Matched Dose-Response Analysis — Exposure-Gradient Assessment in Matched Designs
Matched Dose-Response Analysis in Epidemiology · Also known as: matched trend analysis, dose-response in matched designs, exposure-response analysis with matching, matched exposure-gradient analysis
Matched dose-response analysis evaluates whether increasing levels of exposure are associated with proportionally increasing (or decreasing) risk of an outcome, within a study where cases and controls — or exposed and unexposed individuals — have been deliberately matched on key confounders such as age, sex, or study site. Matching controls residual confounding structurally, while the dose-response component tests whether the exposure-outcome relationship follows a biologically plausible gradient, strengthening causal inference.
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
Use matched dose-response analysis when you have a matched study design (matched case-control or matched cohort) and the primary exposure variable is quantifiable in ordered levels or dose units, and when your research question explicitly concerns whether increasing exposure is associated with increasing risk — a key criterion for establishing biological plausibility and supporting causality. It is especially informative in occupational epidemiology, pharmacoepidemiology, toxicology, and environmental health studies. Do not use this approach if exposure cannot be meaningfully graded into ordered levels, if the matching was not performed on genuine confounders, or if the matched sets are very small (fewer than 2 controls per case) and exposure prevalence is very low — in those situations, power is insufficient to detect any gradient. Also avoid if the primary interest is prevalence rather than incidence, where a cross-sectional design may be more efficient.
Strengths & limitations
- Combines structural confounding control (matching) with causal criterion testing (dose-response gradient), yielding stronger evidence than either approach alone.
- A statistically and biologically consistent dose-response gradient satisfies one of the Bradford Hill criteria for causality, strengthening causal inference beyond a simple exposure-yes/no contrast.
- Conditional logistic regression efficiently uses all matched-set information while properly respecting the matched design structure.
- Well-suited to regulatory and occupational contexts where threshold and gradient analysis are required to set safe exposure limits.
- Allows detection of non-linear or threshold dose-response patterns through spline or polynomial extensions.
- Matching on too many variables or on variables not truly confounding can over-match, reducing efficiency and narrowing generalizability.
- Requires an adequate and meaningful exposure measurement; misclassification of dose (especially if differential between cases and controls) can distort the shape of the gradient.
- The matched design prevents straightforward estimation of the matched variables' own effects, since they are conditioned out of the conditional likelihood.
- Power to detect a dose-response trend declines sharply with few matched sets or extreme exposure imbalance across categories.
- Defining exposure categories post-hoc inflates type I error; categories must be pre-specified.
Frequently asked
Why must I use conditional logistic regression rather than ordinary logistic regression in a matched case-control design?
Conditional logistic regression conditions the likelihood on each matched set, effectively removing the matched confounders from the estimation. Ordinary logistic regression treats matched sets as independent observations, ignoring the pairing. In a 1:1 matched design, this produces biased odds ratio estimates — typically attenuated toward the null — and artificially anti-conservative standard errors for the trend test. The matched structure must always be respected analytically.
How many exposure categories should I use?
Three to five ordered categories are typical. Too few categories (only two) reduce the dose-response test to a simple exposed/unexposed contrast and hide the gradient shape. Too many categories create sparse cells within matched sets, especially if the matched sets are small, and inflate variance. Tertiles or quartiles based on the exposure distribution in controls are a common and defensible choice, pre-specified before unblinding outcome data.
What if the dose-response relationship appears non-linear or U-shaped?
Non-linear patterns are clinically important and should not be dismissed. Restricted cubic splines or fractional polynomial terms entered into conditional logistic regression can model non-linearity without forcing a linear or even monotone assumption. Always plot the fitted dose-response curve with confidence bands. A U-shaped or threshold pattern may have distinct biological implications and should be reported and interpreted, not corrected away.
Does a significant dose-response trend prove causality?
No. A dose-response gradient satisfies one of Bradford Hill's nine criteria for causal inference but is neither necessary nor sufficient by itself. Confounding by an unmeasured variable that is correlated with dose could produce a spurious gradient. Biological plausibility, consistency across study populations, temporality (exposure precedes outcome), and experiment evidence are needed alongside the gradient to build a causal argument.
Can I apply this method to a matched cohort design, not just case-control?
Yes. In a matched cohort study, the analysis typically uses Cox proportional hazards regression stratified on matched sets (equivalent to conditional Cox regression), with the exposure dose entered as an ordinal or continuous predictor. The trend test logic is identical: estimate hazard ratios per exposure category and test whether they increase monotonically. The matched-cohort version is preferred when outcome incidence is high enough to require proper time-to-event modeling.
Sources
- Rothman, K.J., Greenland, S., & Lash, T.L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins. ISBN: 978-0781755641
- Breslow, N.E., & Day, N.E. (1980). Statistical Methods in Cancer Research, Vol. 1: The Analysis of Case-Control Studies. IARC Scientific Publications No. 32. International Agency for Research on Cancer. link ↗
How to cite this page
ScholarGate. (2026, June 3). Matched Dose-Response Analysis in Epidemiology. ScholarGate. https://scholargate.app/en/epidemiology/matched-dose-response-analysis
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.
- Case-control studyEpidemiology↔ compare
- Cohort StudyEpidemiology↔ compare
- Dose-Response AnalysisEpidemiology↔ compare
- Matched case-control studyEpidemiology↔ compare