Pathologic N0 Status in Clinical T1N0M0 Lung Adenocarcinoma is Predictable by the Solid Component Proportion with Quantitative CT Number Analysis

Correctly predicting pathologic regional node-negative (pN0) disease in patients with lung cancer before operation may avoid unnecessary mediastinal lymph node dissection (MLND). In this study, we analyze the value of the radiographic and histopathological features of primary tumors for predicting pN0 status in cT1N0M0 lung adenocarcinoma and to establish an optimal surgical strategy for avoiding MLND in cT1N0M0 lung adenocarcinoma patients. We retrospectively investigated the histopathological and radiographic data of 348 surgically resected cT1N0M0 lung adenocarcinoma patients with systematic lymph node dissection from January 2005 to December 2012. Histopathological features and radiographic features were analyzed. Multivariable analysis was used to identify significant predictors of pN0 disease. Our results showed that pN0 disease was detected in 306 patients (87.9%) among the 348 patients with cT1N0M0 lung adenocarcinoma. A decreasing trend of the pN0 disease proportion was observed with both increasing histological grade and decreased differentiation (P < 0.001). In multivariable analysis, the solid component proportion was a significant predictor of pN0 disease. Among 110 patients with a solid component proportion of no more than 21.3%, mediastinal lymph node involvement was not observed. Patients who meet this criterion may be successfully managed with lung resection without MLND.

Recent advancements in the quality of high-resolution CT (HRCT) and the widespread practice of low-dose CT (LDCT) for lung cancer screening have resulted in a greater increase in the early detection of smaller-sized NSCLC, especially lung adenocarcinoma appearing as subsolid nodule 11,12 . As a result, limited resection like wedge resection or segmenectomy is more favored and the necessity of standard lobectomy begin to be questioned for small or subsolid types of adenocarcinoma 2 . Meanwhile, the degree to which MLND is necessary also becomes a challenging issue because the biological behavior of small lung adenocarcinomas showing as subsolid nodules on CT is less aggressive and usually pN 0 diseases. These patients are good candidates for the omission of MLND. Therefore, it is very important to establish reliable prediction criteria for pN 0 disease preoperatively especially using quantitative radiographic imaging to avoid performing MLND.
In this study, we attempt to analyze the value of radiographic and histopathological features of primary tumors for predicting pN 0 disease in cT 1 N 0 M 0 lung adenocarcinoma to establish an optimal surgical strategy for avoiding unnecessary MLND in these patients. We analyzed both HRCT and PET/CT indexes, and we also proposed three-dimensional computer quantified CT number analysis to calculate the solid component (Sc) proportion and ground glass opacity component (GGOc) proportion.
Logistic regression analysis demonstrated that the Sc proportion was only significant predictor of pN 0 disease (β = 6.153, P = 0.001; OR = 470.085, 95% CI: 12.291~17978.987). With the increase of Sc proportion, the chance of pN 0 disease is reduced (Fig. 3). The positive predictive value (PPV) of pN 0 disease was 100% for patients with a tumor Sc proportion of no more than 21.3%. Of the 348 patients, only 137 underwent PET/CT, and thus SUV max was removed from the multivariate analysis.

Discussion
A randomized, prospective clinical trial has suggested that MLND can be avoided if there is sufficient evidence of pN 0 disease 13 . Although multiple studies have attempted to identify clinical, pathological and radiographic factors of lymph node status in early-stage adenocarcinoma 6,7,14-18 , our study differs from these previous studies in several ways. First, this study made cT 1 N 0 M 0 lung adenocarcinoma a separate object of study. Second, all resected specimens were reviewed pathologically according to the 2011 IASLC/ATS/ERS classification, and a tumor-grading system was established based on histopathological subtypes according to the newly proposed classification 19 . Third and most important, quantitative radiographic analysis method was introduced, the Sc and GGOc proportion calculated using three-dimensional computer-quantified CT number analysis was adopted to qualitatively reflect the tumor internal appearance, resulting in more objective and convenient selection criteria for clinical application.
Our study results included three main findings: (a) the Sc proportion was an independent predictor of pN 0 disease in cT 1 N 0 M 0 lung adenocarcinoma; (b) all cT 1 N 0 M 0 lung adenocarcinoma patients with an Sc proportion of no more than 21.3% had no lymph node metastasis; and (c) the pN 0 disease proportion exhibited a decreasing trend with both increasing histological grade and differentiation grade. Our results may have clinical implications. With the aid of preoperative quantitative radiographic features, we differentiated cT 1 N 0 M 0 lung adenocarcinoma patients with or without pN 0 disease, thus reducing unnecessary MLND.
