- Email: [email protected]
The Lateral View
The Lateral View: A Screening Method for Knee Trauma1 Adarsh Verma, Albert Su, MD, Alex M. Golin, MD, Brian O’Marrah, Judith K. Amorosa, MD
Rationale and Objectives. The authors performed this study to evaluate whether any one conventional radiographic view is sufficient as a screening method in the detection of acute knee trauma. Materials and Methods. The authors prospectively assessed the efficacy of a single conventional radiograph of the knee in the detection of signs of knee fractures in 214 consecutive adults with acute knee trauma. The evaluated efficacy parameters included specificity, sensitivity, negative predictive value (NPV), and positive predictive value. The percentage reduction in the need for additional conventional radiographs was also calculated. Results. Fifty-three of the 214 patients (24.8%) had a knee fracture. The sensitivity of the single lateral view in the detection of knee fractures was 100% (95% confidence interval [CI] ⫽ 94.3, 100). The lateral view of the traumatized knee was normal in 143 patients (66.8%). The probability of not having a fracture if the lateral view was normal (NPV) was also 100% (95% CI ⫽ 97.9, 100). The need for additional radiographs was reduced 67%. Conclusion. A single lateral view as a screening tool for knee fractures has a very high sensitivity and NPV. Because more than 65% of the patients had a normal lateral view in this study, there can be a considerable amount of savings in terms of radiology services for these patients. Key Word. Knee, fractures.
There are 1.3 million annual visits to U.S. emergency departments because of knee trauma, with more than $1 billion spent on radiography of the knee for these injuries; 90%–92% of all knee examinations show no fracture (1). The high cost and radiation exposure of knee imaging combined with the low yield create a need to reevaluate the current imaging method used to screen for knee injury. The traditional method of evaluating the injured knee has been a knee series, which varies greatly from institution to institution and within a given institution, depend-
Acad Radiol 2001; 8:392–397 1 From the University of Medicine and Dentistry of New Jersey-Robert Wood Johnson Medical School, One Robert Wood Johnson Place, Box 19, New Brunswick, NJ 08903-0019. Received September 6, 2000; revision requested October 25; revision received December 6; accepted January 4, 2001. Address correspondence to J.K.A.
©
AUR, 2001
392
ing on the requesting physician’s preference at a particular moment. The practice of obtaining a radiograph of the traumatized knee is universal, although according to Fishwick et al (2) up to 25% of knee radiographs obtained for acute trauma do not correlate with clinical findings. Fishwick et al concluded that radiographs obtained for acute trauma do not reliably depict all important knee injuries. Magnetic resonance (MR) imaging appears to be a superior method for imaging the soft tissues, cartilaginous joint surfaces, and bone around the knee when compared with conventional radiography of the knee (3). The prohibitive cost of MR imaging, however, does not make this technology a feasible frontline tool for screening acute knee trauma. Clinical decision rules, such as the Ottawa Knee Rules developed by Stiell and colleagues (4), are a powerful way to improve the quality, efficiency, and cost-effectiveness of health care delivery. According to the Ottawa Knee Rules, radiography is required in patients with acute
Academic Radiology, Vol 8, No 5, May 2001
knee injury only if (a) the patient is 55 years or older, (b) there is tenderness at the head of the fibula, (c) there is isolated tenderness of the patella, (d) the patient is unable to flex the knee to 90°, and (e) the patient is unable to walk four steps both at the time of injury and at evaluation. If radiography is not required (on the basis of the physician’s usual practice), a follow-up call is made in 2 weeks to ensure that there is no fracture. These rules are a practical way of reducing unnecessary radiographic examinations without reducing diagnostic accuracy. In addition, they encourage physicians to be more selective in their use of radiography in patients with acute knee injuries on the basis of a set of predetermined epidemiologic and clinical criteria. A delayed diagnosis of a missed knee fracture, however, can result in a poor clinical outcome for the patient. For physicians in the United States, malpractice damage awards due to failure to diagnose a fracture averaged $55,600 in 1995 (5). Consequently, the patients’ insistence on diagnostic imaging for their acute injuries, combined with the physicians’ fear of lawsuits for missed fractures, make the implementation of clinical decision rules virtually nonexistent in the United States. Therefore, we set out to determine an efficient and cost-effective screening method for detecting fractures in patients with acute knee trauma. Part of our goal was to decrease the need for unnecessary radiographic views of the injured knee while preserving the patients’ and physicians’ desire to radiographically rule out a fracture. The purpose of this study was to test the theory that a single conventional radiograph may suffice to screen for acute knee trauma if it has a reasonably high sensitivity and negative predictive value (NPV). The lateral view was chosen as the focus of our study because, in our experience, it most often shows the abnormality in trauma cases. We propose that, on the basis of a thorough clinical examination of the knee and a single lateral view, the patient’s knee can be categorized as (a) normal, with no further radiographs necessary; (b) abnormal, with the need to obtain additional plain radiographs for further evaluation; or (c) abnormal, with the need to perform MR imaging or, occasionally, computed tomography (CT) for further evaluation. CT may be the best method of evaluation when surgical intervention is indicated. If meniscal or ligamentous injury is suspected by the orthopedic consultant, MR imaging may be the best method of further evaluation. In case of dislocation of the knee joint where vascular injury is suspected, arteriography (or CT arte-
LATERAL VIEW FOR KNEE TRAUMA
riography or MR arteriography) would be the recommended next step in evaluation.
