Musculoskeletal Imaging
Original Research
Diagnostic Performance of CT for Occult Proximal Femoral Fractures: A Systematic Review and Meta-Analysis
Audio Available | 
Share
OBJECTIVE. The purpose of this study was to assess the diagnostic performance of CT for detection of occult proximal femoral fracture.
MATERIALS AND METHODS. A systematic search adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines was performed for studies of the sensitivity and specificity of CT for detection of hip fracture. Two independent investigators extracted data and assessed the quality of each study using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). A qualitative systematic review and quantitative meta-analysis were performed. A Bayesian bivariate random effects meta-analysis model with integrated nested Laplace approximation was used to estimate sensitivity and specificity.
RESULTS. Thirteen heterogeneously reported studies were assessed that included 1248 patients (496 with a hip fracture and 752 without) with MRI or clinical follow-up as the reference standard. There were 50 false-negative examinations. The summary estimate of sensitivity was 94% and of specificity was 100%.
CONCLUSION. CT can be considered a reasonable choice when occult proximal femoral fracture is suspected clinically in patients for whom MRI is contraindicated or not readily available. However, for patients with ongoing clinical concern about hip fracture despite normal CT findings, MRI should be performed.
Keywords: CT, meta-analysis, occult hip fracture, sensitivity, specificity
Proximal femoral fractures are associated with high disability rates, health care costs, and mortality rates [1]. Delayed diagnosis may necessitate more extensive surgical reconstruction and can result in avascular necrosis, nonunion, and thromboembolic complications [2].
The initial imaging evaluation of hip fracture typically consists of an anteroposterior radiograph of the pelvis and frontal and lateral radiographs of the hip in question [3]. In most cases, this standard radiographic series is sufficient for diagnosis and characterization of proximal femoral fractures, allowing definitive treatment. In a minority of patients, high clinical suspicion of fracture and normal radiographic findings lead to additional imaging to exclude a radiographically occult fracture. For an independently mobile patient with normal radiographic findings who is unable to bear weight or has persistent pain after trauma, clinical examination has been found unreliable for distinguishing between patients with and those without fracture [4].
A number of patient and technical factors may lead to a fracture being considered occult on radiographs, such as bony demineralization, fracture nondisplacement or impaction, overlying soft tissues, and suboptimal positioning. Occult fractures have been reported to be as common as 3–10% among all normal radiographic findings obtained for trauma [5–7]. In patient cohorts with normal radiographic findings and ongoing clinical concern about fracture, the incidence of proximal femoral fracture is 7–40% when MRI is used as the reference standard [5, 7–10].
Given its high diagnostic performance in detecting and excluding occult proximal femoral fractures, MRI has been the modality of choice when occult fracture is suspected. However, barriers to MRI include inconsistent availability, expense, contraindications in patients with incompatible devices, and susceptibility to motion. MRI also generally involves longer delay before imaging is performed and longer examination times than for CT. An abbreviated MRI protocol consisting of coronal STIR and coronal T1-weighted sequences can be used to mitigate some of these challenges and has high diagnostic accuracy [11]. Dual-energy CT also has high sensitivity and specificity for occult hip fractures and increased reader confidence [12]; however, this modality is inconsistently available. Therefore, conventional CT remains the modality of choice in most emergency departments for the detection or exclusion of radiographically occult hip fractures because it is nearly universally available at every institution.
Reports on the diagnostic performance of CT for assessment of occult proximal femoral fractures are conflicting. Some authors [13–17] have suggested near 100% accuracy of conventional CT, whereas others [18–21] have found MRI superior by a substantial margin. There has also been variable use of MRI and clinical follow-up or repeat CT as the reference standard, which may contribute to the conflicting reports of the diagnostic accuracy of CT. The purpose of this study was to perform a systematic review and meta-analysis of the current literature to evaluate the diagnostic performance of CT in the evaluation of clinically suspected occult hip fracture.
| Materials and Methods |   |
|---|
A systematic search was performed of three online research databases, adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The search included all articles in the PubMed, Web of Science, and Embase databases from inception through December 1, 2018. The following query was performed in each database separately by two of the authors: “(Computed tomography OR CT) AND (hip OR femur OR femoral neck OR intertrochanteric OR peritrochanteric OR subtrochanteric OR greater trochanter) AND (occult fracture OR nondisplaced fracture).”
