Candidate model |
|---|---|---|
| Response decreases | Production is inhibited | Type I |
| Response decreases | Loss is stimulated | Type II |
| Response increases | Production is stimulated | Type III |
| Response increases | Loss is inhibited | Type IV |
Other considerations include the biology of the endpoint, experimental evidence, dosing regimen, sampling frequency, prior knowledge, and parameter identifiability.
When the data cannot distinguish two mechanisms, the distinction should not be overstated. Two models may produce similar observed trajectories even though their mechanistic interpretations differ.
13. Emax and Imax Functions
The turnover model specifies where the drug acts. A concentration-effect function specifies how strongly the drug changes that process.
A common inhibitory function is:
A common stimulatory function is:
These functions have several useful properties. When C = 0, the drug effect is zero. As concentration increases, the effect approaches its maximum asymptotically.
For example, when C = IC50:
Similarly, when C = EC50:
The Emax relationship is not the only possible drug-effect function. Depending on the biology and data, Hill-type functions, linear relationships, sigmoid models, or mechanistically specified functions may be appropriate.
14. Indirect Response and Hysteresis
Hysteresis describes a situation in which the observed relationship between concentration and effect differs depending on whether concentration is increasing or decreasing.
With a rapidly changing drug concentration and a slower response turnover process, the same concentration can correspond to different response values at different times.
Plotting response against concentration can therefore produce a loop rather than a single concentration-effect curve.
A delayed indirect response can generate hysteresis when response turnover is slower than changes in drug concentration.
Hysteresis can also arise from other mechanisms, including effect compartments, tolerance, active metabolites, and complex biological feedback. Therefore, a hysteresis loop is evidence of temporal separation between concentration and effect, but it does not by itself identify the mechanism.
15. Indirect-Response Models vs Effect-Compartment Models
Both indirect-response and effect-compartment models can describe delayed drug effects, but they represent different mechanisms.
| Feature | Indirect-response model | Effect-compartment model |
|---|---|---|
| Intermediate quantity | Response turnover | Effect-site concentration |
| Drug acts by | Changing production or loss of response | Changing concentration at a hypothetical effect site |
| Primary delay mechanism | Response turnover | Distribution between plasma and effect site |
| Typical state variable | R(t) | Ce(t) |
| Biological interpretation | Turnover of the measured response | Delay between plasma and effect-site concentration |
Neither approach should be selected simply because it produces a visually appealing delay. The choice should reflect the scientific question and the biological interpretation that the data can support.
16. Worked Example: A Drug That Inhibits Biomarker Production
Consider a hypothetical drug that inhibits production of an endogenous biomarker. Suppose the baseline biomarker concentration is 100 units/L, and the response turnover rate constant is 0.20 h−1.
Assume the drug follows a Type I indirect-response model with:
- Imax = 1
- IC50 = 2 mg/L
- A constant drug concentration of 2 mg/L
Step 1: Calculate baseline production
At baseline:
Thus, before drug exposure, the system produces approximately 20 units/L per hour while losing the same amount at steady state.
Step 2: Calculate the fractional inhibition
Using the Imax function:
The drug therefore produces a 50% inhibition of the baseline production rate at this concentration.
Step 3: Calculate the drug-modified production rate
Step 4: Calculate the new steady-state response
Under constant concentration, the drug-modified steady state is:
Thus, the biomarker is predicted to move from a baseline of 100 units/L toward a new steady state of 50 units/L.
Step 5: Determine the response turnover half-life
The response therefore changes on a time scale of approximately 3.5 hours. The new steady state is not reached immediately when the drug concentration changes.
For a constant drug concentration introduced at time zero, the solution can be written as:
Substituting the values:
At 3.47 hours, approximately one response half-life has elapsed, so the remaining difference from the new steady state is approximately one-half of its initial value. The predicted response is therefore approximately:
This illustrates the central feature of indirect-response modeling: the drug effect on production is immediate in the model, but the observed response changes progressively because the biomarker itself has turnover.
17. What Happens Under Repeated Dosing?
Under repeated dosing, the drug concentration changes continuously according to the PK model. The indirect-response model then receives that changing concentration as its driver.
This can produce response patterns that differ substantially from the concentration profile.
