Predicting malignancy in mediastinal lymph nodes by endobronchial ultrasound: A new ultrasound scoring system

Transcription

1 bs_bs_banner ORIGINAL ARTICLE Predicting malignancy in mediastinal lymph nodes by endobronchial ultrasound: A new ultrasound scoring system GERALD SCHMID-BINDERT, 1 HONGBIN JIANG, 5 GEORG KÄHLER, 1 JOACHIM SAUR, 2 THOMAS HENZLER, 3 HAO WANG, 6 SHENGXIANG REN, 7 CAICUN ZHOU 7 AND LOTHAR R. PILZ 4 1 Department of Surgery, 2 1st Department of Medicine, 3 Institute of Clinical Radiology and Nuclear Medicine, University Medical Center Mannheim, 4 Medical Faculty Mannheim of Heidelberg University, Mannheim, Germany, 5 Emergency Department, 6 Department of Surgery, and 7 Department of Medical Oncology, Cancer Institute, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai, China ABSTRACT Background and objective: Endobronchial ultrasound (EBUS) is now widely used in patients with resectable non-small-cell lung cancer to sample mediastinal lymph nodes (LN) for preoperative staging. The aim of this study was to investigate prospectively the utility of six ultrasound criteria to predict malignancy in mediastinal LN. Methods: EBUS was performed in patients with mediastinal lymphadenopathy irrespective of the underlying disease. The following criteria were expected to predict malignancy: short axis >1 cm, heterogeneous pattern, round shape, distinct margin, absence of a central hilar structure and high blood flow in a LN. A sum score prediction model for malignancy was built. If more than two criteria were present,ln was classified as high risk for malignancy. Moreover, interrater variability of two blinded investigators was evaluated. Results: Two hundred eighty-one LN in 145 patients were analysed. Forty-four percent of LN were found malignant, 10% revealed sarcoidosis, and 10% revealed tuberculosis. Interobserver agreement was very good. Positive was best for heterogeneity (73%), with a negative of more than 80%. The sum score resulted in an odds ratio of 15.5 if more than two criteria were positive (P < ). Conclusions: The assessment of ultrasound criteria during routine EBUS examinations is feasible and reproducible with very good interrater agreement. If less than three of the described criteria are present, a LN has a very low chance of being malignant. The best single criterion to predict malignancy is heterogeneity. The introduction of the sum score of ultrasound criteria could potentially increase diagnostic accuracy. Correspondence: Hongbin Jiang, Emergency Department, Shanghai Pulmonary Hospital, Tongji University School of Medicine, no. 507 Zhengmin Road, Shanghai , China. jiang-r-d@hotmail.com Received 14 December 2011; Invited to revise 26 January 2012, 17 April 2012; revised 20 March 2012, 11 May 2012; accepted 24 May 2012 (Associate Editor: David Feller-Kopman). SUMMARY AT A GLANCE EBUS-guided biopsies were performed in 281 lymph nodes of 145 patients. A new ultrasound scoring system was tested prospectively to investigate the utility of ultrasound criteria predictive of malignancy. The method presented could provide a useful additional tool for LN staging of the mediastinum. Key words: endobronchial ultrasound, mediastinal lymph node, non-small-cell lung cancer, sarcoidosis, tuberculosis. INTRODUCTION Endobronchial ultrasound (EBUS)-guided transbronchial needle aspiration (TBNA) has become a useful tool to improve the diagnostic yield for lymph node (LN) staging in lung cancer patients. 1 7 In a recent meta-analysis, it was stated that LN sampling additionally based on computed tomography or positron emission tomography (PET)-positive LN findings will result in a higher sensitivity compared with LN sampling without considering this information. 8 However, systematic sampling of multiple LN can be very timeconsuming. The broad application of EBUS in daily clinical practice opens new opportunities for ultrasound-guided LN sampling. Several studies have shown that sonographical findings seem to be associated with malignancy in LN, for example, in head and neck, breast, cervix, or oesophageal cancer In a retrospective lung cancer study, four sonographical features were found to be predictive of malignancy in mediastinal LN (round shape, distinct margin, heterogeneous echogenicity, presence of coagulation necrosis sign). 17 From analysis of several studies, the criteria most promising for prospective evaluation were: round doi: /j x

