- Email: [email protected]
Computed tomographic staging of esophageal carcinoma: a study on interobserver variation and correlation with pathological findings
0 1992 Elsevier Science Publishers
40
EURRAD
European Journal of Radiology, 15 (1992) 40-44 B.V. All rights reserved. 0720-048X/92/$05.00
00264
Computed tomographic staging of esophageal carcinoma: a study on interobserver variation and correlation with pathological findings Reginald Goei”, Rob J.S. Lamersb, Harry A. Engelshove”,
Khiam T. Oeib
“Department of Radiology,De Wever Hospital,Heerlen. Netherlands bDepartment of Radiology. UniversityHospitalMaastricht, Universityof L&burg, Limburg, Netherlands (Received 9 October
Key words: Esophagus,
neoplasm;
1991; accepted after revision 26 November
Esophagus,
CT; Efftcacy study, CT; Diagnostic
1991)
Radiology, observer performance
Abstract Despite numerous reports on the efficacy of CT in the staging of esophageal carcinoma, no data are available on the reproducibility of the procedure. Three experienced radiologists independently reviewed the CT scans of 35 patients retrospectively, Calculation of interobserver variation was performed using the K statistic. The CT findings of each observer were subsequently correlated with the surgical and pathological findings of 17 patients. There was a large interobserver variation concerning involvement of the aorta, pulmonary vessels, vertebral column, stomach and lymph nodes, ranging from poor to excellent agreement. Agreement between observers on extension of the disease to the tracheobronchial tree, pericardium and liver was good or excellent. Agreement between the CT findings of all observers and the surgical findings for invasive growth was poor. CT pathological correlation of the three observers showed sensitivities ranging from 50 to 57x, specificities ranging from 50 to 60% and accuracies ranging from 46 to 71%. It can be concluded from this study that patients with positive CT findings for involvement of the tracheobronchial tree, the pericardium and the liver should be considered unresectable for cure. Negative findings, however, should be interpreted with caution, because involvement of other structures may still be present. Despite optimistic reports on the efficacy of CT in the pretherapy staging of esophageal neoplasms, this modality has its limitations.
Introduction Although numerous reports recommend computed tomography (CT) as a useful procedure in the preoperative staging of esophageal carcinoma [ l-71, others have questioned its accuracy [ 8,9]. The mortality rate for curative surgery of squamous carcinoma of the esophagus is 29% and represents the highest of any surgical procedure, whereas patients who undergo palliative surgery have about a 50 y0 chance of surviving the procedure [lo]. Evaluation of esophageal carcinoma prior to the selection of treatment is therefore of utmost importance. The accuracy of CT in the assessment of tumor invasion into mediastinal structures, lymphadenopathy, and distant metastasis, has been the subject of multiple Correspondence to: Reginald Goei, M.D., Department ofRadiology, De Wever Hospital, P.O. Box 4446,640l CX Heerlen, The Netherlands.
research series [l-4,6,7]. To our knowledge, however, no data are available in the literature on the consistency of observers’ findings regarding tumor staging with CT. In this study, the reproducibility of CT in staging esophageal cancer is measured by assessing the interobserver variation between three experienced radiologists. In our opinion, surgical findings are subject to personal interpretation and consequently cannot be used as the gold standard. Therefore, to get an impression of the accuracy of the observers, unlike other studies, their CT findings were correlated to the findings at pathological examination of the resected specimens. Materials and Methods Between 1984 and 1989,35 patients with endoscopitally detected esophageal carcinoma underwent pretherapy CT scanning within 1 month after presentation. Twenty-five patients had squamous cell carcinoma and
41
10 patients had adenocarcinoma. There were 32 males and 3 females ranging in age from 38 to 78 years (mean 63 years). Twenty four patients were examined with a Philips Tomoscan 350 and 11 patients with a Siemens Somatom Plus scanner. Intravenous contrast material was used routinely and the scans were made in the supine position using 6 or 9 mm collimation at contiguous intervals of 9 or 15 mm. All examinations were performed through the entire chest and the upper abdomen, including the liver. To measure interobserver variability, three radiologists were asked to review independently the films of all patients retrospectively. All observers had comparable experience of more than 5 years in the interpretation of CT scans and were not aware of surgical or pathological findings. To stage the neoplasm the CT oriented parameters described by Moss et al. [l] were used: stage 1, intraluminal mass without wall thickening; stage 2, intraluminal mass with wall thickening of more than 5 mm; stage 3, evidence of extraesophageal spread into the aorta, trachea, bronchi, pericardium, pulmonary vessels and the vertebral column; stage 4, any tumor with distant metastasis such as liver metastasis, gastric extension, enlargement of mediastinal, left