Use of CT for Head Trauma: 2007–2015

We conducted a repeated cross-sectional analysis of the National Hospital Ambulatory Medical Care Survey (NHAMCS) ED data sets from 2007 to 2015. The NHAMCS is conducted annually by the Centers for Disease Control and Prevention National Center for Health Statistics, and statistics from 2015 were published in the most recent release (release date: November 20, 2017). In the survey, a multistage probability sampling design is used to collect data on ∼30 000 annual visits to some 300 randomly selected US EDs. Each patient visit is the basic sampling unit and is assigned a weight to allow for the generation of nationally representative population-level estimates. Visit information is collected by trained staff members during a randomly assigned 4-week reporting period, monitored for accuracy by NHAMCS field representatives, and processed centrally with manual and computerized verification algorithms. NHAMCS sampling data are collected from within geographically defined areas, hospitals within those areas, EDs within those hospitals, and on patients who have visited those EDs. More information regarding the NHAMCS sampling methodology can be found on the Web site for the Centers for Disease Control and Prevention.²⁴ NHAMCS data sets are publicly available and deidentified and as such, were deemed exempt from review by the Harvard University Institutional Review Board.

The eligible study population included all pediatric patients <18 years of age presenting with a chief complaint or discharge diagnosis of head injury, which were identified by reason for visit and International Classification of Diseases, Ninth Revision, Clinical Modification codes, respectively. Patient visits were identified by reason for visit codes, including injury of head (55050); contusion, abrasion, or bruise of head (54050); and fracture or dislocation of head and face (50050), as well as by using the International Classification of Diseases, Ninth Revision, Clinical Modification codes for skull fracture (800.xx to 804.xx), concussion (850.xx), intracranial hemorrhage (851.xx to 853.xx), other brain injury (854.xx), and head injury, unspecified (959.01). Analysis was restricted to patients for whom an injury was indicated by using the question, “Is this visit related to an injury?” a variable available in the NHAMCS. Visits recorded with a disposition of “dead on arrival” or “left without been seen” were excluded.

The primary outcome was the annual proportion of children with acute head injuries who received a CT scan in the ED. To explore the latency of knowledge translation, secondary multivariable analyses were conducted to evaluate the association between CT use in the period before and after the publication of the PECARN prediction rules. For secondary analyses, binary variables were created to compare neuroimaging for visits from 2007 to 2009 (before the PECARN period [published October 2009]) with visits from 2010 to 2015 (after PECARN period) as well as for additional sensitivity analyses by sequentially omitting the first years after PECARN publication. Beginning in 2007 (for the outcome of interest), NHAMCS data include specific information that identifies whether CT imaging is neuroimaging or another body system. Additional data were collected on covariates that were shown previously to be associated with CT neuroimaging,^7,25,26 including patient demographic information (age, sex, race, and insurance provider), visit characteristics (triage acuity level and requirement for admission), and ED characteristics (pediatric and teaching status). Missing data within the NHAMCS were coded as unknown or unavailable. Age was treated as a binary variable for children <2 years old and children 2 to 18 years old, consistent with the PECARN rule dichotomization.¹³ Race was classified as white, African American, or other as per the NHAMCS recoded race variable. Insurance provider was categorized as private, Medicaid or Medicare, self-pay, or other. Triage acuity was classified as immediate, emergent, urgent, semiurgent, nonurgent, and unknown or unavailable, as designated in the NHAMCS, and modeled as a categorical variable. Consistent with previous work, EDs were classified as pediatric hospitals if ≥85% of all visits were for patients <21 years and as teaching hospitals if ≥25% of all patients were evaluated by a resident physician.⁷ 