2 EBUS in mediastinal lymph nodes 1191 shape, distinct margin, heterogeneous ultrasound pattern, absence of the central hilar structure (CHS), short axis >1 cm and high blood flow in a LN (measured by colour power Doppler index (CPDI)). Hence, the aim of our study was to investigate prospectively the utility of those ultrasound criteria, which might predict malignancy in mediastinal LN. METHODS Patient selection This study was performed in a cooperative research project of Pulmonary Hospital Shanghai, Tongji University China and University Medical Center Mannheim of Heidelberg University, Germany. Between October 2010 and January 2011, all patients were included, who had mediastinal lymphadenopathy in a chest scan and who were planned for EBUS-TBNA in the Pulmonary Hospital Shanghai. The study was approved by the internal review board of Shanghai Pulmonary Hospital, Tongji University. Informed consent was obtained from all patients who participated in this study. EBUS All examinations were performed by the same three experienced investigators (HW, HJ, GS-B). Bronchoscopy and EBUS were performed in general anaesthesia through a laryngeal mask. The procedures were recorded as digital movies and collected prospectively. We used a convex probe EBUS (BF-UC260F- OL8; Olympus, Tokyo, Japan, 7.5 MHz) and the ultrasound processor EU-C2000 (Olympus). For TBNA, a 22-gauge needle was used (NA-201SX-4022; Olympus). The puncture was done according to the standard specifications for the procedure, described in previous publications. 1,9,18,19 The aspirated material was first smeared onto glass slides, afterwards air-dried and immediately stained with standard haematoxylin and eosin for immediate interpretation by a nearby cytopathologist to confirm the diagnosis. The next aspirated cores were collected in liquid formalin for additional histological evaluation by the pathologist. Two raters (HJ, GS-B) evaluated independently the blinded movies (names of patients and histology were not known to the raters) using the predefined ultrasound criteria. Ultrasound criteria LN was expected to be malignant in the presence of the following features: Round Shape defined as a ratio of two perpendicular diameters <1.5. Distinct Margin in case of a distinguished sharp white line confining the LN (Fig. 1a). Echogenicity categorized as heterogeneous by the investigators based on standard ultrasound criteria (Fig. 1b). Absence of any CHS visible as a linear structure with high echogenicity in the centre of the LN. CHS were documented and evaluated separately if there was a blood vessel in the centre of the LN (Fig. 1c,d). Short axis with a diameter >1 cm. CPDI of grade 2 or 3 (Fig. 1e). (Power Doppler mode was used to grade the blood flow in LN. The amount of blood flow was classified from 0 to 3, according to Wang et al. 12 : grade 0, no blood flow; grade 1, small amount of flow, one or two punctiforms or short rod-shapes of colour flow signal; grade 2, medium amount of flow, one main vessel or a few small vessels could be found as a long strip of a curve; grade 3, rich flow.) Diagnostic tests for sonographical factors Diagnostic tests were performed for the dichotomized variates: short axis <1 cm versus 1 cm; echogenicity heterogeneous versus homogeneous; round shape yes versus no; margin distinct yes versus no; CHS with blood vessels yes versus no; CHS without blood vessels yes versus no; CPDI (0 or 1) versus (2 or 3). Tested were sensitivity, specificity, positive and negative s, diagnostic accuracy, and diagnosis odds ratio for raters 1 and 2, and for rater consensus, respectively. Results are given with 95% confidence interval (CI) (Wald), and in case of diagnosis odds ratio, the P-value is given. 20 Statistics Interobserver variability of the two raters was calculated using standard definitions. 21,22 Sensitivity, specificity, positive and negative s, and diagnosis accuracy were calculated, and odds ratio was given with the corresponding P-value of odds ratio Consensus between raters was stated if both raters were in concordance. To improve predictability, sum scores of positive criteria of EBUS for malignant and negative LN were calculated and compared by Fisher s exact or c 2 -test (two-sided, significance level a=0.05). Receiver operating characteristic curves were used to analyse the probability of malignancy in dependence of number of positive criteria. Area under curve and 95% CI were given. 27 An exact logistic regression model was used to identify factors closely related to histological findings (LogXact 9, 2010, binary logit model, Cytel, Cambridge, MA, USA). 28 Beside exact logistic regression, SAS 9.2, 2010 (SAS Institute, Cary, NC, USA) was used for analysis. RESULTS Patient characteristics and histology results One hundred forty-five patients were recruited between October 2010 and January 2011 at Shanghai