MATERIALS AND METHODS Conventional radiographs (anteroposterior [AP], lateral, oblique, and tunnel views) of 214 consecutive adult patients (age range, 18 –97 years; mean age ⫾ standard deviation, 42 years ⫾ 18) with knee trauma were prospectively interpreted during a period of 10 months. All radiographs were read by one of the authors (J.K.A.), a boardcertified radiologist with special expertise in thoracic radiology and extensive daily experience with conventional radiography and CT for more than 20 years, during the assigned emergency department rotation at a level I trauma center. All patients at least 18 years of age with acute knee trauma were included in the study. AP and lateral radiographs of the traumatized knee were obtained in all patients. One hundred sixty-two oblique and 74 tunnel radiographs were also obtained in a part of the study group. Each image was displayed on the view boxes and analyzed. In all cases, the lateral view was read first, followed by all other views. The reader was thus blinded to the findings in the other views. All findings were tabulated at the reading and subsequently entered into a database. Knee fractures were defined as radiographic evidence of fracture of the patella, the head and neck of the fibula, the proximal 8.0 cm of the tibia, or the distal 8.0 cm of the femur. These true cases of knee fracture served as the basis for statistical analysis. Radiographs with signs of a knee fracture, joint effusion, lipohemarthrosis, air in the joint space, loose bodies, and bone fragments were considered to be “abnormal.” Evidence of osteoarthritis, such as osteophytes, joint space narrowing, and articular cartilage calcification, were considered to be “normal” findings not associated with acute knee injury. According to our algorithm, a lateral radiograph showing any of the above abnormalities was considered “test positive,” and those patients underwent additional radiologic studies for further evaluation (eg, conventional radiography with other views, CT, or MR imaging). Patients who were found to have a fracture on these additional images were classified as having a true-positive finding, and those without any further evidence of a fracture were classified as having a false-positive finding. All normal lateral radiographs were considered to be “test negative,” and pa-
393
VERMA ET AL
tients with no further evidence of a fracture on additional radiologic studies were classified as having a true-negative finding. Patients with a normal lateral view in whom a knee fracture was diagnosed on other radiographs were classified as having a false-negative finding. On the basis of these criteria, the sensitivity, specificity, positive predictive value (PPV), and NPV with the 95% confidence interval (CI) were calculated for each conventional radiographic view (lateral, AP, tunnel, oblique, and lateral and AP combined). Because none of the fractures were missed with a single lateral radiograph, calculation of the long-run risk of missing a fracture (with a larger sample size) within a 95% CI was performed by using the method described by Hanley and Lippman-Hand (6) (see Appendix). Long-run risk can be defined as a calculation of the risk of an event occurring if the event of interest fails to arise in a limited number of subjects but may arise with further testing in a larger patient population. A calculation of the estimated savings in radiologic services was also performed. Standard receiver operating characteristic (ROC) curve analysis (Analyse-It; Analyse-It Software, Leeds, England) was performed on the lateral views to evaluate disease detectability independent of the effects of disease prevalence and decision threshold (7). A five-point confidence rating scale was used to rate the lateral view, as follows: 1 ⫽ definitely normal, 2 ⫽ probably normal, 3 ⫽ indeterminate, 4 ⫽ probably abnormal, and 5 ⫽ definitely abnormal. The area under the ROC curve (Az) was then calculated. RESULTS Two hundred fourteen adult patients were included in this study. There were 121 men (56.5%) and 93 women (43.5%) aged 18 –97 years (mean age ⫾ standard deviation, 42 years ⫾ 18; median age, 39 years). Of the 214 patients, 53 (24.8%) had knee fractures. The various types of knee fracture are given in Table 1. The mechanism of injury for the knee trauma was a motor vehicle accident in 78 patients (36.4%) and a fall in 17 (8%); in 121 patients (56.5%), the injury was caused by other trauma or unknown factors. Both lateral and AP radiographs were obtained in all patients. Six of the lateral images were horizontal cross-table views. An oblique view was obtained in 162 patients (75.7%), and a tunnel view was obtained in 74 (34.6%). The lateral view was normal in 143 of the 214 patients (66.8%), and the AP view was normal in 158 (73.8%).
394
Academic Radiology, Vol 8, No 5, May 2001
Table 1 Prevalence of Fractures in the Study Population
Location of Fracture
No. of Patients (n ⫽ 214)
Patellar Distal femoral Proximal fibular Proximal tibial Tibial plateau Proximal tibial and tibial plateau Proximal tibial and fibular Tibial plateau and fibular Femoral and patellar Patellar and fibular
13 (24.6) 7 (13.3) 4 (7.5) 4 (7.5) 9 (17.0) 4 (7.5) 4 (7.5) 6 (11.3) 1 (1.9) 1 (1.9)
Total
53
Note.—Numbers in parentheses are percentages.
