Figure 1

Flowchart of the developed pipeline for the estimation of continuous reference intervals from routine data. Based on the raw input data (a), overlapping age groups with at least N = 1000 samples are defined (b). An indirect method (here refineR) is applied to each defined age group to estimate a model describing the non-pathological distribution (c). This model is then used to compute a probability of being non-pathological along the concentration range (Eq. (1)), i.e. that a data point with that concentration originated from the non-pathological distribution (d). The light green colored area represents a probability of 1 (most likely non-pathological), and the dark blue colored region a probability of 0 (most likely pathological) (d). These estimated weights are then assigned to each data point (e). Following that, a statistical method to estimate smooth curves (here GAMLSS) is applied to the weighted input data set. The resulting parametric model enables the derivation of continuous reference intervals and percentile charts (f). To interpret and analyze pediatric data in more detail, percentile charts can also be presented with alternative scaling of the covariate age (f inset).