3 1192 G Schmid-Bindert et al. (a) (b) (c) (d) (e) Figure 1 A series of ultrasound images illustrating (a) an example of lymph node classified as margin distinct, (b) an example of lymph node classified as heterogeneous, (c) an example of lymph node classified as central hilar structure (CHS) present, homogeneous and margin indistinct, (d) an example of lymph node classified as CHS with blood vessel, and (e) an example of lymph node with colour power Doppler index grade 3. Pulmonary Hospital. Median age of the study population was 58 years (15 82); 74% of patients were male, and 51% of patients were non-smokers. In the 145 patients, 281 LN were assessed by ultrasound. Most of the 281 LN were found in position 4R (120 lesions, 43%), 7 (99 lesions, 35%) and 4L (23 lesions, 8%), respectively. Forty-four percent of LN were diagnosed as malignant (adenocarcinoma, squamous cell carcinoma, non-small-cell lung cancer not otherwise specified, small-cell lung cancer and Non-Hodgkin lymphoma) (Table 1). Occurrence of sarcoidosis and tuberculosis was 10%, respectively. Preliminarily negative LN (n = 162) was confirmed by follow-up computed tomography scans up to 6 months (n = 14) or surgery (n = 148). In seven LN, diagnosis changed after surgery from negative to malignant (Fig. 2). Thus, EBUS-TBNA had a sensitivity of 0.93 in our study. Interobserver variability Substantial raw agreement was seen in CPDI with 80% and almost perfect agreement in all other factors (>88%) (Table 2). Chance-corrected agreement (k) and chance-independent agreement (F) was also calculated (not shown here). In total, agreement can be stated as given. Table 1 Histology Histology results of all sampled lymph nodes Ultrasound tests No. of lesions %of lesions Adenocarcinoma Squamous 21 7 NSCLC 21 7 NOS Non-Hodgkin lymphoma 3 1 Sarcoidosis TBC Negative ALL NOS, not otherwise specified; NSCLC, non-small-cell lung cancer; TBC, tuberculosis. Short axis as a discriminator showed a diagnosis accuracy of about 54%, indicating a high sensitivity but low specificity. The diagnosis odds ratio with (P = ) indicates some discrimination potential (Tables 3 5). For the variates round shape and CHS (with and without blood vessel), sensitivity is high (>0.95), with

4 EBUS in mediastinal lymph nodes 1193 Figure 2 Follow-up results of all lymph nodes (LN). EBUS, endobronchial ultrasound. Table 2 Raw agreement between the two raters for all ultrasound criteria Factor Raters 1 versus 2 Raw agreement a low specificity. odds ratios were found to be statistically significant (>5.588). Distinct margin showed less sensitivity, with a specificity around odds ratio was 0.50, (P = 0.023) for rater 1 and 0.74 (P = 0.33) for rater 2, reflecting a greater variability for this factor by rater 2. CPDI results revealed that there seems to be no potential for prediction of malignancy. Echogenicity demonstrated best performance of all variates with quite high sensitivity, specificity, and positive and negative s. The diagnosis odds ratio for the rater consensus is (P < ). Echogenicity is depicting good values of all diagnostic tests. Score of combined sonographical factors 95% CI (Wald) Round shape ( ) Distinct margin ( ) Echogenecity ( ) Central hilar structure with ( ) blood vessels Central hilar structure without ( ) blood vessels Colour power Doppler index ( ) CI, confidence interval. To see whether a combination of positive criteria will increase the detection of malignant LN, we investigated a model of sum scores where the number of positive criteria is counted (Table 6). If the two groups negative and malignant LN were compared directly, they were significantly different (P < , c 2 -test). If the presence of 3 6 positive variates is seen as high risk and 1 2 is seen as low risk, the odds ratio for malignancy is 15.5 with a 95% CI of , showing that the presence of more than two variates indicates malignancy of an LN. The difference between the groups determined by Fisher s exact test is highly significant with P < The classifier counts of 2 6 positive criteria summarize the predictive power of all possible probabilities, and the receiver operating characteristic curve reflects the accuracy of the diagnostic test (Fig. 3). The estimate of the odds ratio of the cumulated criteria (score) is 3.529, with a 95% CI of (Wald) and a P < The profile likelihood CI for odds ratios in steps of one unit (one additional positive criterion) is in the 95% setting. The impact of numbers of positive EBUS criteria is shown in Figure 4. Here, the predicted probabilities for malignancy in dependence of number of positive scores met are drawn in this figure exceeding about 80% (actually as seen above 79%) if all criteria are hit. Logistic regression of sonographical factors All available factors were used in the exact logit model for rater 1 and rater 2, respectively. The factor echogenicity is the only candidate of prediction of high significance with an odds ratio of (95% CI , P < ) and (95% CI , P < ) for raters 1 and 2, respectively (Table 7). DISCUSSION International guidelines recommend evaluation of mediastinal LN by mediastinoscopy or EBUS for all lung cancer patients who are candidates for surgery. 