Caitlin H Daly1, Victoria Higgins2, Khosrow Adeli2, Vijay L Grey3, Jemila S Hamid4. 1. Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada. 2. Clinical Biochemistry, Department of Pediatric Laboratory Medicine, The Hospital for Sick Children, Toronto, Canada. 3. Department of Pathology and Molecular Medicine, McMaster University, Hamilton, Ontario, Canada. 4. Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada; Knowledge Translation Program, Li Ka Shing Knowledge Institute, St. Michael's Hospital, Toronto, Ontario, Canada. Electronic address: jhamid@mcmaster.ca.
Abstract
OBJECTIVE: To statistically compare and evaluate commonly used methods of estimating reference intervals and to determine which method is best based on characteristics of the distribution of various data sets. DESIGN AND METHODS: Three approaches for estimating reference intervals, i.e. parametric, non-parametric, and robust, were compared with simulated Gaussian and non-Gaussian data. The hierarchy of the performances of each method was examined based on bias and measures of precision. The findings of the simulation study were illustrated through real data sets. RESULTS: In all Gaussian scenarios, the parametric approach provided the least biased and most precise estimates. In non-Gaussian scenarios, no single method provided the least biased and most precise estimates for both limits of a reference interval across all sample sizes, although the non-parametric approach performed the best for most scenarios. The hierarchy of the performances of the three methods was only impacted by sample size and skewness. Differences between reference interval estimates established by the three methods were inflated by variability. CONCLUSIONS: Whenever possible, laboratories should attempt to transform data to a Gaussian distribution and use the parametric approach to obtain the most optimal reference intervals. When this is not possible, laboratories should consider sample size and skewness as factors in their choice of reference interval estimation method. The consequences of false positives or false negatives may also serve as factors in this decision.
OBJECTIVE: To statistically compare and evaluate commonly used methods of estimating reference intervals and to determine which method is best based on characteristics of the distribution of various data sets. DESIGN AND METHODS: Three approaches for estimating reference intervals, i.e. parametric, non-parametric, and robust, were compared with simulated Gaussian and non-Gaussian data. The hierarchy of the performances of each method was examined based on bias and measures of precision. The findings of the simulation study were illustrated through real data sets. RESULTS: In all Gaussian scenarios, the parametric approach provided the least biased and most precise estimates. In non-Gaussian scenarios, no single method provided the least biased and most precise estimates for both limits of a reference interval across all sample sizes, although the non-parametric approach performed the best for most scenarios. The hierarchy of the performances of the three methods was only impacted by sample size and skewness. Differences between reference interval estimates established by the three methods were inflated by variability. CONCLUSIONS: Whenever possible, laboratories should attempt to transform data to a Gaussian distribution and use the parametric approach to obtain the most optimal reference intervals. When this is not possible, laboratories should consider sample size and skewness as factors in their choice of reference interval estimation method. The consequences of false positives or false negatives may also serve as factors in this decision.
Authors: Jemila S Hamid; Eshetu G Atenafu; Cornelia M Borkhoff; Catherine S Birken; Jonathon L Maguire; Mary Kathryn Bohn; Khosrow Adeli; Mohamed Abdelhaleem; Patricia C Parkin Journal: BMC Pediatr Date: 2021-05-19 Impact factor: 2.125