Intracluster Correlation Coefficient and Multicenter Childhood Injury Research

Purpose Design of a clustered childhood trauma study requires having an estimate of the variability in the outcome of interest between and within the clusters. “Intracluster correlation coefficient (ICC)” is a standard measure that captures these variabilities. Ignoring the ICC in the sample size calculation might result in a severely underpowered study. Unfortunately, in spite of the importance of the subject, ICC has not been reported for any childhood-related trauma outcome. Methods We used the data from 246 level I and II trauma centers contributed data to the United States, National Trauma Data Bank (NTDB) to calculate the ICC for three of the most commonly used outcomes of interest in childhood trauma research (age≤15 years). The evaluated outcomes were: a) emergency department (ED) shock rate; b) early trauma death (i.e. death during the first 24 hours after admission); and c) in-hospital trauma death. Results From 2000 to 2004, 13% of the 952,242 patients in NTDB were ≤ 15 years old. Approximately 17,000 of these children suffered severe injuries, of them 84% were hospitalized at a level I or II trauma center. In general, the ICC for ED shock rate (0.005, 95% CI: 0.000–0.010) was significantly less than the ICC for in-hospital trauma fatality rate (0.023, 95% CI: 0.013–0.033). A similar pattern was observed when we compared the ICC of ED shock rate with the ICC of in-hospital trauma fatality rate for different genders or types of injury. Furthermore, for all three outcomes of interest, the ICC was larger for adults, in comparison to children, reflecting more variability in adultʹs care. However, the observed differences in the ICCs between children and adults were not statistically significant. Conclusion The estimated ICCs should be incorporated in sample size calculation for any multicenter trauma study that uses exclusively exposed or unexposed clusters of patients. Although the lack of difference between children and adults could be attributed to the similarity of heterogeneity in hospital and pre-hospital care between these two groups, the wide confidence intervals for the point estimates should not be overlooked.