We included original research investigations that evaluated CT for assessment of occult proximal femoral fracture in association with trauma (occult proximal femoral fracture was defined as a fracture not apparent or equivocal on radiographs), included sufficient information to complete a 2 × 2 contingency table, and used either MRI or clinical follow-up as the reference standard. Articles describing only pelvic and acetabular fractures, review articles, and articles on nonhuman subjects were excluded. There were no language exclusions.
Patient data from each of the articles were extracted separately by each of the two authors and compiled as a final list of patients in a spreadsheet. For each patient, the result of the CT examination and of the study reference standard with respect to hip fracture was recorded. When available, patient demographics recorded included age, sex, type of hip fracture according to CT findings, type of hip fracture according to the reference standard, and treatment. For each study, the CT scanner type, CT technique, and method of interpretation (number and type of readers), and study reference standard were also recorded. Hip fractures were defined as fractures involving the proximal femur. Acetabular and other pelvic fractures were not included. Each author's extracted data were then compared, and discrepancies between authors were resolved by consensus. Quality assessment was performed independently by two investigators using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool [22], in which the risks of bias and applicability concerns were evaluated in the domains of patient selection, flow of patients, performance of the index test, and standard of reference.
A qualitative and descriptive systematic review was based on the extracted data. A meta-analysis to evaluate the diagnostic performance of CT for detection of occult hip fracture was performed by means of a Bayesian bivariate random effects method with integrated nested Laplace approximation [23, 24]. For each study, summary estimates of sensitivity and specificity were modeled with 95% CIs, and corresponding forest plots were drawn. The prior distribution was the penalized complexity priors. The ROC AUC was calculated, and a summary ROC curve was drawn. To explore heterogeneity among the studies, I2 values were calculated; I2 greater than 50% was designated as significant heterogeneity. Publication bias was evaluated by visual inspection of Deeks funnel plots [25, 26]. All statistical analyses were performed with the meta4diag, integrated nested Laplace approximation (INLA), and mada packages of R software (version 3.5.2, R Foundation for Statistical Computing).
| Results | |
|---|
The initial search query yielded a total of 517 articles; 127 were removed as duplicates (Fig. 1). After review of all titles, the abstracts of 41 were reviewed, and 27 articles received a complete full-text review. Of these, 13 studies comprising 1248 patients met all inclusion criteria and were used in the final analysis. All studies were in English. The references of these studies were searched, and no additional studies were found.
 View larger version (83K) | Fig. 1 —Chart shows literature search results and process for identifying articles for inclusion. |
The 13 articles included in this meta-analysis were published between 2005 and 2018 (Table 1). Countries of origin included the United Kingdom (five articles), Sweden (three articles), United States, Canada, Denmark, Israel, and Romania (one article each). Study populations undergoing CT ranged from six patients [18] to 231 patients [13]. One study population consisted of only patients with fractures [21]. All but one of the studies, that by Haubro et al. [28], which was prospective, were retrospective. The results of quality analysis are shown in Figure 2.
 View larger version (68K) | Fig. 2A —Quality assessment of included primary articles according to revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. A, Chart shows results for individual studies. |
 View larger version (43K) | Fig. 2B —Quality assessment of included primary articles according to revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. B, Chart shows proportion of studies with low (green), high (orange), and unclear (blue) risk of bias. |
 View larger version (33K) | Fig. 2C —Quality assessment of included primary articles according to revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. C, Chart shows proportion of studies with low (green), high (orange), and unclear (blue) concerns about applicability. |
Data extraction yielded a total of 1248 patients (694 women, 364 men, 190 sex not specified; mean age, 81 years; mean age of women, 82 years; mean age of men, 80 years). Mean age was not specified for the 190 patients whose sex was not specified. A total of 496 patients had proximal femoral fractures, and 752 did not. The most common proximal femoral fracture was of the femoral neck (155) followed by intertrochanteric ridge (134), and greater trochanter (72). One patient had a femoral head fracture and one a subtrochanteric fracture. In 124 patients, the hip location was unspecified, including five fractures documented as extracapsular.
