1. What Are Tumor Growth and Tumor Response Models?
Tumor growth models describe how tumor burden changes over time in the absence of treatment or under a specified natural-history assumption. Tumor response models extend that framework by describing how treatment changes the tumor trajectory.
In oncology pharmacometrics, the tumor burden may be represented by a measurable quantity such as tumor volume, the sum of target-lesion diameters, tumor weight, or another longitudinal biomarker. The model then describes how that quantity evolves through time.
A tumor-response model is a mathematical representation of how tumor burden changes through time as a consequence of underlying growth processes and treatment effects.
2. What Questions Can Tumor Models Help Answer?
Tumor-growth and tumor-response models can address several related scientific questions.
| Question | Model component | What it describes |
|---|---|---|
| How would the tumor change without treatment? | Growth model | Underlying tumor expansion or natural history |
| How quickly does treatment reduce tumor burden? | Treatment-effect model | Magnitude and time course of tumor killing or inhibition |
| Does the treatment effect depend on exposure? | PK/PD relationship | Relationship between concentration or exposure and effect |
| Why does the tumor begin growing again? | Regrowth / resistance model | Loss of treatment effect or emergence of resistant disease |
| How variable are responses among patients? | Population model | Between-patient variability and covariate effects |
| What might happen under a different dosing regimen? | Simulation model | Predicted tumor trajectories under alternative exposure scenarios |
These questions are related but distinct. A model that accurately describes tumor growth may not automatically explain treatment response, and a treatment-response model may require a separate model for drug exposure.
3. What Is Tumor Burden?
The first step in tumor modeling is defining the quantity that represents tumor burden. In clinical studies, this may be derived from imaging or another longitudinal measurement.
For example, let \(T(t)\) represent tumor burden at time \(t\). Depending on the application, \(T(t)\) might represent tumor volume, total lesion diameter, or another quantitative tumor measurement.
The observed data can then be viewed as measurements of an underlying latent trajectory:
where \(Y_i\) is the observed tumor measurement at time \(t_i\), and \(\varepsilon_i\) represents measurement or residual variability.
4. Modeling Tumor Growth
A tumor-growth model specifies how tumor burden changes in the absence of a treatment effect. The simplest models assume that the instantaneous growth rate is proportional to tumor burden.
An exponential growth model can be written as:
where \(r\) is the intrinsic growth-rate constant.
The corresponding solution is:
Under this model, the tumor grows at a constant fractional rate. As tumor burden increases, the absolute rate of increase also increases.
Exponential growth assumes a constant fractional growth rate. It can be useful over a limited range but may become unrealistic when applied over long periods.
More flexible growth models can incorporate slowing growth as tumor burden increases. Two commonly used alternatives are the logistic and Gompertz models.
5. The Logistic Tumor-Growth Model
The logistic model assumes that growth slows as tumor burden approaches a limiting level.
Here, \(r\) is a growth-rate parameter and \(K\) represents a carrying-capacity-like parameter within the mathematical model.
When \(T\) is small relative to \(K\), the term \(1-T/K\) is close to 1 and growth is approximately exponential. As \(T\) approaches \(K\), the growth rate decreases.
The logistic model is therefore useful when tumor growth appears to decelerate with increasing burden and when an asymptotic growth pattern is scientifically reasonable.
6. The Gompertz Tumor-Growth Model
The Gompertz model provides another way to represent growth that slows as tumor burden increases.
As with the logistic model, \(K\) represents an asymptotic scale within the mathematical formulation. The growth rate depends on both current tumor burden and its relationship to \(K\).
A useful feature of the Gompertz formulation is that its relative growth rate declines progressively as the tumor becomes larger.
| Growth model | Basic assumption | Characteristic behavior |
|---|---|---|
| Exponential | Constant fractional growth rate | Growth accelerates in absolute terms without an upper limit |
| Logistic | Growth slows with increasing burden | Approaches an asymptotic level |
| Gompertz | Relative growth rate declines with burden | Strong deceleration as tumor burden increases |
No growth model is universally correct. Model choice should be driven by the scientific question, observed trajectory, study duration, identifiability, and intended use of the model.
7. From Tumor Growth to Tumor Response
A treatment-response model adds a process that modifies the underlying tumor trajectory.
A simple conceptual model separates growth and treatment-induced loss:
where \(G(T)\) represents tumor growth and \(K(T,C)\) represents treatment-induced tumor loss as a function of tumor burden and drug concentration \(C\).
For example, if growth is exponential and drug killing is proportional to both tumor burden and concentration, a simple model could be written as:
or equivalently:
The sign and magnitude of \(r-k(C)\) determine the instantaneous direction of tumor change in this simplified model.
