Differentiation of Benign From Metastatic Adrenal Masses in Patients With Renal Cell Carcinoma on Contrast-Enhanced CT

Transcription

1 Genitourinary Imaging Original Research Sasaguri et al. CT Differentiation of Adrenal Masses in Patients With RCC Genitourinary Imaging Original Research Kohei Sasaguri 1,2 Naoki Takahashi 1 Mitsuru Takeuchi 1,3 Rickey E. Carter 4 Bradley C. Leibovich 5 Akira Kawashima 1 Sasaguri K, Takahashi N, Takeuchi M, Carter RE, Leibovich BC, Kawashima A Keywords: adrenal, CT, diagnosis, metastasis, renal cell carcinoma DOI: /AJR Received February 1, 2016; accepted after revision April 22, Department of Radiology, Mayo Clinic, 200 First St SW, Rochester, MN Address correspondence to N. Takahashi (Takahashi.Naoki@mayo.edu). 2 Present address: Department of Radiology, Faculty of Medicine, Saga University, Saga, Japan. 3 Present address: Department of Radiology, Nagoya City University Graduate School of Medical Sciences and Medical School, Nagoya, Japan. 4 Department of Health Sciences Research, Mayo Clinic, Rochester, MN. 5 Department of Urology, Mayo Clinic, Rochester, MN. AJR 2016; 207: X/16/ American Roentgen Ray Society Differentiation of Benign From Metastatic Adrenal Masses in Patients With Renal Cell Carcinoma on Contrast-Enhanced CT OBJECTIVE. The purpose of this article was to determine whether imaging features on contrast-enhanced CT can differentiate benign from metastatic adrenal masses in patients with renal cell carcinoma (RCC). MATERIALS AND METHODS. Between January 2008 and January 2014, 135 patients with untreated RCC were found to have 163 adrenal masses (102 benign and 61 metastatic) on contrast-enhanced CT including the corticomedullary phase (66 benign and 42 metastatic) or nephrographic phase (56 benign and 33 metastatic) or both. Imaging features of renal and adrenal masses were recorded, including T and N staging components of renal masses, internal texture, CT attenuation values, and attenuation differences between renal and adrenal masses. Logistic regression diagnostic models to differentiate benign from metastatic adrenal mass were constructed using independently significant imaging parameters in the respective corticomedullary and nephrographic phases (corticomedullary phase model and nephrographic phase model). Diagnostic performance of the models was evaluated by ROC analysis. RESULTS. Statistically significant variables for the models were regional lymphadenopathy, perirenal or renal sinus fat invasion (corticomedullary phase model only), adrenal mass size, CT attenuation value of adrenal mass, and absolute value of attenuation difference between renal and adrenal masses. Both models had excellent diagnostic performance; the AUC and optimal sensitivity and specificity for the diagnosis of metastasis were 0.991, 100%, and 92.4%, respectively, in the corticomedullary phase model and 0.947, 81.8%, and 96.4%, respectively, in the nephrographic phase model. CONCLUSION. Differentiation between benign and metastatic adrenal masses in patients with RCC can be achieved accurately by combining multiple imaging features on contrast-enhanced CT. A drenal metastasis is reported to occur in % of patients with renal cell carcinoma (RCC) [1, 2]. On the other hand, adrenal incidentalomas, which are discovered at radiologic examinations conducted for indications other than adrenal disease, are relatively common, occurring in approximately 3 7% of the adult population [3], and most are benign. Given the frequency of adrenal metastasis in patients with RCC and the frequency of incidentaloma in the general population, it is expected that approximately half of adrenal masses in patients with RCC are metastases. Previous studies showed that the presence of adrenal abnormalities on CT or MRI had a 11 93% chance of being metastasis [4 12]. There are two established diagnostic methods to differentiate benign from malignant adrenal masses using CT. The first method relies on the presence of a large amount of intracytoplasmic lipid in the adrenal adenoma, and the generally accepted criterion is 10 HU or less on unenhanced CT [13]. Although the criterion is specific for benign adrenal mass, about 30% of benign masses are lipid poor (> 10 HU on unenhanced CT) and remain indeterminate [3]. The second method relies on faster washout of contrast material on delayed contrastenhanced CT due to the small interstitial space in the adrenal adenoma [14]. This method is generally accurate in differentiating benign from malignant adrenal masses [13, 14]. However, adrenal metastasis from hypervascular primary extraadrenal malignancies, such as RCC or hepatocellular carcinoma, may show a washout pattern similar to that of the adrenal adenoma [15]. RCC is frequently found on single-phase contrast-enhanced CT for nonurologic indications. Therefore, a diagnostic AJR:207, November

