Summary
Objective: Record linkage may create powerful datasets with which investigators can conduct
comparative effectiveness studies evaluating the impact of tests or interventions
on health. All linkages of health care data files to date have used protected health
information (PHI) in their linkage variables. A technique to link datasets without
using PHI would be advantageous both to preserve privacy and to increase the number
of potential linkages.
Methods: We applied probabilistic linkage to records of injured children in the National Trauma
Data Bank (NTDB, N = 156,357) and the Pediatric Health Information Systems (PHIS,
N = 104,049) databases from 2007 to 2010. 49 match variables without PHI were used,
many of them administrative variables and indicators for procedures recorded as International
Classification of Diseases, 9th revision, Clinical Modification codes. We validated
the accuracy of the linkage using identified data from a single center that submits
to both databases.
Results: We accurately linked the PHIS and NTDB records for 69% of children with any injury,
and 88% of those with severe traumatic brain injury eligible for a study of intervention
effectiveness (positive predictive value of 98%, specificity of 99.99%). Accurate
linkage was associated with longer lengths of stay, more severe injuries, and multiple
injuries.
Conclusion: In populations with substantial illness or injury severity, accurate record linkage
may be possible in the absence of PHI. This methodology may enable linkages and, in
turn, comparative effectiveness studies that would be unlikely or impossible otherwise.
Keywords
Medical record linkage - pediatrics - brain injuries
