Presented at the American Academy of Pediatrics, Section on Emergency Medicine, National Conference and Exhibition, Washington, DC, October 2009.
Bedside Ultrasound Diagnosis of Clavicle Fractures in the Pediatric Emergency Department
Article first published online: 6 JUL 2010
© 2010 by the Society for Academic Emergency Medicine
Academic Emergency Medicine
Volume 17, Issue 7, pages 687–693, July 2010
How to Cite
Cross, K. P., Warkentine, F. H., Kim, I. K., Gracely, E. and Paul, R. I. (2010), Bedside Ultrasound Diagnosis of Clavicle Fractures in the Pediatric Emergency Department. Academic Emergency Medicine, 17: 687–693. doi: 10.1111/j.1553-2712.2010.00788.x
The study was funded by an intradepartmental grant from the University of Louisville, Department of Pediatrics (No. GR1719-CROS01). It covered supplies, statistical consulting, and training; there was no salary reimbursement.
Disclosure: Dr. Gracely was a paid statistical consultant.
Supervising Editor: Thomas G. Costantino, MD.
- Issue published online: 6 JUL 2010
- Article first published online: 6 JUL 2010
- Received January 22, 2010; revision received April 2, 2010; accepted April 12, 2010.
ACADEMIC EMERGENCY MEDICINE 2010; 17:687–693 © 2010 by the Society for Academic Emergency Medicine
Objectives: Clavicle fractures are among the most common orthopedic injuries in children. Diagnosis typically involves radiographs, which expose children to radiation and may consume significant time and resources. Our objective was to determine if bedside emergency department (ED) ultrasound (US) is an accurate alternative to radiography.
Methods: This was a prospective study of bedside US for diagnosing clavicle fractures. A convenience sample of children ages 1–18 years with shoulder injuries requiring radiographs was enrolled. Bedside US imaging and an unblinded interpretation were completed by a pediatric emergency physician (EP) prior to radiographs. A second interpreter, a pediatric EP attending physician with extensive US experience, determined a final interpretation of the US images at a later date. This final interpretation was blinded to both clinical and radiography outcomes. The reference standard was an attending radiologist’s interpretation of radiographs. The primary outcome was the accuracy of the blinded US interpretation for detecting clavicle fractures compared to the reference standard. Secondary outcome measures included the interrater reliability of the unblinded bedside and the blinded physicians’ interpretations and the FACES pain scores (range, 0–5) for US and radiograph imaging.
Results: One-hundred patients were included in the study, of whom 43 had clavicle fractures by radiography. The final US interpretation had 95% sensitivity (95% confidence interval [CI] = 83% to 99%) and 96% specificity (95% CI = 87% to 99%), and overall accuracy was 96%, with 96 congruent readings. Positive and negative predictive values (PPVs and NPVs, respectively) were 95% (95% CI = 83% to 99%) and 96% (95% CI = 87% to 99%), respectively. Interrater reliability (kappa) was 0.74 (95% CI = 0.60 to 0.88). FACES pain scores were available for the 86 subjects who were at least 5 years old. Pain scores were similar during US and radiography.
Conclusions: Compared to radiographs, bedside US can accurately diagnose pediatric clavicle fractures. US causes no more discomfort than radiography when detecting clavicle fractures. Given US’s advantage of no radiation, pediatric EPs should consider this application.
Clavicle fractures are a common pediatric injury in emergency departments (EDs). The incidence of clavicle fractures in two studies was estimated at between 5 and 12 per 10,000 person-years.1,2 Their management is usually straightforward, but their diagnosis typically requires confirmation by imaging.3,4 Plain radiographs are the test of choice. However, radiographs take time, cost money, and expose patients to radiation.5–7
An alternative to radiographs for diagnosis of clavicle fractures is ultrasound (US). US is already used to diagnose neonatal clavicle fractures due to birth trauma.8–10 Because US may show poorly ossified neonatal bones better than radiographs, and because it lacks ionizing radiation, it is well suited to this application.11 There are published case series with high diagnostic success rates (100%) of US for neonatal clavicle fractures.9,10 In addition, another case series describes US diagnosis of clavicle fractures in older children in an orthopedic clinic, again with high diagnostic accuracy: 96% for US versus 91% for radiographs.12 A series of 653 pediatric general trauma patients found US performed by radiologists to be as accurate as radiographs when screening for any suspected fracture and superior to radiographs for detecting clavicle fractures.13
The accuracy of US for clavicle fractures in these settings suggests its potential use to diagnose clavicle fractures in the pediatric ED. We hypothesized that bedside US can reliably diagnose clavicle fractures when compared to the reference standard of plain radiographs.