2 Sasaguri et al. method that does not rely on CT densitometry or contrast material washout characteristics of adrenal masses is clinically desired for the evaluation of patients with suspected RCC. The purpose of this study was to determine whether imaging features on contrast-enhanced CT can differentiate benign from malignant adrenal masses in patients with RCC. Materials and Methods This retrospective study was approved by the institutional review board of Mayo Clinic and was HIPAA compliant. All patients had previously consented to the use of their medical records for research purposes. Patients We performed a cross-referencing search of our medical database using the search terms renal cell carcinoma or kidney and carcinoma in a pathologic record and adrenal in pathologic reports or medical records between January 2008 and January Five hundred sixty-three patients with RCC were identified by the search. We reviewed the radiologic records of the patients and included patients who had adrenal masses 1 cm or larger and who underwent contrast-enhanced CT in the corticomedullary phase or nephrographic phase or both. If an adrenal gland had more than one mass, the largest mass was evaluated as an index lesion. As a result, 211 patients with 242 adrenal masses remained. Then, 79 adrenal masses in 76 patients were excluded, as shown in Figure 1. There was a patient with RCC in the right kidney and bilateral adrenal masses, one of which was adreno cortical carcinoma and another was adenoma by pathologic evaluation. We did not include the adrenocortical carcinoma but included the adrenal adenoma for analyses. There was another patient with a pheochromocytoma that was not included in the analyses. The data for the adrenocortical carcinoma and pheochromocytoma were shown separately. Finally, 163 adrenal masses (102 benign and 61 malignant) in 135 patients were used for the analyses. CT Protocol Before treatment for RCC, all patients underwent contrast-enhanced CT including the corticomedullary phase or nephrographic phase or both, but CT protocols were variable, and 85 of 135 (63%) CT examinations were performed at outside institutions. Therefore, adequacy of the phases (corticomedullary or nephrographic) was determined objectively. A circular ROI was placed on the renal cortex and medulla, and when the attenuation difference (i.e., attenuation of the cortex minus that of the medulla) was more than 50 HU, it was considered the corticomedullary phase; otherwise, it Search terms renal cell carcinoma OR kidney and carcinoma in a pathologic record between January 2008 and January 2014 was considered the nephrographic phase. The cutoff value of 50 HU was determined according to a previous study [16]. The ROI placement was performed by a radiologist with 9 years of experience in abdominal imaging using an Advantage Windows workstation (version 4.6, GE Healthcare). The mean slice thickness was 4.1 mm (median, 5.0 mm; range, mm). The dose and injection rate of contrast material were variable. Ninety-nine of 135 (73%) patients underwent unenhanced CT. 563 Patients identified by the cross-referencing search 211 Patients with 242 adrenal masses 1 cm with corticomedullary or nephrographic phase or both 135 Patients with 163 adrenal masses 1 cm (102 benign and 61 malignant) Search terms adrenal in pathologic reports or medical records between January 2008 and January 2014 Excluded (352 patients) CT without corticomedullary or nephrographic phase (62 patients) No CT examination (11 patients) No adrenal mass 1 cm (279 patients) Excluded (79 masses) Direct extension of RCC to adrenal gland (24 masses) 10 HU on corticomedullary or nephrographic phase (10 masses) No imaging follow-up (41 masses) Adrenal metastasis from nonrenal origin (2 masses) Adrenocortical carcinoma (1 mass)* Pheochromocytoma (1 mass) Fig. 1 Flowchart of patient selection. Asterisk denotes that adrenocortical carcinoma was excluded but patient was not excluded because patient had benign adrenal mass > 1 cm in contralateral adrenal gland. RCC = renal cell carcinoma. Image Analysis CT images were reviewed by the same radiologist and using the same workstation as mentioned in the previous paragraph. The corticomedullary and nephrographic phases were reviewed at different sessions. The reviewer was aware that the patients had both renal and adrenal masses but was blinded to the pathologic and clinical information. Renal masses were reviewed for T and N staging, including maximal mass size, presence of perirenal or renal sinus fat invasion, thrombosis in the renal vein or inferior vena cava, and regional lymphadenopathy [17]. Perirenal or renal sinus fat invasion was judged as positive when the interface between the renal mass and perirenal or sinus fat was irregular, nodular, or lobulated (Fig. 2). Regional lymphadenopathy was judged as positive when the short axis of the largest lymph node measured more than 1 cm without fatty hilum. The reviewer also recorded the internal texture (homogeneous or heterogeneous) and