A combined approach to generate laboratory reference intervals using unbalanced longitudinal data

• Published in: Journal of Pediatric Endocrinology & Metabolism Vol. 30; no. 7; pp. 767 - 773
• Main Authors: Vogel, Mandy, Kirsten, Toralf, Kratzsch, Jürgen, Engel, Christoph, Kiess, Wieland
• Format: Journal Article
• Language: English
• Published: Germany De Gruyter 26.07.2017; Walter de Gruyter GmbH
• Subjects: Adolescent; biological variation; Child; Child, Preschool; Cholesterol, HDL - blood; Clinical Laboratory Techniques - standards; epidemiology; Female; Follow-Up Studies; Humans; Infant; Infant, Newborn; Laboratories; LMS method; Longitudinal Studies; Male; Prospective Studies; reference intervals, reference limits; Reference Values; Reproducibility of Results; Statistics as Topic; epidemiology; biological variation; LMS method; reference intervals, reference limits
• ISSN: 0334-018X; 2191-0251; 2191-0251
• DOI: 10.1515/jpem-2017-0171

The interpretation of individual laboratory test results requires the availability of population-based reference intervals. In children, reference interval estimation has to consider frequently the strong age-dependency. Generally, for the construction of reference intervals, a sufficiently large number of independent measurement values is required. Data selections from hospitals or cohort studies often comprise dependencies violating the independence assumption. In this article, we propose a combination of LMS-like (mean, M; coefficient of variation, S; skewness, λ or L) and resampling methods to overcome this drawback. The former is recommended by the World Health Organization (WHO) for the construction of continuous reference intervals of anthropometric measurements in children. The approach allows the inclusion of dependent measurements, for example, repeated measurements per subject. It also provides pointwise confidence envelopes as a measure of reliability. The combination of LMS-type methods and resampling provides a feasible approach to estimate age-dependent percentiles and reference intervals using unbalanced longitudinal data.