Methods We are conducting a multicenter clinical trial in which we implement clinical prediction rules for children with minor blunt head trauma using EHR-based clinical decision support. We developed a clinical trial alert to promote data collection in a head trauma assessment tool for eligible patients in the EHRs (all Epic®). To trigger this alert, we created site-specific groupings of structured chief complaints, with the aim of attaining at least 90% sensitivity and as high as possible specificity to correctly identify those with blunt head trauma who were eligible for study. At 4 sites, we collected chief complaint frequencies for patient encounters matched against relevant ICD 9 head injury codes. Based on this list, we constructed multiple possible chief complaint groupings for each site, consisting of combinations such as: all those having any injury-related complaints, the 15 and 30 most frequent complaints for head injuries, all injury and head-injury related complaints, and other combinations based on site specifics. We created a de-identified reporting process at one lead site using Crystal Reports® 2008 that indicated whether at least one chief complaint was absent or present within each grouping for each eligible patient encounter (the test variable). We also created an outcome variable to denote presence or absence of a head injury, based on ICD 9 code (the reference standard), for each patient encounter. Each site then ran reports using this process. We analyzed all groupings for sensitivity and specificity using SAS® 9.3.
Results Each site tested 17-26 combinations of chief complaint groupings. There was substantial variation in sensitivity and specificity for each studied chief complaint grouping. Collaboratively, study investigators selected the final groupings which had the test characteristics displayed in the table.
Table 1: Final Grouping Statistics to Identify Eligible Children with Head Trauma
|Site||Sensitivity (%)||Specificity (%)||Grouping Description|
|1||95.1||83.4||All injury plus 'headache'|
|2||97.8||83.8||Based on top 30 head injury complaints|
|3||96.6||88.6||Based on top 30 head injury complaints|
|4||95.5||75.3||Based on top 30 head injury complaints|
Conclusion We created site-specific chief complaint groupings in a multicenter trial to optimize accurate identification of eligible children with head injury through triggering a clinical trial alert. The methods were easily adapted across sites, with a centralized process for creating reports, analyzing de-identified data and selecting the chief complaint groupings. With clinical trial data collection ongoing, we will now re-assess and refine the groupings to further improve sensitivity and specificity.