location (upper pole or other) of the mass. The presence of metastasis outside the adrenal gland was not evaluated. Adrenal masses were evaluated for size, location (ipsilateral or contralateral to the renal mass), laterality (bilateral or unilateral), margin (smooth margin or irregular or lobulated margin), and internal texture (homogeneous or heterogeneous). CT attenuation values (in Hounsfield units) of the renal and adrenal masses were measured by placing a circular ROI on the axial plane. If a mass had a homogeneous appearance, the largest possible ROI was placed on the mass while avoiding partial volume artifacts. If the mass had a heterogeneous appearance, two or three small ( cm 2 ) ROIs were placed on the most avid AJR:207, November 2016

3 CT Differentiation of Adrenal Masses in Patients With RCC ly enhancing parts, and the mean attenuation value was recorded (Fig. 3). Calcification, blood vessels, and cystic or necrotic areas were excluded for the ROI measurements. In addition, the attenuation value of the abdominal aorta was measured to compare the background enhancement between benign and metastatic groups. When unenhanced CT was available, the attenuation value of the adrenal mass on unenhanced CT was also recorded. Standard of Reference The standard of reference for adrenal mass was pathologic diagnosis (official pathology report), by surgery or percutaneous biopsy, or on imaging follow-up. If the mass was stable in size for at least 1 year, it was considered benign [3]. If the mass increased in size (> 2 mm/year) or showed interval regression in size after chemotherapy, it was considered metastatic [18, 19]. The size measurement of adrenal masses was performed by a separate radiologist with 14 years of experience in abdominal imaging who did not participate in the image analysis. Statistical Analysis Statistical analyses were performed using statistical software (Matlab, version 2014b, Math- Works; and SPSS version 20, IBM-SPSS). A A Univariate analyses were performed to compare patients demographics and imaging features between benign and metastatic adrenal masses. In addition to the imaging features described already, differences in CT attenuation values between renal and adrenal masses (CT attenuation value of renal mass minus that of adrenal mass), both simply subtracted values and absolute values, were calculated for analyses on the hypothesis that the attenuation difference would be smaller in metastatic adrenal masses than in nonmetastatic masses. Categoric data were compared with the Fisher exact test. Continuous data were tested using the independent t test if normal distribution was achieved; otherwise, the nonparametric Mann-Whitney U test was used. A p < 0.05 was considered statistically significant. Multivariate logistic regression analyses were performed to construct diagnostic models to differentiate benign from metastatic adrenal masses. Variables from the corticomedullary and nephrographic phases were analyzed separately; therefore, two logistic regression models were constructed (corticomedullary phase model and nephrographic phase model). The diagnostic models predicted probability of metastasis by the following equation [20]: B B Fig. 2 Two patients with renal cell carcinoma (RCC) who underwent assessment of perirenal or renal sinus fat invasion on CT. A, 73-year-old woman with RCC in left kidney. On axial contrast-enhanced CT image, perirenal and sinus fat invasion was judged as positive because of surface irregularity. Perirenal fat invasion was observed on pathologic evaluation. B, 43-year-old man with RCC in left kidney. On axial contrast-enhanced CT image, perirenal fat invasion was judged as positive because of surface nodularity, although finding was negative on pathologic diagnosis. exp (Intercept + β1 X1 + β2 X βi Xi) p = 1 + exp (Intercept + β1 X1 + β2 X βi Xi), where p is the probability of metastasis (range, 0 1), βi and Xi are coefficients calculated by the multivariate logistic regression analysis, and the observed value of variable i is selected by the following method. The modeling was performed using a crossvalidation method [20]. First, variables were entered into the model in a stepwise manner, and a model to minimize the deviance was constructed at each model flexibility level (number of variables). This means that the same number of models as the number of variables was constructed. Then, the diagnostic performance of the model at each model flexibility level was evaluated by a fivefold cross-validation method. Patients were randomly divided into five groups, and then four of the five sets were used to construct a model. The model was applied to the remaining set, and the diagnosis performance was evaluated by the AUC. The fivefold cross-validation was repeated 20 times, and the mean AUC and its standard error were calculated. The model flexibility to maximize AUC was first determined. Then, the simplest model whose AUC was within one standard error of the maximal AUC was chosen as the opti- Fig year-old man with clear cell renal cell carcinoma with left adrenal metastasis. Axial contrast-enhanced CT images were obtained. A, Left renal mass shows heterogeneous appearance, and small circular ROI is placed on avidly enhanced area. B, Adrenal mass also shows heterogeneous appearance, and small circular ROI is placed on periphery of adrenal mass avoiding central low attenuation area. AJR:207, November

