Imputing Longitudinal Growth Data in International Pediatric Studies: Does CDC Reference Suffice?
: Li Zhingue, Toppari Jorma, Lundgren Markus, Frohnert Brigette I, Achenbach Peter, Veijola Riitta, Anand Vibha; T1DI study group
: AMIA Annual Symposium
Publisher: American Medical Informatics Association
: 2021
: AMIA ... Annual Symposium proceedings. AMIA Symposium
: AMIA Annual Symposium Proceedings 2021
: AMIA ... Annual Symposium proceedings. AMIA Symposium
: AMIA Annu Symp Proc
: AMIA ... Annual Symposium proceedings
: 2021
: 754
: 762
: 1559-4076
: 1942-597X
: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8861671
: https://research.utu.fi/converis/portal/detail/Publication/174959224
This study investigates a missing value imputation approach for longitudinal growth data in pediatric studies from multiple countries. We analyzed a combined cohort from five natural history studies of type 1 diabetes (T1D) in the US and EU with longitudinal growth measurements for 23,201 subjects. We developed a multiple imputation methodology using LMS parameters of CDC reference data. We measured imputation errors on both combined and individual cohorts using mean absolute percentage error (MAPE) and normalized root-mean-square error (NRMSE). Our results show low imputation errors using CDC reference. Overall height imputation errors were lower than for weight. The largest MAPE for weight and height among all age groups was 4.8% and 1.7%, respectively. When comparing performance between CDC reference and country-specific growth charts, we found no significant differences for height (CDC vs. German: p =0.993, CDC vs. Swedish: p=0.368) and for weight (CDC vs. Swedish: p=0.513) for all ages.
UTU Research Portal