In our study, tumors were graded according to four classifications based on histopathological subtypes and prognosis. Sica et al. 20 first proposed a 3-tier grading system for lung adenocarcinoma in 2010: Grade 1, a pattern with low metastatic potential (bronchioloalveolar carcinoma, BAC); Grade 2, patterns with intermediate metastatic potential (acinar and papillary); and Grade 3, patterns with high metastatic potential (solid and micropapillary). However, in 2011, the newly proposed lung adenocarcinoma classification discontinued the use of "BAC", and other changes in uniform terminology, such as adenocarcinoma in situ (AIS) and minimally invasive adenocarcinoma (MIA), and diagnostic criteria were implemented 19 . Studies 21,22 have shown that this new classification system is a stage-independent predictor of survival. Five-year disease-free survival is 100% or near 100% in patients with AIS and MIA, more than 85% in lepidic predominant invasive adenocarcinoma, 65% in acinar and papillary predominant invasive adenocarcinoma, and 40% in solid and micropapillary predominant invasive adenocarcinoma. The tumor-grading system in this study is based on a combination of the classification of Sica et al. and the lung adenocarcinoma classification system proposed in 2011. A higher grade represents a worse prognosis.
A decreasing trend in the pN 0 disease proportion with increasing tumor grade and decreasing tumor differentiation was observed in this study, consistent with the worse prognosis of lung adenocarcinoma patients with higher tumor grade and poorer tumor differentiation. In this study, all cT 1 N 0 M 0 lung adenocarcinoma patients with grade 1 tumors (AIS or MIA) had pN 0 disease, whereas only 97.7% of patients with well-differentiated tumors had pN 0 disease. Therefore, we believe that tumor grade has stronger predictive ability for lymph node status than tumor differentiation and that tumor grade 1 is a significant predictor of pN 0 disease. However, because of histological heterogeneity, AIS and MIA can only be diagnosed in resected specimens and not in small biopsies or cytology specimens 23 . Consequently, histopathological subtypes have limited significance in predicting node metastasis preoperatively and determining the most appropriate surgical procedures.
In univariate analysis of radiographic features, tumor size, size stratification, location, contour, appearance, Sc proportion, GGOc proportion and SUV max were associated with pN 0 disease. In this study, all 21 patients with a tumor size of no more than 1 cm had pN 0 disease, in accordance with previous study 14 . We also found that compared with patients with node-positive disease, primary tumors in patients with pN 0 disease are more likely to present a smooth margin and peripheral location. Smooth margin and peripheral location were frequently proven to be benign signs in lung nodule diagnosis. Tumor appearance has been proven to be a significant predictor of lymph node status. Researchers have found that most patients with a dominant GGO component on CT have no lymph node metastasis 15 . In this study, all 52 patients with pGGO had pN 0 disease, consistent with previous studies 6,16 . However, tumor appearance is only used as a subjective and visual characteristic in the past study, and thus is not sufficiently objective and accurate for prediction of pN 0 disease. The great progresses of computer make quantitative imaging increasingly applying in modern radiology practice, assisting in the clinical assessment and providing a source of biomarkers for a spectrum of diseases 24 . In the current study, the Sc proportion and GGOc proportion calculated by three-dimensional computer-quantified CT number analysis were adopted to qualitatively reflect tumor appearance innovatively, resulting in more objective and accurate selection criteria     for pN 0 disease. The Sc proportion and GGOc proportion were all associated with lymph node status. Primary tumors in patients with pN 0 disease had lower Sc proportions and higher GGOc proportions than patients with node-positive disease. Studies 7,17,18 have demonstrated that the solid component is a powerful predictor of lymph node metastasis, consistent with our study. SUV max is a semiquantitative index of FDG uptake of PET/CT, which is associated with glucose metabolism in tissue. SUV max was significantly lower in patients with pN 0 disease than patients with node-positive disease in this study, consistent with previous studies 7, [25][26][27] . We analyzed all factors except SUV max by logistic multivariate analysis to assess the joint effects and interactions of the variables on pN 0 disease. Our results showed that only the Sc proportion retained statistical significance, whereas tumor size, size stratification, location, contour, appearance and GGOc proportion did not differ significantly in multivariate analysis. We believe that the effects of these factors were substituted by the enrolled index. When Sc proportion ≤21.3% was used as the criterion, the PPV for pN 0 disease was 100%. On the basis of our result, patients of cT 1 N 0 M 0 lung adenocarcinoma who meet this reliable prediction criterion may omit MLND when surgical treatment.
Our study has several limitations. First, this is a single study; further studies at multiple centers are needed to confirm our results, especially the predictive value of Sc. Second, this is a retrospective study, and a prospective study is needed to verify this conclusion. Finally, we did not perform survival analysis because the follow-up time was too short in some cases; advanced study is warranted to analyze patient survival.
In conclusion, we respectively reviewed 348 cT 1 N 0 M 0 lung adenocarcinoma patients who underwent surgical pulmonary resection with systematic MLND at our hospital in an attempt to investigate independent predictors of pN 0 disease. The Sc proportion was identified as statistically significant predictive risk factor. The PPV of pN 0 disease was 100% for patients with a tumor Sc proportion of no more than 21.3%. Patients who meet this criterion may be managed with lung resection without MLND.

Materials and Methods
Patients. This prospective study was approved by the institutional review board at the Cancer Hospital of the Chinese Academy of Medical Sciences, and the informed consent was waived due to the retrospective nature of the study and the data were going to be analyzed anonymously. This study was conducted in accordance with the Declaration of Helsinki.