8. Linking Drug Exposure to Tumor Response
In pharmacometric models, the treatment effect is often linked to drug concentration or exposure through a pharmacodynamic relationship.
A simple \(E_{\max}\) model is:
Here, \(E_{\max}\) is the maximum modeled effect and \(EC_{50}\) is the concentration associated with half of the maximum effect under the model.
The tumor model can then be expressed conceptually as:
The drug concentration \(C(t)\) may itself come from a PK model:
This creates a mechanistic chain connecting dose to exposure and then exposure to tumor response.
9. Modeling Treatment-Induced Tumor Kill
Many tumor-response models represent treatment as an additional loss process. One simple formulation is:
where \(k_{\mathrm{kill}}(C)\) is a concentration-dependent killing rate.
For example, an \(E_{\max}\)-type killing rate can be written as:
Combining the two equations gives:
This model has an intuitive interpretation: the tumor grows at rate \(r\), while treatment produces a concentration-dependent loss rate.
However, real treatment effects may not be instantaneous. A delay between drug exposure and tumor response can motivate the use of an effect compartment, indirect-response model, transit compartments, or other delay mechanisms.
10. Why Do Tumors Shrink and Then Regrow?
A common longitudinal pattern in oncology is an initial decrease in tumor burden followed by stabilization or regrowth.
A tumor-response model can represent this pattern through several mechanisms, including:
- declining drug exposure;
- time-varying treatment effect;
- emergence of resistant tumor cells;
- different growth rates among tumor-cell subpopulations;
- adaptation of the tumor microenvironment; or
- an inadequate treatment-effect model.
A simple resistance model can divide tumor burden into sensitive and resistant components:
where \(T_S\) represents treatment-sensitive tumor cells and \(T_R\) represents a resistant population.
A conceptual two-population system could be:
If treatment strongly suppresses \(T_S\) but has little effect on \(T_R\), the resistant population can eventually dominate the total tumor burden and produce apparent regrowth.
11. Accounting for Delayed Tumor Response
Drug concentration and tumor response often operate on different time scales. A drug may reach peak concentration quickly while measurable tumor shrinkage occurs much later.
One way to represent this is with an effect compartment:
where \(C_e\) is an effect-site concentration and \(k_{eo}\) controls equilibration between plasma concentration \(C\) and the effect compartment.
The tumor effect can then depend on \(C_e\) rather than directly on plasma concentration:
Other approaches use chains of transit compartments or explicit biological delay mechanisms.
The important principle is that a delayed response should not automatically be interpreted as evidence for delayed drug absorption. The delay may arise from distribution, intracellular processes, downstream signaling, cell-cycle effects, or the time required for affected cells to disappear.
12. Common Tumor-Growth and Response Model Structures
| Model structure | What it represents | Typical use |
|---|---|---|
| Exponential growth | Constant fractional growth | Simple natural-history description |
| Logistic growth | Growth slowing toward an asymptotic level | Density-limited or saturating growth representation |
| Gompertz growth | Progressively declining relative growth rate | Tumor-growth trajectories with strong deceleration |
| Growth + linear kill | Growth opposed by treatment-induced loss | Simple tumor shrinkage models |
| Growth + \(E_{\max}\) kill | Concentration-dependent treatment effect | Exposure-response modeling |
| Delayed-effect model | Separates drug concentration from effect timing | Delayed tumor response |
| Resistant subpopulation model | Sensitive and resistant tumor populations | Response followed by regrowth |
| Population tumor model | Typical trajectory plus variability and covariates | Clinical oncology datasets |
The appropriate level of complexity depends on the data and scientific purpose. A highly mechanistic model is not automatically preferable if the available data cannot identify its parameters.
13. Connecting Model Behavior to Observed Response Patterns
Different combinations of growth and treatment-effect parameters can produce qualitatively different tumor trajectories.
Conceptual trajectories illustrate that shrinkage, stabilization, and regrowth can emerge from different balances between tumor growth and treatment effect.
Importantly, these curves are model outputs, not direct explanations of biology. Multiple mechanisms can produce similar observed patterns over a limited observation period.
14. Key Parameters in Tumor-Response Models
| Parameter | Interpretation | Influence on the trajectory |
|---|---|---|
| \(r\) | Intrinsic tumor-growth rate | Controls how rapidly tumor burden increases |
| \(K\) | Asymptotic scale in logistic/Gompertz formulations | Controls the degree of growth limitation |
| \(k_{\max}\) | Maximum modeled drug-induced kill rate | Controls the maximum strength of treatment effect |
| \(EC_{50}\) | Exposure producing half-maximal effect | Controls exposure sensitivity |
| \(k_{eo}\) | Effect-site equilibration rate | Controls delay between plasma exposure and modeled effect |
| \(r_R\) | Growth rate of resistant cells | Influences the speed of regrowth after treatment suppression |
Parameter interpretation always depends on the structural model. For example, an apparent growth-rate parameter in a Gompertz model does not have exactly the same interpretation as \(r\) in an exponential model.
