Figure 2
From: refineR: A Novel Algorithm for Reference Interval Estimation from Real-World Data
Description of generating simulated test cases and benchmarking the resulting estimations using an example picked from Fig. 3b (‘CREA’). (a) Example of designed ground truth distribution for ‘CREA’ in green with added pathological distributions left (25–50) (15%) and right (70–126) (20%) of the non-pathological distribution in red. (b) Simulated test datasets using 100 different seeds that serve as input to the algorithms. (c) Results obtained using the refineR algorithm on datasets shown in (b) with the green horizontal dashed lines showing the estimated reference interval and the green curve showing the estimated non-pathological distribution. (d) Visualization of the results for the simulated datasets shown in (b) with 100 different seeds and applying the algorithm on the various datasets (c). The obtained estimated reference intervals are grouped into the color-coded categories regarding their deviation from the ground truth (Table 1). L lower reference limit, U upper reference limit, TE total error, RI reference interval.
