Intracluster correlation coefficients for neonatal trials: an analysis of national population level clinical data
File(s)
Accepted version
Author(s)
Jawad, sena
Prevost, Toby
Gale, Chris
Type
Journal Article
Abstract
Background
Clustering of outcomes occurs naturally in neonatal data due to multiple births clustered by mother, and infants clustered within the neonatal unit administering their care. Estimating these cluster effects is important for neonatal study design and impacts upon sample sizes for neonatal trials.
Methods
We analysed retrospective data from all neonatal admissions in England and Wales between January 2016 and January 2020 held in the National Neonatal Research Database. Intracluster correlation coefficients (ICCs) and their 95% confidence intervals were calculated for core neonatal outcomes and those commonly used in trials, at the level of both the neonatal unit and mother. Results were stratified by gestational age and neonatal unit level of birth. To illustrate the impact of clustering the design effect was estimated for a theoretical cluster trial.
Results
Intracluster correlation coefficients varied between outcomes and gestational age groups. Neonatal unit level intracluster correlation coefficients were low for mortality (0.0054, 95% CI 0.0039, 0.0068) and other core outcomes (severe necrotising enterocolitis 0.0042, 95% CI 0.0020, 0.0063) and were higher for outcomes related to care delivery (duration of intensive care 0.0237, 95% CI 0.0177, 0.0298; duration receiving parenteral nutrition 0.0265, 95% CI 0.0197, 0.0332). Gestation at birth was inversely correlated with neonatal unit ICC estimates. At the level of the mother ICC estimates were generally larger, especially for preterm infants.
Conclusions
We have estimated ICCs for key neonatal outcomes at the level of both neonatal unit and mother with high precision using national, population level routinely recorded data. These ICC estimates can be used to inform future neonatal studies and sample size calculations for neonatal trials.
Clustering of outcomes occurs naturally in neonatal data due to multiple births clustered by mother, and infants clustered within the neonatal unit administering their care. Estimating these cluster effects is important for neonatal study design and impacts upon sample sizes for neonatal trials.
Methods
We analysed retrospective data from all neonatal admissions in England and Wales between January 2016 and January 2020 held in the National Neonatal Research Database. Intracluster correlation coefficients (ICCs) and their 95% confidence intervals were calculated for core neonatal outcomes and those commonly used in trials, at the level of both the neonatal unit and mother. Results were stratified by gestational age and neonatal unit level of birth. To illustrate the impact of clustering the design effect was estimated for a theoretical cluster trial.
Results
Intracluster correlation coefficients varied between outcomes and gestational age groups. Neonatal unit level intracluster correlation coefficients were low for mortality (0.0054, 95% CI 0.0039, 0.0068) and other core outcomes (severe necrotising enterocolitis 0.0042, 95% CI 0.0020, 0.0063) and were higher for outcomes related to care delivery (duration of intensive care 0.0237, 95% CI 0.0177, 0.0298; duration receiving parenteral nutrition 0.0265, 95% CI 0.0197, 0.0332). Gestation at birth was inversely correlated with neonatal unit ICC estimates. At the level of the mother ICC estimates were generally larger, especially for preterm infants.
Conclusions
We have estimated ICCs for key neonatal outcomes at the level of both neonatal unit and mother with high precision using national, population level routinely recorded data. These ICC estimates can be used to inform future neonatal studies and sample size calculations for neonatal trials.
Date Acceptance
2026-07-06
Citation
Trials
ISSN
1745-6215
Publisher
BMC
Journal / Book Title
Trials
Copyright Statement
Copyright This paper is embargoed until publication. Once published the Version of Record (VoR) will be available on immediate open access.
License URL
Publication Status
Accepted