4 Sasaguri et al. mal model. The final models were constructed using all datasets, and the diagnostic performance was evaluated by ROC analyses; the optimal cutoff point was determined by the Yoden index, and sensitivity, specificity, positive predictive value, and negative predictive value for the diagnosis of metastasis were calculated. To evaluate the interobserver agreement and reliability of imaging features, which consist of TABLE 1: Univariate Analyses of Imaging Features Imaging Features diagnostic models, another radiologist with 8 years of experience in abdominal radiology reviewed 40 randomly selected cases (20 cases each were selected from the corticomedullary phase and nephrographic phase cohorts). Interobserver agreement and reliability were assessed using kappa statistics for categoric data and intraclass correlation coefficients for continuous data [21]. Corticomedullary Phase Subgroup Analysis We evaluated the diagnostic performance of the models in a subgroup of adrenal masses with attenuation of more than 10 HU on unenhanced CT. Results Among the 135 patients with 163 adrenal masses (102 benign and 61 metastatic), 88 patients with 108 adrenal masses (66 benign Nephrographic Phase Benign (n = 66) Metastatic (n = 42) p Benign (n = 56) Metastatic (n = 33) p Attenuation values of abdominal aorta (HU) ± 66.3 (98 347) ± 64.9 (93 355) ± 21.6 (85 177) ± 30.7 (92 202) 0.14 Renal mass size (mm) 46.4 ± 25.3 (10 115) 82.4 ± 31.5 (36 195) < ± 34.4 (15 190) 83.9 ± 36.8 (34 195) < Perirenal or sinus fat invasion < Negative 43 (65.2) 4 (9.5) 32 (57.1) 6 (18.2) Positive 23 (34.8) 38 (90.5) 24 (42.9) 27 (81.2) Renal vein or inferior vena cava thrombosis < Negative 62 (93.9) 24 (57.1) 46 (82.1) 19 (57.6) Positive 4 (6.1) 18 (42.9) 10 (17.9) 14 (42.4) Regional lymphadenopathy < Negative 61 (92.4) 25 (59.5) 51 (91.1) 23 (69.7) Positive 5 (7.6) 17 (40.5) 5 (8.9) 10 (30.3) Location of renal mass Upper pole 15 (22.7) 12 (28.6) 14 (25.0) 15 (45.5) Others 51 (77.3) 30 (71.4) 42 (75.0) 18 (54.6) Internal texture of renal mass < Homogeneous 20 (30.3) 0 7 (12.5) 0 Heterogeneous 46 (69.7) 42 (100) 49 (87.5) 33 (100) Adrenal mass size (mm) 17.6 ± 6.1 (10 40) 35.8 ± 22.4 (10 116) < ± 8.0 (10 60) 34.2 ± 23.7 (10 112) < Adrenal mass location to renal mass Ipsilateral 35 (53.0) 22 (52.4) 27 (48.2) 21 (63.6) Contralateral 31 (47.0) 20 (47.6) 29 (51.8) 12 (36.4) Laterality of adrenal mass Unilateral 33 (50.0) 16 (38.1) 33 (58.9) 16 (48.5) Bilateral 33 (50.0) 26 (61.9) 23 (41.1) 17 (51.5) Margin of adrenal mass < Smooth 65 (98.5) 32 (76.2) 55 (98.2) 29 (87.9) Irregular or lobulated 1 (1.5) 10 (23.8) 1 (1.8) 4 (12.1) Internal texture of adrenal mass < < Homogeneous 51 (77.3) 14 (33.3) 45 (80.4) 12 (36.4) Heterogeneous 15 (22.7) 28 (66.7) 11 (19.6) 21 (63.6) Attenuation values of adrenal mass (HU) 59.1 ± 24.4 (12 123) 99.1 ± 43.0 (25 195) < ± 24.6 (17 134) 83.3 ± 19.6 (50 130) < Attenuation difference between renal and 44.9 ± 47.6 ( 55 to 164) 5.5 ± 23.9 ( 59 to 99) < ± 32.2 ( 35 to 114) 16.2 ± 21.7 ( 8 to 81) < adrenal masses (HU) a Attenuation difference (absolute value) (HU) 52.9 ± 38.4 (1 164) 16.6 ± 17.9 (2 99) < ± 27.6 (4 114) 17.2 ± 20.9 (0 81) < Note Categoric data are shown as raw numbers with percentage in parentheses, and continuous data are shown as mean ± SD (range). a CT attenuation value of renal mass minus that of adrenal mass AJR:207, November 2016

5 CT Differentiation of Adrenal Masses in Patients With RCC and 42 metastatic) and 76 patients with 89 adrenal masses (56 benign and 33 metastatic) were evaluated using contrast-enhanced CT in the corticomedullary and nephrographic phases, respectively. Note that 20 (20%) benign adrenal lesions and 14 (23%) metastases were imaged on biphasic contrast-enhanced CT that included both the corticomedullary phase and nephrographic phase. Twenty-nine of 102 adrenal masses (28%) were diagnosed as benign (17 adrenal adenoma, eight normal adrenal tissue, two nodular hyperplasia, one myelolipoma, and one fibrotic and necrotic tissue), and 39 of 61 masses (64%) were diagnosed as metastatic by pathologic evaluation. The remaining 95 masses (73 benign and 22 metastatic) were diagnosed by imaging follow-up. The mean size change of benign adrenal masses was 1 mm (range, 1 to 5 mm), with a median duration of follow-up of 40 months (range, months) and a mean growth rate of 1 mm/year (range, 0 2 mm/year). Sixteen of the 22 metastases showed an increase in size, with