This was a prospective evaluation of a diagnostic test (US) against a reference standard (radiographs). The project was approved by the local university institutional review board and funded by an internal departmental grant. The study was registered at ClinicalTrials.gov. Informed consent from parents or guardians, and assent from children age 7 years and older, was obtained for all subjects. The study enrolled patients from March 2008 through September 2009.
Study Setting and Population
Pediatric patients presenting to a single ED at an urban, tertiary care children’s hospital were recruited as a convenience sample when the study investigators were available. Triage nurses and attending physicians routinely paged the investigators for likely study candidates at any time of the day or night. Additionally, research assistants who work from noon to midnight notified investigators when a candidate patient arrived.
Patients age 1 to 18 years presenting with shoulder or clavicle pain due to recent trauma were eligible for the study if they had a normal neurovascular examination and were to have radiographs as part of their evaluation. A parent or guardian had to be available to give formal consent without unduly delaying the patient’s routine care.
Patients were excluded from the study if they had altered mental status, hemodynamic instability, multisystem trauma, open shoulder wounds, significant developmental delay, inability to understand and give informed consent in English, allergy to US gel, prior radiographs at another institution, or ipsilateral clavicle fracture in the prior 24 months.
All patients were offered ibuprofen (10 mg/kg, maximum 400 mg) or oxycodone (0.1 mg/kg, maximum 10 mg) for injury pain prior to enrollment as part of our institution’s routine triage nursing practice for patients presenting with presumed orthopedic injuries. After consent and age-appropriate assent were obtained, a study investigator performed a brief history and clinical examination of the subject. Basic demographic information and the investigator’s pretest clinical suspicion for clavicle fracture were recorded. After at least 30 minutes had passed since pain medication had been dosed, a standardized US examination was performed by one of the study investigators (all but three were performed by KPC). A 10- to 15-MHz high-frequency linear transducer connected to a Sonosite M-Turbo (Bothell, WA) US machine was used to visualize the clavicle bone starting at the sternal junction and moving laterally. Images were recorded of the medial, middle, and lateral clavicle for later review (Figure 1). Once the clavicle was examined, US images of the proximal humerus and the ipsilateral pleural sliding were viewed to document concomitant findings such as humerus fracture or pneumothorax. Diagnosis of clavicle fractures was based primarily on seeing cortical bone disruption, but also could be diagnosed by seeing a callus, hematoma, or bone motion with respiratory cycle. These US findings have been documented as characteristic of bone fractures in several prior studies.14–17
At the end of the US examination, patients age 5 or older were asked to self-rate how much pain had occurred during imaging on a Wong-Baker FACES pain scale. This pain score and the bedside US interpretation (“fracture” or “no fracture”) were recorded.
Patients then proceeded to radiology for plain films as ordered by the treating emergency physician (EP). After completion of the plain radiographs, patients age 5 years and older were asked to rate their pain during radiographic imaging on the FACES pain scale. Disposition of patients from this point was at the discretion of the treating EP.
Plain films were interpreted per routine by our institution’s radiology staff or its overnight radiology service. Attending radiologists performed all readings. The interpretations were recorded on the study database as either fracture or no fracture, which served as the study’s reference standard.
At a later date, recorded US images were reviewed on a blinded basis by a different study investigator (FHW) and a determination of fracture or no fracture was recorded for each case. For cases of disagreement between the radiology reading and the blinded US interpretation, a second, blinded review of radiographs was performed by a single senior pediatric radiologist.
The primary outcome was the agreement between a blinded EP’s interpretation of US images and a blinded radiologist’s interpretation of radiographs. The only clinical information these reviewers had was the age of the patient and a one-sentence description of the mechanism of injury. The physician who performed the blinded reviews (FHW) did not provide care to any study patients to maintain blinding. In addition, he had extensive experience with bedside US of both adults and children, although not for this application. The choice of a blinded US interpretation as the primary outcome was made to focus the study on the ability of the information contained in the US images themselves to make an accurate diagnosis.
There were several secondary outcomes for this study. The first was a comparison of the unblinded bedside physician’s interpretation of US images with the radiologist’s reading of radiographs. This outcome was chosen because it was felt to reflect daily practical use of this application. Another secondary outcome was the interrater reliability between the blinded US reviewer and the unblinded bedside physician. The study also investigated the pain scores for children ages 5 years and older for both US and radiographic imaging. Finally, the pretest clinical probability of clavicle fracture (coded as “high,”“medium,” or “low”) was recorded before imaging and then compared with the final radiograph diagnosis.