The oblique view was normal in 136 of the 162 patients (84.0%), and the tunnel view was normal in 66 of the 74 patients (89.2%). Seventy-one patients (33.2%) had an abnormal lateral view, 56 (26.2%) had an abnormal AP view, 26 (16.0%) had an abnormal oblique view, and eight (10.8%) had an abnormal tunnel view. The abnormalities included patellar dislocation, patellar fracture, proximal fibular fracture, tibial plateau fracture, distal femoral fracture, femoral condyle fracture, joint effusion, lipohemarthrosis, air in the joint space, soft-tissue swelling, loose bodies, and bone fragments. In nine patients, only the lateral view showed the abnormality. Eight of these patients had a joint effusion, and one had a patellar fracture. All nine of these patients had normal AP, oblique, and tunnel views. One patient with normal findings on the lateral view had a medial joint separation on the AP and oblique views. Another patient with normal findings on the lateral view had a soft-tissue injury on the AP view. Neither of these patients had knee fractures. Results of statistical analysis of these findings indicate that the lateral view alone is very sensitive for screening for acute trauma of the knee. The sensitivity of the single lateral view in the detection of knee fractures was 100% (long-run risk with 95% CI ⫽ 94.3, 100), the specificity was 88.8% (95% CI ⫽ 83.9, 93.7), the PPV was 74.6% (95% CI ⫽ 64.5, 84.7), and the NPV was 100% (long-run risk with 95% CI ⫽ 97.9, 100). The sensitivity, specificity, NPV, and PPV for the various views are given in Table 2. For AP and lateral conventional radiographs combined, which are very commonly obtained in the detection of
LATERAL VIEW FOR KNEE TRAUMA
Academic Radiology, Vol 8, No 5, May 2001
Table 2 Sensitivity, Specificity, NPV, and PPV for the Different Views
No. of Patients*
No. of Patients with Normal Findings*
Lateral
214 (100)
143 (66.8)
AP
214 (100)
158 (73.8)
AP and lateral
214 (100)
141 (65.9)
Oblique
162 (75.7)
136 (84.0)
Tunnel
74 (34.6)
66 (89.2)
View
Sensitivity (%)†
Specificity (%)†
PPV (%)†
NPV (%)†
100 (53/53) [94.3, 100]‡ 90.6 (48/53) [82.7, 98.5] 100 (53/53) [94.3, 100]‡ 71.4 (15/21) [52.1, 90.7] 87.5 (7/8) [64.6, 100]
88.8 (143/161) [83.9, 93.7] 95.0 (153/161) [91.6, 98.4] 87.6 (141/161) [82.5, 92.7] 92.2 (130/141) [87.8, 96.6] 98.5 (65/66) [95.6, 100]
74.6 (53/71) [64.5, 84.7] 85.7 (48/56) [76.5, 94.9] 72.6 (53/73) [62.4, 82.8] 57.7 (15/26) [38.7, 76.7] 87.5 (7/8) [64.6, 100]
100 (143/143) [97.9, 100]‡ 96.8 (153/158) [94.1, 99.5] 100.0 (141/141) [97.9, 100]‡ 95.6 (130/136) [92.2, 99.0] 98.5 (65/66) [95.6, 100]
* Numbers in parentheses are percentages. † Numbers in parentheses are the numbers of patients. Numbers in brackets are the 95% CI. ‡ Long-run risk range.
None of the lateral views were classified as showing “indeterminate” or “probably normal” findings for ROC analysis. The results yielded an Az of 0.97 in the use of the lateral view to screen for any abnormality (Figure). Finally, according to our proposal, it would not be necessary to obtain additional conventional radiographic views in patients with a normal lateral view (NPV, 100%). Consequently, there would be a 66.8% reduction (66.8% of lateral views were negative in our study) in the need for additional conventional radiographic views. DISCUSSION
Graph shows the ROC curve for the detection of an abnormality on a lateral view of the knee. “No discrimination” is a reference ROC curve for a test that is unable to discriminate a population with disease from a population without disease.
knee fractures, the sensitivity was 100% (long-run risk with 95% CI ⫽ 94.3, 100), the specificity was 87.6% (95% CI ⫽ 82.5, 92.7), the PPV was 72.6% (95% CI ⫽ 62.4, 82.8), and the NPV was 100% (long-run risk with 95% CI ⫽ 97.9, 100). The sensitivity, specificity, PPV, and NPV for the single lateral view alone were not statistically different from those for the lateral and AP views combined.
The conventional radiographic knee series has traditionally been the method of choice in the initial radiologic evaluation of patients with acute knee injury. The definition of which radiographs make up a complete knee series varies greatly among institutions and among individual physicians. Patients may commonly undergo imaging with two to six views. The high cost of a conventional radiographic series containing multiple views, combined with the low yield of positive pathologic findings on these plain radiographs, creates a need to rethink the radiologic screening method for acute knee trauma. Recently, a report has been published about clinical decision rules for the use of radiography in acute knee injury (8). A review of current literature, however, suggests that no conclusive algorithm exists for choosing the conventional radiographic views that must be included in a knee series. In our study, we attempted to compare the various conventional radiographic views of the knee to
395
VERMA ET AL
determine which views are the most productive for discovering injury, specifically knee fractures. We found that the sensitivity of the lateral view for depicting knee fractures was 100%, with an NPV of 100%, in a study population with a 24.8% prevalence of knee fractures. This means that a single lateral view will depict 100% of knee fractures, and, if the lateral view is negative, the likelihood that the patient does not have a fracture (NPV) is 100%. Specificity was not an aim of this study. According to our proposal, the presence of any sign of injury on a lateral view would necessitate further evaluation with other diagnostic imaging techniques (ie, conventional radiography with other views, CT, or MR imaging) even though there may not be any clear evidence of a fracture. On the basis of these practices, this screening method has relatively low specificity for a knee fracture. A specificity of 88.8%, however, is acceptable in this study, where our goal was to show the efficacy of the lateral radiograph for depicting all the true cases of a fracture (high sensitivity) and not missing any fractures when the radiograph is normal (high NPV). In our study, knee joint effusion or lipohemarthrosis were detectable only on the lateral or cross-table lateral view and not on any of the other conventional radiographic views. The effusion may be detected on the lateral view as a widening of the opaque suprapatellar strip or as an increase in the fat pad separation at the base of the suprapatellar pouch. Hall (9) has previously shown that the accuracy of the fat pad separation sign in the diagnosis and exclusion of an effusion was 88% and 90%, respectively. Lipohemarthrosis is also a valuable finding identifiable on the lateral view (10). Lee et al (11) concluded that the horizontal cross-table lateral view is the best method for demonstrating the fat-fluid level in lipohemarthrosis of the knee. The presence of a fat-fluid level in the suprapatellar bursa is virtually pathognomonic for intraarticular bone fracture (3,12). Although the actual injury (eg, tibial plateau fracture) may be seen more definitively on an AP or oblique view, the identification of lipohemarthrosis on the lateral view is an excellent indicator of the need to image the knee further in search of specific fracture. At that point, additional conventional radiographs with multiple projections, CT scans, or MR images may be obtained as the next step. Some of the limitations of our study include the fact that only one radiologist interpreted the conventional radiographs; thus, the results are based on one reader’s in-
396
Academic Radiology, Vol 8, No 5, May 2001
terpretations. Although it is true that interobserver correlation cannot be evaluated when only one reader is used, this is the way in which extremity and most other conventional radiographs are interpreted in most clinical settings. Another limitation of the study was that there was no patient follow-up to determine the presence of a knee fracture that might have been missed during the initial evaluation. Finally, one could argue that there would be additional cost (eg, room cost, time) associated with a second trip to the radiology department for those patients who had an abnormal lateral view and would require additional views as per our algorithm. Only one-third of our 214 patients, however, had abnormal lateral views that necessitated additional radiographic views. We believe that the savings would still substantially outweigh the indirect costs described earlier. We propose that although conventional radiographs may not depict all specific knee injuries, the lateral conventional radiograph is a sensitive and cost-effective way to screen for general indicators of a serious knee injury (or fracture) such as joint effusion and lipohemarthrosis (13). Our data suggest that if the lateral view is negative for any sign of injury, including joint effusion and lipohemarthrosis, we can be 95% confident that the remainder of the conventional radiographic views are 98%–100% likely to be negative, as well. Thus, only the lateral view should be obtained during the initial screening examination. A negative lateral view would eliminate the need to obtain any other conventional radiographs because they too would probably be negative. In cases where the clinical suspicion of injury is high despite the presence of a negative lateral conventional radiograph, the knee should then be imaged further with additional radiography or a more definitive imaging method such as MR imaging. If some findings of injury do become evident on a lateral view screening radiograph, then a decision to obtain other conventional radiographic views or to perform CT or MR imaging can be entertained. During the past 90 years, more special views have been developed for skeletal radiology than for any other radiologic subspecialty (14). As long as conventional radiography and arthrography were the only available methods for imaging the knee, additional conventional radiographic views were required for more definitive evaluation of injury. Now that there are new sensitive methods of evaluation for knee injury (eg, MR imaging and CT), it is time to rethink the routine use of multiple-projection conventional radiography.
