From Drug Exposure to Tumor Control
In an oncology PK/PD analysis, a concentration measurement is not itself evidence that a drug has engaged its target, and target engagement is not itself evidence of meaningful tumor control. A useful analysis makes these links explicit: exposure is related to receptor occupancy, and occupancy is related to the rate of tumor growth or shrinkage. Each link has assumptions that must be checked against how the measurements were collected.
The case below is hypothetical. Concentrations are measured in plasma, receptor occupancy is recorded as a fraction from 0 to 1, and tumor volume is measured in mm3. The model is a compact way to organize the evidence; its fitted relationships do not by themselves establish that occupancy is a validated surrogate for clinical benefit.
The occupancy equation assumes that the concentration is the free concentration at the receptor, binding is at equilibrium, and a single binding site with a single \(K_D\) is an adequate approximation. In practice, occupancy may be measured directly, for example in a tumor biopsy or a suitable target-engagement assay. If so, the measured \(R\) can be used in the tumor model without first inferring occupancy from plasma concentration. The concentration-to-occupancy equation remains a check on whether the exposure data and occupancy measurements are mutually plausible.
The tumor equation assumes that occupancy is effectively constant over the interval being modeled, the untreated growth rate is constant, and the drug effect reduces the net growth rate rather than changing tumor volume by an instantaneous jump. With constant parameters, integrating the equation gives \(V(t)=V(0)\exp\{[k_g-k_{\max}R/(I_{50}+R)]t\}\). Thus the log volume ratio divided by elapsed time estimates the net growth rate.
Worked Example: Converting Concentration to Occupancy
Suppose a hypothetical drug has \(K_D=30\) mg/L, and a patient has a relevant free concentration of \(C=30\) mg/L. Under the one-site equilibrium model, calculate occupancy.
Substitute into the occupancy equation:
The model predicts 50.0% occupancy at this concentration. The concentration units cancel, leaving a dimensionless fraction. This prediction is not automatically the tumor occupancy: plasma concentration must represent concentration at the target site, and the equilibrium and binding assumptions must be reasonable. If a tumor assay instead reports \(R=0.42\), that measurement should not be silently replaced by 0.50; the discrepancy could reflect distribution, timing, assay error, or model inadequacy.
Estimating the Occupancy-Response Relationship
For tumor response, define the drug-associated reduction in growth rate as \(E(R)=k_{\max}R/(I_{50}+R)\). It is nonnegative when the parameters are positive, equals zero at zero occupancy, and approaches \(k_{\max}\) only as occupancy increases without bound. Because physical occupancy cannot exceed 1, the largest reduction allowed by this model over feasible occupancy is \(E(1)=k_{\max}/(I_{50}+1)\), not \(k_{\max}\). This distinction matters when asking whether tumor stabilization is even possible.
An untreated or control group can help estimate \(k_g\). For a treated group with approximately constant occupancy over an interval of length \(T\), calculate the observed net rate as \(\log[V(T)/V(0)]/T\). The estimated drug-associated reduction is then \(k_g\) minus that net rate. This calculation requires comparable tumor measurement timing and a defensible untreated growth-rate estimate; changes due to surgery, concurrent therapy, or different disease status can make the comparison misleading.
Worked Example: Estimating the Occupancy-Response Parameters
Consider a hypothetical case series with an untreated growth-rate estimate \(k_g=0.0800\) per week. Occupancy is measured during each five-week interval and treated as constant. Each tumor starts at 100 mm3. The following observed values are constructed to illustrate parameter estimation.
| Group | Occupancy \(R\) | Week 0 volume (mm3) | Week 5 volume (mm3) |
|---|---|---|---|
| Untreated | 0 | 100 | 149.2 |
| Drug A | 0.25 | 100 | 110.5 |
| Drug B | 0.50 | 100 | 92.31 |
| Drug C | 0.75 | 100 | 81.87 |
For Drug A, the observed net rate is \(\log(110.5/100)/5\), approximately \(0.0200\) per week. Its estimated drug-associated reduction is \(0.0800-0.0200=0.0600\) per week. For Drug C, the net rate is \(\log(81.87/100)/5\), approximately \(-0.0400\) per week; the estimated reduction is \(0.0800-(-0.0400)=0.1200\) per week.
Use these two points to solve for \(k_{\max}\) and \(I_{50}\):
Dividing the first equation by the second eliminates \(k_{\max}\):
The left side is 0.5, so \(2(I_{50}+0.75)=3(I_{50}+0.25)\). Expanding gives \(2I_{50}+1.50=3I_{50}+0.75\), hence \(I_{50}=0.750\). Substituting into the Drug A equation gives \(0.0600=k_{\max}(0.25/1.00)=0.25k_{\max}\), so \(k_{\max}=0.240\) per week.
The model predicts a reduction of \(0.240(0.50)/(0.750+0.50)=0.0960\) per week for Drug B. Its predicted net rate is \(0.0800-0.0960=-0.0160\) per week. Over five weeks, the predicted volume is \(100\exp(-0.0160\times5)=92.31\) mm3, matching the constructed observation. This agreement is an internal check, not independent validation: two points determined the two parameters, and Drug B is consistent with the resulting curve.
Derivation and Feasible Tumor Control
The constant-occupancy tumor equation can be solved by separating variables. For positive tumor volume, divide both sides by \(V(t)\) and multiply by \(dt\). Integrating from 0 to \(t\) gives the log-volume relation below. It justifies using the log ratio to estimate the net rate.
