Table 5.

Comparison of fractional polynomials and splines

Fractional polynomial	Linear splines
Automated procedure for selecting the best-fitting model	‘Rule of thumb’ – no consensus on the best method to choose the number and position of knot points
Needs constant term (fixed and random effect), plus 1 fixed and 1 random effect per degree (e.g. a 2nd-degree polynomial needs 3 fixed effects and 3 random effects)	Needs constant term (fixed and random effect), plus 1 fixed and 1 random effect, plus 1 fixed and random effect per knot point (e.g. a model with 2 knot points requires 4 fixed and 4 random effects)
Smooth curve	Biologically implausible broken-stick model
Harder to interpret coefficients directly	Easy to interpret coefficients