Automated age and sex partitioning of reference intervals based on routine laboratory data

Abstract Objectives In this study we evaluated the performance of the decision and regression tree–based algorithm rpart for age- and sex-partitioning of reference intervals (RIs) in an indirect setting, and compared its results with those derived from the CALIPER study. Methods Routine laboratory data for γ-glutamyltransferase (GGT), creatinine (CREA), and alkaline phosphatase (AP) were analyzed. Indirect RIs were estimated using the reflimR package, with partitions defined either by CALIPER or suggested by rpart . Corresponding partitions were compared using the Harris and Boyd method and assessed for clinical plausibility. Results rpart produced clinically and statistically meaningful partitioning schemes in mixed populations with unknown proportions of pathological values. Compared with CALIPER, rpart often yielded comparable RIs but avoided non-significant splits. Notable differences were observed in specific age groups, particularly in early childhood partitions for GGT and neonatal partitions for AP. Validation against original CALIPER RIs showed good agreement in most cases, with some broader intervals in the indirect approach. Conclusions These findings support the use of rpart not only in carefully preselected reference cohorts but also in heterogeneous routine laboratory datasets. Combined with indirect RI estimation via reflimR, automated partitioning offers a practical and reproducible approach to establishing clinically relevant RIs. Further multicenter studies should be conducted to confirm these results.