Autor:innen:
T. Ammer (Penzberg, DE)
A. Schützenmeister (Penzberg, DE)
H. Prokosch (Erlangen, DE)
M. Rauh (Erlangen, DE)
C. Rank (Penzberg, DE)
J. Zierk (Erlangen, DE)
Introduction
Most biomarkers and their corresponding reference intervals depend on covariates, such as sex, age, or ethnicity. While the derivation of reference intervals for categorical covariates like sex is straightforward, it is substantially more challenging to correctly represent the influence of continuous covariates. Most pronounced during physiological development in children, but extending beyond adulthood, many analytes show clinically relevant age-dependent dynamics. One current approach for the estimation of age-specific reference intervals is a discretization of the continuous covariate. However, this approach leads to unnatural, discontinuous transitions between reference intervals estimated for the artificial age groups. Therefore, continuous reference intervals are needed to adequately assess test results of these analytes. Here, we propose a novel method leveraging routine measurements and a recently published algorithm, refineR, to establish smooth, continuous reference intervals and percentile charts for biomarkers of interest.
Methods
We developed a fully automated pipeline for the generation of continuous reference intervals utilizing solely an indirect method (refineR) and real-world data. First, the input data is divided into fine-grained subgroups, ensuring sufficient amount of data points per group. Second, we apply the refineR algorithm to each group and use the estimated model to determine reference intervals and percentiles for each age bin. These percentiles are then automatically smoothed using a median and bilateral filter. Subsequently, the refineR algorithm is applied in an iterative way alternating between model estimation and smoothing of the percentile curves. The smoothed percentiles from iteration i then serve as regularization for model estimation in iteration i+1 to finally generate smooth, continuous reference intervals. The presented pipeline was applied to three important biomarkers with extensive pediatric dynamics (hemoglobin, alkaline phosphatase, and creatinine) using data obtained during patient care.
Results
The calculated percentile charts for hemoglobin, alkaline phosphatase and creatinine from birth to 18 years of age are in accordance with previously established reference intervals, demonstrating that the presented approach correctly models age-dependency and generates valid continuous reference intervals.
Conclusions
The provided pipeline enables the fully automated generation of high-precision percentile charts using real-world data, while requiring no additional tool, except the refineR algorithm. Providing precise percentile charts allows for accurately capturing the pronounced age-dependent dynamics that occur in many biomarkers, facilitating the interpretation of test results and ultimately improving patient care.