Direct-Effect Emax Models for Biomarker Response

From Exposure to a Saturable Biomarker Change

When a biomarker changes with drug concentration, a simple straight-line relationship can be a useful first description, but it may imply that the change grows without limit as concentration increases. A saturable model instead allows the drug-attributable change to approach a finite maximum. The direct-effect Emax model expresses that idea with a small number of interpretable parameters: a baseline response, a maximum drug-attributable change, and the concentration associated with half that change.

“Direct effect” means that the modeled response at a given observation is determined by the drug concentration at the corresponding time, without an explicit delay between concentration and response. This is an assumption about the exposure-response relationship, not a claim that every biomarker responds instantaneously in biology. The earlier material on linking biomarker changes to individual drug exposure motivates using a concentration relevant to each response observation; the model here specifies the shape of that relationship.

Definition: Let $C$ be the drug concentration associated with a biomarker observation, $R_0$ the reference baseline response, and $\Delta R(C)$ the drug-attributable change from baseline. The direct-effect Emax model is $$ \Delta R(C)=\frac{E_{\max}C}{EC_{50}+C}, \qquad R(C)=R_0+\Delta R(C). $$ Here, $E_{\max}$ is the signed maximum drug-attributable change in the biomarker, and $EC_{50}>0$ is the concentration at which the modeled change is half of $E_{\max}$. Concentration and $EC_{50}$ have the same units; $E_{\max}$, $\Delta R$, and $R_0$ are expressed in biomarker units.

The sign of $E_{\max}$ gives the direction of the modeled effect. If $E_{\max}<0$, concentration is associated with a decrease from baseline; if $E_{\max}>0$, it is associated with an increase. The model’s maximum is therefore signed: a negative $E_{\max}$ is the limiting decrease, not a positive magnitude. At $C=0$, the predicted drug-attributable change is zero and the predicted response is $R_0$. As concentration increases without bound, $\Delta R(C)$ approaches $E_{\max}$.

Deriving and Reading the Emax Curve

The model can be built by describing the fraction of the maximum effect attained at concentration $C$. For the basic Emax model, take that fraction to be $C/(EC_{50}+C)$. It is zero at zero concentration and approaches one as concentration grows. At $C=EC_{50}$, the fraction is exactly one-half:

\frac{EC_{50}}{EC_{50}+EC_{50}}=\frac{1}{2}.

Multiplying this fraction by the signed maximum change $E_{\max}$ gives the drug-attributable change. Adding the reference baseline gives the predicted response. This construction explains why $EC_{50}$ refers to half of the drug-attributable maximum change, rather than half of the total response $R(C)$.

The assumptions needed to interpret this curve include a well-defined baseline on the same measurement scale as the response; nonnegative concentrations; a positive $EC_{50}$; and a response that moves monotonically toward one finite limiting change over the concentration range of interest. The direct-effect form also assumes that the selected concentration corresponds to the response time closely enough that an explicit delay is not needed. To use the model as a conditional mean with an observation-error model, one may write $R_i=R_0+E_{\max}C_i/(EC_{50}+C_i)+\varepsilon_i$, where $\varepsilon_i$ has mean zero conditional on concentration. A normal, constant-variance error assumption may be used for a basic nonlinear regression analysis, but it is an additional assumption, not part of the Emax curve itself.

Worked Example: Predicting a Decrease From Baseline

Suppose a hypothetical biomarker has baseline $R_0=100$ pg/mL. A direct-effect Emax model has $E_{\max}=-60$ pg/mL and $EC_{50}=2$ mg/L. Predict the response at $C=3$ mg/L.

The model first predicts the change from baseline:

\Delta R(3)=\frac{(-60\ \text{pg/mL})(3\ \text{mg/L})}{2\ \text{mg/L}+3\ \text{mg/L}} =\frac{-180}{5}\ \text{pg/mL} =-36\ \text{pg/mL}.

The predicted response is $R(3)=100-36=64$ pg/mL. The negative $E_{\max}$ indicates a decrease; at 3 mg/L the model predicts 60% of the maximum decrease, because $3/(2+3)=0.60$. At $C=2$ mg/L, the predicted change would be $-60/2=-30$ pg/mL, half of $E_{\max}$. These are model predictions, not evidence that the biomarker change causes a clinical outcome.

The curve’s low- and high-concentration behavior clarifies when the model can and cannot be distinguished from a simpler line. At concentrations much smaller than $EC_{50}$, $EC_{50}+C$ is approximately $EC_{50}$, so $\Delta R(C)$ is approximately $(E_{\max}/EC_{50})C$. At concentrations much larger than $EC_{50}$, the response is close to its plateau. Thus, low concentrations mainly inform the ratio $E_{\max}/EC_{50}$, while observations near and above the bend are important for separating the parameters and estimating the plateau.