a mean size change of 18 mm (range, 5 55 mm), a median duration of follow-up of 10 months (range, 1 14 months), and a mean growth rate of 29 mm/year (range, 6 66 mm/year). Six of the 22 metastases showed a decrease in size after chemotherapy, with a mean size change of 16 mm (range, 26 to 8 mm), a median duration of follow-up of 4 months (range, 3 4 months). The mean patient age was 62 years (range, years) in the benign group and 61 years (range, years) in the metastatic group (p = 0.3). The ratio of men to women was 61:28 in the benign group and 32:14 in the metastatic group (p = 1). The results of univariate analyses of the imaging features are shown in Table 1. There were no significant differences in the background enhancement (attenuation values of abdominal aorta) between benign and metastatic groups in both the corticomedullary phase and nephrographic phase. Renal masses in the adrenal metastasis group were more advanced than those in the benign group; all variables related to staging of renal mass (size, perirenal or renal sinus fat invasion, renal vein or inferior vena cava thrombosis, and regional lymphadenopathy) had statistically significant differences. Metastatic adrenal masses tended to be larger, more irregular, and more heterogeneous and showed higher CT attenuation values in both the corticomedullary and nephrographic phases. CT attenuation differences between renal and adrenal masses were statistically significantly larger in the benign group than in the metastatic group in both the corticomedullary and nephrographic phases. Some imaging features had high predictive values (> 90%) for metastasis or benignity, although a limited percentage of cases met the criteria (Table 2). Table 3 shows variables in the logistic regression diagnostic models that were constructed from all datasets. Probabilities of metastasis (range, 0 1) were calculated from the models shown in Appendix 1. The models had excellent diagnostic performance: the corticomedullary phase model had an AUC of (95% CI, ), and the fivefold cross-validated AUC was 0.976; the nephrographic phase model had an AUC of (95% CI, ), and the fivefold cross-validated AUC was For the diagnosis of metastasis at the optimal cutoff point by the Yoden index, sensitivity was 100% (42/42; 95% CI, %), specificity was 92.4% (61/66; 95% CI, %), positive TABLE 2: Imaging Features With High Predictive Values for Metastasis or Benignity Imaging Features Predictive Values for Metastasis Predictive Values for Benignity predictive value was 89.4% (42/47; 95% CI, %), and negative predictive value was 100% (61/61; 95% CI, %) in the corticomedullary phase model (cutoff point, 0.29); sensitivity was 81.8% (27/33; 95% CI, %), specificity was 96.4% (54/56; 95% CI, %), positive predictive value was 93.1% (27/29; 95% CI, %), and negative predictive value was 90.0% (54/60; 95% CI, %) in the nephrographic phase model (cutoff point, 0.54). Subgroup Analysis Unenhanced CT examinations were available for 117 of 163 (72%) adrenal masses (84 benign and 33 metastatic) in 99 patients. The mean CT attenuation value was 9.3 HU (range, 12 to 52 HU) for benign adrenal masses and 34.1 HU (range, HU) for metastatic adrenal masses. The frequencies of adrenal masses with attenuation of more than 10 HU were 36.9% (31/84) in benign adrenal masses and 100% (33/33) in metastatic adrenal masses. The diagnostic models also showed excellent diagnostic performance when applied to adrenal masses with attenuation more than 10 HU on unenhanced CT, with an AUC of (95% CI, ) in the corticomedullary phase model (22 benign and 26 metastatic) and (95% CI, ) in the nephrographic phase model (17 benign and 13 metastatic). Adrenocortical Carcinoma and Pheochromocytoma A case of adrenocortical carcinoma scanned in the corticomedullary phase was diagnosed as metastasis, whereas a case of pheochromocytoma scanned in the nephrographic phase was diagnosed as benign. Positive Rate in Metastatic Adrenal Masses Positive Rate in Benign Adrenal Masses Adrenal mass > 4 cm 95.5 (21/22) [ ] 34.4 (21/61) [ ] 1.0 (1/102) [0 5.3] Attenuation values of adrenal mass in corticomedullary phase > 130 HU 100 (11/11) [ ] 26.2 (11/42) [ ] 0 (0/66) [0 4.4] Renal mass 4 cm 93.8 (45/48) [ ] 4.9 (3/61) [ ] 44.1 (45/102) [ ] Homogeneous renal mass 100 (25/25) [ ] 0 (0/61) [0 4.8] 24.5 (25/102) [ ] Attenuation of adrenal mass in corticomedullary phase 95.5 (20/21) [ ] 2.4 (1/42) [ ] 30.3 (20/66) [ ] less than 40 HU Attenuation of adrenal mass in nephrographic 100 (23/23) [ ] 0 (0/33) [0 8.7] 41.1 (23/56) [ ] phase < 50 HU Attenuation difference between renal and adrenal 97.1 (34/35) [ ] 2.4 (1/42) [ ] 51.5 (34/66) [ ] masses > 40 HU in corticomedullary phase Note Data are percentage (number/total) [95% CI]. AJR:207, November

