CT characteristics in pulmonary adenocarcinoma with epidermal growth factor receptor mutation

Comprehensively investigate the association of CT morphology and clinical findings of adenocarcinoma with EGFR mutation status. Retrospectively included 282 patients who was pathologically proved as lung adenocarcinoma with known EGFR mutation status (mutations: 138 patients, female: 86, median age: 66 years; wildtype: 144 patients, female: 67, median age: 62 years) and their pre-treatment CT scans were analyzed. CT findings and clinical information were collected. Univariate and multivariable logistic regression analysis were performed. Adjusted for age, gender and smoking history of two groups, significantly more patients with pleural tags, pleural and liver metastases were found in the EGFR mutated group (P = 0.007, 0.004, and 0.043, respectively). Multivariable logistic regression analysis found that the model included age, gender, smoking history, air bronchogram, pleural tags, pleural and liver metastasis had a moderate predictive value for EGFR mutation status (AUC = 0.741, P < .0001). Exon-19 deletion was associated with air bronchogram which adjusted for age, gender and smoking history (P = 0.007, OR: 2.91, 95%CI: 1.25–7.79). The evidence of pleural tags, pleural and liver metastases go along with a higher probability of EGFR mutation in adenocarcinoma patients and air bronchogram is positively associated with Exon-19 deletion mutation.

Introduction Lung cancer remains the leading cause of cancer deaths for both men and women in the worldwide [1]. Many advances have been made in the understanding of the pathogenesis and management of lung cancer, particularly of adenocarcinoma (ADC). Specifically, the discovery of epidermal growth factor receptor (EGFR) mutations has changed lung cancer treatment. EGFR mutation are associated with a dramatic clinical response to the EGFR tyrosine kinase inhibitors (EGFR TKIs) gefitinib and erlotinib [2][3][4].
EGFR mutation testing is usually based on formaline fixed and paraffin embedded tumor specimens [5]. Approximately two thirds of non-small cell lung cancer (NSCLC) patients are diagnosed at an advanced stage of the disease [6] where only limited tumor specimens (biopsies, cytology) can be obtained in contrast to complete tumor resection. These limited tissue/ cytologic samples are not always available or evaluable for diagnosis and mutation testing, leaving some patients unable to have the EGFR mutation status of their tumors determined [7]. Tumor heterogeneity [8] and the presence of lesions that are inaccessible to needle biopsy challenge the tumor biopsies as well. These challenges are accentuated in a later line setting because re-biopsy may not be feasible and tumor heterogeneity may be greater.
Therefore, a less invasive procedure to increase the pre-test probability for EGFR mutation analysis would be helpful. It has already been shown that non-smoking status, female and East Asian ethnicity are correlated with EGFR mutation, but they are not sufficient to select or exclude patients for EGFR mutation testing [9][10]. Several studies have demonstrated that circulating free tumor-derived DNA (ctDNA), which can be isolated from the plasma or serum of patients with NSCLC, is feasible to assess EGFR mutation status [11][12]. However, ctDNA analysis is technically challenging, the suitability and performance of ctDNA testing varies significantly between different geographic regions and different laboratories [13][14].
Computed tomography (CT) is wildly used in clinic to evaluate lung cancer patients and few studies have been carried out to investigate the imaging features of ADC with EGFR mutations. The results are controversial and the correlation of imaging features with EGFR mutation is unclear [15][16][17][18]. We hypothesize that there were some CT characteristics might correlate with EGFR mutation status and those CT characteristics might serve as a complementary way to suggest the EGFR mutation status.
In order to identify imaging characteristics of EGFR mutated ADC, we retrospectively analyzed computed tomography (CT) images of a cohort suffering from ADC with known EGFR mutation status.

Materials and methods
This study was in compliance with the Health Insurance Portability and Accountability Act (HIPAA) regulations. Informed written consent for examinations (including CT, PET-CT and pathology examinations) were obtained from all patients. Clinical records of included patients suffering from ADC admitted to our 3rd level throacic hospital between February 2006 and October 2013 were reviewed retrospectively. The retrospective analysis was approved by the ethics committee of the medical school of the University of Heidelberg (IRB approval number S-048/2012). All patient records were anonymized and de-identified prior to analysis.