gastric or celiac lymph nodes. Involvement of the aorta was assessed according to the criteria described by Picus et al. [2] : an angle of abutment between the tumor and the aorta of less than 45 ’ was considered as evidence of non invasion; when the angle varied between 45 ’ and 90’ the findings were considered indeterminate; an angle of contact of 90’ or greater was considered consistent with direct invasion. Invasion of the tumor into the trachea or bronchial tree was considered positive when compression or indentation of these structures was present. Obliteration of the fat planes between the tumor and the pericardium was considered as evidence of infiltration into the heart. Involvement of other paraesophageal structures such as the vertebral column and pulmonary vessels was considered likely when there was a close area of contact with the tumor. Enlargement of mediastinal or abdominal lymph nodes of greater than 1 cm was considered as nodal metastasis. Patients with stage 1 and 2 tumors were classified as candidates for curative resection, whereas patients having stage 3 and 4 neoplasms were considered unresectable for cure. To determine the agreement between observers, the K statistic was used. The rc value was calculated with the following formula: IC = (p0 - pc)/( 1 - PC), where p0 is the observed proportion of agreement and pc is the proportion expected by chance. Kappa can take values
from - 1 to + 1, and the rating by Fleiss [ 1 l] which characterizes the strength of agreement was used: K I 0.40 = poor, 0.40 - 0.75 = fair to good and 2 0.75 = excellent agreement beyond chance. The observers were asked to score each parameter with “present”, “absent” or “equivocal”. Only the “present” and “absent” scores were used to calculate the rc; values, whereas the “equivocal” scores were regarded as inconclusive. Based on the clinical status at presentation and the initial CT report, 17 patients underwent surgery through a thoraco-abdominal approach. All surgical specimens were sent for pathological examination. Since the pathological findings were considered as the only definitive confirmation, correlation between the CT findings of each observer and the findings at surgery was also performed using the rc statistic. The CT pathological correlation, however, was expressed in sensitivity, specificity and accuracy. Results The results of the calculations of the K; values between the pairs of observers are listed in Table 1. None of the observers were able to stage all 35 patients conclusively. Aortic invasion appeared to be the most difficult parameter to assess; in only 37 - 57% of cases were both observers of each pair able to measure the angle of abutment easily and there was a wide variation of agreement between pairs of observers. There are also wide ranges of ICvalues between the observers concerning invasion into pulmonary vessels, gastric extension and lymphadenopathy, whereas agreement in case of involvement of the vertebral column was no better than chance. Interobserver agreement regarding tracheobronchial invasion, pericardial invasion and detection of liver metastasis was good or excellent. The final agreement as to whether or not the neoplasm is resectable for cure was poor. Of 17 patients who underwent surgery, in 12, 13 and 14 cases the CT findings of the observers could be correlated with the findings at surgery (Table 2). There is poor agreement between the CT findings of all observers and the surgical findings. Table 3 shows that CT pathological correlation was possible in respectively 12,13 and 14 of 17 patients whose surgical specimens were sent for pathological examination. All observers have low sensitivities, specificities and accuracies for extraesophageal spread of the tumor into adjacent mediastinal structures. In Table 4 the CT stages of three observers in all 35
42 TABLE 1 K values of interobserver
variation in CT staging of esophageal K
carcinoma
values
Observers l-2 0.32 0.58 0.57 0.00 0.00 0.34 0.60 - 0.59 1.00 0.00 - 0.06
Aortic invasion Tracheobronchial invasion Pericardial invasion Pulmonary vessels invasion Vertebral column involvement Gastric extension Liver metastasis Mediastinal lymphadenopathy Celiac lymphadenopathy Left gastric lymphadenopathy Resectability for cure
(57%)’ (86%) (86%) (83 %) (83%) (80%) (80%) (77%) (74%) (77%) (83%)
Observers 1-3
Observers 2-3
0.00 (43%) 0.66 (83%) 0.66 (83 %) 0.00 (80%) 0.00 (80%) 0.76 (74%) 1.00 (80%) 0.38 (38%) 0.14 (80%) 0.00 (80%) 0.04 (71%)
0.82 (37%) 0.80 (77%) 0.70 (74%) 0.44 (80%) 0.00 (71%) 0.59 (80%) 1.00 (74%) 0.29 (71%) 0.21 (77%) 0.59 (74%) 0.33 (74%)
K: GO.40 = poor; 0.40-0.75 = fair/good; 20.75 = excellent. between parentheses are the proportion of patients whose or “absent”. Calculation of K was based on the following formulas: p0 = (a + d)/n, pc = (a + b)(a + c)/n’ + (b + d)(c + d)/n*, and K = (p0 In case of observers l-2: a = “present” for both observers, b = “present” 1 and “present” for observer 2, and d = “absent” for both observers. In case of resectability for cure a positive finding means resectable and
’ Percentages
Correlation
K = K =
of CT pathological
-
Figures between parentheses are the number of patients who had surgery and whose CT was scored by the respective observers with presence or absence of invasive growth.
