Clin Chem Lab Med. 2026 Apr 15. doi: 10.1515/cclm-2026-0262. Online ahead of print.
ABSTRACT
OBJECTIVES: To establish discrete and continuous age-specific reference intervals for serum alpha-fetoprotein in children and to evaluate whether continuous modeling improves interpretation during early infancy, when concentrations decline rapidly.
METHODS: In this prospective, cross-sectional study, serum alpha-fetoprotein was measured in 3,680 apparently healthy Vietnamese children from birth to <19 years using the Roche Cobas Pro platform. Discrete age-specific reference intervals were derived nonparametrically. For ages 0-3 years, continuous age-specific reference intervals were modeled using GAMLSS (generalized additive model for location, scale, and shape), and age-specific percentile tables were generated from fitted curves. Sex- and weight-related effects were assessed, and clinical utility was illustrated with a representative case.
RESULTS: Alpha-fetoprotein concentrations were highest in the neonatal period and declined steeply during the first months of life, then stabilized at low levels in later childhood. Sex differences were minimal and did not support sex-specific reference intervals. Continuous reference intervals for ages 0-<3 years represented the central 95 % of AFP concentrations and provided higher temporal resolution than discrete partitions in early infancy. A high-resolution percentile table (daily from 0 to 90 days; monthly from 4 to 12 months) enabled age-resolved interpretation. After age adjustment, body weight was not independently associated with alpha-fetoprotein. In the clinical case, serial values intermittently exceeded discrete upper limits yet remained within continuous age-specific percentiles, highlighting limitations of discrete partitions.
CONCLUSIONS: Continuous age-specific reference intervals improve clinically meaningful interpretation during early childhood, whereas discrete reference intervals remain appropriate for older children with stable alpha-fetoprotein concentrations.
PMID:41996693 | DOI:10.1515/cclm-2026-0262