The mean time between CT and initial normal radiographic findings was less than 24 hours in three studies [19, 20, 27] and between 24 and 48 hours in three studies [14, 28, 29] (Table 2). One article [16] reported that CT was performed less than 24 hours after radiography in 67.3% of cases. The time elapsed between initial radiography and CT was not specified in six studies [12, 13, 15, 17, 18, 21]. Reports of 11 studies specified the type of CT scanner used, all of which were 16-MDCT or greater. Specifically, three studies [18, 20, 21] were performed with 16-MDCT; three studies [13, 17, 27] with two CT scanners, including one 16-MDCT and one newer-generation scanner; three studies [16, 28, 29] with 64-MDCT scanners; one study [19] with a 128-MDCT scanner; and one study [12] with a dual-energy CT scanner with only the single-source images analyzed.
The CT examinations were interpreted by a variety of methods. The clinical report was generated by a single radiologist [15, 19, 21], two observers [16, 20, 28], or three or more observers [12, 13, 27] (Table 2). Three reports [14, 17, 29] did not specify how the CT examinations were interpreted. CT interpretation was reported as fracture or no fracture in all but one study [13], which reported ancillary findings, such as marrow edema, ef-fusion, and lipohemarthrosis.
In three studies, including 44 patients [20], 67 patients [28], and six patients [18], all patients underwent MRI as the reference standard (Table 2). In three studies, including 231 patients [13], 61 patients [17], and 118 patients [12], clinical follow-up was the only reference standard. In six retrospective studies [15, 16, 19, 21, 27, 29] clinical follow-up was used, and MRI was performed for a minority of patients. The reason for performing MRI for specific patients was rarely individually described.
CT identified 446 of 496 (90.0%) fractures. Fifty of the 1248 CT examinations had false-negative findings. These included nine femoral neck fractures, 10 intertrochanteric fractures, one greater trochanter fracture, and 30 unspecified proximal femoral fractures. The false-negative findings were diagnosed with MRI in 35 cases (nine by Collin et al. [20], 14 by Deleanu et al. [21], two by Dunker et al. [27], three by Haubro et al. [28], four by Lubovsky et al. [18], two by Mandell et al. [19], and one by Sadozai et al. [29]) with follow-up CT in nine cases [21], with subsequent radiography in five cases (two by Dunker et al. [27], three by Sadozai et al. [29]), and with dual-energy CT in one case [12].
Because of heterogeneity in data reporting and study designs, formal statistical analyses were not performed on the parameters in the systematic review, including prevalence of hip fractures, CT detector type, method of CT interpretation, and delay between radiography and CT. Nor was a formal analysis performed on the reference standard used (MRI or clinical follow-up), owing to high risk of bias and variable proportion of MRI examinations performed in each study, each MRI in the retrospective studies being performed for specific clinical reasons unique to each patient.
A meta-analysis was performed on the pooled diagnostic performance of CT for assessment of nondisplaced proximal femoral fracture. The estimated summary sensitivity of CT for detection of nondisplaced hip fracture was 94% (95% credible interval, 83–99%), and the specificity was 100% (95% credible interval, 99–100%). Forest plots of pooled sensitivity and specificity are shown in Figure 3. The AUC value was 0.981. The summary ROC curve is shown in Figure 4.
 View larger version (48K) | Fig. 3 —Forest plot shows pooled sensitivity and specificity in included primary articles. TP = true-positive, FP = false-positive, TN = true-negative, FN = false-negative. |
 View larger version (38K) | Fig. 4 —Graph shows summary ROC curve (bivariate model) for pooled data from included primary articles. |
The Higgins I2 was 10.98%, indicating absence of significant heterogeneity. The forest plot did show some variability in sensitivity and specificity, but this variability may be due to several factors. A variety of techniques were used for CT image acquisition and interpretation. CT scans were interpreted by varying numbers of radiologists of varying subspecialty and experience, and four articles did not specify the interpreters of the CT scans. The reference standard was variable between studies and also within individual studies. Five articles described the use of a combination of clinical follow-up and either MRI or follow-up CT as the reference standard in cases of high suspicion. The specific circumstances of each patient undergoing MRI in these studies with mixed reference standards were usually not specified. Five articles described the use of clinical follow-up alone, and duration varied from 30 days to 12 months. Several articles did not define a specific duration of follow-up. MRI was the reference standard in three articles, which included six patients [18], 44 patients [20], and 67 patients [28]. Most of the studies were retrospective; only one [28] was prospective.