All analyses were performed by using Stata v.14.1 (StataCorp, College Station, TX) with survey procedures for weights, stratum, and primary sampling units to account for the multistage sampling design of the NHAMCS and to generate population estimates. The annual population-level CT trend was analyzed by using a Pearson’s χ² global test for differences among survey-weighted proportions. Multivariable logistic regression was used to test the strength of association of factors associated with CT neuroimaging and to compare CT use before and after the PECARN rules. Missing data were greatest for triage acuity (14.9% of observations), such that sensitivity analyses were performed by complete case analysis with these observations removed as well as with all unavailable triage values recoded as highest acuity (immediate) and recoded as lowest acuity (nonurgent). Population estimates are presented in millions unless otherwise noted, and percentages are expressed as survey-weighted proportions with 95% confidence intervals (CIs). A 2-tailed P value <.05 was considered statistically significant.

During the 9-year study period, there were 269 721 ED visits in the NHAMCS, of which 59 921 were for children <18 years old. There were 3089 observations that met inclusion criteria, of which 35 were excluded for a disposition of “dead on arrival” or “left without being seen.” The final analysis included 3054 unweighted observations, representing an estimated 14.3 million (95% CI: 13.0–15.7 million) pediatric ED visits for head trauma. Nationwide estimates for pediatric head trauma increased from 1.31 million in 2007 to 1.91 million in 2015. The median patient age was 6 years (interquartile range: 2–13 years old); 61.4% of patients were boys (95% CI: 59.0%–63.7%), 1.6% of patients were hospitalized (95% CI: 1.2%–2.2%), and a majority of patients were evaluated in nonteaching (88.2%; 95% CI: 85.2%–90.7%) and nonpediatric hospitals (88.7%; 95% CI: 84.2%–92.0%). Demographics and visit characteristics for the study population are shown in Table 1.

Overall, 32.0% (95% CI: 29.0%–34.6%) of children with head injuries underwent CT neuroimaging. Figure 1 reveals the population-weighted annual trend of CT neuroimaging in the study population by year, which ranged from 26.4% to 36.0%. Over the course of the 9-year study period, there was no significant linear trend in the annual estimates for CT neuroimaging among children with head injuries (P trend = .50). Restricting this analysis to only the lowest acuity categories (semiurgent or nonurgent), CT use was 16.0% in 2007 and 26.1% in 2015 (P trend = .69). Approximately 84.6% of children with head injuries who received CT imaging underwent isolated neuroimaging without CT of any other body region. Excluding children who received both head and any nonhead CT imaging, overall CT use was 29.5% with no linear trend between 2007 (27.6%; 95% CI: 19.8%–37.1%) and 2015 (30.2%; 95% CI: 26.8%–32.4%) (P trend = .39).

Multivariable analysis results are presented in Table 2. Similar to the linear trend analysis of weighted annual CT use, a multivariate analysis revealed no significant association between CT neuroimaging and year (adjusted odds ratio [aOR]: 1.02; 95% CI: 0.97–1.07; P = .50) after adjustment for patient age, sex, race, insurance provider, pediatric or teaching hospital, and triage acuity. Sensitivity analyses similarly revealed no significant association with year when observations missing acuity data (coded as unknown or unavailable in NHAMCS) were excluded (aOR: 1.02; 95% CI: 0.96–1.08; P = .53), or when all these missing values were recoded as the highest (immediate) acuity level (aOR: 1.01; 95% CI: 0.96–1.06; P = .78) or as the lowest (nonurgent) acuity level (aOR: 1.02; 95% CI: 0.97–1.08; P = .40).

Several factors were found to be significantly associated with increased CT neuroimaging (Table 2), including age ≥2 years, white race, acuity categories urgent or higher, and presentation to a nonteaching or nonpediatric hospital.

The unadjusted proportion of children with head injuries who underwent CT neuroimaging did not differ when comparing the periods before (2007–2009; 33.2%) and after the PECARN rules (2010–2015; 31.4%; P = .49). Similarly, there was no difference when adjusted for potential confounding in a multivariate analysis (aOR: 1.07 after PECARN versus before; 95% CI: 0.84–1.38; P = .59). To allow for a possible transitional wash-in period, sensitivity analyses were performed by sequentially omitting the first years after the PECARN publication, and each also revealed no change in CT neuroimaging in the 6-year period after the PECARN rules (Table 3).

