Dose-Response Experimental Design and Analysis
Also known as: dose-response analysis, dose-response curve, Doz-Yanıt Tasarımı ve Analizi (Dose-Response), ED50 analysis, 4PL model, 5PL model
Dose-response design is a framework for planning and analysing experiments that characterise the relationship between the amount of a stimulus — such as a drug dose or a chemical concentration — and the magnitude of a biological or physiological response. Formalised in regulatory guidance by the ICH E4 guideline (1994) and extensively developed in the statistical literature by Ritz et al. (2015), the framework covers experiment design, four-parameter and five-parameter logistic curve fitting, key benchmark estimates (ED50/EC50, NOAEL, LOAEL), and monotone trend testing via the Williams procedure.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Apply dose-response design whenever you need to map a graded stimulus-response relationship across at least four or five dose levels, with a minimum of approximately 20 observations in total. The approach suits pharmacology, toxicology, ecotoxicology, agronomy, and any field where regulatory thresholds such as NOAEL and LOAEL must be estimated. The dose-response relationship may be monotone increasing, monotone decreasing, or even U-shaped (hormesis). A log-transformation of the dose axis typically linearises the middle portion of the sigmoidal curve and improves model fit. No normality requirement applies at the design level, though residual diagnostics after fitting are advisable.
Strengths & limitations
- Provides biologically interpretable benchmark estimates (EC50, NOAEL, LOAEL) with associated confidence intervals.
- The 4PL/5PL model is highly flexible and captures a wide range of sigmoidal curve shapes.
- Does not require a normality assumption; residuals can be examined after fitting.
- Williams test is specifically designed for monotone dose-response hypotheses, giving it more power than general ANOVA for this pattern.
- Applicable to continuous and ordinal response variables across diverse scientific disciplines.
- Requires at least four or five dose levels spread across the curve; too few levels prevent reliable parameter estimation.
- The 4PL model assumes a symmetric sigmoidal shape; asymmetric curves require the more complex 5PL form.
- Nonlinear fitting can be sensitive to poor starting values, yielding convergence failures or physiologically implausible parameter estimates.
- Extrapolation beyond the observed dose range is unreliable.
- NOAEL estimation depends heavily on the spacing of dose levels chosen at the design stage.
Frequently asked
What is the difference between EC50 and ED50?
EC50 is the concentration (typically in vitro) that produces 50 % of the maximum effect in a continuous assay. ED50 is the dose that produces a defined effect (often quantal, such as protection from death) in 50 % of a population in an in vivo setting. Both are estimated from dose-response curves but in different experimental contexts.
When should I use a 5PL instead of a 4PL model?
Use the 5PL model when the dose-response curve is asymmetric around the inflection point — that is, the slope rises more steeply on one side of EC50 than on the other. Fit both models and compare them with an F-test or AIC; if the extra asymmetry parameter is statistically justified, prefer the 5PL.
Can dose-response analysis handle U-shaped (hormesis) data?
Yes, but the standard 4PL/5PL models assume a monotone relationship, so they cannot fit a hormesis curve directly. For hormetic data you need a biphasic or brain-Cousens model. Importantly, the Williams monotone-trend test will have low power against a U-shaped alternative; graphical inspection should precede model selection.
How many dose levels and replicates are required?
ICH E4 and practical guidance recommend at least four to five dose levels spanning the range from no effect to maximum effect, plus a control group. Per-group replication of at least four to six observations improves parameter precision. The minimum total sample of around 20 is a lower bound; larger studies with more levels and replicates yield narrower confidence intervals for EC50 and NOAEL.
Sources
- Ritz, C., Baty, F., Streibig, J. C., & Gerhard, D. (2015). Dose-Response Analysis Using R. PLOS ONE, 10(12), e0146021. DOI: 10.1371/journal.pone.0146021 ↗
- ICH E4 (1994). Dose-Response Information to Support Drug Registration. International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. link ↗
How to cite this page
ScholarGate. (2026, June 1). Dose-Response Experimental Design and Analysis. ScholarGate. https://scholargate.app/en/experimental-design/dose-response-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Full Factorial DesignExperimental design↔ compare
- Logistic RegressionResearch Statistics↔ compare
- One-way ANOVAStatistics↔ compare
- Repeated-measures ANOVAStatistics↔ compare