Risk-adjusted dose-response analysis
Risk-Adjusted Dose-Response Analysis · Also known as: confounder-adjusted dose-response, covariate-adjusted dose-response modeling, risk-stratified dose-response analysis, adjusted exposure-response analysis
Risk-adjusted dose-response analysis quantifies the relationship between increasing levels of an exposure (dose) and the probability or magnitude of an outcome (response), while simultaneously controlling for baseline risk factors that could confound or modify this relationship. The method is widely applied in clinical epidemiology, pharmacoepidemiology, and environmental health research to isolate the causal contribution of exposure intensity from background risk heterogeneity among participants.
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 risk-adjusted dose-response analysis when the research question concerns how outcome risk changes across a gradient of exposure intensity and the study population is heterogeneous with respect to baseline risk. It is appropriate for observational cohorts, nested case-control studies, and randomized trials with active dose variation. It is not appropriate when the exposure is binary with no dose information, when the sample is too small to support a flexible non-linear model (generally fewer than 100 outcome events per dose level), or when important confounders are unmeasured and cannot be proxied — in that setting, residual confounding may dominate any dose-response signal.
Strengths & limitations
- Separates the pure exposure-intensity effect from baseline risk heterogeneity, yielding less biased effect estimates in observational settings.
- Accommodates non-linear dose-response shapes through spline or polynomial modeling rather than forcing linearity.
- Applicable across a wide range of outcome types — binary, count, time-to-event — through the appropriate regression family.
- Produces clinically interpretable dose-specific risk estimates that can inform exposure limits or treatment thresholds.
- Can be combined with propensity-score methods for doubly robust estimation.
- Effectiveness of risk adjustment depends on the quality and completeness of confounder measurement; unmeasured confounders can still bias estimates substantially.
- Flexible non-linear models require adequate sample size and sufficient spread of observations across the dose range — sparse data at extreme doses produce wide, unstable confidence intervals.
- The choice of functional form for the dose variable (spline knots, polynomial degree) can influence conclusions; results may be sensitive to these modeling decisions.
- Risk adjustment does not resolve confounding by indication in pharmacoepidemiology, where sicker patients may systematically receive higher doses.
Frequently asked
How is this different from a simple dose-response analysis without adjustment?
An unadjusted dose-response analysis estimates the total association between exposure level and outcome, including the portion attributable to confounding variables. When exposure level is correlated with baseline risk characteristics — as is common in observational studies — the unadjusted curve conflates the true dose effect with background risk gradients. Risk adjustment isolates the exposure effect by holding confounder values constant across dose levels, producing an estimate that more closely approximates the causal dose-response relationship.
Should I use propensity scores or regression adjustment to control for confounders?
Both approaches are valid; the choice depends on context. Regression adjustment is simpler and more efficient when the outcome model is correctly specified. Propensity-score weighting (IPTW) places the confounders in the exposure model and can better balance covariate distributions across dose levels, but requires a continuous dose propensity model (generalized propensity score). Doubly robust estimators combining both approaches offer protection against misspecification of either model and are increasingly recommended.
How many knots should I use for restricted cubic spline modeling of the dose variable?
The standard recommendation is three to five knots placed at fixed quantiles of the dose distribution (e.g., 5th, 27.5th, 50th, 72.5th, 95th percentiles for five knots). Fewer than three knots forces linearity; more than five rarely adds biological insight and risks overfitting. The number should be pre-specified based on sample size and the expected complexity of the dose-response shape, not chosen post-hoc to optimize visual appeal.
Can I apply this approach to randomized controlled trials?
Yes, and adjustment is still valuable in RCTs when dose varies continuously within or across randomized arms. Randomization removes confounding of the treatment assignment but not necessarily of the dose actually received — compliance, metabolic differences, and dose modifications can reintroduce covariate imbalance across dose levels within a treatment arm. Adjusting for baseline characteristics in the dose-response model increases precision and addresses any residual imbalance.
What is the E-value and why is it relevant here?
The E-value quantifies how strong an unmeasured confounder would need to be — in terms of its association with both the dose variable and the outcome — to fully explain away an observed adjusted dose-response estimate. It was proposed by VanderWeele and Ding (2017) as a sensitivity analysis tool. Computing E-values for the adjusted risk estimates at key dose levels communicates the robustness of the dose-response conclusions to potential residual confounding.
Sources
- Greenland, S. (1995). Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology, 6(4), 356-365. DOI: 10.1097/00001648-199507000-00005 ↗
- Rothman, K. J., Greenland, S., & Lash, T. L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins. ISBN: 978-0781755641
How to cite this page
ScholarGate. (2026, June 3). Risk-Adjusted Dose-Response Analysis. ScholarGate. https://scholargate.app/en/epidemiology/risk-adjusted-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.
- Cohort StudyEpidemiology↔ compare
- Cox proportional hazardsEpidemiology↔ compare
- Dose-Response AnalysisEpidemiology↔ compare
- Logistic RegressionResearch Statistics↔ compare