7,29,30 EBUS is rapidly becoming the preferred option because it is a minimally invasive procedure that is generally available, easy to use and with minimal additional costs. It is generally accepted that patients with PET-negative mediastinum do not have to be further evaluated before surgery. However, in a study by Herth et al., it has been shown that in 97 patients with PET-negative LN, EBUS-TBNA revealed malignancy in nine patients showing that systematic

5 1194 G Schmid-Bindert et al. Table 3 Results of diagnosis test with 95% CI (Wald) and odds ratio, Rater 1 Lymph node variate Sensitivity Specificity Rater 1 Positive Negative accuracy odds ratio P-value (odds ratio) Short axis <10 mm ( ) ( ) ( ) ( ) ( ) ( ) Echogenicity ( ) ( ) ( ) ( ) ( ) ( ) < heterogeneous versus homogeneous Round shape yes/no ( ) ( ) ( ) ( ) ( ) ( ) < Margin distinct yes/no ( ) ( ) ( ) ( ) ( ) ( ) CHS with central blood ( ) ( ) ( ) ( ) ( ) ( ) vessel yes/no CHS without central blood vessel yes/no Colour power Doppler index (0,1) versus (2,3) ( ) ( ) ( ) ( ) ( ) ( ) < ( ) ( ) ( ) ( ) ( ) ( ) Short axis was measured once, no rating. Values are given with 95% confidence interval. CHS, central hilar structure. Table 4 Results of diagnosis test with 95% CI (Wald) and odds ratio, Rater 2 Lymph node variate Sensitivity Specificity Rater 2 Positive Negative accuracy odds ratio P-value (odds ratio) Short axis <10 mm ( ) ( ) ( ) ( ) ( ) ( ) Echogenicity ( ) ( ) ( ) ( ) ( ) ( ) < heterogeneous versus homogeneous Round shape yes/no ( ) ( ) ( ) ( ) ( ) ( ) < Margin distinct yes/no ( ) ( ) ( ) ( ) ( ) ( ) CHS with central blood ( ) ( ) ( ) ( ) ( ) ( ) vessel yes/no CHS without central blood vessel yes/no Colour power Doppler index (0,1) versus (2,3) ( ) ( ) ( ) ( ) ( ) < ( ) ( ) ( ) ( ) ( ) ( ) Short axis was measured once, no rating. Values are given with 95% confidence interval. CHS, central hilar structure.

6 EBUS in mediastinal lymph nodes 1195 Table 5 Results of diagnosis test with 95% CI (Wald) and odds ratio, Consensus Rater RATER: Consensus P-value (odds ratio) odds ratio accuracy Negative Positive Specificity Sensitivity Lymph node variate Short axis <10 mm ( ) ( ) ( ) ( ) ( ) ( ) Echogenicity ( ) ( ) ( ) ( ) ( ) ( ) < heterogeneous versus homogeneous Round shape yes/no ( ) ( ) ( ) ( ) ( ) ( ) < Margin distinct yes/no ( ) ( ) ( ) ( ) ( ) ( ) CHS with central blood ( ) ( ) ( ) ( ) ( ) ( ) < vessels yes/no CHS without central ( ) ( ) ( ) ( ) ( )) < blood vessel yes/no Colour power Doppler ( ) ( ) ( ) ( ) ( ) ( ) index (0,1) versus (2,3) Short axis was measured once, no rating. Values are given with 95% confidence interval. CHS, central hilar structure. Table 6 Number of patients with accordant number of positive criteria Score (= number of positive criteria) Negative lymph nodes No. of points Malignant lymph nodes No. of points SUM Figure 3 Receiver operating characteristic (ROC) graph, showing five classifiers according to number of positive criteria. LN sampling seems to be useful even in PET-negative LN to improve accuracy in preoperative staging. 31 The aim of our study was to find decision criteria based on ultrasound to assess malignancy in LN. Several ultrasound criteria associated with malignancy of LN have been described in the literature. However, in most studies, analyses were done retrospectively, and the results were sometimes conflicting. For example, echogenicity was not consistently correlated to malignancy in all studies. 10,17 Another limitation of previous studies was that no patients with benign diseases were included. It was stated by Holty et al. that sensitivity of EBUS-TBNA depends on the prevalence of mediastinal LN metastases. 32 It is not known whether the described criteria may also be associated with benign diseases like sarcoidosis or tuberculosis raising the question whether ultrasound features could help to discriminate between malignant and benign LN.