6 Sasaguri et al. TABLE 3: Variables in the Logistic Regression Diagnostic Models to Predict Probability of Metastasis for Adrenal Mass in Patient With Renal Cell Carcinoma Interobserver Agreement The results of interobserver reliability are shown in Table 4. All of the coefficients were more than 0.8, which were considered very good interobserver agreement and reliability. Variables Estimate Standard Error Discussion Adrenal mass characterization on CT is done with attenuation values on unenhanced CT correlating the amount of intracytoplasmic fat detection and with calculating contrast enhancement washout ratios [13]. In general, the methods are highly reliable, but Choi et al. [15] reported that adrenal metastases from hypervascular primary extraadrenal malignancies, such as RCC or hepatocellular carcinoma, showed a washout pattern similar to that of adrenal adenoma. The presence of signal drop on chemical shift MRI, which is used to document the increased amount of intracellular lipids, is another useful method to differentiate adrenal adenoma from nonadenoma. However, Rodacki et al. [22] reported that three of 14 adrenal metastases from RCCs showed a signal drop on chemical shift MRI. Therefore, the diagnostic methods other than unenhanced CT might be unreliable to differentiate benign from metastatic adrenal masses in patients with RCC. In this study, we evaluated whether imaging features on contrast-enhanced CT other than contrast material washout characteristics can differentiate benign from metastatic adrenal masses in patients with RCC. On univariate analyses, there were several imaging features with high predictive values for metastasis or benignity. However, as shown in the results, the percentage of cases that met the criteria was low, ranging from 24.5% to 51.5%. We then constructed multivariate logistic regression models to differentiate benign from metastatic adrenal masses, combining imaging features obtained from contrast-enhanced CT in either the corticomedullary or nephrographic phase model. Both models had excellent diagnostic performance and also were useful when the population was limited to indeterminate adrenal masses on unenhanced CT (> 10 HU). Not surprisingly, factors related to RCC stage, including renal mass size, perirenal or sinus fat invasion, renal vein or inferior vena cava thrombosis, and lymph node metastasis, are associated with adrenal involvement from RCC [2, 23]. Our study showed that all imaging features associated with T or N staging were more advanced in the metastatic group than in the benign group on univariate analyses. Of those, regional lymph- p Odds Ratio (95% CI) Corticomedullary phase model Regional lymphadenopathy ( ) Perirenal or renal sinus fat invasion ( ) Adrenal mass size (mm) ( ) Attenuation values of adrenal mass in corticomedullary phase (HU) ( ) Attenuation difference between renal and adrenal masses in corticomedullary phase (absolute values) (HU) ( ) Intercept Nephrographic phase model Regional lymphadenopathy ( ) Adrenal mass size (mm) ( ) Attenuation values of adrenal mass in nephrographic phase (HU) < ( ) Attenuation difference between renal and adrenal masses in nephrographic phase (absolute values) (HU) ( ) Intercept TABLE 4: Interreader Agreement and Reliability of Variables for the Logistic Regression Models Variables Coefficient a (95% CI) Corticomedullary phase model Regional lymphadenopathy 0.86 (0.58 1) Perirenal or sinus fat invasion 0.90 (0.71 1) Adrenal mass size 0.90 ( ) Attenuation values of adrenal mass in corticomedullary phase 0.97 ( ) Attenuation difference between renal and adrenal masses in corticomedullary phase 0.84 ( ) (absolute value) Probability of metastasis by corticomedullary phase model 0.89 ( ) Nephrographic phase model Regional lymphadenopathy 1 (1 1) Adrenal mass size 0.85 ( ) Attenuation values of adrenal mass in nephrographic phase 0.94 ( ) Attenuation difference between renal and adrenal masses in nephrographic phase 0.95 ( ) (absolute value) Probability of metastasis by nephrographic phase model 0.97 ( ) a Data are kappa coefficient for categoric data and intraclass correlation coefficient for continuous data AJR:207, November 2016

7 CT Differentiation of Adrenal Masses in Patients With RCC adenopathy and perirenal or renal sinus fat invasion were significant variables in the corticomedullary phase model, and regional lymphadenopathy was significant in the nephrographic phase model for predicting adrenal metastasis. The size of the adrenal mass is an important variable in predicting malignancy [13, 24]. If an adrenal mass is larger than 4 cm, it is usually thought to be malignant [3, 13]. The adrenal masses larger than 4 cm in our population also had high predictive value for metastasis. On the contrary, smaller adrenal mass tended to be benign. However, there was substantial overlap between benign and metastatic adrenal masses with small size. The CT attenuation values of metastatic adrenal masses were statistically significantly higher than those of benign masses in both the corticomedullary phase and nephrographic phase. According to