ETA2023 45th Annual Meeting of the European Thyroid Association ETA 2023 Oral Session 6: Pregnancy (5 abstracts)
Optimizing the dutch newborn screening for congenital hyothyroidism by using amino acids and acylcarnitines via a machine learning based approach
Heleen Jansen 1 , Marije van Haeringen 2 , Marelle Bouva 3 , Wendy den Elzen 4 , Eveline Bruinstroop 5 , Catharina van der Ploeg 6 , Paul van Trotsenburg 7 , Nitash Zwaveling-Soonawala 7 , Annemieke Heijboer 8 , Annet Bosch 9 , Robert de Jonge 10 , Mark Hoogendoorn 11 & Anita Boelen 12
1Amsterdam Umc, Vrije Universiteit and University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism, Department of Laboratory Medicine, Endocrine Laboratory, Amsterdam, Netherlands; 2Amsterdam Umc, Vrije Universiteit Amsterdam, Department of Laboratory Medicine, Endocrine Laboratory and Department of Computer Science, Amsterdam, Netherlands; 3Reference Laboratory Neonatal Screening, Center for Health Protection, National Institute for Public Health and the Environment, Bilthoven, Netherlands; 4Amsterdam Umc, University of Amsterdam, Amsterdam Public Health Research, Department of Laboratory Medicine, Amsterdam, Netherlands; 5Amsterdam Umc, University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism, Department of Endocrinology and Metabolism, Amsterdam, Netherlands; 6Tno - Child Healthy, Leiden, Netherlands; 7Amsterdam Umc, University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism, Department of Paediatric Endocrinology, Emma Children’s Hospital, Amsterdam, Netherlands; 8Amsterdam Umc, Vrije Universiteit and University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism and Amsterdam Reproduction and Development Research Institute, Department of Laboratory Medicine, Endocrine Laboratory, Amsterdam, Netherlands; 9Amsterdam Umc, University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism, Department of Pediatrics, Division of Metabolic Disorders, Emma Children’s Hospital, Amsterdam, Netherlands; 10Amsterdam Umc, Vrije Universiteit and University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism, Department of Laboratory Medicine, Amsterdam, Netherlands; 11Vrije Universiteit Amsterdam, Department of Computer Science, Amsterdam, Netherlands; 12Amsterdam Umc, University of Amsterdam, Amsterdam Gastroenterology, Endocrinology and Metabolism and Amsterdam Reproduction and Development Research Institute, Department of Laboratory Medicine, Endocrine Laboratory, Amsterdam, Netherlands
Objectives: Congenital hypothyroidism (CH) is an inborn thyroid hormone (TH) deficiency mostly caused by disturbances at the thyroid level (thyroidal CH, CH-T). Less frequently, but equally important, CH can be the result of hypothalamic/pituitary dysfunction (central CH, CH-C). Most CH newborn screening (NBS) programs are based on the measurement of thyroid-stimulating hormone (TSH), thereby only detecting CH-T. The Dutch NBS detects CH by measuring total T4 concentrations in newborns as a first tier followed by TSH in the 20% lowest daily T4 concentrations. Thyroxine binding globulin (TBG) is measured in the 5% lowest T4 concentrations to exclude false-positive referrals due to (partial) TBG deficiency. The T4/TBG ratio is calculated as an indirect measure of free T4. The Dutch T4-TSH-TBG algorithm effectively detects both CH-T and CH-C, however, at the cost of a low positive predictive value (PPV) of 21% in the period 2007-2017. A slightly higher PPV of 26% was yielded when using a machine-based learning model on the adjusted dataset described below (methods) based on the original parameters of the Dutch CH NBS. Recent studies describe an association between THs and amino acids (AAs) and acylcarnitines (ACs) of which several are measured during NBS for other diseases. Therefore, we aimed to investigate whether AAs and ACs contribute to discriminate newborns with and without CH using a machine-based learning model which leads to reduction of false-positive referrals.
Methods: Dutch NBS data between 2007-2017 (sex, gestational age and weight, age at NBS, T4, TSH, TBG, T4/TBG ratio, AAs, ACs) from 1079 false-positive referrals and newborns with CH-T (431) and CH-C (84), as well as data from 1842 newborns with a normal CH screening (from 2019) were used. A Random Forest model including all these data was developed and the PPV and area under the ROC curve (AUROC) of this model to predict CH were calculated.
Results: The Random Forest model yielded an artificial sensitivity of 100%, while obtaining a PPV of 48% and AUROC of 0.99. Besides TSH and T4, tyrosine and succinylacetone were the main parameters contributing to the model’s performance. A second model emphasizing parameters contributing to predict CH-C showed that T4/TBG ratio contributed most, TSH did not contribute but gestational age, C16:1, phenylalanine, and C2 were important parameters.
Conclusions: Adding several AAs and ACs to this machine-based learning model led to a significant improvement of the PPV (26% to 48%) suggesting AAs and ACs benefit the current algorithm.
Article tools
My recent searches
My recently viewed abstracts
Authors
Endocrine Abstracts
ISSN 1470-3947 (print) | ISSN 1479-6848 (online)
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more