15. How Are Tumor-Response Models Estimated?
Tumor models are generally estimated by comparing observed longitudinal tumor measurements with predictions from the proposed model.
- Define the scientific question. Decide whether the goal is to describe natural history, quantify treatment effect, predict response, characterize resistance, or support simulation.
- Assemble longitudinal tumor data. Record tumor measurements and their observation times, together with dosing and treatment information.
- Characterize drug exposure. When appropriate, use measured concentrations or a PK model to generate the exposure driving the tumor effect.
- Select a candidate growth model. Consider exponential, logistic, Gompertz, or another scientifically justified structure.
- Add a treatment-effect model. Specify how treatment modifies tumor growth or loss.
- Specify the observation model. Account for residual variability and the measurement scale of the tumor data.
- Estimate parameters. Use an appropriate estimation framework, potentially including nonlinear mixed-effects methods for population data.
- Evaluate model adequacy. Examine fits, residuals, parameter plausibility, predictive checks, and sensitivity to structural assumptions.
- Use the model for prediction or simulation. Clearly distinguish observed data from model-based predictions.
16. Worked Example: Growth Versus Treatment Effect
Consider a simplified tumor model with an exponential baseline growth rate of \(r=0.02\) day\(^{-1}\). Suppose treatment produces an effective kill rate of \(k=0.05\) day\(^{-1}\) during a period of constant drug exposure.
Step 1: Specify the model
Step 2: Calculate the net rate
Because the net rate is negative, the simplified model predicts tumor shrinkage during this treatment period.
Step 3: Predict tumor burden
Suppose the initial tumor burden is \(T_0=100\) units. The solution is:
Step 4: Evaluate the tumor after 20 days
The model therefore predicts a tumor burden of approximately 54.9 units after 20 days under the stated constant-effect assumption.
Step 5: What happens when treatment effect falls?
If the effective treatment rate later decreases to \(k=0.01\) day\(^{-1}\), then:
The net rate becomes positive, so tumor burden begins increasing again.
The example is intentionally simplified. Real tumor-response models may incorporate time-varying exposure, delayed effects, nonlinear drug effects, multiple tumor populations, patient variability, and measurement error.
17. Population Tumor-Response Models
Clinical oncology datasets contain measurements from multiple patients, and tumor trajectories can differ substantially between individuals.
A population model can represent an individual parameter as:
where \(\theta_{\mathrm{pop}}\) is the typical population value and \(\eta_i\) represents an individual-specific deviation.
Covariates can also be incorporated when there is a scientific rationale for doing so. For example, a growth parameter might depend on a patient characteristic:
where \(X_i\) is a covariate and \(f(\cdot)\) describes the modeled relationship.
This framework allows the model to distinguish:
- typical tumor-growth behavior;
- between-patient variability;
- residual measurement variability; and
- systematic relationships with patient or treatment characteristics.
18. Why Identifiability Matters
Tumor-response models can become structurally complex very quickly. A model containing growth, treatment effect, delay, resistance, and multiple populations may have many parameters.
Not every parameter is necessarily identifiable from a particular dataset.
For example, if only a few tumor measurements are available after treatment begins, it may be difficult to separately estimate:
- the baseline growth rate;
- the maximum treatment effect;
- the exposure-response parameter;
- the delay between exposure and effect; and
- the growth rate of a resistant population.
Identifiability should therefore be considered before interpreting complex parameters as independently estimated biological quantities.
19. How Should Tumor Models Be Evaluated?
A tumor-response model should be evaluated against the scientific purpose for which it was developed.
Useful evaluations can include:
- Observed-versus-predicted plots: compare model predictions with observed tumor measurements.
- Individual trajectory plots: examine whether the model reproduces longitudinal patterns for individual patients.
- Residual diagnostics: identify systematic deviations from model predictions.
- Prediction-corrected visual predictive checks: compare observed data with simulations from the model when appropriate.
- Parameter plausibility: determine whether estimated values are consistent with the intended model and available evidence.
- Sensitivity analysis: examine how predictions change when important assumptions or parameters change.
- External or temporal validation: evaluate predictive performance using data not used to construct the model when such data are available.
A model can fit the observed data well and still be unsuitable for extrapolation or mechanistic interpretation. Evaluation should therefore consider the intended use of the model, not just goodness of fit.
20. What Can Tumor Models Be Used to Predict?
Once a model has been developed and evaluated, it can be used to simulate tumor trajectories under specified conditions.