Choi et al. [15], the attenuation values of metastases from RCCs (n = 16) and hepatocellular carcinomas (n = 3) on 1-minute contrast-enhanced CT were significantly higher than those of adenomas. They also reported that attenuation values of greater than 140 HU included 37% (7/19) or 32% (6/19) of metastases, whereas no cases of adenomas were included. Our observation was similar to their result; attenuation values greater than 130 HU in the corticomedullary phase had high predictive values of metastasis (100% [11/11]), although only 26.2% (11/42) of metastatic adrenal masses met the criteria. We calculated the attenuation difference between renal and adrenal masses on the hypothesis that the difference would be smaller in metastases than other adrenal masses because the metastatic site usually follows the characteristics of the primary tumor [25]. As we speculated, the attenuation difference was statistically significantly smaller in metastatic adrenal masses than in benign adrenal masses in both the corticomedullary phase and nephrographic phase. In the univariate analysis, the attenuation difference was useful for the prediction of benignity when the difference was large; the attenuation difference larger than 40 HU in the corticomedullary phase had high predictive values of benignity. In addition, the absolute values of the attenuation difference were statistically significant variables in the multivariate logistic regression models of both the corticomedullary phase and nephrographic phase. On the other hand, the attenuation difference was less useful in the prediction of metastasis when the difference was small because of large overlap with benign adrenal masses. There are some limitations to this study. First, the study was retrospective in nature. CT techniques were variable (CT examinations were performed at outside institutions in 63% of cases), which resulted in differences in slice thickness, amount and injection rate of iodinated contrast material, and scanning delay. Second, not all adrenal masses were pathologically confirmed in this study. Although a standard of reference based on stability in size has been an accepted criterion [13], it might cause misclassification [26]. However, our population would be largely different from the general population if just a pathologic proof were adopted as the standard of reference. Therefore, we consider this is an inevitable limitation for studies on adrenal mass characterization. Third, although the diagnostic performance of our proposed models was excellent, this study had an exploratory design and lacked external validations. Fourth, although the corticomedullary phase model seemed to be superior to the nephrographic phase model, the superiority could not be confirmed because the overlapped population who underwent both corticomedullary phase and nephrographic phase scans was small (21% [29/135] of the total population). In addition, we did not have a reasonable explanation for why perirenal or renal sinus fat invasion was not a significant variable for the nephrographic phase model but was significant for the corticomedullary phase model. It might be related to differences in the patient populations because there were fewer patients in the nephrographic phase model compared with the corticomedullary phase model. Thus, it would be possible that the variable could become significant in the nephrographic phase model if the model were constructed with a larger population. Finally, our diagnostic models were aimed to differentiate benign from metastatic adrenal masses but were not aimed to predict the pathologic subtypes. The role of our models in the diagnosis of rare incidental malignant adrenal masses (i.e., adrenal cortical carcinoma, malignant lymphoma, or angiosarcoma) or pheochromocytoma, which is generally indicated for surgery, was not evaluated. In conclusion, differentiation between benign and metastatic adrenal masses in patients with RCC can be achieved accurately by combining multiple imaging features on contrast-enhanced CT in either the corticomedullary or nephrographic phase. References 1. Kuczyk M, Wegener G, Jonas U. The therapeutic value of adrenalectomy in case of solitary metastatic spread originating from primary renal cell cancer. Eur Urol 2005; 48: O Malley RL, Godoy G, Kanofsky JA, Taneja SS. The necessity of adrenalectomy at the time of radical nephrectomy: a systematic review. J Urol 2009; 181: Berland LL, Silverman SG, Gore RM, et al. Managing incidental findings on abdominal CT: white paper of the ACR incidental findings committee. J Am Coll Radiol 2010; 7: Sawai Y, Kinouchi T, Mano M, et al. Ipsilateral adrenal involvement from renal cell carcinoma: retrospective study of the predictive value of computed tomography. Urology 2002; 59: Antonelli A, Cozzoli A, Simeone C, et al. Surgical treatment of adrenal metastasis from renal cell carcinoma: a single-centre experience of 45 patients. BJU