Data were recorded in Excel (Microsoft Corp., Redmond, WA) and transferred to SPSS version 17 (SPSS Inc., Chicago, IL) for statistical analysis. The outcomes for the blinded review and the bedside interpretation of US images were compared with the radiologist’s interpretation to calculate sensitivity, specificity, and positive and negative likelihood ratios with 95% confidence intervals (CI). Positive and negative predictive values (PPVs and NPVs, respectively) were also calculated; however, these statistics are determined in part by the actual prevalence of the condition being tested (here, clavicle fractures). Because that prevalence may differ from one setting to another, PPVs and NPVs are site specific; our results are reported here for comparison purposes only. Interrater reliability was calculated using a kappa statistic. Pain score interpolated medians were calculated and reported, along with means (which provide different information and are more familiar). Differences in pain scores were compared between procedures with a Wilcoxon signed ranks test. Pain scores were also categorized into low pain (scores of 0 or 1) and high pain (scores of 2–5) to report aggregate frequencies. For all statistical analysis comparisons, a Type I error level of 0.05 was the level for significance.
Based on 2006 historical data at our institution, 45% of patients presenting with shoulder injuries are ultimately diagnosed with clavicle fractures. From prior published studies, the accuracy of US for diagnosing clavicle fractures was estimated at 95% or better.10,12 Using this information, we estimated CIs around point estimates for sensitivity and specificity of US using a binomial distribution and different sample sizes. A sample size of 100 subjects was found to have 95% CIs of less than 20%. Furthermore, a review of recent studies of bedside US diagnostic applications published in peer-reviewed journals found sample sizes ranging from 68 to 118, giving similar CI estimates.18–24 This study therefore planned to recruit 100 subjects.
The study enrolled 103 subjects, and 100 completed all aspects of the study. Three patients were withdrawn after enrollment: one for extreme agitation that prevented obtaining any US imaging, one for technical difficulties recording images, and one for the postenrollment discovery of a recent ipsilateral clavicle fracture that met exclusion criteria. Three other patients were offered enrollment but declined to participate in the study: one parent was concerned about the time commitment, and two children refused to give assent for fear of pain.
The mean (± standard deviation [SD]) age of subjects was 10.7 ±4.3 years with a range of 1 to 17 years. There were 75 boys and 25 girls. Eighty-six children were age 5 years or older, and self-reported pain scores were obtained from them. There were 43 patients with clavicle fracture as defined by the reference standard radiographs and 57 patients with no clavicle fracture. No pneumothoraces were seen during the study. There were three proximal humerus fractures, all of which were seen on US before radiographs. For this study of clavicle fracture diagnosis, patients with humerus fractures were reported purely on the basis of their clavicle findings (none of which were fractured).
Table 1 shows the primary outcome of the blinded reviewer’s interpretation of US images for patients who did and did not have a clavicle fracture on radiograph. Based on these results, the sensitivity of US to radiographically detected fracture was 95% (95% CI = 83% to 99%), and the specificity was 96% (95% CI = 87% to 99%). The likelihood ratio of a positive US result was 27 (95% CI = 7 to 106) and of a negative US result was 0.05 (95% CI = 0.01 to 0.19). The positive PPV was 95% (95% CI = 83% to 99%) and the NPV was 96% (95% CI = 87% to 99%).
There were two false-negative cases—fractures seen on radiograph that were not seen on US. The first case was a 14-year-old boy who had fallen on his shoulder and had a faint hairline fracture visible on radiograph. This fracture was missed by the blinded US reviewer, but not by the physician who did the bedside US. The second false-negative was a 4-year-old boy who fell on a waterslide. He also had a hairline fracture visible on radiograph. It was not identified by either US interpretation. Follow-up with this patient’s family revealed that within 24 hours the child had abandoned his sling and swathe. His father reported that the next day he “was playing in the backyard like nothing had happened.” The two false-positive cases were not followed after discharged from the ED.
The four cases of disagreement between the blinded US reviewer and the radiologist (false-positives and false-negatives) underwent a second review of the radiograph findings. None of these reviews by a senior pediatric radiologist resulted in a change in the original radiologist’s interpretation.
Table 2 shows the results for the secondary outcome of the unblinded bedside US interpretation for patients who did and did not have a clavicle fracture on radiograph. Based on these results, the unblinded bedside US interpretation had a sensitivity of 93% (95% CI = 80% to 98%) and specificity of 86% (95% CI = 74% to 93%). The likelihood ratio of a positive US result was 6.6 (95% CI = 3.5 to 12.7) and of a negative US result was 0.08 (95% CI = 0.03 to 0.24). The PPV was 83% (95% CI = 69% to 92%) and the NPV was 94% (95% CI = 83% to 98%). The bedside interpretation had lower specificity and positive likelihood ratio results because of a higher rate of false positives. The interrater reliability kappa statistic was 0.74 (95% CI = 0.60 to 0.88) for the comparison of the blinded reviewer’s interpretation with the unblinded bedside interpretation of US images.