A total of 348 consecutive patients with surgically resected cT 1 N 0 M 0 lung adenocarcinoma at our hospital between January 2005 and December 2012 were included. The inclusion criteria for our study were as follows: (a) single adenocarcinomas (3 cm or less in diameter at HRCT) with no evidence of malignant satellite nodules (as proven by prior imaging study or lung biopsy); (b) no hilar or mediastinal lymphadenopathy on imaging study or at mediastinoscopy; a short lymph node diameter greater than 1 cm on CT axial image with window level in 380 HU and window within 40 HU or maximum SUV (SUV max ) on PET/CT image more than 2.5 was determined as lymphadenopathy; (c) first treatment with surgical pulmonary resection plus systematic MLND (more than 6 hilar or mediastinal lymph nodes resected); (d) either integrated 18   HRCT images were assessed retrospectively for morphological features and three-dimensional computer-quantified CT number analysis. Two radiologists (M.L. and L.Z., 11 and 8 years of experience, respectively), who were informed that the involved patients had surgically treated adenocarcinoma but were blinded to histological subtypes, analyzed the morphological features on Carestream GCRIS 2.1 PACS workstation (Carestream Health) in consensus. The morphological features included tumor size (the longest tumor diameter on the transverse lung window image in which the largest nodule dimension appeared), appearance (solid nodule [SN], part-solid nodule [PSN], or pure ground-glass-opacity [pGGO]) 28 , location (center or periphery, periphery was defined as within 3 cm of the pleura) 29 , and contour (smooth, lobular or spiculated), necrosis (necrosis of tumor was defined as low attenuation in the tumor). Three-dimensional computer-quantified CT number analysis was performed on an ADW 4.6 workstation (GE Medical Systems) to measure the proportion of GGOc and Sc. This measurement was performed in two steps. For the first step, the entire tumor mass was separated from surrounding anatomic structures using computer-aided volume measurement software (Auto Contour in Volume Rendering, GE Medical Systems). The computer generated tumor boundaries were then visually inspected by a radiologist (L.Z.) for accuracy and consistency. If any segmentation results were considered suboptimal, the tumor contours that were superimposed on the original images were edited by the same radiologist (L.Z.), and then the mean CT value of tumor was automatically calculated by computer. Threshold values of GGOc and Sc were obtained by receiver operating characteristic (ROC) curves. As the second step, the entire tumor mass was separated using the same method in the first step, and the computer automatically generated the CT number histogram of the tumor. The proportion of GGOc/Sc was calculated based on the threshold values obtained in the first step. The calculation methods were also described in detail in our previous report 30 .
Two chest radiologists with PET/CT diagnostic experience prospectively evaluated the integrated PET/CT images. For semiquantitative analysis of 18 F-FDG uptakes, a region of interest (ROI) was placed over the tumor site on the hottest trans-axial slice. In some patients, nodular FDG uptake could not be identified on the PET component images of the PET/CT study. In those patients, a ROI was drawn in a presumed nodular location, taking into account the CT component images of PET/CT. The SUV max within the ROI was used as the reference measurement.
Pathologic Evaluation. All resected specimens were reviewed pathologically by an experienced lung pathologist (S.W., 8 years of experience in lung pathology) according to the 2011 International association for the study of lung cancer/American thoracic society/European respiratory society (IASLC/ATS/ERS) classification 19 . For difficult cases, histological subtypes were assessed by consultation with pathological experts. Based on the reported prognosis of lung adenocarcinoma [31][32][33][34] , tumors were graded according to four classifications. Grade 1 included histological subtypes of adenocarcinoma in situ (AIS) and minimally invasive adenocarcinoma (MIA). Grade 2 corresponded to the lepidic pattern of invasive adenocarcinoma. Grade 3 corresponded to tumors that showed acinar or papillary patterns. Grade 4 corresponded to tumors that showed micropapillary or solid patterns. When tumors were composed predominantly of a variant pattern, the tumors were removed from this classification system because they were uncommon and heterogeneous in terms of biological behavior and prognosis. Considering the proximity of tumor cells and normal cells under the microscope, tumors were divided into the well-differentiated group, moderately differentiated group and poorly differentiated group. Statistical analysis. The trends of pN 0 disease proportion among the different tumor grades and tumor differentiation degrees were assessed by chi-square trend test. The prevalences of nominal variables (e.g., tumor appearance, location, contour and intratumoral necrosis) were compared using Fisher's exact test. Differences in the mean values of continuous variables (e.g., tumor size, GGOc proportion, Sc proportion and SUV max ) were compared using the Independent-Student's t test. Multivariate logistic regression analysis was used to identify significant predictors of pathologic node-negative disease. All statistical analyses were performed using a commercial software package (SPSS, Inc., an IBM Company, Chicago, IL, USA). A P value of less than 0.05 was considered to indicate a significant difference. Data availability. All data generated or analysed during this study are included in this published article (and its Supplementary Information files).