Int 2006; 97: Autorino R, Di Lorenzo G, Damiano R, et al. Adrenal sparing surgery in the treatment of renal cell carcinoma: when is it possible? World J Urol 2003; 21: De Sio M, Autorino R, Di Lorenzo G, et al. Adrenalectomy: defining its role in the surgical treatment of renal cell carcinoma. Urol Int 2003; 71: Kletscher BA, Qian J, Bostwick DG, Blute ML, Zincke H. Prospective analysis of the incidence of ipsilateral adrenal metastasis in localized renal cell carcinoma. J Urol 1996; 155: Leibovitch I, Raviv G, Mor Y, Nativ O, Goldwasser B. Reconsidering the necessity of ipsilateral adrenalectomy during radical nephrectomy for renal cell carcinoma. Urology 1995; 46: Tsui KH, Shvarts O, Barbaric Z, Figlin R, de Kernion JB, Belldegrun A. Is adrenalectomy a necessary component of radical nephrectomy? UCLA experience with 511 radical nephrectomies. J Urol 2000; 163: Moudouni SM, En-Nia I, Patard JJ, Manunta A, Guille F, Lobel B. Real indications for adrenalectomy in renal cell carcinoma. Scand J Urol Nephrol 2002; 36: Gill IS, McClennan BL, Kerbl K, Carbone JM, Wick M, Clayman RV. Adrenal involvement from renal cell carcinoma: predictive value of computerized tomography. J Urol 1994; 152: Boland GW, Blake MA, Hahn PF, Mayo-Smith WW. Incidental adrenal lesions: principles, techniques, and algorithms for imaging characterization. Radiology 2008; 249: Korobkin M, Brodeur FJ, Francis IR, Quint LE, Dunnick NR, Londy F. CT time-attenuation washout curves of adrenal adenomas and nonadenomas. AJR 1998; 170: Choi YA, Kim CK, Park BK, Kim B. Evaluation of AJR:207, November

8 Sasaguri et al. adrenal metastases from renal cell carcinoma and hepatocellular carcinoma: use of delayed contrastenhanced CT. Radiology 2013; 266: Szolar DH, Kammerhuber F, Altziebler S, et al. Multiphasic helical CT of the kidney: increased conspicuity for detection and characterization of small (< 3-cm) renal masses. Radiology 1997; 202: Eggener S. TNM staging for renal cell carcinoma: time for a new method. Eur Urol 2010; 58: Woo S, Cho JY, Kim SY, Kim SH. Adrenal adenoma and metastasis from clear cell renal cell carcinoma: can they be differentiated using standard MR techniques? Acta Radiol 2014; 55: Inan N, Arslan A, Akansel G, Anik Y, Balci NC, Demirci A. Dynamic contrast enhanced MRI in the APPENDIX 1: Calculation of Probability of Metastasis Using Logistic Regression Diagnostic Models The probability of metastasis by corticomedullary phase model was calculated by the following equation: p = exp ( X X X X X5) 1 + exp ( X X X X X5), where p is the probability of metastasis (range, 0 1), X1 denotes regional lymphadenopathy (absent = 0, present = 1), X2 denotes perirenal or sinus fat invasion (absent = 0, present = 1), X3 denotes adrenal size in millimeters, X4 denotes CT attenuation values (Hounsfield units) of adrenal mass in corticomedullary phase, and X5 denotes the attenuation difference between renal and adrenal masses in corticomedullary phase (Hounsfield units). The probability of metastasis by the nephrographic phase model was calculated by the following equation p = differential diagnosis of adrenal adenomas and malignant adrenal masses. Eur J Radiol 2008; 65: James G, Witten D, Hastie T, Tibshirani R. Linear model selection and regularization. In: James G, Witten D, Hastie T, Tibshirani R, eds. An introduction to statistical learning: with applications in R. New York, NY: Springer-Verlag, 2013: Gisev N, Bell JS, Chen TF. Interrater agreement and interrater reliability: key concepts, approaches, and applications. Res Social Adm Pharm 2013; 9: Rodacki K, Ramalho M, Dale BN, et al. Combined chemical shift imaging with early dynamic serial gadolinium-enhanced MRI in the characterization of adrenal lesions. AJR 2014; 203: Ito K, Nakazawa H, Marumo K, et al. Risk factors for ipsilateral adrenal involvement in renal cell exp ( X X X X4) 1 + exp ( X X X X4), carcinoma. Urology 2008; 72: Koo HJ, Choi HJ, Kim HJ, Kim SO, Cho KS. The value of 15-minute delayed contrast-enhanced CT to differentiate hyperattenuating adrenal masses compared with chemical shift MR imaging. Eur Radiol 2014; 24: Coquia SF, Johnson PT, Ahmed S, Fishmas EK. MDCT imaging following nephrectomy for renal cell carcinoma: protocol optimization and patterns of tumor recurrence. World J Radiol 2013; 5: Hamrahian AH, Ioachimescu AG, Remer EM, et al. Clinical utility of noncontrast computed tomography attenuation value (Hounsfield units) to differentiate adrenal adenomas/hyperplasias from nonadenomas: Cleveland Clinic experience. J Clin Endocrinol Metab 2005; 90: where p is the probability of metastasis (range, 0 1), X1 denotes regional lymphadenopathy (absent = 0, present = 1), X2 denotes adrenal size in millimeters, X3 denotes CT attenuation values (Hounsfield units) of adrenal mass in the nephrographic phase, and X4 denotes attenuation difference between renal and adrenal masses in nephrographic phase (Hounsfield units) AJR:207, November 2016