For example, a drug may have a relatively short plasma half-life but produce a prolonged biomarker response if the biomarker turns over slowly. Conversely, a rapidly turning-over biomarker may respond relatively quickly to changes in concentration.
Repeated dosing can therefore create:
- Accumulation of drug concentration.
- Accumulation or persistence of pharmacodynamic response.
- Delayed attainment of pharmacodynamic steady state.
- Different apparent time courses for concentration and response.
Importantly, PK steady state and PD steady state do not necessarily occur on the same time scale.
18. Indirect-Response Models in Population PK/PD
In population PK/PD modeling, turnover and drug-effect parameters can vary among individuals. A population model can represent typical behavior while estimating between-subject variability.
For example, the turnover rate could be modeled as:
where kout,pop is the typical population value and η represents an individual-specific deviation.
Covariates can also be incorporated when scientifically justified. For example, a biomarker's baseline or turnover rate might depend on a patient characteristic.
The resulting model can be conceptualized as:
Population PK/PD models are especially useful when concentration and response observations are sparse or collected across heterogeneous patients, because the model can combine information across individuals while explicitly representing variability.
19. Identifiability and Study Design
Indirect-response models can contain parameters that are difficult to estimate when the sampling design does not adequately capture the relevant time scales.
For example, if biomarker samples are collected only long after dosing, it may be difficult to distinguish a rapidly turning-over response from a slowly turning-over response. Similarly, sparse concentration data can make it difficult to characterize the PK driver that feeds the PD model.
Important design considerations include:
- Sampling during both rising and falling concentration phases.
- Sampling frequently enough to characterize response turnover.
- Observing baseline response before treatment.
- Collecting enough post-dose observations to characterize the delayed response.
- Using dosing regimens that provide information about the relevant concentration range.
- Including sufficient subjects or observations to estimate between-subject variability when using population models.
20. How Should an Indirect-Response Model Be Evaluated?
Model evaluation should examine both the concentration and response components.
| Evaluation | Question |
|---|---|
| Observed vs predicted concentrations | Does the PK component adequately describe the drug concentration data? |
| Observed vs predicted responses | Does the full PK/PD model reproduce the response trajectory? |
| Residual diagnostics | Are systematic errors apparent in the observation model? |
| Time-course plots | Does the model reproduce onset, peak, delay, and recovery? |
| Parameter plausibility | Are estimates consistent with biological and pharmacologic knowledge? |
| Visual predictive checks | Can simulated data reproduce important features of the observed population? |
| Sensitivity analysis | Do conclusions depend strongly on uncertain assumptions? |
A model that reproduces the observed response but requires implausible parameters or fails to describe important features of the concentration data may not be scientifically adequate.
21. What Indirect-Response Models Do Not Tell Us Automatically
Indirect-response models are useful mechanistic abstractions, but their parameters should not automatically be interpreted as direct measurements of underlying biology.
- Turnover parameters are model-dependent. kout describes the turnover process represented by the model.
- Production and loss mechanisms may not be uniquely identifiable. Type I and Type II models can sometimes generate similar response profiles.
- A fitted delay does not prove a particular biological mechanism. Other mechanisms can also create temporal separation between concentration and effect.
- Parameter estimates depend on the PK driver. Misspecification of the PK model can propagate into the PD model.
- Baseline assumptions matter. A changing untreated response may require a more complex turnover or disease-progression model.
- Extrapolation requires caution. Predictions at concentrations or time scales outside the observed data depend strongly on model assumptions.
22. A Practical Indirect-Response PK/PD Workflow
- Define the scientific question. Determine what response needs to be explained or predicted.
- Characterize the PK. Develop an adequate model for the concentration-time data.
- Characterize baseline response. Determine whether the response is approximately stable or changes naturally over time.
- Inspect concentration and response profiles together. Look for delays, hysteresis, and differences in time scales.
- Identify the likely turnover mechanism. Determine whether the drug plausibly affects production, loss, or another process.
- Select a candidate indirect-response structure. Consider Types I–IV or an appropriate extension.
- Specify the concentration-effect relationship. Use an appropriate Emax, Imax, Hill, linear, or mechanistic function.
- Estimate the model. Fit the PK and PD components using an appropriate modeling framework.
- Evaluate diagnostics. Examine predictions, residuals, time-course behavior, and parameter plausibility.
