Exposure–response modeling
Describe a direct saturating relationship between concentration and pharmacodynamic response.
Concepts and reasoning
Modeling a saturating response
A direct pharmacodynamic model relates concentration to an expected response at the same time. Many biological responses saturate: increasing concentration produces smaller additional changes as a limiting response is approached. An Emax model describes this pattern with a baseline, a maximum change, and a concentration scale.
E₀ is the expected response at zero concentration. Emax is the change from baseline to the high-concentration asymptote. The asymptotic response is E₀ + Emax, not necessarily Emax itself. A negative Emax can describe a decreasing response under an appropriate parameterization.
EC50 is the concentration at which the modeled change reaches half of Emax. It is not automatically a clinically optimal concentration or a recommended dose. Translating concentration to dose requires pharmacokinetics, and a useful clinical target requires benefit–risk and endpoint considerations.
Direct response versus delayed response
A direct model assumes current concentration is sufficient to determine the expected response. Some effects lag behind concentration because of distribution to an effect site, receptor dynamics, or biological turnover. Then equal measured concentrations on the rising and falling parts of a profile can have different responses, producing hysteresis.
A sigmoid Emax model adds a Hill coefficient controlling steepness. Greater flexibility can improve fit but also makes parameter estimation harder when the concentration range is narrow. The introductory pharmacodynamic modeling tutorial describes direct models and extensions; the model should follow the response mechanism and available data.
Mathematical foundations
The Emax family
The direct Emax model is
The sigmoid version is
where C is nonnegative concentration, EC50 is positive, and h is a positive Hill coefficient. At C = EC50, the fraction is one-half for any h. At zero concentration, the predicted response is E₀.
For the direct model at concentrations much smaller than EC50,
Only the ratio Emax/EC50 strongly determines the low-concentration slope. Many large Emax and EC50 pairs can therefore fit the same low-range data. Observing a nearly straight line does not identify the eventual maximum.
For h = 1, solving for a fraction f of the maximum change gives C = EC50 × f/(1 − f). Reaching 90% of the modeled maximum change requires 9 × EC50 under that specific model, not 1.8 × EC50.
Worked example
Worked example: baseline and half-maximal change
Let E₀ = 10, Emax = 20, and EC50 = 5 concentration units. For the direct model:
| Concentration | Fraction of maximum change | Expected response |
|---|---|---|
| 0 | 0.000 | 10.000 |
| 1 | 0.167 | 13.333 |
| 5 | 0.500 | 20.000 |
| 20 | 0.800 | 26.000 |
| 45 | 0.900 | 28.000 |
At concentration 5, the expected response is 20, which is halfway between baseline 10 and asymptote 30. It is not half of the asymptote, 15.
Now use a Hill coefficient of 2 with the same baseline, Emax, and EC50. At concentration 1, the fraction becomes 1/(25 + 1) = 0.0385 and the response is 10.769. At concentration 20, the fraction is 400/(25 + 400) = 0.9412 and the response is 28.824. The curve is steeper around EC50 while retaining the same midpoint.
Suppose the experiment only measures concentrations between 0 and 0.5. A direct model with Emax = 20 and EC50 = 5 has initial slope 4. A model with Emax = 40 and EC50 = 10 has the same initial slope. Their predictions are close over that limited range but their asymptotes are 30 and 50. Extrapolating the maximum from those data would be unreliable.
This example also shows why confidence intervals for nonlinear parameters can be asymmetric and strongly correlated. A plausible-looking fitted curve within the observed range does not imply precise individual parameters.
Interpreting the results
Distinguish baseline, maximal change, and potency
An Emax model describes a response approaching a plateau with increasing concentration. Baseline is the response at zero concentration; Emax is the change from baseline to the asymptote, so the limiting response is baseline plus Emax. Emax can be negative when exposure suppresses the response.
EC50 is the concentration producing half the modeled maximal change. It is not necessarily an effective clinical dose or a safe exposure threshold. A sigmoid model adds a Hill coefficient that controls steepness near the transition; it does not independently establish a biological mechanism.
Parameter intervals and predicted responses are conditional on the direct concentration–response form. If observations cover only low concentrations, a large Emax and large EC50 can trade off to fit nearly the same slope. Apparent precision in the observed range may therefore coexist with great uncertainty about the plateau.
Assumptions and limitations
A direct-effect model omits time-dependent mechanisms
The model assumes response is described by the contemporaneous concentration. Delayed effects, tolerance, turnover, or hysteresis can give different responses at the same concentration on rising and falling exposure paths. Those mechanisms need time-aware models rather than a more flexible Hill slope alone.
Additive constant-variance residuals are assumed by ordinary least squares. Response bounds, changing measurement precision, and repeated observations within individuals may require different likelihoods or dependence structures. Concentration measurement error is also not automatically represented.
A single monotone saturating relationship must be plausible over the modeled range. Biphasic effects or multiple mechanisms may violate it. Baseline, plateau, and transition regions need sufficient data for separate estimation; numerical convergence cannot create information about unobserved concentration ranges.
Diagnostics and sensitivity analysis
Check whether the plateau and transition are observed
Plot response against concentration and inspect low, middle, and high exposure regions. Determine whether the estimated plateau is anchored by data or extrapolated far beyond them. Compare ordinary Emax and sigmoid forms only when the added shape parameter is supported.
Inspect residual patterns, unusually influential concentrations, parameter intervals, and the Jacobian condition number. Large changes in Emax or EC50 under small data perturbations suggest weak identification. A low residual standard deviation alone does not solve that problem.
When sampling times are available, check whether equal concentrations have different responses during exposure rise and decline. Such hysteresis is evidence to investigate delay rather than forcing a direct-effect curve. Confirm useful predictions over the intended exposure range and report both the response scale and concentration units.
References and other material
Read the methodological references and supporting tutorials below. The examples in this lesson are original teaching calculations, not findings from these sources.
- Introduction to pharmacodynamic modeling methods — A free tutorial covering direct concentration–effect relationships and extensions.