LATERAL VIEW FOR KNEE TRAUMA
Academic Radiology, Vol 8, No 5, May 2001
In conclusion, a single lateral view of the knee should be used as a screening imaging method for detecting a knee fracture. If signs of injury are absent from the lateral view, then no additional conventional radiographic views are indicated, as they are also likely to be negative for pathologic findings. If clinical suspicion persists, MR imaging would then be a better choice for definitive imaging of the knee. Resorting to MR imaging in that situation would at least eliminate the cost of a multiple-view, conventional radiographic knee series. If the lateral view reveals signs of injury, however, additional conventional radiographic views may be used to categorize the injury better. In as far as positive conventional radiographs may be relied on when making clinical decisions, a full multiple-view knee series following a positive lateral view may avoid the demand for subsequent use of the expensive MR technology. Because most conventional radiographs obtained after acute knee trauma are negative for findings of injury, the single lateral view would eliminate the overuse of multiple conventional radiographs in the imaging of the posttraumatic knee, resulting in considerable cost savings. APPENDIX Calculation of the Long-Run Risk If the event of interest fails to occur in a finite number of subjects, there is still a certain probability or long-run risk that it may occur with further testing in a larger patient population (6). In making inferences from observations with zero numerators (such as zero missed fractures with the lateral view in our study), we still must calculate a true or long-term risk between zero and some upper limit. To be 95% confident of the validity of the calculated long-run risk, the “rule of three” may be applied. The rule of three states that if none of the n patients shows the event of concern (missed fracture in all patients with fracture in this case), we can be 95% confident that the chance of this event happening is at most three in n (3/n, for n greater than 30).
Calculation The lateral view depicted all 53 fractures (100% sensitivity, n ⫽ 53). The long-run risk of missing a clinically important fracture is 3/53 (5.7%). This means that the sensitivity range with the long-run risk (95% CI) is 94.3%–100%. Similarly, the NPV was 100%, which means that there were no cases of missed fracture when the lateral radiograph was negative (no false-negative findings) in 143 patients (n ⫽ 141). The long-run risk of a false-negative result is 3/141 (2.1%). Therefore, the NPV range with the long-run risk (95% CI) is 97.9%– 100%. REFERENCES 1. Stevermer JJ, Chambliss ML. Validation of decision rules for radiography in knee injuries. J Fam Prac 1996; 42:564 –565. 2. Fishwick NG, Learmonth DJ, Finlay DB. Knee effusions, radiology and acute knee trauma. Br J Radiol 1994; 67:934 –937. 3. Juhl JH, Crummy AB. Essentials of radiologic imaging. 6th ed. Philadelphia, Pa: Lippincott, 1993; 76 – 81. 4. Stiell IG, Greenberg GH, Wells GA, et al. Prospective validation of a decision rule for the use of radiography in acute knee injuries. JAMA 1996; 275:611– 615. 5. Nichol G, Stiell IG, Wells GA, Juergensen LS, Laupacis A. An economic analysis of the Ottawa Knee Rule. Ann Emerg Med 1999; 34: 438 – 447. 6. Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? JAMA 1983; 249:1743–1745. 7. Metz CE. Basic principles of ROC analysis. Semin Nucl Med 1978: 7:283–298. 8. Seaberg DC, Yealy DM, Lukens T, Auble T, Mathias S. Multicenter comparison of two clinical decision rules for the use of radiography in acute, high-risk knee injuries. Ann Emerg Med 1998; 32:8 –13. 9. Hall FM. Radiographic diagnosis and accuracy in knee joint effusions. Radiology 1975; 115:49 –54. 10. Singer AM, Naimark A, Felson D, Shapiro JH. Comparison of overhead and cross-table lateral views for detection of knee joint effusion. AJR Am J Roentgenol 1985; 144:973–975. 11. Lee JH, Weissman BN, Nikpoor N, Aliabadi P, Sosman JL. Lipohemarthrosis of the knee: a review of recent experiences. Radiology 1989; 173:189 –191. 12. McCort JJ, Mindelzun RE. Trauma radiology. New York, NY: Churchill Livingstone, 1990; 444 – 455. 13. Ballinger PW. Merrill’s atlas of radiographic positions and radiologic procedures. 8th ed. New York, NY: Mosby, 1995; 240 –249. 14. Putman CE, Ravin CE. Textbook of diagnostic imaging. Philadelphia, Pa: Saunders, 1988; 1405.