Using a large, nationally representative database of US ED visits, we found no change in CT neuroimaging among children with head injuries from 2007 to 2015. CT use remained high (32% overall) throughout the study period, and there was no decrease in linear trend over time or in multivariable analyses after adjustment for potential confounders. Moreover, CT use did not decrease among children who were receiving only isolated neuroimaging nor among low acuity triage categories of those least likely to meet the PECARN high-risk criteria. The finding of no decrease in CT neuroimaging during the 9-year study period, and particularly after the publication of the PECARN rules in 2009, was counter to the a priori hypothesis of this study.

Rates of CT neuroimaging presented here are similar to other investigations. A study of 25 US tertiary pediatric EDs from 2004 to 2006 and another of 40 US tertiary pediatric EDs from 2005 to 2009 both revealed mean rates of 35% for children with head injuries who underwent CT neuroimaging.^8,27 Both studies were limited to children’s hospitals rather than nationally representative data. In an analysis of the NHAMCS from 1995 to 2003, researchers reported a peak rate of 29% of children with head injuries who underwent CT in 2000.⁷ A more recent NHAMCS study from 2002 to 2006 revealed 39% CT use among pediatric visits for head injury in the pooled 5-year sample.²⁶ Of note, previous studies in which the NHAMCS database was queried predated 2007 data capture changes that specify the CT type into head versus nonhead scans and precluded any definitive knowledge of the body region that underwent CT imaging. In a study of 324 435 pediatric head trauma visits to 848 general EDs, researchers used the Nationwide ED Sample to find that CT neuroimaging ranged from 53.7% to 55.2% between 2006 and 2010.⁹ The elevated use of CT found in that study may be because the authors did not apply sampling weights and excluded pediatric EDs in the analysis. Importantly, the majority of estimates for CT use in children with head injuries have been from children’s hospitals, and no national estimates have been reported beyond 2010.

Excess CT use contributes to rising health care costs, exposes many children to the harmful effects of ionizing radiation and for some, the additional risks of sedation required for imaging. With a long-standing recognition of the lifetime fatal malignancy risk that is associated with CT imaging, considerable efforts have been made, championing the “as low as reasonably achievable” principle and a reduction of unnecessary CT scans.¹⁵ Beginning in 2007, the Image Gently Alliance national educational campaign has been raising awareness among health care providers and the public regarding the radiation-associated risks of diagnostic imaging.^28,–30 Lodwick et al³¹ found that the overall use of CT imaging decreased from 2009 to 2013 at 30 tertiary pediatric hospitals. Use of CT neuroimaging declined an average of 6.7% per year; however, this was for all CT head indications, and even among these tertiary pediatric centers, there was an approximate fourfold range in scanning between the hospitals with the lowest and highest rates.³¹ So, although there is some evidence of more conservative use of CT imaging, the current study suggests that this may be predominantly among pediatric centers and may not be for children with head injuries per se. Reducing unnecessary CT neuroimaging specifically among children with head injuries has been a Choosing Wisely priority of the American Academy of Pediatrics since 2013.¹⁷ 

The PECARN decision rules identify children for whom CT can be avoided with a high sensitivity and negative predictive value.^11,13 No broad assessment of CT use has been undertaken since the derivation of these rules in 2009, and we found that the use of CT neuroimaging did not decrease in the 6-year period after the publication of the PECARN rules. This steady elevated use of CT neuroimaging reveals the well-established challenge of translating high quality evidence into practice.³² Knowledge translation of health research has been reported to take an estimated average of 17 years to enter day-to-day clinical practice,^33,34 although there is no consensus on how to precisely measure these translational delays nor how they can be reduced.³³ A 2017 survey of ED physicians revealed that despite familiarity of Choosing Wisely recommendations, over half of respondents reported that they sometimes, often, or always perform CT for patients with head injuries who are at low risk per decision rules.³⁵ Barriers to reducing CT use for minor head injury in the ED have been qualitatively explored. Melnick et al³⁶ described factors that promote or hinder appropriate use of CT within 5 core domains: the need to establish trust, patient and provider anxiety, ED-related practice constraints, the influence of others, and patient expectations. The authors concluded that systems to establish trust and manage uncertainty could optimize CT use in the ED.³⁶ 