The Hill Model: Allowing a Different Curve Shape

The Hill model generalizes the concentration-response fraction by raising concentration and $EC_{50}$ to a positive power $\gamma$, called the Hill coefficient. When $\gamma=1$, it reduces to the basic Emax model. Values other than one alter how gradually or sharply the curve transitions from low response to saturation; they do not change the definition of $EC_{50}$.

Definition: The direct-effect Hill model for biomarker change is $$ \Delta R(C)=\frac{E_{\max}C^\gamma}{EC_{50}^{\gamma}+C^\gamma}, \qquad R(C)=R_0+\Delta R(C), \qquad \gamma>0. $$ For $C=EC_{50}$, the concentration-dependent fraction is $1/2$, so the predicted change is $E_{\max}/2$. The units of $EC_{50}$ are concentration units, and $\gamma$ is dimensionless.

One way to derive this form is to posit that the fraction of maximum change is $f(C)=C^\gamma/(EC_{50}^{\gamma}+C^\gamma)$. At $C=EC_{50}$, numerator and the two denominator terms are each $EC_{50}^{\gamma}$, giving $f=1/2$. As $C$ approaches zero, $f$ approaches zero; as $C$ grows without bound, $f$ approaches one. Multiplication by $E_{\max}$ and addition of $R_0$ then give the Hill model.

The coefficient $\gamma$ controls steepness around $EC_{50}$. To see this, differentiate the fraction with respect to log concentration. At $C=EC_{50}$, the slope of the fraction against $\log C$ is $\gamma/4$, so the corresponding slope of the predicted change is $E_{\max}\gamma/4$. A larger $\gamma$ produces a sharper transition near $EC_{50}$, while a smaller positive $\gamma$ produces a more gradual transition. The coefficient is a curve-shape parameter; it should not automatically be interpreted as a count of binding sites or as a mechanistic explanation.

f(C)=\frac{C^\gamma}{EC_{50}^{\gamma}+C^\gamma}, \qquad \frac{df}{d\log C} =\frac{\gamma C^\gamma EC_{50}^{\gamma}} {(EC_{50}^{\gamma}+C^\gamma)^2}, \qquad \left.\frac{df}{d\log C}\right|_{C=EC_{50}}=\frac{\gamma}{4}.

Worked Example: Interpreting a Hill Coefficient

Consider a hypothetical biomarker with $R_0=25$ units, $E_{\max}=40$ units, $EC_{50}=4$ mg/L, and $\gamma=2$. Predict the response at 2 mg/L and at 8 mg/L.

At 2 mg/L, the fraction of maximum change is $2^2/(4^2+2^2)=4/20=0.20$. Therefore, $\Delta R(2)=40(0.20)=8$ units, and $R(2)=25+8=33$ units.

At 8 mg/L, the fraction is $8^2/(4^2+8^2)=64/80=0.80$. Thus, $\Delta R(8)=40(0.80)=32$ units, and $R(8)=25+32=57$ units. At $C=4$ mg/L, the fraction is $16/(16+16)=0.50$, so the change is 20 units. The value $\gamma=2$ makes the change rise more sharply around 4 mg/L than the basic model with $\gamma=1$; it does not move the half-maximum concentration.

Estimating and Interpreting Parameters

In practice, observed biomarker responses will not generally fall exactly on a smooth model curve. Parameters are estimated by fitting the model to observations, often using nonlinear regression or a likelihood-based method under a stated error model. The fitted curve describes the expected response conditional on concentration under that model; individual observations may differ because of measurement error and other unexplained variation.

The parameters should be interpreted jointly and in context. $E_{\max}$ is a limiting drug-attributable change, not the maximum total biomarker value; the total response tends to $R_0+E_{\max}$. $EC_{50}$ is a concentration, not a dose, and is conditional on the model, biomarker scale, and exposure metric used. For $\gamma=1$, the initial slope against concentration is $E_{\max}/EC_{50}$; a steep low-concentration slope can therefore result from a larger $E_{\max}$, a smaller $EC_{50}$, or both.

Worked Example: Recovering EC50 From a Biomarker Change

In a hypothetical setting, let $R_0=80$ units and $E_{\max}=-50$ units, with a basic Emax model. At a measured concentration of 3 mg/L, the observed response is 60 units. Treat this value as lying on the model curve for the purpose of calculating the implied $EC_{50}$.