397
Rationale and Objectives. The authors performed this study to evaluate whether any one conventional radiographic view is sufficient as a screening method in the detection of acute knee trauma. Materials and Methods. The authors prospectively assessed the efficacy of a single conventional radiograph of the knee in the detection of signs of knee fractures in 214 consecutive adults with acute knee trauma. The evaluated efficacy parameters included specificity, sensitivity, negative predictive value (NPV), and positive predictive value. The percentage reduction in the need for additional conventional radiographs was also calculated. Results. Fifty-three of the 214 patients (24.8%) had a knee fracture. The sensitivity of the single lateral view in the detection of knee fractures was 100% (95% confidence interval [CI] ⫽ 94.3, 100). The lateral view of the traumatized knee was normal in 143 patients (66.8%). The probability of not having a fracture if the lateral view was normal (NPV) was also 100% (95% CI ⫽ 97.9, 100). The need for additional radiographs was reduced 67%. Conclusion. A single lateral view as a screening tool for knee fractures has a very high sensitivity and NPV. Because more than 65% of the patients had a normal lateral view in this study, there can be a considerable amount of savings in terms of radiology services for these patients. Key Word. Knee, fractures.
There are 1.3 million annual visits to U.S. emergency departments because of knee trauma, with more than $1 billion spent on radiography of the knee for these injuries; 90%–92% of all knee examinations show no fracture (1). The high cost and radiation exposure of knee imaging combined with the low yield create a need to reevaluate the current imaging method used to screen for knee injury. The traditional method of evaluating the injured knee has been a knee series, which varies greatly from institution to institution and within a given institution, depend-
Acad Radiol 2001; 8:392–397 1 From the University of Medicine and Dentistry of New Jersey-Robert Wood Johnson Medical School, One Robert Wood Johnson Place, Box 19, New Brunswick, NJ 08903-0019. Received September 6, 2000; revision requested October 25; revision received December 6; accepted January 4, 2001. Address correspondence to J.K.A.
©
AUR, 2001
392
ing on the requesting physician’s preference at a particular moment. The practice of obtaining a radiograph of the traumatized knee is universal, although according to Fishwick et al (2) up to 25% of knee radiographs obtained for acute trauma do not correlate with clinical findings. Fishwick et al concluded that radiographs obtained for acute trauma do not reliably depict all important knee injuries. Magnetic resonance (MR) imaging appears to be a superior method for imaging the soft tissues, cartilaginous joint surfaces, and bone around the knee when compared with conventional radiography of the knee (3). The prohibitive cost of MR imaging, however, does not make this technology a feasible frontline tool for screening acute knee trauma. Clinical decision rules, such as the Ottawa Knee Rules developed by Stiell and colleagues (4), are a powerful way to improve the quality, efficiency, and cost-effectiveness of health care delivery. According to the Ottawa Knee Rules, radiography is required in patients with acute
Academic Radiology, Vol 8, No 5, May 2001
knee injury only if (a) the patient is 55 years or older, (b) there is tenderness at the head of the fibula, (c) there is isolated tenderness of the patella, (d) the patient is unable to flex the knee to 90°, and (e) the patient is unable to walk four steps both at the time of injury and at evaluation. If radiography is not required (on the basis of the physician’s usual practice), a follow-up call is made in 2 weeks to ensure that there is no fracture. These rules are a practical way of reducing unnecessary radiographic examinations without reducing diagnostic accuracy. In addition, they encourage physicians to be more selective in their use of radiography in patients with acute knee injuries on the basis of a set of predetermined epidemiologic and clinical criteria. A delayed diagnosis of a missed knee fracture, however, can result in a poor clinical outcome for the patient. For physicians in the United States, malpractice damage awards due to failure to diagnose a fracture averaged $55,600 in 1995 (5). Consequently, the patients’ insistence on diagnostic imaging for their acute injuries, combined with the physicians’ fear of lawsuits for missed fractures, make the implementation of clinical decision rules virtually nonexistent in the United States. Therefore, we set out to determine an efficient and cost-effective screening method for detecting fractures in patients with acute knee trauma. Part of our goal was to decrease the need for unnecessary radiographic views of the injured knee while preserving the patients’ and physicians’ desire to radiographically rule out a fracture. The purpose of this study was to test the theory that a single conventional radiograph may suffice to screen for acute knee trauma if it has a reasonably high sensitivity and negative predictive value (NPV). The lateral view was chosen as the focus of our study because, in our experience, it most often shows the abnormality in trauma cases. We propose that, on the basis of a thorough clinical examination of the knee and a single lateral view, the patient’s knee can be categorized as (a) normal, with no further radiographs necessary; (b) abnormal, with the need to obtain additional plain radiographs for further evaluation; or (c) abnormal, with the need to perform MR imaging or, occasionally, computed tomography (CT) for further evaluation. CT may be the best method of evaluation when surgical intervention is indicated. If meniscal or ligamentous injury is suspected by the orthopedic consultant, MR imaging may be the best method of further evaluation. In case of dislocation of the knee joint where vascular injury is suspected, arteriography (or CT arte-
LATERAL VIEW FOR KNEE TRAUMA
riography or MR arteriography) would be the recommended next step in evaluation.