There were 86 patients 5 years of age or older who self-reported pain scores for both US and radiographic imaging. The distribution of pain scores is shown in Table 3. On a scale where 0 is “no hurt” and 5 is “hurts worst,” the interpolated median FACES pain score was 0.6 for US and 0.8 for radiographs, with corresponding means of 1 and 1.52, respectively (p = 0.01 by Wilcoxon signed ranks test). Both median and mean procedure differences were small (0.2 and 0.52, respectively) and slightly favored US. Pain scores of 0 or 1 (low pain) were reported 71% of the time for US and 58% of the time for radiographs.
|Pain Score (No. of Respondents)||US||Radiograph|
|0 (no hurt)||41||40|
|1 (hurts little bit)||20||10|
|2 (hurts little more)||14||13|
|3 (hurts even more)||7||7|
|4 (hurts whole lot)||3||6|
|5 (hurts worst)||1||10|
The results for pretest clinical suspicion for clavicle fracture compared to the final radiographic diagnosis are shown in Table 4. The PPV of high clinical suspicion was 80% (95% CI = 64% to 90%). The NPV of low clinical suspicion was 95% (95% CI = 82% to 99%). Additionally, clinical suspicion could be combined with either the blinded reviewer US result or the bedside physician US interpretation. When using the blinded reviewer US result, the PPV of high clinical suspicion combined with a fracture seen on US was 97% (95% CI = 83% to 99%), while the NPV of low clinical suspicion with no fracture seen on US was 100% (95% CI = 89% to 100%). When using the bedside physician US interpretation, the PPV of high clinical suspicion and a fracture on US was 85% (95% CI = 70% to 94%), while the NPV of low clinical suspicion and no fracture seen on US was 97% (95% CI = 85% to 99%).
|Pretest Clinical Suspicion||Radiograph|
This study demonstrates that clavicle fractures can be reliably diagnosed in the pediatric ED using bedside US. In particular, US was most effective when it confirmed a low clinical suspicion of fracture or when a fracture was clearly seen in a patient with high clinical pretest probability of a fracture. There was a tendency to overcall fractures in patients with medium or high clinical suspicion and unclear US findings at the bedside.
Our findings are consistent with the results of a recent study of US screening of 653 pediatric trauma patients for multiple types of fractures followed by radiographs. It concluded that the two imaging methods had equivalent sensitivity (93%) and specificity (99%) in general and found US to be more sensitive than radiographs for detecting clavicle fractures.13 That study differed from our current study in that US imaging was performed by pediatric radiologists rather than EPs.
The question of whether any clavicle imaging is needed at all in cases of high (or low) clinical suspicion has been raised in light of imaging’s associated costs and minimal effect on diagnosis, treatment, and outcomes.25,26 While our results show clinical suspicion to be reasonably accurate, its PPVs and NPVs were improved in all cases with the addition of US results. This improvement ranged from 2% to 17%, with a number needed to US of between 6 and 50 to correct one inaccurate clinical diagnosis.
For medicolegal reasons, many pediatric EPs in the United States may confirm clavicle fractures with some form of imaging, rather than relying purely on clinical judgment. To satisfy these medicolegal concerns, bedside US offers an attractive alternative to radiographs. Additionally, in our experience, bedside imaging that parents witness is often reassuring to families far beyond its quantifiable diagnostic accuracy—and most children enjoy taking home a printed US picture of their fracture.
Ultrasound was no more painful than radiographs in this study. Indeed, the small but statistically significant difference favored US as less painful. The median pain scores were similar for the two methods, and the difference in mean scores was clinically small (less than 1 “face” on the pain scale). An inspection of the distribution of pain scores (Table 3) suggests that in most patients, both techniques cause little pain. However, there was a clear subset of patients who found radiographs extremely painful: 10 of 86 gave it the highest possible pain score. In comparison, only one patient gave US imaging the highest pain score. In these patients, we believe it was less painful to move an US probe for different US views than to reposition the patient’s injured shoulder for different radiographic views.
