The use of CT in the emergency setting might not vary quite as much as initially reported, according to an article published online July 5 in the American Journal of Roentgenology. Researchers found that taking into account hundreds of factors revealed much lower variability between emergency sites.
CT utilization in emergency departments has not only increased dramatically in recent years but also varied widely, even for patients presenting with a similar condition within the same hospital network. However, assessments of CT use have rarely considered the impact of factors that could affect the likelihood of a CT scan being ordered.
To address concerns over the marked variability in CT usage, Dr. Arjun Venkatesh and colleagues from Yale University reviewed CT use at emergency departments within the Yale New Haven Hospital network between July 2013 and 2014. Clinicians performed more than 34,000 CT exams for the nearly 137,000 emergency cases.
To help them assess variation, the researchers developed a machine-learning algorithm that calculated the likelihood -- or risk -- that a patient would undergo a CT exam. This "risk standardization" algorithm adjusted for 567 variables believed to contribute to CT use, such as patient age and sex, as well as medical history, emergency severity, and time of day.
An initial analysis of the data before the algorithm was applied suggested a high variability in emergency CT use for specific patient conditions. The researchers calculated variation using the statistical concept coefficient of variation, a ratio of the standard deviation to the mean, in which higher numbers represented higher levels of variation. Their initial analysis showed a coefficient of variation ranging from a high of 57.9 for abdominal trauma cases to a low of 30.2 for head nontrauma studies.
This number dropped almost twofold after they applied the risk standardization algorithm. For abdominal trauma cases, the coefficient of variation fell to 22.3, while for head nontrauma cases it dropped to 19.5. The reduction in CT usage variability applied regardless of anatomical region or type of condition (traumatic or nontraumatic).
|Change in estimate of CT variation after risk standardization algorithm
|Before risk standardization
|After risk standardization
The smallest degrees of risk-standardized variation were in nontraumatic chest and traumatic head CT exams, which are generally associated with well-established clinical decision protocols, the authors wrote. This tendency suggests that analyzing variation in CT use may be most relevant for clinical indications without firm clinical decision rules.
"Our results show the importance of risk standardization for utilization measures to be valid indicators of provider performance," they wrote. "Despite the analytic burden, administrators should seek to incorporate risk adjustment into local quality improvement to improve clinician engagement and ensure credibility of efforts to reduce patient exposure to ionizing radiation ... and achieve improvements in emergency care value."