MATERIALS AND METHODS Conventional radiographs (anteroposterior [AP], lateral, oblique, and tunnel views) of 214 consecutive adult patients (age range, 18 –97 years; mean age ⫾ standard deviation, 42 years ⫾ 18) with knee trauma were prospectively interpreted during a period of 10 months. All radiographs were read by one of the authors (J.K.A.), a boardcertified radiologist with special expertise in thoracic radiology and extensive daily experience with conventional radiography and CT for more than 20 years, during the assigned emergency department rotation at a level I trauma center. All patients at least 18 years of age with acute knee trauma were included in the study. AP and lateral radiographs of the traumatized knee were obtained in all patients. One hundred sixty-two oblique and 74 tunnel radiographs were also obtained in a part of the study group. Each image was displayed on the view boxes and analyzed. In all cases, the lateral view was read first, followed by all other views. The reader was thus blinded to the findings in the other views. All findings were tabulated at the reading and subsequently entered into a database. Knee fractures were defined as radiographic evidence of fracture of the patella, the head and neck of the fibula, the proximal 8.0 cm of the tibia, or the distal 8.0 cm of the femur. These true cases of knee fracture served as the basis for statistical analysis. Radiographs with signs of a knee fracture, joint effusion, lipohemarthrosis, air in the joint space, loose bodies, and bone fragments were considered to be “abnormal.” Evidence of osteoarthritis, such as osteophytes, joint space narrowing, and articular cartilage calcification, were considered to be “normal” findings not associated with acute knee injury. According to our algorithm, a lateral radiograph showing any of the above abnormalities was considered “test positive,” and those patients underwent additional radiologic studies for further evaluation (eg, conventional radiography with other views, CT, or MR imaging). Patients who were found to have a fracture on these additional images were classified as having a true-positive finding, and those without any further evidence of a fracture were classified as having a false-positive finding. All normal lateral radiographs were considered to be “test negative,” and pa-
393
VERMA ET AL
tients with no further evidence of a fracture on additional radiologic studies were classified as having a true-negative finding. Patients with a normal lateral view in whom a knee fracture was diagnosed on other radiographs were classified as having a false-negative finding. On the basis of these criteria, the sensitivity, specificity, positive predictive value (PPV), and NPV with the 95% confidence interval (CI) were calculated for each conventional radiographic view (lateral, AP, tunnel, oblique, and lateral and AP combined). Because none of the fractures were missed with a single lateral radiograph, calculation of the long-run risk of missing a fracture (with a larger sample size) within a 95% CI was performed by using the method described by Hanley and Lippman-Hand (6) (see Appendix). Long-run risk can be defined as a calculation of the risk of an event occurring if the event of interest fails to arise in a limited number of subjects but may arise with further testing in a larger patient population. A calculation of the estimated savings in radiologic services was also performed. Standard receiver operating characteristic (ROC) curve analysis (Analyse-It; Analyse-It Software, Leeds, England) was performed on the lateral views to evaluate disease detectability independent of the effects of disease prevalence and decision threshold (7). A five-point confidence rating scale was used to rate the lateral view, as follows: 1 ⫽ definitely normal, 2 ⫽ probably normal, 3 ⫽ indeterminate, 4 ⫽ probably abnormal, and 5 ⫽ definitely abnormal. The area under the ROC curve (Az) was then calculated. RESULTS Two hundred fourteen adult patients were included in this study. There were 121 men (56.5%) and 93 women (43.5%) aged 18 –97 years (mean age ⫾ standard deviation, 42 years ⫾ 18; median age, 39 years). Of the 214 patients, 53 (24.8%) had knee fractures. The various types of knee fracture are given in Table 1. The mechanism of injury for the knee trauma was a motor vehicle accident in 78 patients (36.4%) and a fall in 17 (8%); in 121 patients (56.5%), the injury was caused by other trauma or unknown factors. Both lateral and AP radiographs were obtained in all patients. Six of the lateral images were horizontal cross-table views. An oblique view was obtained in 162 patients (75.7%), and a tunnel view was obtained in 74 (34.6%). The lateral view was normal in 143 of the 214 patients (66.8%), and the AP view was normal in 158 (73.8%).
394
Academic Radiology, Vol 8, No 5, May 2001
Table 1 Prevalence of Fractures in the Study Population
Location of Fracture
No. of Patients (n ⫽ 214)
Patellar Distal femoral Proximal fibular Proximal tibial Tibial plateau Proximal tibial and tibial plateau Proximal tibial and fibular Tibial plateau and fibular Femoral and patellar Patellar and fibular
13 (24.6) 7 (13.3) 4 (7.5) 4 (7.5) 9 (17.0) 4 (7.5) 4 (7.5) 6 (11.3) 1 (1.9) 1 (1.9)
Total
53
Note.—Numbers in parentheses are percentages.
The oblique view was normal in 136 of the 162 patients (84.0%), and the tunnel view was normal in 66 of the 74 patients (89.2%). Seventy-one patients (33.2%) had an abnormal lateral view, 56 (26.2%) had an abnormal AP view, 26 (16.0%) had an abnormal oblique view, and eight (10.8%) had an abnormal tunnel view. The abnormalities included patellar dislocation, patellar fracture, proximal fibular fracture, tibial plateau fracture, distal femoral fracture, femoral condyle fracture, joint effusion, lipohemarthrosis, air in the joint space, soft-tissue swelling, loose bodies, and bone fragments. In nine patients, only the lateral view showed the abnormality. Eight of these patients had a joint effusion, and one had a patellar fracture. All nine of these patients had normal AP, oblique, and tunnel views. One patient with normal findings on the lateral view had a medial joint separation on the AP and oblique views. Another patient with normal findings on the lateral view had a soft-tissue injury on the AP view. Neither of these patients had knee fractures. Results of statistical analysis of these findings indicate that the lateral view alone is very sensitive for screening for acute trauma of the knee. The sensitivity of the single lateral view in the detection of knee fractures was 100% (long-run risk with 95% CI ⫽ 94.3, 100), the specificity was 88.8% (95% CI ⫽ 83.9, 93.7), the PPV was 74.6% (95% CI ⫽ 64.5, 84.7), and the NPV was 100% (long-run risk with 95% CI ⫽ 97.9, 100). The sensitivity, specificity, NPV, and PPV for the various views are given in Table 2. For AP and lateral conventional radiographs combined, which are very commonly obtained in the detection of
LATERAL VIEW FOR KNEE TRAUMA
Academic Radiology, Vol 8, No 5, May 2001
Table 2 Sensitivity, Specificity, NPV, and PPV for the Different Views
No. of Patients*
No. of Patients with Normal Findings*
Lateral
214 (100)
143 (66.8)
AP
214 (100)
158 (73.8)
AP and lateral
214 (100)
141 (65.9)
Oblique
162 (75.7)
136 (84.0)
Tunnel
74 (34.6)
66 (89.2)
View
Sensitivity (%)†
Specificity (%)†
PPV (%)†
NPV (%)†
100 (53/53) [94.3, 100]‡ 90.6 (48/53) [82.7, 98.5] 100 (53/53) [94.3, 100]‡ 71.4 (15/21) [52.1, 90.7] 87.5 (7/8) [64.6, 100]
88.8 (143/161) [83.9, 93.7] 95.0 (153/161) [91.6, 98.4] 87.6 (141/161) [82.5, 92.7] 92.2 (130/141) [87.8, 96.6] 98.5 (65/66) [95.6, 100]
74.6 (53/71) [64.5, 84.7] 85.7 (48/56) [76.5, 94.9] 72.6 (53/73) [62.4, 82.8] 57.7 (15/26) [38.7, 76.7] 87.5 (7/8) [64.6, 100]
100 (143/143) [97.9, 100]‡ 96.8 (153/158) [94.1, 99.5] 100.0 (141/141) [97.9, 100]‡ 95.6 (130/136) [92.2, 99.0] 98.5 (65/66) [95.6, 100]
* Numbers in parentheses are percentages. † Numbers in parentheses are the numbers of patients. Numbers in brackets are the 95% CI. ‡ Long-run risk range.