The observed change from baseline is $60-80=-20$ units. The model equation is $-20=(-50)(3)/(EC_{50}+3)$. Dividing both sides by $-50$ gives a fraction of $0.40$, so:

0.40=\frac{3}{EC_{50}+3}, \qquad 0.40(EC_{50}+3)=3, \qquad EC_{50}=4.50\ \text{mg/L}.

As a check, at 3 mg/L the model predicts $\Delta R=-50(3)/(4.5+3)=-20$ units and $R=60$ units. At 9 mg/L, the predicted change is $-50(9)/(4.5+9)=-33.3$ units, giving a response of approximately $46.7$ units. This calculation illustrates parameter interpretation; a single observation does not establish that the Emax model is appropriate or provide a reliable estimate of all its parameters.

Assumptions, Diagnostics, and Alternatives

A direct-effect Emax or Hill curve is most defensible when the response changes monotonically with the selected concentration and shows evidence of saturation, and when the data cover enough of the concentration range to identify the relevant parameters. If all observed concentrations are far below $EC_{50}$, the data may support a roughly linear rise but provide little information to distinguish $E_{\max}$ from $EC_{50}$. If there are no observations near a plateau, the fitted $E_{\max}$ can depend heavily on extrapolation. A fitted value alone does not demonstrate that saturation was observed.

Useful checks include plotting observed response changes against concentration with the fitted curve, examining whether residuals show systematic curvature, and checking whether predicted responses are plausible on the biomarker’s measurement scale. An apparent trend in residuals, a plateau inconsistent with the observations, or estimates that change greatly when one concentration group is omitted can signal weak model support. Assay limitations, an unstable baseline, sparse concentration coverage, or an incorrectly paired concentration and response can also make the apparent curve misleading. These checks assess the adequacy of this response model; they do not establish a biological mechanism.

A linear exposure-response model is a simpler alternative and may be adequate over a narrow range, especially when observed concentrations are well below the apparent bend. Unlike Emax, an unrestricted line does not approach a finite plateau. The Hill model is an alternative when the transition appears steeper or more gradual than the basic Emax shape, but estimating $\gamma$ adds flexibility and can be difficult with sparse data. Choosing the more flexible curve simply because it fits a small data set better risks overinterpreting poorly identified parameters. Neither curve addresses a delayed response: if response depends on prior exposure or shows a time-dependent concentration-response pattern, an instantaneous direct-effect assumption is not sufficient.

A common mistake is to call $EC_{50}$ the concentration that gives half the total baseline-plus-drug response. That is incorrect because the model defines half of the drug-attributable change: $R(EC_{50})=R_0+E_{\max}/2$. For example, with $R_0=100$ pg/mL and $E_{\max}=-60$ pg/mL, the response at $EC_{50}$ is 70 pg/mL, not half of the total response. Always subtract the modeled baseline before interpreting the fraction of $E_{\max}$ attained.

Key takeaway: A direct-effect Emax curve describes a saturable, signed change from baseline as a function of concentration. Interpret $E_{\max}$ as the limiting change, $EC_{50}$ as the half-maximum concentration, and $\gamma$ as a shape parameter when using the Hill extension. These interpretations depend on adequate concentration coverage and a reasonable direct-effect assumption.

Check Your Understanding

Answer each question using the model definitions above. Solutions are included for checking your reasoning.

For $R_0=30$ units, $E_{\max}=-12$ units, and $EC_{50}=3$ mg/L, what response does the basic Emax model predict at $C=3$ mg/L?
In a Hill model, what fraction of $E_{\max}$ is attained at $C=EC_{50}$, and does changing $\gamma$ alter this fraction?
A fitted model has $E_{\max}=25$ units and $R_0=40$ units. What response does it approach as concentration becomes very large?
Why might observations only at concentrations much lower than $EC_{50}$ fail to identify $E_{\max}$ and $EC_{50}$ separately?
An analyst says $EC_{50}$ is the concentration where total biomarker response is half of baseline. Explain the correction.

Solutions. (1) The change is $(-12)(3)/(3+3)=-6$ units, so the response is $30-6=24$ units. (2) The fraction is $1/2$, since $EC_{50}^{\gamma}/(EC_{50}^{\gamma}+EC_{50}^{\gamma})=1/2$; changing positive $\gamma$ changes steepness but not the half-maximum point. (3) The response approaches $R_0+E_{\max}=65$ units. (4) At low concentrations, the basic model is approximately $(E_{\max}/EC_{50})C$, so the data mainly inform the ratio, not each parameter independently. (5) $EC_{50}$ is the concentration at which the drug-attributable change is $E_{\max}/2$; the total response there is $R_0+E_{\max}/2$, not half of baseline.