7 1196 G Schmid-Bindert et al. Figure 4 Predicted probability of malignancy in relation to number of positive ultrasound criteria. Table 7 Probable predictable factors in sonographic measurements Exact logit model Rater 1 Rater 2 Model term Odds ratio 95% CI P-value Odds ratio 95% CI P-value ShortAxis: 10 mm versus <10 mm ( ) ( ) RoundShape2: round versus not round ( ) ( ) MarginDist2: distinct versus not distinct ( ) ( ) Echo2: heterogeneous ( ) < ( ) < versus homogeneous CHS_BV2: no versus yes ( ) ( ) CHS_NOBV2: no versus yes ( ) ( ) CPDI2: 2 or 3 versus 0 or ( ) ( ) CI, confidence interval; CPDI, colour power Doppler index; CHI, central hilar structure. Another important issue is that interobserver variability has never been reported. So far, it was unclear whether different investigators can assess these criteria reliably. The design of our study took into account the previously mentioned shortcomings, concerning prospective data, prevalence of benign disease and reliability. To obtain a representative study population, all patients who were planned for diagnostic EBUS- TBNA at Shanghai Pulmonary Hospital between October 2010 and January 2011 were included, irrespective of their underlying diagnosis. Six predefined ultrasound features were assessed systematically in all detected LN. To get information about the interobserver variability, two blinded physicians analysed all recorded EBUS videos. In total, our results demonstrated very good reliability of all EBUS criteria, except CPDI showing only moderate agreement between investigators. Ultrasound criteria obtained during routine EBUSprocedures can reliably be reproduced. Thus, our study confirmed that the assessment of these criteria is a feasible method in clinical practice. In concordance with previous studies our results show, that ultrasound criteria are mainly useful due to their negative. If two or less of the defined criteria were present, the chance of malignancy in an LN was very low (around 5%). The implication is if an LN had a size of less than 1 cm in short axis, a homogenous pattern, shape other than round, an indistinct margin and showed a CHS, then this LN is most probably benign. On the other hand, the best single criterion to predict malignancy was heterogeneity. If this feature was present, 85% of LN were malignant. Specificity was around 80%, and positive and negative predictive values for heterogeneity had a range of , respectively, showing that this parameter is the best predictive factor (P < ). Additionally, we looked at sum scores of the sonographical criteria as a prediction model for malignancy. If more than two positive variates were present, there was a 15.5 times higher probability for malignancy (P < ). Because the various factors of sonographical measurement have different s, an alternative approach would be to weight the input of each

8 EBUS in mediastinal lymph nodes 1197 factor into the scoring system. Thus, the strongest factors heterogeneity as positive and presence of hilar structure as negative predictive factors for malignancy would then have a greater impact in the scoring system. When we tested this weighted scoring model, the area under curve in the receiver operating characteristic model was only marginally increased in comparison with the dichotomized model (about 2%, not shown here). Another aspect we investigated was blood flow in the LN. It is known, that angiogenesis is linked to tumour growth and metastasis We assumed therefore that the amount of vascularization in a LN should be correlated to malignancy in mediastinal LN and CPDI was chosen as surrogate measure for vascularization. As a result, we found that reliability was only moderate in our study. This could be due to technical reasons. To measure CPDI properly, a calibration of the power Doppler mode magnitude has to be performed before each patient s EBUS examination. In our study, there have been some artefacts related to the power Doppler imaging, which may have influenced reliability. Thus, these CPDI results have to be interpreted with caution. Despite this fact, the absence of