None of the lateral views were classified as showing “indeterminate” or “probably normal” findings for ROC analysis. The results yielded an Az of 0.97 in the use of the lateral view to screen for any abnormality (Figure). Finally, according to our proposal, it would not be necessary to obtain additional conventional radiographic views in patients with a normal lateral view (NPV, 100%). Consequently, there would be a 66.8% reduction (66.8% of lateral views were negative in our study) in the need for additional conventional radiographic views. DISCUSSION
Graph shows the ROC curve for the detection of an abnormality on a lateral view of the knee. “No discrimination” is a reference ROC curve for a test that is unable to discriminate a population with disease from a population without disease.
knee fractures, the sensitivity was 100% (long-run risk with 95% CI ⫽ 94.3, 100), the specificity was 87.6% (95% CI ⫽ 82.5, 92.7), the PPV was 72.6% (95% CI ⫽ 62.4, 82.8), and the NPV was 100% (long-run risk with 95% CI ⫽ 97.9, 100). The sensitivity, specificity, PPV, and NPV for the single lateral view alone were not statistically different from those for the lateral and AP views combined.
The conventional radiographic knee series has traditionally been the method of choice in the initial radiologic evaluation of patients with acute knee injury. The definition of which radiographs make up a complete knee series varies greatly among institutions and among individual physicians. Patients may commonly undergo imaging with two to six views. The high cost of a conventional radiographic series containing multiple views, combined with the low yield of positive pathologic findings on these plain radiographs, creates a need to rethink the radiologic screening method for acute knee trauma. Recently, a report has been published about clinical decision rules for the use of radiography in acute knee injury (8). A review of current literature, however, suggests that no conclusive algorithm exists for choosing the conventional radiographic views that must be included in a knee series. In our study, we attempted to compare the various conventional radiographic views of the knee to
395
VERMA ET AL
determine which views are the most productive for discovering injury, specifically knee fractures. We found that the sensitivity of the lateral view for depicting knee fractures was 100%, with an NPV of 100%, in a study population with a 24.8% prevalence of knee fractures. This means that a single lateral view will depict 100% of knee fractures, and, if the lateral view is negative, the likelihood that the patient does not have a fracture (NPV) is 100%. Specificity was not an aim of this study. According to our proposal, the presence of any sign of injury on a lateral view would necessitate further evaluation with other diagnostic imaging techniques (ie, conventional radiography with other views, CT, or MR imaging) even though there may not be any clear evidence of a fracture. On the basis of these practices, this screening method has relatively low specificity for a knee fracture. A specificity of 88.8%, however, is acceptable in this study, where our goal was to show the efficacy of the lateral radiograph for depicting all the true cases of a fracture (high sensitivity) and not missing any fractures when the radiograph is normal (high NPV). In our study, knee joint effusion or lipohemarthrosis were detectable only on the lateral or cross-table lateral view and not on any of the other conventional radiographic views. The effusion may be detected on the lateral view as a widening of the opaque suprapatellar strip or as an increase in the fat pad separation at the base of the suprapatellar pouch. Hall (9) has previously shown that the accuracy of the fat pad separation sign in the diagnosis and exclusion of an effusion was 88% and 90%, respectively. Lipohemarthrosis is also a valuable finding identifiable on the lateral view (10). Lee et al (11) concluded that the horizontal cross-table lateral view is the best method for demonstrating the fat-fluid level in lipohemarthrosis of the knee. The presence of a fat-fluid level in the suprapatellar bursa is virtually pathognomonic for intraarticular bone fracture (3,12). Although the actual injury (eg, tibial plateau fracture) may be seen more definitively on an AP or oblique view, the identification of lipohemarthrosis on the lateral view is an excellent indicator of the need to image the knee further in search of specific fracture. At that point, additional conventional radiographs with multiple projections, CT scans, or MR images may be obtained as the next step. Some of the limitations of our study include the fact that only one radiologist interpreted the conventional radiographs; thus, the results are based on one reader’s in-
396
Academic Radiology, Vol 8, No 5, May 2001
terpretations. Although it is true that interobserver correlation cannot be evaluated when only one reader is used, this is the way in which extremity and most other conventional radiographs are interpreted in most clinical settings. Another limitation of the study was that there was no patient follow-up to determine the presence of a knee fracture that might have been missed during the initial evaluation. Finally, one could argue that there would be additional cost (eg, room cost, time) associated with a second trip to the radiology department for those patients who had an abnormal lateral view and would require additional views as per our algorithm. Only one-third of our 214 patients, however, had abnormal lateral views that necessitated additional radiographic views. We believe that the savings would still substantially outweigh the indirect costs described earlier. We propose that although conventional radiographs may not depict all specific knee injuries, the lateral conventional radiograph is a sensitive and cost-effective way to screen for general indicators of a serious knee injury (or fracture) such as joint effusion and lipohemarthrosis (13). Our data suggest that if the lateral view is negative for any sign of injury, including joint effusion and lipohemarthrosis, we can be 95% confident that the remainder of the conventional radiographic views are 98%–100% likely to be negative, as well. Thus, only the lateral view should be obtained during the initial screening examination. A negative lateral view would eliminate the need to obtain any other conventional radiographs because they too would probably be negative. In cases where the clinical suspicion of injury is high despite the presence of a negative lateral conventional radiograph, the knee should then be imaged further with additional radiography or a more definitive imaging method such as MR imaging. If some findings of injury do become evident on a lateral view screening radiograph, then a decision to obtain other conventional radiographic views or to perform CT or MR imaging can be entertained. During the past 90 years, more special views have been developed for skeletal radiology than for any other radiologic subspecialty (14). As long as conventional radiography and arthrography were the only available methods for imaging the knee, additional conventional radiographic views were required for more definitive evaluation of injury. Now that there are new sensitive methods of evaluation for knee injury (eg, MR imaging and CT), it is time to rethink the routine use of multiple-projection conventional radiography.