Interrater reliability was good to excellent with a kappa of 0.74 (95% CI = 0.60 to 0.88). For comparison with two recent radiology studies of interrater reliability, the kappa statistic for two radiologists reading chest radiographs to diagnose pneumonia was 0.53 (95% CI = 0.37 to 0.69) and acute respiratory distress syndrome was 0.37 (95% CI = 0.22 to 0.52).27,28
In many EDs, bedside US is available for a variety of other uses: FAST examinations of trauma patients, imaging of cardiac effusion and function, guiding vascular access, evaluation of abscesses and foreign bodies, bladder catheterizations, diagnosing forearm fractures, estimating volume status, and placing nerve blocks.18,29–32 Growing familiarity with US in the ED highlights the opportunity for US diagnosis of clavicle fractures.
Although not specifically studied, it is our opinion that clavicle fracture is an excellent application for novice ultrasonographers in the ED. The patients commonly present in stable condition, the exam is brief and easy to perform, the images are usually straightforward to interpret, and radiographs may be ordered on any uncertain case to confirm the US findings. We believe most novice operators will find—as we did—that uncertainty diminishes over time and radiographs are needed less frequently.
Such uncertainty may be reflected in the higher number of false-positive readings by the unblinded bedside ultrasonographer in this study, who was a less experienced physician. These errors represent a conservative approach to shoulder injuries, as there is little downside to placing an injured shoulder in a sling-and-swathe for several days and discontinuing it if symptoms resolve.
In published studies of a variety of different applications, bedside emergency US has been shown to shorten length of stay.32–36 It can also lower costs.36–38 Furthermore, it avoids radiation exposure (concerning principally for the child’s thyroid gland), albeit at a relatively low dose in typical radiographs.5,6,39 We believe that these benefits of time, cost, and radiation avoidance may also accrue to patients and providers when US is adopted as the primary diagnostic test for clavicle fractures, with radiographs reserved for indeterminate cases.
This study was conducted at a single site by a small number of investigators, all of whom had prior experience with US. All but three of the US examinations were performed by a single physician. The blinded reviewer had extensive US experience with both adult and pediatric US, albeit for other applications. Patient-reported pain scores were collected by the study investigator, not by a neutral third party, and therefore may be biased by a patient’s desire to please the investigator or by the investigator’s phrasing of pain score questions.
Limitations also included convenience sampling rather than consecutive sampling of patients with shoulder injuries. However, we have no reason to believe patients presenting at times when investigators were unavailable were different from enrolled subjects.
Financial costs (or savings) related to the choice of imaging modality were not studied. Because the study involved US followed by radiographs for all subjects, it was not able to compare the speed of diagnosis of one modality versus the other. We were also unable to quantify the effect of imaging modality on overall length of stay in the ED. However, it is our opinion that bedside US appeared notably faster than radiographs in most instances.
This study’s reference standard was the patient’s radiograph obtained at the initial visit. There was no specific follow-up on patients after discharge, and therefore occult clavicle fractures diagnosed a week or more after initial presentation may have been missed by our reference standard. This limitation in particular affects the interpretation of false-positive US findings. It is possible some of these false-positives were actually occult clavicle fractures for which US was more sensitive than radiographs—a possibility suggested by other studies.12,13
The choice of radiographic studies to order was left to the treating EPs in each case. While this reflects real-world variations in practice, it also meant some patients received “shoulder series” views, while others had “clavicle series” views. As a practical matter, the diagnosis of clavicle fracture could be made or excluded to the attending radiologist’s satisfaction in all cases. However, it is possible that different outcomes for the reference standard would have occurred if all subjects had identical radiographic studies. Additionally, the choice of radiographic studies and variations in the style of care provided by different radiology technicians may influence pain scores for radiography.
The pretest probability of fracture was reported only as high, medium, or low without more specific definitions of what these terms mean. Consequently, high or low probability for our setting may differ from that of other physicians in other settings.
Finally, US interpretations were recorded only as fracture or no fracture without recording the confidence level of each reading. Many fractures were simple to see on US—we had numerous occasions of untrained parents watching the imaging in process and exclaiming, “Oh look! It’s broken right there!” In contrast, there was a subset of patients with subtle, less obvious findings. Had we recorded a confidence level for each US interpretation, we would be able to quantify what we observed anecdotally: that US was adequate in most—but not all—cases.
This study showed that bedside ultrasound in the pediatric ED can accurately diagnose clavicle fractures when compared to plain radiographs. Given its diagnostic accuracy, minimal pain during examination, and lack of ionizing radiation, physicians should consider US for bedside diagnosis of pediatric clavicle fractures.
- 25Prospective evaluation of clinical assessment in the diagnosis and treatment of clavicle fracture: are radiographs really necessary? Can J Emerg Med. 2003; 5:309–13., , , et al.