Study design
This study was conducted in accordance with the Declaration of Helsinki. The Institutional Review Board and the Ethics Committee of Shanghai Chest Hospital approved the study, and waived the requirement for patient consent due to its retrospective design.
Patients
A search of the electronic medical records of Shanghai Chest Hospital yielded 5,748 consecutive patients with a pathological diagnosis of lung cancer who had undergone surgical resection at the Department of Thoracic Surgery between January 2014 and December 2015. Inclusion criteria were (1) availability of pre-resection CT; (2) lesions manifesting as pGGNs on CT; (3) lesions < 3 cm in diameter as determined via CT; and (4) pathologically proven lung adenocarcinoma. Exclusion criteria were: (1) no pre-resection CT available; (2) lesions manifesting as part-solid or solid nodules on CT; (3) lesions >3 cm in diameter on CT; and (4) histological diagnosis other than lung adenocarcinoma, such as AAH or squamous cell carcinoma. The following clinical data were recorded for all included patients: age, sex, smoking history (never or current/former), symptoms (none, or any of cough, shortness of breath, fever, hemoptysis, chest pain, recurrent pulmonary infection), serum carcinoembryonic antigen (CEA) concentration, and CT features. All patients underwent preoperative thoracic unenhanced CT examination at our hospital.
CT imaging
Lung CT scans were performed with a Somatom Sensation-64 (Siemens Medical Systems, Forchheim, Germany) with 120 kVp and 100 mAs. All CT examinations included the entire thorax at full suspended inspiration with the patient lying supine. When a nodule was identified, target CT was performed with the following parameters: pitch, 0.64; 1-3 second scan time; matrix size, 1024*1024; FOV, 180mm.The lung window width was consistently 1600 Hounsfield units (HU), and the window level was −600 HU.
Nodule size was expressed as maximal (longest diameter on axial images) and mean (average of the maximum length and width of the nodule) diameters [12]. CT density was defined as the average CT attenuation (HU) within the nodule that did not contain blood vessels or bronchioles. Two readers (radiologist J.G. with 16 years of experience in chest CT and surgeon J.F. with 12 years of experience in thoracic surgery) who were blinded to the histopathological results and clinical data evaluated all CT scans independently. Each reader measured the sizes and d of the lesions in the lung window setting on the transverse CT section that displayed the largest nodule dimensions. The average of the measurements obtained by each of the two reviewers was used for analysis. The observers also recorded the presence of particular signs such as pleural retraction, air bronchogram, bubble lucency, and spiculated margins. Pleural retraction was defined as linear attenuation heading toward the pleura or the major or minor fissure from a pGGN. Air bronchogram was defined as air-filled bronchi within a pGGN. Bubble lucency was defined as the presence of small spots of round or ovoid air attenuation within a pGGN. Spiculated margins were defined as the presence of strands extending from a nodule margin into the lung parenchyma without reaching the pleural surface [4]. The results were compared between two readers. If the results were discrepant, the two readers reevaluated the scan to reach a consensus. If no consensus was reached, another radiologist (Z.X.G. with 24 years of experience) was consulted and their decision was deemed final.
Histopathological findings
To ensure that the resected nodules corresponded to nodules observed inCT scans, the radiologic and surgical procedures were conducted on the same day; CT-guided microcoils were inserted to mark both the nodule and the visceral pleural surface. Intra-operative fluoroscopy was then used to identify the microcoils, and thus the nodule to be resected. Two chest pathologists (F.X.J., with 12 years of experience, and Z.H., with 23 years of experience), who were blinded to all clinical information reviewed the pathological specimens independently and classified the lesions as atypical adenomatous hyperplasia, AIS, MIA, or IAC in accordance with the criteria described in the 2015 World Health Organization Classification of Lung Tumors [13-14];Disagreements were resolved by consensus. They also classified the histological subtypes of IACs as lepidic predominant, acinar predominant, papillary predominant, micropapillary predominant, solid predominant, or invasive mucinous adenocarcinoma and then classified all tumors as pre-IAC (including AIS and MIA) or IAC.
Statistical analysis
For continuous predictors, optimal cutoff values were determined via the maximal Youden index (sensitivity + specificity − 1). Receiver operating characteristic curves, and corresponding areas under the curves, sensitivities and specificities were presented for these dichotomized predictors using the optimal cutoffs.
Integer cutoffs can be more convenient in clinical practice; thus, approximate integer of optimal cutoffs were applied for maximal and mean diameters (both 10 mm), CT‑determined density (−600 HU), and age (55 years). Although our cutoff value for CEA was 2.19 µg/L, we used the upper limit of the normal range of 5 µg/L as our standard for stratification.
These dichotomized predictors were described using frequencies and percentages, differences between IAC groups were compared using the χ2 test. Some patients had multiple nodules, to assess relationships between clinicopathological factors and lung IAC, generalized estimating equation (GEE) was applied for the correlated data using with a logit link function and a binomial distribution. The working correlation structure was selected according to the quasi-likelihood under the independence model criterion minimum principle. In multivariable model 1, variables with p < 0.1 in univariable analysis were included. In multivariable model 2, all independent variables were included to assess the stability of the results. Lastly, a multicollinearity test was performed on all independent variables. All statistical analyses were performed with SPSS Statistics, version 17.0(SPSS Inc., Chicago, IL, USA), and p < 0.05 was deemed to indicate statistical significance.