Benefits of linking routine medical records to the GUiNZ longitudinal birth cohort: Childhood injury predictors

Luam Ghebreab, Data Maager – Global Vaccine Data Network (GVDN), Pro VC Bridget Kool, Education; Arier Lee, Biostatistician, Epidemiology and Biostatistics; Honorary Prof Susan Morton, Population Health.

Home

•

Project

•

Benefits of linking routine medical records to the GUiNZ longitudinal birth cohort: Childhood injury predictors

Introduction

Combining self-reported longitudinal cohort data with routine healthcare data can help mitigate the limitations of each data type. Linking these datasets addresses missing data issues, reduces recall and information biases, and validates both data sources. Administrative data linkage has enhanced data completeness in longitudinal studies, making it a popular approach in national and international studies.

This study involved children from the Growing Up in New Zealand (GUiNZ) longitudinal birth cohort, whose mothers or primary caregivers responded to the injury questions over the preschool period and who had consented for their child’s data to be linked to routinely collected injury records from New Zealand’s Accident Compensation Corporation (ACC). The GUiNZ, initiated in 2009, is a contemporary longitudinal birth cohort that aims to understand multidisciplinary determinants of children’s health and development and Childhood injury is a signiﬁcant public health concern, leading to hospitalisations and deaths in New Zealand and worldwide. Reliable data are essential for identifying high-risk groups and planning eﬀective prevention strategies. Much research on childhood injuries relies on parental recall or routinely collected health service data. This study is part of a more signiﬁcant project on “predictors of injury among young children in New Zealand, where Hosking et al. developed an “ecological model of injury across the life course” to explore multiple factors surrounding preschool child injury. Following the linkage of maternal-reported child injury data with administrative injury records, this novel approach allowed us to compare (i) injury incidences from maternal recall with those recorded in ACC injury claims and (ii) the demographic characteristics of concordant and discordant injury reports and assess the level of agreement.

Linkage approach

This study used a “Master” linkage structure to deﬁne the outcome of “unintentional child injury”, where the linkage of GUiNZ cohort participants with an ACC claim is meaningfully interpreted as a cohort participant having an injury. The deterministic approach was employed using the study participant’s unique National Health identiﬁer (NHI) to minimise the false-matching and missed-matching rates. Individuals’ NHI within the GUiNZ study came from linkage in the perinatal period and with explicit consent. Additional identiﬁers such as gender and month of birth of the cohort were provided as accessory identifying variables for linkage to check for correct matches in case the NHI was corrupted. A manual review of records was used to determine whether the linked ACC injury data matched an individual record in the GUiNZ data, ensuring the date of birth and gender-matched. Injury claims that occurred outside the period of interest were excluded.

Findings

The comparison between injury data obtained from maternal reports in the GUiNZ study and claims from the Accident Compensation Corporation (ACC) reveals several noteworthy ﬁndings. Firstly, mothers tended to underreport injury events, and this underreporting increased as children grew older. The timing of data collection during diﬀerent Data Collection Waves (DCWs), each with varying recall periods (9 months, 15 months, and 30 months), contributed to potential underreporting, as more extended recall periods led to diﬃculties in remembering events over extended timeframes.

Mothers tended to underreport injury events, lowering sensitivity, high speciﬁcity, and generally lower kappa agreement. The level of agreement between maternal injury recall and ACC injury record was above 90% at 9M with good-adjusted Kappa agreement (k = 0.83) between the two data resources, but despite relatively high levels of agreement (79% and 71% at 24M and 54M respectively), the adjusted Kappa level remained moderate (k = 0.42-0.57).

Additionally, injury rates increased as children aged, with a higher likelihood of underreporting as children reached the preschool stage. This underreporting could often be due to caregivers not witnessing the injury, perceiving minor injuries as unimportant, or a combination of these factors.

The data linkage process, which merges rich primary information from the GUiNZ cohort with detailed ACC injury data, has signiﬁcant implications for future planning. Linking survey data with routine details makes accessing a wide range of additional injury-related data possible, thereby supporting injury prevention eﬀorts and enriching the existing evidence base for more informed policy initiatives. This linkage also beneﬁts researchers by reducing participant interview time, minimising their burden, lowering costs, ensuring accurate information, and optimising resource utilisation.

This work is published in BMC Medical Research Methodology 23, Article number: 91 (2023)
https://doi.org/10.1186/s12874-023-01900-0