Diagnostic accuracy of contrast enhanced ultrasound in patients with blunt abdominal trauma presenting to the emergency department: a systematic review and meta-analysis

We aimed to investigate the diagnostic accuracy of contrast-enhanced ultrasound (CEUS) in evaluating blunt abdominal trauma for patients presenting to the emergency department. Electronic search of Scopus and Pubmed was performed from inception to September 2016. Human studies investigating the diagnostic accuracy of CEUS in identifying abdominal solid organ injuries were included. Risk of bias was assessed using the QUADAS tool. A total of 10 studies were included in the study and 9 of them were included for meta-analysis. The log(DOR) values ranged from 3.80 (95% CI: 2.81–4.79) to 8.52 (95% CI: 4.58–12.47) in component studies. The combined log(DOR) was 6.56 (95% CI: 5.66–7.45). The Cochran’s Q was 11.265 (p = 0.793 with 16 degrees of freedom), and the Higgins’ I2 was 0%. The CEUS had a sensitivity of 0.981 (95% CI: 0.868–0.950) and a false positive rate of 0.018 (95% CI: 0.010–0.032) for identifying parenchymal injuries, with an AUC of 0.984. CEUS performed at emergency department had good diagnostic accuracy in identifying abdominal solid organ injuries. CEUS can be recommended in monitoring solid organ injuries, especially for patients managed with non-operative strategy.


Risk of bias.
Risks of bias of individual studies were assessed using the QUADAS tool ( Table 1). The tool consisted of 14 items, each of which should be scored as "yes", "no" or "unclear" for an individual study. Item 1 was to assess spectrum bias in an individual study and it was scored "yes" if the individual study stated it included patients with blunt abdominal trauma. Item 2 was to assess whether the study explicitly define the inclusion and exclusion criteria. Because CT was considered to be able to correctly identify parenchymal injury, item 3 was scored "yes" if they used CT as reference standard. If an individual study described that CT scan was performed immediately after CEUS, the item 4 was scored as "yes". If all participants received reference standard test for an individual study, item 5 was scored as "yes". Item 6 was to exclude differential verification bias and this was "yes" if all participants received reference test irrespective of the CEUS results. Item 7 described the incorporation bias, and if review authors believed that interpretation of CT is independent of the CEUS results, item 7 was considered as "yes". If sufficient details on how to perform CEUS and CT scan were described, item 8 and 9 should be "yes". If there was sufficient evidence that CEUS and CT results were interpreted independently, items 10 and 11 were scored as "yes". Clinical data such as history of trauma, patient demographics were available during the study and clinical practice, thus item 12 was scored "yes" for all individual studies. If no intermediate/uninterpretable test results were reported, item 13 was scored as "no". If both test results were obtained, item 14 was scored as "no" for the individual study 17 .
Synthesis of results. The sensitivity and specificity are two interrelated quantities commonly reported for diagnostic test. Bivariate approach to the meta-analysis of diagnostic accuracy was employed in the study 18 . Quantities such as sensitivity, false positive rate and area under curve (AUC), as well as associated 95% confidence intervals were reported. Diagnostic odds ratio (DOR) was also reported for component studies, as well as for the pooled result. In analogy to meta-analysis of odds ratio (OR), meta-analysis of DOR was performed according to the method developed by Glas and colleagues 19 . Cochran's Q statistic and Higgins I 2 were reported for the assessment of heterogeneity of included studies 20 . A forest plot of the log(DOR) values together with the summary estimate was also presented. Subgroup analysis was performed by restricting dataset to the type of solid organs (e,g, spleen, liver and kidney). All statistical analyses were performed using R (version 3.3.2) 21 .

Results
Study selection. The initial search identified 421 citations (Fig. 1). After removing duplicates, there remained 295 citations. A number of 281 citations were excluded because they were experimental studies, reviews, commentaries, case reports, and non-trauma conditions. The remaining 14 citations were reviewed for the full text. Four citations were excluded because two studies included only patients with CT-confirmed solid organ injury 22,23 , one was duplicated report 16 , and one is a case series report 24 . As a result, a total of 10 studies were included in the study and 9 were included for meta-analysis 14, 15, 25-32 . Study characteristics. Characteristics of included studies are shown in Table 2. All studies were published after the year 2000, and five studies were published after 2010. The sample sizes of included studies ranged from 22 to 392. Four studies utilized prospective design, four studies employed retrospective design and the remaining two did not explicitly report the design. All studies included blunt abdominal trauma and the severity of trauma citations were reviewed for the full text. Four citations were excluded because two studies included only patients with CT-confirmed solid organ injury, one was duplicated report, and one is a case series report. As a result, a total of 10 studies were included in the study and 9 were included for meta-analysis.
was mild to moderate. They used different words to describe the severity such as "low energy", "minor", "isolated" and "hemodynamically stable". Most studies investigated the three major solid organs including kidney, spleen and liver. One study investigated diagnostic accuracy of CEUS on adrenals and found that CEUS missed two adrenal lesions as compared with CT 14 . Valentino's study included patients with blunt abdominal trauma in which they among other injuries identified lesions in the pancreas and adrenals 29 . The included studies reported a wide range in the study participants' age. One study included pediatric patients 26 , and two studies included both adult and pediatric patients 14,15 . Three studies did not report summary statistics for the age [28][29][30] . Timing of CEUS was reported in two studies 14,27 , and the remaining studies did not explicitly report the timing of CEUS. All studies reported the CT as reference standard, against which CEUS was compared.
Risk of bias for individual study. The risk of bias was assessed using the QUADAS tool (  Table 3. Quality assessment of included studies using QUADAS tools. Pooled analysis of diagnostic accuracy of CEUS was performed using the bivariate approach. The included studies provided 17 datasets for pooled analysis. Overall, the CEUS had a sensitivity of 0.981 (95% CI: 0.868-0.950) and a false positive rate of 0.018 (95% CI: 0.010-0.032) for identifying parenchymal injuries ( Table 4). The area under curve was 0.984. In subgroup analysis restricting to the type of solid organs the diagnostic accuracy of CEUS was excellent. The sensitivity and false positive rate for spleen were 0.904 (95% CI: 0.829-0.947) and 0.028 (95% CI: 0.007-0.099), respectively. The sensitivity and false positive rate for liver were 0.941 (95% CI: 0.784-0.986) and 0.011 (95% CI: 0.004-0.035), respectively. The sensitivity and false positive rate for kidney were 0.910 (95% CI: 0.616-0.984) and 0.011 (95% CI: 0.004-0.035), respectively. The summary receiver operating characteristic curves (SROC) were shown in Fig. 3 and subgroup analysis was shown in Fig. 4.