patients and in the subgroup of 17 operated patients is shown. It can be derived from this table that there are also differences between observers in the CT staging of esophageal carcinoma in the operated group, particu-
0.24 0.06 0.05 0.15 0.00 0.12 0.13 0.31 0.28 0.20 0.11
a negative finding means unresectable.
Correlation between CT findings of three observers and surgical findings for invasive growth of esophageal carcinoma (n = 17) 0.04 - 0.09 0.32
0.38 0.68 0.66 0.15 0.00 0.56 0.87 0.03 0.45 0.20 0.10
- pc)/(l - PC). for observer 1 and “absent” for observer 2, c = “absent” for observer
TABLE 3
K =
SE
CTs were scored by the respective pairs of observers with “present”
TABLE 2
Observer 1, Surgical findings (14) Observer 2, Surgical findings (12) Observer 3, Surgical findings (13)
Mean
Observer 1 (14) Observer 2 (12) Observer 3 (13)
findings of three observers (n = 17)
Sensitivity
Specificity
Accuracy
(%)
(%)
(%)
50 57 50
50 60 60
71 50 46
Figures between parentheses are the number of patients whose surgical specimen was evaluated pathologically and whose CT was scored by the respective observers with presence or absence of invasive growth.
larly with respect to stage l-2 tumors. Furthermore, of 17 patients had stage 3-4 tumors at operation.
9
TABLE 4 CT staging of three observers
of the total group (n = 35) and the operated
Observer 1 Observer 2 Observer 3 Operated patients with pathological confirmation Figures between parentheses
represent
group (n = 17)
Stage l-2
Stage 3-4
Inconclusive
10 (10) 9 (5) 9 (5) 8
19 (4) 16 (7) 15 (5) 9
6 (3) 10 (5) 11 (4)
the operated
patients.
scores
Total 35 (17) 35 (17) 35 (17) 17
43
Discussion Although most esophageal tumors are in an advanced stage at the time of detection, numerous efforts has been made to stage the disease by means of CT. Despite encouraging reports [l-7] on the ability of CT to accurately stage esophageal carcinoma, the procedure remains controversial and critical reports exist in the surgical literature [ 8,9]. The question remains as to whether or not CT signs of extraesophageal spread of the tumor are reliable. Among the criteria for invasion of esophageal carcinoma into mediastinal structures, involvement of the aorta is the only parameter which has been quantified [ 21. Nevertheless, this study showed that in many cases all observers had difficulties in measuring the angle of contact between the tumor and the aorta. Moreover, from the wide variation of the K; values between the pairs of observers, it becomes clear that the angle of abutment is,in fact, not a reproducible sign. This observation is in agreement with those made by Inculet et al. [ 121 who found no correlation between the angle of abutment and resectability, whereas Lehr et al. [9] reported that CT has a sensitivity as low as 6% for aortic infiltration. The poor agreement concerning involvement of the vertebral column and pulmonary vessels is probably due to difficulties in assessing whether obliteration of fat planes between the tumor and these structures is caused by invasive growth or the cachectic condition of the patient. There is also poor agreement between one of three pairs of observers with regard to gastric extension of the tumor. We think this was caused by the difficulty of determining the exact location of the gastroesophageal junction on the CT cuts. Assessment of lymph node metastasis is less reliable and is supported by the study of Halvorsen et al. [ 141 who combined the data from six series and reported a sensitivity of 48 06 for mediastinal adenopathy and 6 1 y0 for abdominal adenopathy, while specificity was 91% and 94 9~) respectively. Interobserver agreement in the assessment of tracheobronchial invasion and pericardial invasion was good. This finding is in accordance with the study of Inculet et al. [ 121 who reported a high predictive value of tracheobronchial invasion of 0.83, and the study of Halvorsen et al. [ 131 who reported the sensitivity of tracheobronchial invasion and pericardial invasion to be 98 and loo%, respectively. Table 2 shows the poor agreement between the CT findings of all observers and the surgical findings. Since only 17 of 35 patients underwent surgery, the value of this finding is limited.