Potential applications include:
- predicting tumor trajectories under alternative dosing regimens;
- quantifying the expected magnitude and duration of tumor response;
- evaluating how changes in exposure may alter tumor response;
- exploring treatment schedules;
- simulating response heterogeneity across a patient population;
- investigating possible resistance scenarios; and
- supporting quantitative decision making in drug development.
These predictions are conditional on the model structure, estimated parameters, input assumptions, and range of applicability.
21. Tumor Models and Clinical Response Endpoints
Longitudinal tumor models and clinical response endpoints answer related but different questions.
A continuous tumor model represents the trajectory of tumor burden through time. Clinical oncology endpoints may instead classify patients according to categories such as response, stable disease, or progression, or summarize time-to-event outcomes.
A longitudinal model can therefore provide information that is lost when continuous tumor measurements are reduced to a categorical endpoint.
| Approach | Primary representation | Information retained |
|---|---|---|
| Longitudinal tumor model | Continuous tumor trajectory | Timing, magnitude, duration, shrinkage, and regrowth |
| Categorical response | Response category | Classification according to predefined criteria |
| Time-to-event analysis | Time until an event | Timing of progression or another defined event |
These approaches can be complementary. A tumor-response model may provide a mechanistic or longitudinal description, while clinical endpoints may provide standardized summaries for clinical interpretation and decision making.
22. What Tumor Models Do Not Tell Us Automatically
Mathematical tumor models are simplifications. Several cautions are important when interpreting them.
- A fitted curve is not automatically a biological mechanism. Different mechanisms can generate similar trajectories.
- A good fit does not establish biological truth. Fit quality is only one part of model evaluation.
- Parameter meaning depends on model structure. A growth parameter in one model may not be directly comparable with a parameter from another model.
- Resistance cannot necessarily be inferred from regrowth alone. Alternative explanations may exist.
- Sparse measurements limit what can be learned. Important features may occur between observation times.
- Measurement conventions matter. Tumor diameter, tumor volume, and other tumor-burden measures are not interchangeable.
- Extrapolation can be sensitive to assumptions. Predictions far outside the observed data range may depend strongly on structural assumptions.
23. A Practical Tumor-Modeling Workflow
- Start with the scientific question. Define whether the objective is natural-history characterization, treatment-effect quantification, prediction, resistance analysis, or simulation.
- Define tumor burden. Establish exactly what longitudinal measurement the model will represent.
- Explore the data. Plot individual and population tumor trajectories against time and treatment exposure.
- Specify the baseline growth model. Consider exponential, logistic, Gompertz, or another appropriate structure.
- Add treatment effect. Connect drug exposure to tumor killing or growth inhibition when supported by the data.
- Consider delays. Add an appropriate delay mechanism when exposure and tumor response occur on different time scales.
- Consider resistance. Introduce additional populations or time-varying mechanisms only when justified by the scientific question and available data.
- Specify variability. Account for residual error and, when appropriate, between-patient variability.
- Evaluate identifiability. Determine whether the available data can support the proposed model complexity.
- Evaluate predictive performance. Use diagnostics, simulations, and validation appropriate to the intended use.
- Simulate and interpret. Use the model to explore clinically relevant scenarios while clearly communicating assumptions and uncertainty.
24. Key Takeaways
- Tumor-growth models describe how tumor burden changes through time under a specified natural-history assumption.
- Common growth structures include exponential, logistic, and Gompertz models.
- Tumor-response models add treatment-induced tumor loss or growth inhibition to the underlying growth process.
- PK/PD-linked tumor models can connect dose to drug exposure and exposure to tumor response.
- Treatment response can be modeled using concentration-dependent functions such as \(E_{\max}\) relationships.
- Delayed effects may require effect compartments, transit compartments, or other mechanisms that separate exposure from response timing.
- Regrowth can be represented phenomenologically or through models containing sensitive and resistant tumor populations.
- Population models can describe typical tumor behavior together with between-patient variability and covariate relationships.
- Complexity should be supported by the available data; additional parameters are useful only when they can be meaningfully estimated and evaluated.
- A good fit does not by itself establish a biological mechanism or guarantee predictive validity.
- Tumor models can support simulation and prediction, but predictions remain conditional on model assumptions and parameter uncertainty.
- The most useful tumor model is one that is sufficiently detailed to answer the scientific question while remaining identifiable, interpretable, and testable.
Where to Go Next
A natural progression is to study PK/PD-linked tumor-growth inhibition models in greater detail, followed by delayed tumor response, tumor-growth inhibition metrics, resistance models, longitudinal nonlinear mixed-effects models, and quantitative systems pharmacology approaches to oncology.
The next tutorial can build directly on this framework by deriving a treatment-response model from a PK model, an \(E_{\max}\) exposure-response relationship, and a tumor-growth equation, then showing how the combined system can reproduce tumor shrinkage, stabilization, and regrowth.