Identification of preoperative prediction factors of tumor subtypes for patients with solitary ground-glass opacity pulmonary nodules

This study was approved by the institutional review board of Tianjin Medical University Cancer Institute and Hospital, Tianjin, China. Between January 2013 and December 2016, 6317 patients with pulmonary nodules underwent surgical resection with curative intent at Cancer Institute and Hospital of Tianjin Medical University. Of these patients, 423 were selected according to inclusion criteria and exclusion criteria. Our inclusion criteria were as follows: (1) patients with solitary GGO nodules on chest CT scan, (2) patients who had their lesions surgically removed and had postoperative pathological diagnosis, (3) R0 resection. Our exclusion criteria included: (1) patients who had no CT scan, (2) pulmonary multiple GGO nodules, mixed GGO with consolidation to the maximum tumor diameter greater than 0.75 or pure solid tumors, (3) patients with history of lung cancer or who had a malignancy elsewhere. All patients underwent lobectomy or limited resection (segmentectomy or wedge resection) with hilar and mediastinal lymphadenectomy or lymph node sampling. There were no objective criteria for limited resection, and the indications depended on each surgeon’s preference. A patient with a GGO > 5-8 mm may be subjected to surgical treatment and with a GGO less than 20 mm may be given the limited resection, including segmentectomy and wedge resection.

For each patient, age, gender, smoking status, a family history of cancer, location of tumor, the serum tumor markers: carcinoembryonic antigen (CEA) and carbohydrate antigen 19–9 (CA19–9) and histological subtypes were extracted from patient medical records. Classification of lung adenocarcinoma was assessed by two pathologists in accordance with the new interdisciplinary IASLC/ATS/ERS classification of lung adenocarcinoma. We classified all patients into two groups: AIS /MIA group and invasive adenocarcinoma (IA) group.

Chest CT scans were performed before surgery by using one of three multi-detector CT systems: Somatom Sensation 64 (Siemens Medical Solutions, Forchheim, Germany), Light speed 16, and Discovery CT750 HD (GE Healthcare, Milwaukee, WI, USA) scanner. Scanning parameters were as follows: 120kVp with tube current adjusted automatically, 1.5 mm reconstructionthickness with 1.5 mm reconstruction interval for 64-detector scanner; and 120kVp, 150–200 mA, 1.25 mm reconstruction thickness with 1.25 mm reconstruction interval for the other two scanners. GGO was defined as a hazy increase in lung attenuation without obscuring the underlying vascular marking. Two observers who were unaware of pathologic staging viewed each CT scan of 423 solitary GGO nodules and assessed nodules morphology blindly. Morphology included density, size, air bronchogram, bubble-like sign, spicule sign, pleural tag or lobular. Based on the density via CT, GGO nodules were classified into two groups: pure GGO (pGGO) (a tumor without solid component), and mixed GGO (mGGO) (a tumor with both GGO and solid components).

About 3 ml of peripheral blood was collected from each case in coagulated tube. The serum was separated by centrifuging at 3000×g for 5 min, and then transferred to a new Eppendorf tube, and stored at − 80 °C for further analysis. Tumor biomarkers, including CEA, CA19–9 were measured using electrochemiluminescence immunoassays according to standard procedure of Roche Company’s kit and Roche E170 automatic immunity analyzer. The cut-off points for each tumor biomarker determined by the manufacturer, were as follows: CEA, 5 μg/l; CA19–9, 39 U/ml.

Statistical analysis performed by using SPSS 24.0 software. The T test, Mann-Whitney U test, χ² test or Fisher’s exact test were used to test for difference in clinical factors and imaging characters between different pathology groups, asappropriate. T test was used for categorical variables fitted normal distribution and expressed by x ± s. Mann-Whitney U test was used for categorical variablesfitted non-normal distribution and expressed by M (P25, P75). χ² test or Fisher’s exact test was used for categorical variables. Receiver operating characteristics (ROC) curves were generated to evaluate the predictive potential of identified clinical and imaging signatures for IA, then combined all identified factors to predict histological types by adding all five factors to a bivariable-adjusted logistic regression model. Optimal cut-off values were calculated by ROC cures. Univariate logistic regression and binary logistic regression analysis were also performed to assess the diagnostic accuracy by using cut-off values. All statistical tests were two-sided, and a p value ≤0.05 was considered statistically significant.

There were total 423 patients with solitory GGO nodules. Among them, 393 had adenocarcinoma and 30 had benign nodules. The clinicopathologic characteristics of 393 (92.9%) patients with adenocarcinoma were summarized in Table 1. There were 117 male and 276 female patients (median age, 57 years; range: 27–78 years). Two hundred and ninety-six of 393 (75.3%) patients were never active smokers. Sixty-six patients had a family history of other cancers (such as liver cancer, gastric cancer, colorectal cancer and bone cancer) and 50 (43.1%) patients had a family history of lung cancer. Of 393 patients, preoperative serum CEA and CA19–9 were tested in 379 patients. The number of patients with elevated CEA and CA19–9 were 25 (6.6%) and 81 (21.4%), respectively. GGO nodules were often found in the superior lobe of right lung (n = 163), followed by the superior lobe of left lung (n = 98). The histological types according to IASLC/ATS/ERS classification were as follows: 269 (68.4%) patients with IA, 115 (29.3%) patients with MIA, 9 (2.3%) patients with AIS. Lobectomy, segmentectomy and wedge resection were performed in 349 (88.8%), 21 (5.3%) and 23 (5.9%) patients, respectively.

Age, family history of cancer, Serum CEA level, GGO size, type of GGO and bubble-like sign differed significantly between patients with AIS/MIA and those with IA (p: 0.008, 0.046, 0.000, 0.000, 0.000, and 0.001). (Table 2). By the bivariate logistic analysis, tumor size, mixed GGO and bubble-like sign were independent predictors of IA (p:0.000, 0.000 and 0.021, respectively). ROC curves were generated to assess the IA prediction accuracy of the six factors identified in univariate analysis. It showed that a family history of cancer had a poor accuracy (p = 0.1148). Therefore, this parameter was not brought into muti-factor combination. Figure 1 shows the true-positive ratios (sensitivity) and false-positive ratios (1 minus specificity) for age, CEA, tumor size, GGO type and bubble-like sign. The areas under curves (AUCs) for age, serum CEA level, tumor size, GGO type and bubble-like sign were 0.59, 0.62, 0.87, 0.72, and 0.58 (p: 0.0058, 0.0002, < 0.0001, < 0.0001, and 0.001, respectively), with sensitivities of 57%, 54%, 86%, 71%, 35% and specificities of 59%, 65%, 72%, 73%, 83%. We also assessed the IA prediction accuracy of these five factors in combination via bivariate logistic regression analysis. The AUC of the five-factor combination was 0.91, with a sensitivity of 82% and a specificity of 87% (p < 0.0001). (Table 3). To further distinguish the IA from AIS/MIA, we performed univariate and multivariate analyses using the optimal cut-off values calculated from the ROC curves (Table 4). According to the univariate analysis, patients were more likely to be diagnosed with IA if they had these factors: more than 58.5 years, a serum CEA level > 1.970 μg/L, a tumor size > 13.50 mm, mGGO and bubble-like sign (p: 0.005, 0.001, 0.000, 0.000, and 0.001, respectively). According to the bivariate analyses, the combination of these five factors was an independent diagnostic factor for IA (p:0.000).