LATERAL VIEW FOR KNEE TRAUMA
Academic Radiology, Vol 8, No 5, May 2001
In conclusion, a single lateral view of the knee should be used as a screening imaging method for detecting a knee fracture. If signs of injury are absent from the lateral view, then no additional conventional radiographic views are indicated, as they are also likely to be negative for pathologic findings. If clinical suspicion persists, MR imaging would then be a better choice for definitive imaging of the knee. Resorting to MR imaging in that situation would at least eliminate the cost of a multiple-view, conventional radiographic knee series. If the lateral view reveals signs of injury, however, additional conventional radiographic views may be used to categorize the injury better. In as far as positive conventional radiographs may be relied on when making clinical decisions, a full multiple-view knee series following a positive lateral view may avoid the demand for subsequent use of the expensive MR technology. Because most conventional radiographs obtained after acute knee trauma are negative for findings of injury, the single lateral view would eliminate the overuse of multiple conventional radiographs in the imaging of the posttraumatic knee, resulting in considerable cost savings. APPENDIX Calculation of the Long-Run Risk If the event of interest fails to occur in a finite number of subjects, there is still a certain probability or long-run risk that it may occur with further testing in a larger patient population (6). In making inferences from observations with zero numerators (such as zero missed fractures with the lateral view in our study), we still must calculate a true or long-term risk between zero and some upper limit. To be 95% confident of the validity of the calculated long-run risk, the “rule of three” may be applied. The rule of three states that if none of the n patients shows the event of concern (missed fracture in all patients with fracture in this case), we can be 95% confident that the chance of this event happening is at most three in n (3/n, for n greater than 30).
Calculation The lateral view depicted all 53 fractures (100% sensitivity, n ⫽ 53). The long-run risk of missing a clinically important fracture is 3/53 (5.7%). This means that the sensitivity range with the long-run risk (95% CI) is 94.3%–100%. Similarly, the NPV was 100%, which means that there were no cases of missed fracture when the lateral radiograph was negative (no false-negative findings) in 143 patients (n ⫽ 141). The long-run risk of a false-negative result is 3/141 (2.1%). Therefore, the NPV range with the long-run risk (95% CI) is 97.9%– 100%. REFERENCES 1. Stevermer JJ, Chambliss ML. Validation of decision rules for radiography in knee injuries. J Fam Prac 1996; 42:564 –565. 2. Fishwick NG, Learmonth DJ, Finlay DB. Knee effusions, radiology and acute knee trauma. Br J Radiol 1994; 67:934 –937. 3. Juhl JH, Crummy AB. Essentials of radiologic imaging. 6th ed. Philadelphia, Pa: Lippincott, 1993; 76 – 81. 4. Stiell IG, Greenberg GH, Wells GA, et al. Prospective validation of a decision rule for the use of radiography in acute knee injuries. JAMA 1996; 275:611– 615. 5. Nichol G, Stiell IG, Wells GA, Juergensen LS, Laupacis A. An economic analysis of the Ottawa Knee Rule. Ann Emerg Med 1999; 34: 438 – 447. 6. Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? JAMA 1983; 249:1743–1745. 7. Metz CE. Basic principles of ROC analysis. Semin Nucl Med 1978: 7:283–298. 8. Seaberg DC, Yealy DM, Lukens T, Auble T, Mathias S. Multicenter comparison of two clinical decision rules for the use of radiography in acute, high-risk knee injuries. Ann Emerg Med 1998; 32:8 –13. 9. Hall FM. Radiographic diagnosis and accuracy in knee joint effusions. Radiology 1975; 115:49 –54. 10. Singer AM, Naimark A, Felson D, Shapiro JH. Comparison of overhead and cross-table lateral views for detection of knee joint effusion. AJR Am J Roentgenol 1985; 144:973–975. 11. Lee JH, Weissman BN, Nikpoor N, Aliabadi P, Sosman JL. Lipohemarthrosis of the knee: a review of recent experiences. Radiology 1989; 173:189 –191. 12. McCort JJ, Mindelzun RE. Trauma radiology. New York, NY: Churchill Livingstone, 1990; 444 – 455. 13. Ballinger PW. Merrill’s atlas of radiographic positions and radiologic procedures. 8th ed. New York, NY: Mosby, 1995; 240 –249. 14. Putman CE, Ravin CE. Textbook of diagnostic imaging. Philadelphia, Pa: Saunders, 1988; 1405.
397