The funnel plot (Fig. 5) shows mild asymmetry, suggesting that some publication bias was present.
 View larger version (37K) | Fig. 5 —Funnel plot shows results of evaluation of publication bias among included primary articles. Extremes of whiskers indicate 95% confidence limits. |
| Discussion | |
|---|
Hip fractures lead to a large number of orthopedic admissions, particularly in the elderly population. Timely diagnosis of these fractures is clinically important, because delayed diagnosis results in increased morbidity and mortality and increased health care costs [1, 2].
Because conventional CT remains the modality of choice in most emergency departments for the detection or exclusion of radiographically occult hip fractures, owing to its easy availability and rapid image acquisition, accurate assessment of the diagnostic performance of CT is needed to guide patient care, particularly when CT findings are normal. To our knowledge, ours is the first meta-analysis aimed at evaluating the diagnostic performance of CT for assessment of radiographically occult or equivocal proximal femoral fractures on the basis of current primary literature.
Thirteen heterogeneously reported studies were included in which the diagnostic performance of CT was either directly reported or could be calculated on the basis of reported data. Study populations undergoing CT ranged from six patients [18] to 231 patients [13]. Our meta-analysis population consisted of 1248 patients, 496 of whom had occult proximal femoral fractures. In this population, Bayesian modeled estimates of sensitivity and specificity of CT were 94% and 100%. The AUC was 0.981. Although it is evident that CT does not show all nondisplaced fractures, the current literature supports moderate to high sensitivity and very high specificity.
Limitations of this meta-analysis include heterogeneity in the patient population, study design, and reporting of the included articles. The type of hip fracture was not consistently reported and was not specified for 124 patients, which limits the conclusions about geographic distribution of fractures. Among those reported, femoral neck (155) and inter-trochanteric fractures (134) were the most common. Of the 50 patients with false-negative CT scans, fracture type was unspecified for 30, meaning we cannot draw conclusions about which types of femoral fractures are most likely to be CT occult. In addition, the definition of occult hip fracture varied between articles. Some study populations included only patients who had truly normal radiographic findings (thus, the fractures were radiographically occult), whereas others included populations in which the radiographs were equivocal. Several articles did not explicitly mention the specific radio-graphic findings other than stating that their study investigated occult hip fracture.
The reference standard varied substantially between included studies, only three (117/1248 patients) consistently having MRI as the reference standard. This likely reflects the limited availability of MRI at many institutions and the challenges in performing both MRI and CT in a large cohort of patients. In our meta-analysis, we considered the studies in which MRI was not consistently used as the reference standard to have potential risk of bias because the reference standard for hip fracture diagnosis was not used. We believe, however, that clinical follow-up with or without additional imaging other than MRI is likely to capture the vast majority of patients with hip fractures after normal CT findings are made and would almost certainly capture fractures requiring surgical intervention. Nevertheless, this remains a limitation of this meta-analysis.
It is also difficult to know whether the few false-positive findings in our meta-analysis could be due in part to an inconsistent reference standard. Of the three studies (117 patients) in which MRI was consistently used as the reference standard, there were three false-positive findings, all from the same study [20], which included 44 patients. Only two additional false-positive findings were reported among the studies in which MRI was not consistently used as the reference standard. However, false-positive findings might have been underreported if the supposed fracture called at CT was nonsurgical and follow-up MRI was not performed.
In this meta-analysis, 446 of 496 (90.0%) fractures were identified with CT. Of the 50 false-negative findings, 35 were found at subsequent MRI, 14 were found either on follow-up radiographs or at repeat CT, and one was noted at dual-energy CT. Overall, this meta-analysis shows that CT has high sensitivity (94%) and very high specificity (100%) for occult hip fractures.
| Conclusion | |
|---|
CT can be considered a reasonable choice when occult proximal femoral fracture is suspected, especially if MRI is not readily available or may delay patient care. However, for patients with ongoing clinical concern about hip fracture despite normal CT findings, MRI should be performed, although this clinical scenario affects only approximately 10% of patients undergoing CT for occult proximal femoral fractures.
| References | |
|---|