Tumor stabilization in this model occurs when net growth is zero or negative, so the drug-associated reduction must be at least \(k_g\). With the fitted values, the maximum feasible reduction occurs at \(R=1\):
Since 0.1371 exceeds \(k_g=0.0800\) per week, stabilization is feasible in the model. Solving \(0.240R/(0.750+R)\ge0.0800\) gives \(0.240R\ge0.0600+0.0800R\), hence \(R\ge0.375\). At exactly 37.5% occupancy, the predicted net growth rate is zero. For \(R=0.75\), the predicted net rate is \(-0.0400\) per week, corresponding to \(100[\exp(-0.0400\times5)-1]\approx-18.1\%\) volume change over five weeks.
Worked Example: Does the Predicted Response Meet a Shrinkage Target?
Suppose the hypothetical study prespecifies at least a 20% tumor-volume reduction by week 5 as a meaningful shrinkage target for this analysis. This is an operational example criterion, not a universal oncology definition. Find the occupancy required by the fitted model.
A 20% reduction means \(V(5)/V(0)\le0.80\). Taking logarithms and dividing by five weeks, the net growth rate must be at most \(\log(0.80)/5=-0.04463\) per week. Thus the drug-associated reduction must be at least \(0.0800-(-0.04463)=0.12463\) per week. Solve:
At the target occupancy, the model predicts a concentration of \(C=K_D R/(1-R)\). If \(K_D=30\) mg/L, this is \(30(0.8102)/(1-0.8102)\approx128.1\) mg/L. The Drug C occupancy of 0.75 therefore predicts shrinkage, but not the specified 20% reduction: its predicted decrease is about 18.1%. The target is feasible because 0.8102 is below the physical occupancy limit of 1. Even at full occupancy, the model predicts a five-week volume ratio of \(\exp[(0.0800-0.1371)5]\approx0.751\), a decrease of about 24.9%.
Diagnostics, Alternatives, and Interpretation
A fitted concentration-occupancy-response chain is useful only if its assumptions and predictive behavior are plausible. At a minimum, inspect the following:
- Exposure and occupancy agreement. Compare measured occupancy with the occupancy predicted from concentration, allowing for assay error and timing. Systematic disagreement may indicate that plasma concentration is not a good proxy for tumor exposure or that equilibrium binding is inadequate.
- Time stability. If concentration and occupancy vary substantially during follow-up, a single average occupancy can misrepresent effect. A time-varying model would need to relate the changing \(R(t)\) to the accumulated change in tumor volume.
- Response pattern. Plot observed log-volume changes against occupancy and compare them with the fitted curve. Curvature, plateaus inconsistent with the model, or different patterns across patients can signal misspecification or unmodeled variation.
- Biological and measurement comparability. Confirm that tumor measurements, observation windows, and untreated growth estimates refer to comparable settings. Volume measurement error and changes in tumor composition can distort apparent growth rates.
One alternative is to model tumor volume directly against concentration, bypassing the occupancy measurement. That may be useful when occupancy data are unavailable, but it does not reveal whether exposure reaches the target or whether occupancy is the more proximal predictor. Another approach is to use observed occupancy as the response-model input, as in the case above, while separately checking the concentration-occupancy relation. A more flexible model can estimate parameters using all observations rather than solving equations from selected points; it should also represent measurement and between-subject variability when the data support doing so. More flexibility does not remove the need for diagnostic checks.
A crucial limitation is that an association between occupancy and tumor change in an oncology case does not establish that occupancy predicts treatment benefit across patients, trials, or populations. Occupancy can be a mechanistic biomarker and a useful predictor within a specific analysis without being a validated surrogate endpoint. Treatment allocation, disease severity, tumor biology, measurement error, and post-baseline selection can affect the apparent association. The earlier material on surrogate validation is relevant when making broader claims about clinical benefit; the present model alone cannot justify those claims.
A particularly common mistake is to call \(k_{\max}\) the maximum reduction achievable at a physical occupancy. In this parameterization, it is the asymptotic reduction as \(R\) tends to infinity. Over the feasible range \(0\le R\le1\), the maximum is \(k_{\max}/(I_{50}+1)\). If this feasible maximum is below \(k_g\), the model cannot predict stabilization at any possible occupancy, even though \(k_{\max}\) itself may exceed \(k_g\). Check the feasible maximum before interpreting a fitted parameter as evidence that tumor control is attainable.
Check Your Understanding
Answer each question, then compare with the brief solutions.
- If \(C=15\) mg/L and \(K_D=45\) mg/L, what occupancy does the one-site equilibrium model predict?
- For the fitted model \(k_g=0.0800\), \(k_{\max}=0.240\), and \(I_{50}=0.750\), what is the greatest reduction in growth rate possible over \(0\le R\le1\)?
- At \(R=0.50\), calculate the net tumor growth rate and state whether the model predicts growth or shrinkage.
- Why does agreement between a predicted and observed tumor volume at a point used to estimate model parameters not constitute independent validation?
- If the feasible maximum reduction is less than \(k_g\), what does this model imply about tumor stabilization?
Solutions. (1) \(R=15/(45+15)=0.250\), or 25.0%. (2) \(E(1)=0.240/(0.750+1)=0.1371\) per week. (3) The reduction is \(0.240(0.50)/(0.750+0.50)=0.0960\) per week, so the net rate is \(0.0800-0.0960=-0.0160\) per week; the model predicts shrinkage. (4) The observation helped determine the fitted parameters, so matching it reflects fitting rather than a separate predictive test. (5) Since even the largest physically attainable reduction cannot offset untreated growth, the model predicts positive net growth at every feasible occupancy.