In our opinion the reliability of comparative studies in which the surgical findings are used definitively should be questioned. In this study all observers had low sensitivity, specificity and accuracy for assessing invasive growth of the tumor into adjacent mediastinal structures. This finding contrasts sharply with data reported by several authors [3,6,7,13] who found high accuracy rates. However, Lehr et al. [9] reported an accuracy of 50% for the detection of mediastinal invasion, which is no better than chance, and which is in agreement with the results of this study. Although it may be presumed that the prevalence of stage l-2 tumors are higher in the operated group, this is not the case in the present study. Eight of 17 operated patients had stage l-2 tumors at operation, whereas the remaining 9 had stage 3-4 neoplasms (Table 4). If it is stated that the eight patients with stage l-2 tumors are called the true positive group, the true positive rate (i.e. sensitivity) of the initial decision strategy, which was also based on the CT examination, is 47% and thus is no better than the sensitivities of the observers. It can be concluded that despite optimistic reports on the efficacy of CT in the pretherapy staging of esophageal carcinoma, this modality has its limitations. Tracheobronchial invasion, pericardial invasion and liver metastasis probably are reliable signs, and patients with positive findings with respect to these structures should be considered unresectable for cure. CT criteria of involvement of other structures, such as the aorta, pulmonary vessels, vertebral column, stomach and lymph nodes are difficult to assess and are subject to a wide interobserver variation. Therefore, negative CT findings should be interpreted with caution, because extension of the tumor into these structures may still be present. References Moss AA, Schnyder P, Thoeni RF, Margulis AR. Esophageal carcinoma: Pretherapy staging by computed tomography. AJR 1981; 136: 1051-1056. Picus D, Balfe DM, Koehler RE, Roper CL, Owen JW. Computed Tomography in the staging of esophageal carcinoma. Radiology 1983; 146: 433-438. Thompson WM, Halvorsen Jr RA, Foster Jr WL, Williford ME, Postlethwait RW, Korobkin M. Computed tomography for staging esophageal and gastroesophageal cancer: reevaluation. AJR 1983; 141: 951-958. Quint LE, Glazer GM, Orringer MB, Gross BH. Esophageal carcinoma: CT findings. Radiology 1985; 155: 171-175. Halvorsen Jr RA, Magruder-Habib K, Foster Jr WL, Roberts Jr L, Postlethwait RW, Thompson WM. Esophageal cancer staging by CT: long-term follow-up study. Radiology 1986; 161: 147-151. Takashima S, Takeuchi N, Shiozaki H, Kobayashi K, Moromoto S, Ikezoe J, Tomiyama N, Harada K, Shogen K,
44
7
8
9
10
Kozuka T. Carcinoma of the Esophagus: CT vs. MR imaging in determining resectability. AJR 1990; 156: 297-302. Rasch L, Brenoe J, Olesen KP. Predictability of esophagus- and cardiac-tumor resectability by preoperative computed tomography. Eur J Radio1 1990; 11: 42-45. Lea IV JW, Prager RL, Bender Jr HW. The questionable role of computed tomography in preoperative staging of esophageal cancer. Ann Thorac Surg 1984; 5: 479-481. Lehr L, Rupp N, Siewert JR. Assessment of resectability of esophageal cancer by computed tomography and magnetic resonance imaging. Surgery 1988; 10: 344-350. Earlam R, Cunha-Melo JR. Oesophageal squamous cell car-
11 12
13
14
cinema: I. A critical review of surgery. Brit J Surg 1980; 67:38 l-390. Fleiss JL. Statistical methods for rates and proportions. p. 218. Wiley, New York 1985. Inculet RI, Keller SM, Dwyer A. Evaluation of noninvasive tests for the preoperative staging of carcinoma of the esophagus: A prospective study. Ann Thorac Surg 1985; 40: 561-565. Halvorsen Jr,RA, Thompson WM. Computed tomographic staging of gastrointestinal tract malignancies. Part I. Esophagus and stomach. Invest Radio1 1987; 22: 2-16. Halvorsen Jr RA, Thompson WM. CT ofEsophageal neoplasms. Radio1 Clin N Am 1989; 27: 667-685.
