NEW YORK (Reuters Health) - A machine-learning model correctly identifies more children with head trauma who could avoid CT imaging than does the Pediatric Emergency Care Applied Research Network (PECARN) head-trauma decision rules, according to a database study.
The PECARN rules are easy to use, but their simplicity may limit maximal predictive accuracy. The widespread availability of health information technologies in developed countries could allow application of more accurate clinical decision support tools.
Dr. Thomas A. Trikalinos from Brown University School of Public Health, in Providence, Rhode Island, and colleagues used machine learning and mathematical optimization methods to develop an optimal classification tree (OCT) and compared its ability to identify children at very low risk of clinically important traumatic brain injury (TBI) with that of the PECARN rules.
They examined data on more than 42,000 children with head trauma and without severely altered mental status who were examined between 2004 and 2006 at 25 North American emergency departments participating in PECARN
The OCT included a dozen predictors of clinically important TBI (age, sex, injury severity, loss of consciousness, altered mental status, and so on). The average OCT-predicted risks were less than 0.05% in the very-low-risk, 1.1% to 1.5% in the low-risk and greater than 4% in the higher-risk PECARN categories.
Compared with the PECARN rules, the corresponding OCT identified 33% more children younger than 2 years and predominantly nonverbal as being at very low risk, and 32% fewer patients in the higher-risk category in which CT scanning is recommended. It correctly identifying a similar number of patients with clinically important TBI in the higher-risk category, however.
Among older, predominantly verbal children, the OCT identified 14% more children without clinically important TBI in the very-low-risk stratum and 8% fewer patients in the higher-risk category, compared with the PECARN rules.
In the younger group, the OCT correctly identified all patients with clinically important TBI (while the PECARN misclassified one newborn). In the older group, OCT rules missed 10 and PECARN rules missed nine of 278 patients with clinically important TBI, the researchers report in JAMA Pediatrics, online May 13.
In both groups, the OCT had significantly higher specificity and significantly better positive predictive value and positive likelihood ratio than the PECARN rules. But the two approaches did not differ significantly in terms of sensitivity, negative predictive values and negative likelihood ratios.
"Optimal classification tree-based rules may have better predictive performance and provide personalized and more granular risk predictions than the PECARN rules," the researchers conclude. "However, OCTs are inherently more complicated than the PECARN rules because they include more predictors (e.g., age), encode predictors in several levels instead of dichotomizing them, and examine interactions between predictors. In practice, OCTs would have to be integrated into the electronic health record to provide real-time personalized risk predictions."
"We surmise that large health systems that aim to optimize operations by capitalizing on better predictive performance would consider easy-to-use implementations of the OCTs in their systems," they note.
Dr. Joseph J. Zorc from Children's Hospital of Philadelphia and the University of Pennsylvania Perelman School of Medicine, who co-authored a linked editorial, told Reuters Health by email, "Integrating more complex rules into clinical care offers the potential to improve accuracy of prediction for head injury and other conditions. Using this new rule would require integration into the electronic health record, although the PECARN group has already demonstrated the feasibility of that in prior work."