Discussion
The systematic review included 10 studies, 9 of which were included for meta-analysis. All studies reported good diagnostic accuracy of CEUS in identifying parenchymal lesions in patients with blunt abdominal trauma. There is no evidence of significant heterogeneity among included studies and among different types of solid organs (I 2 = 0%). The result showed that the CEUS had a sensitivity of 0.981 (95% CI: 0.868-0.950) and a false positive rate of 0.018 (95% CI: 0.010-0.032) for identifying parenchymal injuries, and the area under curve was 0.984. The patients who are candidates for examination with CEUS are the hemodynamically stable patients with a history of low-energy blunt abdominal trauma. Since the diagnostic accuracy of CEUS is comparable to the CT scan, it is reasonable to recommend CEUS for assessment of patients presenting to the emergency department with a history of blunt abdominal trauma. Furthermore, CEUS is radiation-free and is suitable for repeated evaluations. For blunt abdominal trauma, delayed hemorrhage is common and thus repeated evaluations with CEUS is valuable in excluding potentially life-threatening conditions. Particularly, since there is trend shifting from operative to  Table 4. Diagnostic performance of CEUS using computed tomography as reference standard. Statistics were pooled using bivariate model. The n value in the first column was the number of datasets available for combination, and thus one study could provide more than one dataset.
non-operative management of blunt splenic trauma, continuous monitoring of the lesion is of paramount importance 33 . CEUS provides a convenient and useful tool for such a purpose 13 . However, CEUS has limited value in evaluating deep and small organs such as adrenals. In Miele's study, all adrenal lesions were missed during CEUS, but they were confirmed with CT scan 14 . At one month follow up, the adrenal lesions were again missed in CEUS examination, but they could be identified by magnetic resonance imaging (MRI). Valentino's study also reported that CEUS did not identify a lesion of the right kidney and an adrenal hematoma, which were later identified with CT 29 . For pancreas, it is reported that CEUS was able to identify 21/22 blunt pancreatic injury, yielding a false negative rate of 4.5%. However, this study only included patients  with CT-confirmed pancreatic injury without reporting the false positive number, failing to provide a whole picture of the diagnostic accuracy 22 . Valentino M and colleagues reported a case of pancreatic injury. While B-mode US failed to demonstrate the injury lesion which was recognized by CT and MRI, CEUS well demonstrated the injury and was useful for monitoring during follow up of the lesion 34 . Although current evidence showed a good diagnostic performance of CEUS in identifying pancreatic lesions, the quality of evidence is limited. Lv's study was retrospective in design that it was largely unknown whether the interpretation of CEUS result was influenced by CT findings or not 22 . In other words, the review bias could not be fully excluded 17 . As a result, the diagnostic accuracy of CEUS on pancreatic injury needs to be further examined in well design studies.
One limitation of the study was that the sample sizes of component studies were generally small. Half of the studies had a sample size below 100, which were traditionally regarded as small studies. There is evidence that small studies may overestimate the effect size in intervention meta-analysis 35 . Although small study effect has not been fully investigated in meta-analysis of diagnostic test, the result of our study should be interpreted with caution. However, studies with large sample size also showed an excellent diagnostic accuracy of the CEUS in identifying solid organ injuries, providing robustness to our results. It should also be noted that many studies were retrospective in design, which means that there was no strict study protocol for the performance of CT and CEUS. The concern with such study design is that the results of CT and CEUS may interfere with each other. In real clinical practice, emergency physicians have access to all imaging information (e.g. including CT findings if available) of a patient and this can influence their interpretation of CEUS findings. The traditional explanatory trials have been criticized for its external validity, because they are typically performed in experienced centers with strict inclusion/exclusion criteria [36][37][38] . In contrast, pragmatic trials allow better generalization of the result to the real world setting. In our example, pragmatic design means that physicians are not blind to imaging results during the performance of CEUS, confounding the diagnostic accuracy of CEUS. It is interesting to investigate how the presence of the results of other imaging studies can influence the interpretation of CEUS findings. It should be acknowledged that B-mode US and CEUS both have limited value in a trauma setting when evaluating bowel or mesenteric lesions. The images cannot be obtained clearly with the interference of bowel gas. Furthermore, the marketed ultrasound contrast agents are not excreted renally and therefore lesions in the collecting ducts are not visualized and could be overlooked 39 .
In conclusion, the study showed that CEUS performed at emergency department had good diagnostic accuracy in identifying solid organ injuries. It is a radiation-free technique that can be considered in monitoring solid organ injuries, especially for patients managed with non-operative strategy.
Data availability statement. All data were available in original articles.