40
EURRAD
European Journal of Radiology, 15 (1992) 40-44 B.V. All rights reserved. 0720-048X/92/$05.00
00264
Computed tomographic staging of esophageal carcinoma: a study on interobserver variation and correlation with pathological findings Reginald Goei”, Rob J.S. Lamersb, Harry A. Engelshove”,
Khiam T. Oeib
“Department of Radiology,De Wever Hospital,Heerlen. Netherlands bDepartment of Radiology. UniversityHospitalMaastricht, Universityof L&burg, Limburg, Netherlands (Received 9 October
Key words: Esophagus,
neoplasm;
1991; accepted after revision 26 November
Esophagus,
CT; Efftcacy study, CT; Diagnostic
1991)
Radiology, observer performance
Abstract Despite numerous reports on the efficacy of CT in the staging of esophageal carcinoma, no data are available on the reproducibility of the procedure. Three experienced radiologists independently reviewed the CT scans of 35 patients retrospectively, Calculation of interobserver variation was performed using the K statistic. The CT findings of each observer were subsequently correlated with the surgical and pathological findings of 17 patients. There was a large interobserver variation concerning involvement of the aorta, pulmonary vessels, vertebral column, stomach and lymph nodes, ranging from poor to excellent agreement. Agreement between observers on extension of the disease to the tracheobronchial tree, pericardium and liver was good or excellent. Agreement between the CT findings of all observers and the surgical findings for invasive growth was poor. CT pathological correlation of the three observers showed sensitivities ranging from 50 to 57x, specificities ranging from 50 to 60% and accuracies ranging from 46 to 71%. It can be concluded from this study that patients with positive CT findings for involvement of the tracheobronchial tree, the pericardium and the liver should be considered unresectable for cure. Negative findings, however, should be interpreted with caution, because involvement of other structures may still be present. Despite optimistic reports on the efficacy of CT in the pretherapy staging of esophageal neoplasms, this modality has its limitations.
Introduction Although numerous reports recommend computed tomography (CT) as a useful procedure in the preoperative staging of esophageal carcinoma [ l-71, others have questioned its accuracy [ 8,9]. The mortality rate for curative surgery of squamous carcinoma of the esophagus is 29% and represents the highest of any surgical procedure, whereas patients who undergo palliative surgery have about a 50 y0 chance of surviving the procedure [lo]. Evaluation of esophageal carcinoma prior to the selection of treatment is therefore of utmost importance. The accuracy of CT in the assessment of tumor invasion into mediastinal structures, lymphadenopathy, and distant metastasis, has been the subject of multiple Correspondence to: Reginald Goei, M.D., Department ofRadiology, De Wever Hospital, P.O. Box 4446,640l CX Heerlen, The Netherlands.
research series [l-4,6,7]. To our knowledge, however, no data are available in the literature on the consistency of observers’ findings regarding tumor staging with CT. In this study, the reproducibility of CT in staging esophageal cancer is measured by assessing the interobserver variation between three experienced radiologists. In our opinion, surgical findings are subject to personal interpretation and consequently cannot be used as the gold standard. Therefore, to get an impression of the accuracy of the observers, unlike other studies, their CT findings were correlated to the findings at pathological examination of the resected specimens. Materials and Methods Between 1984 and 1989,35 patients with endoscopitally detected esophageal carcinoma underwent pretherapy CT scanning within 1 month after presentation. Twenty-five patients had squamous cell carcinoma and
41
10 patients had adenocarcinoma. There were 32 males and 3 females ranging in age from 38 to 78 years (mean 63 years). Twenty four patients were examined with a Philips Tomoscan 350 and 11 patients with a Siemens Somatom Plus scanner. Intravenous contrast material was used routinely and the scans were made in the supine position using 6 or 9 mm collimation at contiguous intervals of 9 or 15 mm. All examinations were performed through the entire chest and the upper abdomen, including the liver. To measure interobserver variability, three radiologists were asked to review independently the films of all patients retrospectively. All observers had comparable experience of more than 5 years in the interpretation of CT scans and were not aware of surgical or pathological findings. To stage the neoplasm the CT oriented parameters described by Moss et al. [l] were used: stage 1, intraluminal mass without wall thickening; stage 2, intraluminal mass with wall thickening of more than 5 mm; stage 3, evidence of extraesophageal spread into the aorta, trachea, bronchi, pericardium, pulmonary vessels and the vertebral column; stage 4, any tumor with distant metastasis such as liver metastasis, gastric extension, enlargement of mediastinal, left gastric or celiac lymph nodes. Involvement of the aorta was assessed according to the criteria described by Picus et al. [2] : an angle of abutment between the tumor and the aorta of less than 45 ’ was considered as evidence of non invasion; when the angle varied between 45 ’ and 90’ the findings were considered indeterminate; an angle of contact of 90’ or greater was considered consistent with direct invasion. Invasion of the tumor into the trachea or bronchial tree was considered positive when compression or indentation of these structures was present. Obliteration of the fat planes between the tumor and the pericardium was considered as evidence of infiltration into the heart. Involvement of other paraesophageal structures such as the vertebral column and pulmonary vessels was considered likely when there was a close area of contact with the tumor. Enlargement of mediastinal or abdominal lymph nodes of greater than 1 cm was considered as nodal metastasis. Patients with stage 1 and 2 tumors were classified as candidates for curative resection, whereas patients having stage 3 and 4 neoplasms were considered unresectable for cure. To determine the agreement between observers, the K statistic was used. The rc value was calculated with the following formula: IC = (p0 - pc)/( 1 - PC), where p0 is the observed proportion of agreement and pc is the proportion expected by chance. Kappa can take values
from - 1 to + 1, and the rating by Fleiss [ 1 l] which characterizes the strength of agreement was used: K I 0.40 = poor, 0.40 - 0.75 = fair to good and 2 0.75 = excellent agreement beyond chance. The observers were asked to score each parameter with “present”, “absent” or “equivocal”. Only the “present” and “absent” scores were used to calculate the rc; values, whereas the “equivocal” scores were regarded as inconclusive. Based on the clinical status at presentation and the initial CT report, 17 patients underwent surgery through a thoraco-abdominal approach. All surgical specimens were sent for pathological examination. Since the pathological findings were considered as the only definitive confirmation, correlation between the CT findings of each observer and the findings at surgery was also performed using the rc statistic. The CT pathological correlation, however, was expressed in sensitivity, specificity and accuracy. Results The results of the calculations of the K; values between the pairs of observers are listed in Table 1. None of the observers were able to stage all 35 patients conclusively. Aortic invasion appeared to be the most difficult parameter to assess; in only 37 - 57% of cases were both observers of each pair able to measure the angle of abutment easily and there was a wide variation of agreement between pairs of observers. There are also wide ranges of ICvalues between the observers concerning invasion into pulmonary vessels, gastric extension and lymphadenopathy, whereas agreement in case of involvement of the vertebral column was no better than chance. Interobserver agreement regarding tracheobronchial invasion, pericardial invasion and detection of liver metastasis was good or excellent. The final agreement as to whether or not the neoplasm is resectable for cure was poor. Of 17 patients who underwent surgery, in 12, 13 and 14 cases the CT findings of the observers could be correlated with the findings at surgery (Table 2). There is poor agreement between the CT findings of all observers and the surgical findings. Table 3 shows that CT pathological correlation was possible in respectively 12,13 and 14 of 17 patients whose surgical specimens were sent for pathological examination. All observers have low sensitivities, specificities and accuracies for extraesophageal spread of the tumor into adjacent mediastinal structures. In Table 4 the CT stages of three observers in all 35
42 TABLE 1 K values of interobserver
variation in CT staging of esophageal K
carcinoma
values
Observers l-2 0.32 0.58 0.57 0.00 0.00 0.34 0.60 - 0.59 1.00 0.00 - 0.06
Aortic invasion Tracheobronchial invasion Pericardial invasion Pulmonary vessels invasion Vertebral column involvement Gastric extension Liver metastasis Mediastinal lymphadenopathy Celiac lymphadenopathy Left gastric lymphadenopathy Resectability for cure
(57%)’ (86%) (86%) (83 %) (83%) (80%) (80%) (77%) (74%) (77%) (83%)
Observers 1-3
Observers 2-3
0.00 (43%) 0.66 (83%) 0.66 (83 %) 0.00 (80%) 0.00 (80%) 0.76 (74%) 1.00 (80%) 0.38 (38%) 0.14 (80%) 0.00 (80%) 0.04 (71%)
0.82 (37%) 0.80 (77%) 0.70 (74%) 0.44 (80%) 0.00 (71%) 0.59 (80%) 1.00 (74%) 0.29 (71%) 0.21 (77%) 0.59 (74%) 0.33 (74%)
K: GO.40 = poor; 0.40-0.75 = fair/good; 20.75 = excellent. between parentheses are the proportion of patients whose or “absent”. Calculation of K was based on the following formulas: p0 = (a + d)/n, pc = (a + b)(a + c)/n’ + (b + d)(c + d)/n*, and K = (p0 In case of observers l-2: a = “present” for both observers, b = “present” 1 and “present” for observer 2, and d = “absent” for both observers. In case of resectability for cure a positive finding means resectable and
’ Percentages
Correlation
K = K =
of CT pathological
-
Figures between parentheses are the number of patients who had surgery and whose CT was scored by the respective observers with presence or absence of invasive growth.
patients and in the subgroup of 17 operated patients is shown. It can be derived from this table that there are also differences between observers in the CT staging of esophageal carcinoma in the operated group, particu-
0.24 0.06 0.05 0.15 0.00 0.12 0.13 0.31 0.28 0.20 0.11
a negative finding means unresectable.
Correlation between CT findings of three observers and surgical findings for invasive growth of esophageal carcinoma (n = 17) 0.04 - 0.09 0.32
0.38 0.68 0.66 0.15 0.00 0.56 0.87 0.03 0.45 0.20 0.10
- pc)/(l - PC). for observer 1 and “absent” for observer 2, c = “absent” for observer
TABLE 3
K =
SE
CTs were scored by the respective pairs of observers with “present”
TABLE 2
Observer 1, Surgical findings (14) Observer 2, Surgical findings (12) Observer 3, Surgical findings (13)
Mean
Observer 1 (14) Observer 2 (12) Observer 3 (13)
findings of three observers (n = 17)
Sensitivity
Specificity
Accuracy
(%)
(%)
(%)
50 57 50
50 60 60
71 50 46
Figures between parentheses are the number of patients whose surgical specimen was evaluated pathologically and whose CT was scored by the respective observers with presence or absence of invasive growth.
larly with respect to stage l-2 tumors. Furthermore, of 17 patients had stage 3-4 tumors at operation.
9
TABLE 4 CT staging of three observers
of the total group (n = 35) and the operated
Observer 1 Observer 2 Observer 3 Operated patients with pathological confirmation Figures between parentheses
represent
group (n = 17)
Stage l-2
Stage 3-4
Inconclusive
10 (10) 9 (5) 9 (5) 8
19 (4) 16 (7) 15 (5) 9
6 (3) 10 (5) 11 (4)
the operated
patients.
scores
Total 35 (17) 35 (17) 35 (17) 17
43
Discussion Although most esophageal tumors are in an advanced stage at the time of detection, numerous efforts has been made to stage the disease by means of CT. Despite encouraging reports [l-7] on the ability of CT to accurately stage esophageal carcinoma, the procedure remains controversial and critical reports exist in the surgical literature [ 8,9]. The question remains as to whether or not CT signs of extraesophageal spread of the tumor are reliable. Among the criteria for invasion of esophageal carcinoma into mediastinal structures, involvement of the aorta is the only parameter which has been quantified [ 21. Nevertheless, this study showed that in many cases all observers had difficulties in measuring the angle of contact between the tumor and the aorta. Moreover, from the wide variation of the K; values between the pairs of observers, it becomes clear that the angle of abutment is,in fact, not a reproducible sign. This observation is in agreement with those made by Inculet et al. [ 121 who found no correlation between the angle of abutment and resectability, whereas Lehr et al. [9] reported that CT has a sensitivity as low as 6% for aortic infiltration. The poor agreement concerning involvement of the vertebral column and pulmonary vessels is probably due to difficulties in assessing whether obliteration of fat planes between the tumor and these structures is caused by invasive growth or the cachectic condition of the patient. There is also poor agreement between one of three pairs of observers with regard to gastric extension of the tumor. We think this was caused by the difficulty of determining the exact location of the gastroesophageal junction on the CT cuts. Assessment of lymph node metastasis is less reliable and is supported by the study of Halvorsen et al. [ 141 who combined the data from six series and reported a sensitivity of 48 06 for mediastinal adenopathy and 6 1 y0 for abdominal adenopathy, while specificity was 91% and 94 9~) respectively. Interobserver agreement in the assessment of tracheobronchial invasion and pericardial invasion was good. This finding is in accordance with the study of Inculet et al. [ 121 who reported a high predictive value of tracheobronchial invasion of 0.83, and the study of Halvorsen et al. [ 131 who reported the sensitivity of tracheobronchial invasion and pericardial invasion to be 98 and loo%, respectively. Table 2 shows the poor agreement between the CT findings of all observers and the surgical findings. Since only 17 of 35 patients underwent surgery, the value of this finding is limited.