Several reports have revealed successful implementation of the PECARN rules, resulting in lower rates of CT imaging for pediatric head trauma without missing neurosurgical TBIs.^21,–23 The implementation of the PECARN rules at an urban tertiary pediatric ED reduced CT scanning from 21% to 15%, with an additional reduction to 9% from individual practitioner performance feedback. There were no increases in ED length of stay, 72-hour return visits, or missed significant TBIs.²¹ Similarly, a hospital-sponsored quality improvement project incorporated the maintenance of certification credits and participant coaching to reduce the CT rate from 29% to 17% in a community ED.²² A multifaceted intervention of 5 pediatric EDs and 8 general EDs that was centered on real-time, computerized PECARN clinical-decision support was associated with decreased CT use among all children with minor blunt head trauma without increasing the rate of missed injuries.²³ Findings from the latter study suggest achievable benchmarks of <15% CT use for all children with minor head trauma and <5% for children who are at low risk per the PECARN rules.²³ An important remaining challenge will be scaling these successes to a national level, particularly given the epidemic of pediatric head injuries.¹ 

In our study, we identify several factors that are associated with an increased use of CT neuroimaging, including age, higher triage acuity, and presentation to nonteaching or nonpediatric hospitals. Each of these variables have been previously shown to be associated with increased CT use.^7,9,26 Moreover, white race was significantly associated with increased odds of CT neuroimaging, adding to a growing body of literature that suggests racial disparities in the care of children with acute head injuries.^26,37 Identifying patient and provider groups that are most likely to benefit from targeted quality initiatives to reduce CT overuse remains an important area of future research.

There are several potential limitations to the current study. The NHAMCS provides high-level, nationally representative data but does not contain important clinical information, such as the severity of the head injury or the Glasgow coma score. Therefore, the true appropriateness of CT use cannot be assessed. We have attempted to control for injury severity using triage acuity in all models. In the NHAMCS, the unit of analysis was the ED visit; we are unable to account for repeat visits by individual patients. Similarly, no patient or provider identifiers are released in the NHAMCS data set, which precludes longitudinal analysis of individual patients or institutions. As such, we are unable to identify hospital-level variation or performance at specific centers. Also, our study is dependent on coding accuracy within the NHAMCS, and we cannot exclude the possibility of misclassification. However, NHAMCS data are processed at a central facility and reviewed manually for accuracy, and the diagnostic codes that are used are those endorsed by the Centers of Disease Control and Prevention for head trauma case inclusion.³⁸ Misclassification is also possible in the ascertainment of hospital pediatric and teaching status, although we have used definitions consistent with those that have been described previously.⁷ 

The use of CT neuroimaging for children with acute head injuries did not decrease over the 9-year study period, which also spanned the publication of the PECARN clinical-decision rules. Results from this large nationally representative study suggest a need for targeted quality improvement initiatives to ensure appropriate CT use in this patient population. There remains an important need for interventions to translate clinical-decision rules into practice.

aOR

adjusted odds ratio

CI

confidence interval

CT

computed tomography

ED

emergency department

NHAMCS

National Hospital Ambulatory Medical Care Survey

PECARN

Pediatric Emergency Care Applied Research Network

TBI

traumatic brain injury

We thank Geetha Iyer (Harvard T.H. Chan School of Public Health) and Raphael Freitas (McGill University Health Centre) for technical assistance.