an association between CPDI and malignancy could be explained by the high prevalence of inflammatory diseases (e.g. tuberculosis), which are also accompanied by high vascularization. To our knowledge, this is the first study to examine a patient population with an approximately equal number of malign and benign lymphadenopathy. Especially, the high prevalence of sarcoidosis and tuberculosis is noteworthy because little is known about ultrasound features of LN in these diseases. Our study suggests that the evaluated criteria are applicable in such a mixed population. How can these results now be integrated into clinical practice? Should PET-negative LN without any ultrasound criteria for malignancy be excluded from biopsy? At the moment, the answer would be no. But, the introduced ultrasound scoring system could be seen as an additional tool for decision-making: for example, if an LN sample obtained by EBUS is histologically negative but has a high sum score of ultrasound criteria, then surgery should be recommended to confirm the diagnosis. If on the other hand the LN has a low risk of malignancy according to the ultrasound scoring system, it would be acceptable considering these additional information to observe the patient only in follow-up computed tomography scans. In clinical practice, EBUS is often also used to obtain tissue for confirmation of diagnosis. Here, the sum score could help to select those LN for histological examination with the highest probability of malignancy. In a subsequent randomized, controlled study, the systematic LN sampling should be tested against a selective sampling based on scoring parameters without biopsying every LN. Additionally, in such a study, the positive of EBUS could be compared with PET. In conclusion, the assessment of ultrasound criteria during routine EBUS examinations is feasible and reproducible with very good interrater agreement. Our results show that if less than three of the described criteria are present, an LN has a very low chance of being malignant. The best single criterion to predict malignancy was heterogeneity, and the best negative predictive factor was CHS. The usage of the introduced sum score can be a method to give a higher probability of malignancy: if more than two criteria are present, odds ratio for malignancy is as high as Further studies are recommended to evaluate the potential role of this scoring system. REFERENCES 1 Yasufuku K, Nakajima T, Motoori K et al. Comparison of endobronchial ultrasound, positron emission tomography, and CT for lymph node staging of lung cancer. Chest 2006; 130: Lee JE, Kim HY, Lim KY et al. Endobronchial ultrasound-guided transbronchial needle aspiration in the diagnosis of lung cancer. Lung Cancer 2010; 70(1): Herth FJ, Krasnik M, Kahn N et al. Combined endoscopicendobronchial ultrasound-guided fine-needle aspiration of mediastinal lymph nodes through a single bronchoscope in 150 patients with suspected lung cancer. Chest 2010; 138: Yasufuku K, Nakajima T, Fujiwara T et al. Role of endobronchial ultrasound-guided transbronchial needle aspiration in the management of lung cancer. Gen. Thorac. Cardiovasc. Surg. 2008; 56: Herth FJ, Eberhardt R, Vilmann P et al. Real-time endobronchial ultrasound guided transbronchial needle aspiration for sampling mediastinal lymph nodes. Thorax 2006; 61: Eloubeidi MA. Endoscopic ultrasound-guided fine-needle aspiration in the staging and diagnosis of patients with lung cancer. Semin. Thorac. Cardiovasc. Surg. 2007; 19: De Leyn P, Lardinois D, Van Schil PE et al. ESTS guidelines for preoperative lymph node staging for non-small cell lung cancer. Eur. J. Cardiothorac. Surg. [Consensus Development Conference Practice Guideline]. 2007; 32(1): Gu P, Zhao YZ, Jiang LY et al. Endobronchial ultrasound-guided transbronchial needle aspiration for staging of lung cancer: a systematic review and meta-analysis. Eur. J. Cancer [Meta- Analysis Review]. 2009; 45: Yasufuku K, Chiyo M, Sekine Y et al. Real-time endobronchial ultrasound-guided transbronchial needle aspiration of mediastinal and hilar lymph nodes. Chest 2004; 126(1): Gill KR, Ghabril MS, Jamil LH et al. Endosonographic features predictive of malignancy in mediastinal lymph nodes in patients with lung cancer. Gastrointest. Endosc. 