In our opinion the reliability of comparative studies in which the surgical findings are used definitively should be questioned. In this study all observers had low sensitivity, specificity and accuracy for assessing invasive growth of the tumor into adjacent mediastinal structures. This finding contrasts sharply with data reported by several authors [3,6,7,13] who found high accuracy rates. However, Lehr et al. [9] reported an accuracy of 50% for the detection of mediastinal invasion, which is no better than chance, and which is in agreement with the results of this study. Although it may be presumed that the prevalence of stage l-2 tumors are higher in the operated group, this is not the case in the present study. Eight of 17 operated patients had stage l-2 tumors at operation, whereas the remaining 9 had stage 3-4 neoplasms (Table 4). If it is stated that the eight patients with stage l-2 tumors are called the true positive group, the true positive rate (i.e. sensitivity) of the initial decision strategy, which was also based on the CT examination, is 47% and thus is no better than the sensitivities of the observers. It can be concluded that despite optimistic reports on the efficacy of CT in the pretherapy staging of esophageal carcinoma, this modality has its limitations. Tracheobronchial invasion, pericardial invasion and liver metastasis probably are reliable signs, and patients with positive findings with respect to these structures should be considered unresectable for cure. CT criteria of involvement of other structures, such as the aorta, pulmonary vessels, vertebral column, stomach and lymph nodes are difficult to assess and are subject to a wide interobserver variation. Therefore, negative CT findings should be interpreted with caution, because extension of the tumor into these structures may still be present. References Moss AA, Schnyder P, Thoeni RF, Margulis AR. Esophageal carcinoma: Pretherapy staging by computed tomography. AJR 1981; 136: 1051-1056. Picus D, Balfe DM, Koehler RE, Roper CL, Owen JW. Computed Tomography in the staging of esophageal carcinoma. Radiology 1983; 146: 433-438. Thompson WM, Halvorsen Jr RA, Foster Jr WL, Williford ME, Postlethwait RW, Korobkin M. Computed tomography for staging esophageal and gastroesophageal cancer: reevaluation. AJR 1983; 141: 951-958. Quint LE, Glazer GM, Orringer MB, Gross BH. Esophageal carcinoma: CT findings. Radiology 1985; 155: 171-175. Halvorsen Jr RA, Magruder-Habib K, Foster Jr WL, Roberts Jr L, Postlethwait RW, Thompson WM. Esophageal cancer staging by CT: long-term follow-up study. Radiology 1986; 161: 147-151. Takashima S, Takeuchi N, Shiozaki H, Kobayashi K, Moromoto S, Ikezoe J, Tomiyama N, Harada K, Shogen K,
44
7
8
9
10
Kozuka T. Carcinoma of the Esophagus: CT vs. MR imaging in determining resectability. AJR 1990; 156: 297-302. Rasch L, Brenoe J, Olesen KP. Predictability of esophagus- and cardiac-tumor resectability by preoperative computed tomography. Eur J Radio1 1990; 11: 42-45. Lea IV JW, Prager RL, Bender Jr HW. The questionable role of computed tomography in preoperative staging of esophageal cancer. Ann Thorac Surg 1984; 5: 479-481. Lehr L, Rupp N, Siewert JR. Assessment of resectability of esophageal cancer by computed tomography and magnetic resonance imaging. Surgery 1988; 10: 344-350. Earlam R, Cunha-Melo JR. Oesophageal squamous cell car-
11 12
13
14
cinema: I. A critical review of surgery. Brit J Surg 1980; 67:38 l-390. Fleiss JL. Statistical methods for rates and proportions. p. 218. Wiley, New York 1985. Inculet RI, Keller SM, Dwyer A. Evaluation of noninvasive tests for the preoperative staging of carcinoma of the esophagus: A prospective study. Ann Thorac Surg 1985; 40: 561-565. Halvorsen Jr,RA, Thompson WM. Computed tomographic staging of gastrointestinal tract malignancies. Part I. Esophagus and stomach. Invest Radio1 1987; 22: 2-16. Halvorsen Jr RA, Thompson WM. CT ofEsophageal neoplasms. Radio1 Clin N Am 1989; 27: 667-685.