Knots divide the predictor range into joined cubic segments, allowing the fitted relationship to change shape across different ranges. At each knot, the curve remains continuous, and its first and second derivatives also match, so transitions are smooth rather than abrupt. Linear behavior beyond the outer knots limits boundary flexibility and reduces unstable curvature at the extremes.
Restricted Cubic Splines use several joined cubic segments rather than forcing one high-degree polynomial to describe the entire predictor range. This segmented representation can accommodate changing slopes while maintaining smooth joins. It therefore offers flexibility for nonlinear patterns without the instability associated with high-degree polynomials or the rigid shape imposed by a straight-line term.
The selected knots determine how the predictor range is partitioned into cubic segments and where the fitted curve can change its local shape. Because the segments must join smoothly, knot placement affects how the model represents bends, thresholds, or plateaus. Knots therefore connect model flexibility with the observed exposure or age range.
A basic workflow starts by identifying a continuous predictor and outcome, selecting knots across the predictor range, and fitting the spline within a regression model. The resulting fitted relationship can then be visualized and interpreted alongside covariate adjustment when adjustment is part of the analysis. This workflow helps assess whether a straight-line assumption is inadequate.
In medicine, the technique can examine dose-response relationships, disease risk across age or exposure levels, and treatment effects that may not follow a straight line. Its fitted curve can show whether the association changes across the predictor range, including threshold-like or plateau behavior, while remaining part of an adjusted regression analysis.
Interpretation should focus on the shape of the fitted association across the predictor range rather than on a single overall slope. A curve may indicate changing risk, dose-response behavior, or treatment effects in different ranges, and visualization can make these patterns apparent. The same display can help communicate where threshold-like or plateau behavior occurs.