- Assess identifiability and sensitivity. Determine whether key parameters and conclusions are supported by the available data.
- Use the model for simulation or prediction. Distinguish clearly between observed data and model-based predictions.
23. Where Are Indirect-Response Models Used?
Indirect-response models are particularly useful when a drug modifies a biological quantity that has its own turnover dynamics.
| Application | Potential modeling role |
|---|---|
| Biomarkers | Describe drug-induced suppression or stimulation of biomarker production or loss |
| Hormones | Represent changes in production or clearance of endogenous hormones |
| Cell counts | Model drug effects on production or loss of circulating cell populations |
| Inflammatory markers | Characterize delayed pharmacodynamic changes in biomarkers |
| Glucose and metabolic endpoints | Represent drug effects on production and utilization processes |
| Enzyme activity | Describe indirect changes when enzyme turnover contributes to delayed response |
| Disease biomarkers | Separate drug-driven changes from baseline turnover or disease progression |
The appropriate structure depends on the biological system. Some applications require extensions involving precursor pools, transit compartments, feedback, tolerance, disease progression, or time-varying baseline behavior.
24. Extensions of Indirect-Response Models
The four classical models are a starting point rather than a complete PK/PD framework. More complex systems can be represented by adding biologically motivated components.
- Transit compartments: represent several sequential stages between drug action and measured response.
- Feedback models: allow the response itself to influence production or loss.
- Precursor-dependent models: represent formation of the measured response through an intermediate biological pool.
- Tolerance models: allow pharmacodynamic sensitivity or response pathways to change during treatment.
- Disease-progression models: allow baseline response to change independently of drug exposure.
- Turnover with delayed drug action: combines response turnover with an additional effect-site or intermediate process.
The purpose of these extensions is not simply to improve numerical fit. Each additional component should correspond to a scientific hypothesis that can be supported by the data.
25. Direct, Effect-Compartment, and Indirect-Response Models
| Model | Core idea | Source of temporal behavior |
|---|---|---|
| Direct Emax | Effect is an immediate function of concentration | Concentration itself |
| Effect compartment | Drug concentration at a hypothetical effect site drives effect | Distributional delay between plasma and effect site |
| Indirect response | Drug changes response production or loss | Turnover of the response |
| Transit/precursor model | Drug effect propagates through intermediate biological states | Multiple biological or kinetic stages |
These models can sometimes produce similar concentration-effect patterns. Their distinguishing feature is the mechanism used to generate the temporal relationship.
26. Key Takeaways
- An indirect-response PK/PD model describes drug effects that arise through changes in the production or loss of a response.
- The basic turnover model is represented by dR/dt = Rin − koutR.
- At baseline steady state, production and loss balance: Rin = koutR0.
- The four classical models are Type I inhibition of production, Type II stimulation of loss, Type III stimulation of production, and Type IV inhibition of loss.
- Types I and II generally decrease the response, while Types III and IV generally increase it.
- The response can lag behind plasma concentration because the response has its own turnover time scale.
- The response turnover half-life is ln(2)/kout in the simple first-order turnover model.
- Emax and Imax functions can describe how concentration modifies production or loss.
- Indirect-response models can generate hysteresis without requiring the drug concentration itself to have a delayed effect-site distribution.
- PK steady state and PD steady state can occur on different time scales.
- Indirect-response models and effect-compartment models can both describe delayed effects, but they represent different mechanisms.
- Production and loss mechanisms may not always be distinguishable from the observed data alone.
- Sampling design is critical because estimating turnover and drug-effect parameters requires observations across the relevant concentration and response time scales.
- Population PK/PD models can incorporate between-subject variability and covariate relationships in turnover and drug-effect parameters.
- The most useful indirect-response model is one that is scientifically plausible, identifiable from the available data, diagnostically adequate, and appropriate for the intended prediction or interpretation.
Where to Go Next
A natural progression is to study effect-compartment models, followed by transit-compartment PK/PD models, turnover models with feedback, time-to-event PK/PD models, and population PK/PD modeling.
The next tutorial can build on the indirect-response framework by showing how effect compartments and hysteresis provide an alternative mechanism for describing delayed pharmacodynamic effects, and how to distinguish that approach from response-turnover models.