2010; 72: Schmulewitz N, Wildi SM, Varadarajulu S et al. Accuracy of EUS criteria and primary tumor site for identification of mediastinal lymph node metastasis from non-small-cell lung cancer. Gastrointest. Endosc. 2004; 59: Wang Y, Dan HJ, Fan JH et al. Evaluation of the correlation between colour power Doppler flow imaging and vascular endothelial growth factor in breast cancer. J. Int. Med. Res. 2010; 38: Ahuja AT, Ying M. Sonographic evaluation of cervical lymph nodes. AJR Am. J. Roentgenol. 2005; 184: Bhutani MS, Hawes RH, Hoffman BJ. A comparison of the accuracy of echo features during endoscopic ultrasound (EUS) and EUS-guided fine-needle aspiration for diagnosis of malignant lymph node invasion. Gastrointest. Endosc. 1997; 45: Catalano MF, Alcocer E, Chak A et al. Evaluation of metastatic celiac axis lymph nodes in patients with esophageal carcinoma: accuracy of EUS. Gastrointest. Endosc. 1999; 50:

9 Lee N, Inoue K, Yamamoto R et al. Patterns of internal echoes in lymph nodes in the diagnosis of lung cancer metastasis. World J. Surg. 1992; 16: ; discussion Fujiwara T, Yasufuku K, Nakajima T et al. The utility of sonographic features during endobronchial ultrasound-guided transbronchial needle aspiration for lymph node staging in patients with lung cancer: a standard endobronchial ultrasound image classification system. Chest 2010; 138: Yasufuku K, Chiyo M, Koh E et al. Endobronchial ultrasound guided transbronchial needle aspiration for staging of lung cancer. Lung Cancer 2005; 50: Herth FJ, Krasnik M, Vilmann P. EBUS-TBNA for the diagnosis and staging of lung cancer. Endoscopy 2006; 38(Suppl. 1): S Glas AS, Lijmer JG, Prins MH et al. The diagnostic odds ratio: a single indicator of test performance. J. Clin. Epidemiol. 2003; 56: Meade MO, Cook RJ, Guyatt GH et al. Interobserver variation in interpreting chest radiographs for the diagnosis of acute respiratory distress syndrome. Am. J. Respir. Crit. Care Med. 2000; 161(1): Cohen J. A Coefficient of agreement for nominal scales. Educ. Psychol. Meas. 1960; 20: Altman DG, Bland JM. Diagnostic tests. 1: sensitivity and specificity. BMJ 1994; 308(6943): Altman DG, Bland JM. Diagnostic tests 2: s. BMJ 1994; 309(6947): Harper R, Reeves B. Reporting of precision of estimates for diagnostic accuracy: a review. BMJ 1999; 318(7194): Armitage P, Berry G. Statistical Methods in Medical Research, 3rd edn. Blackwell, London, 1994; 131. G Schmid-Bindert et al. 27 Zhou X, Obuchowski N, McClish D. Statistical Methods in Diagnostic Medicine, 1st edn. John Wiley & Sons, New York, 2002; Agresti A. Categorical Data Analysis. Wiley-Interscience, New York, Crino L, Weder W, van Meerbeeck J et al. Early stage and locally advanced (non-metastatic) non-small-cell lung cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and followup. Ann. Oncol. [Practice Guideline]. 2010; 21(Suppl. 5): v NCCN. NCCN clinical practice guidelines in oncology nonsmall cell lung cancer, version Herth FJ, Eberhardt R, Krasnik M et al. Endobronchial ultrasound-guided transbronchial needle aspiration of lymph nodes in the radiologically and positron emission tomographynormal mediastinum in patients with lung cancer. Chest [Clinical Trial Research Support, Non-U.S. Gov t]. 2008; 133: Holty JE, Kuschner WG, Gould MK. Accuracy of transbronchial needle aspiration for mediastinal staging of non-small cell lung cancer: a meta-analysis. Thorax [Meta-Analysis Research Support, U.S. Gov t, Non-P.H.S. Review]. 2005; 60: Folkman J. Tumor angiogenesis: therapeutic implications. N. Engl. J. Med. 1971; 285: Figg W, Folkman J. Angiogenesis: An Integrative Approach From Science to Medicine. Springer, New York, Dome B, Hendrix MJ, Paku S et al. Alternative vascularization mechanisms in cancer: pathology and therapeutic implications. Am. J. Pathol. 2007; 170(1): Herbst RS, Onn A, Sandler A. Angiogenesis and lung cancer: prognostic and therapeutic implications. J. Clin. Oncol. 2005; 23: