Dose Response

Dose response describes the relationship between the amount, intensity, or duration of an exposure (the "dose") and the magnitude of the resulting biological or clinical effect (the "response"). Observing a dose response gradient, where increasing levels of exposure produce progressively larger (or smaller) effects, is one of the Bradford Hill criteria for causal inference and strengthens the argument that the observed association between exposure and outcome is causal rather than spurious.^[1][2] Dose response relationships can be linear, curvilinear, threshold based, or U shaped, depending on the nature of the exposure and the biological mechanism.

Dose response evidence can strengthen the case that an observational association is causal, but a gradient is neither necessary nor sufficient for causality. Bradford Hill (1965), who called it a "biological gradient", stressed that none of his criteria can be required for a causal conclusion.^[2] A confounder that itself rises with exposure can produce a spurious gradient, and a real cause can act through a threshold rather than a steady trend.^[1] If a medication produces a greater effect at higher doses, a toxin causes more damage with greater exposure, or an exercise program yields more benefit with increased frequency, the argument for a causal relationship is considerably stronger than if the effect were constant regardless of dose. Regulatory agencies use dose response data to establish safe exposure limits for chemicals and pollutants, to determine optimal drug dosing, and to set public health guidelines. In evidence synthesis, the GRADE approach lists a dose response gradient as one of the factors that can raise the certainty of evidence from observational studies.^[4]

Dose response relationships are characterized using several approaches. In experimental settings, such as pharmacological studies, researchers systematically vary the dose and measure the response, plotting a dose response curve. When the response is plotted against the logarithm of the dose, the curve for many drugs follows a sigmoidal (S shaped) pattern: minimal effect at low doses, a steep increase in response over a middle range, and a plateau at high doses where receptors are saturated. Key parameters include the ED50 (the dose that produces 50% of the maximum effect), the maximum effect (Emax), and the therapeutic window (the range of doses that produce beneficial effects without unacceptable toxicity).

In observational epidemiology, dose response is assessed by dividing the exposure into categories (tertiles, quartiles, or quintiles) and examining whether the outcome increases or decreases monotonically across categories. A test for linear trend across categories (such as the Cochran Armitage test or inclusion of the exposure as a continuous variable in a regression model) provides a formal statistical assessment of dose response. More flexible approaches, such as restricted cubic splines or fractional polynomials, allow the shape of the dose response relationship to be estimated without assuming linearity.^[1][3]

Not all dose response relationships are monotonic. U shaped or J shaped curves, where both very low and very high exposures are associated with adverse outcomes, have been described for alcohol consumption and mortality, vitamin D levels and health outcomes, and body mass index and various diseases. These nonlinear relationships require analytical methods that can detect curvilinearity, and their interpretation requires careful consideration of confounding and reverse causation.

In meta analysis, dose response meta analysis synthesizes data from multiple studies to estimate the pooled dose response curve, combining studies that examined different exposure levels to produce a more comprehensive picture than any single study could provide.^[3][5]

Dose response analysis is appropriate whenever the research question involves the relationship between varying levels of exposure and an outcome. It is standard practice in pharmacology (determining optimal drug dosing), toxicology (establishing safe exposure limits), nutrition (assessing the relationship between nutrient intake and health outcomes), and epidemiology (evaluating the strength of causal claims). It is not appropriate when exposure is binary (present or absent) with no meaningful gradation, or when the available data do not capture sufficient variation in exposure levels to characterize a gradient.

A dose response meta analysis by Smith and colleagues (2016) examined the relationship between physical activity (leisure time or total) and the risk of type 2 diabetes across 28 prospective cohort studies involving about 1.26 million participants. Using restricted cubic splines to allow for nonlinearity, it found a curvilinear relationship. Compared with inactive people, those who achieved 150 minutes of moderate activity per week had a 26% lower risk (relative risk 0.74, 95% CI: 0.69 to 0.80), and those who achieved twice that amount had a 36% lower risk (relative risk 0.64, 95% CI: 0.54 to 0.73). The greatest relative benefits came at low levels of activity, supporting public health messaging that emphasizes the value of even modest activity for sedentary populations, while additional benefits continued well above the recommended 150 minutes.