Patients and clinical assessment
1575 consecutive ADC patients have been analyzed for EGFR mutation status (Exon 18-21), demographics and tumor histopathology. 271 patients (17%) showed EGFR mutations and a similar number of patients (n = 280) with EGFR wildtype from the same database was selected randomly for comparison. According to the inclusion criterion (available CT images before surgery, chemo-and radiotherapy in the Picture Archiving and Communication System, PACS [Synapse, Fuji Medical System]), 282 patients (male: female: 129: 153, mean age: 64 years) were included into the analysis. In a few cases (n = 6), the only pre-treatment imaging available was a positron emission tomogram with a non-enhanced CT (PET-CT) which was deemed adequate for lesion interpretation and characterization.
Gender, age, smoking status (non-smokers were defined as having smoked <100 cigarettes/life, former and active smokers were designated as smokers), malignant tumor history were retrieved from clinical documents ( Table 1).

Histopathologic and EGFR analysis
Pathological diagnosis of the surgical specimens (n = 40), biopsies (n = 235) and cytological specimens (n = 7) were performed by board-certified pathologists according to the criteria of the 2004 WHO and 2011 IASLC/ATS/ERS classification [19]. EGFR mutations in exons 18-21 were determined by direct DNA sequencing as described previously in detail [20].

CT evaluation
All included patients underwent chest CT or PET-CT, which had been conducted within onemonth prior treatment and were interpreted on a PACS reading workstation retrospectively by two experienced chest radiologists (Z.J and D.J) in consensus blinded to the EGFR analysis results. CT examinations were performed at multiple institutions with a variety of helical scanners. The median section thickness used was 3mm (range from 0.5mm to 7mm). Detailed imaging characteristics of the primary lesions, corresponding lymph nodes and distant metastases were recorded (Fig 1), including [21][22][23][24][25][26]

Lymph node (LN):
According to the 7th Edition TNM classification of lung cancer, LN were divided into three levels: N1 -N3). Short axial diameter of the biggest LN in each level was recorded if more than 5 mm as well as the attenuation of the corresponding LN after contrast enhancement (fatty, isodense, hyperdense, necrosis, mixed). A LN with a short axis diameter of more than 10mm was rated as metastasis [27][28].
3. Distant metastases: a) Presence of nodule(s) in the contralateral lung (recording the maximal axial diameter of the biggest nodule and the distribution of pulmonary metastases [regional or random and diffuse]); presence of lymphatic carcinomatosis or pleural carcinosis (all the pleural metastases were either proven by histology or clinical criteria (pleura with obvious irregular thickening or nodules of the pleura which became more irregular and thicker during the follow up); b) Nearly all included patients (n = 267/282) underwent additional imaging studies (abdominal CT or ultrasound, brain CT or magnetic resonance (MR) imaging and whole-body bone scanning with technetium 99m medronate) for tumor staging. All available images and reports were reviewed. The specific distant metastatic organ (brain, liver, adrenal gland etc), the number of corresponding metastases and the total number of distant metastases which were divided into two groups (less or more than five) were recorded.

Statistical analysis
Initially, differences between categorical clinical and CT features were compared by χ2 test, Fisher's exact test and Kruskal-Wallis test. Quantitative continuous variables in clinical and CT data were compared by Mann-Whitney-U test. Cases like "the size of satellite nodule and minor size of N1LN" with single missing data points too much were not included into the logistic regression analysis. Univariate logistic and multivariable logistic regression analysis which adjusted for age, gender and smoking history of two groups were employed to evaluate the relationship between clinical and CT features with EGFR mutation status. The optimized combination of different CT and clinical features to predicting the EGFR mutation status was performed by logistic regression analysis. P < 0.05 was considered to be statistically significant.
Bonferroni adjustment method will be performed in multiple comparisons if necessary. The statistical software (SPSS 16.0; SPSS, Chicago, Ill) was used to perform the analysis and create graphs.   (Fig 2). All the clinical characteristics with EGFR mutation status were recorded in Table 1.

EGFR mutation status and CT features
Primary tumor. No statistically significant differences between EGFR mutation status (mutation: M and wildtype: WT) in tumor location (peripheral, middle and central), size, shape, margin, attenuation, pleural contact and enhancement. However, AB was found in 44% (n M = 55) of the EGFR mutated patients and 31% (n WT = 39) of the patients with wt (P = 0.033). The percentage of EGFR mutated patients (n M = 42 (33%) vs. n WT = 26 (20%), P = 0.013) with more than one pleural tag was significantly higher. EGFR mutated patients showed a significantly higher incidence of satellite nodule (n M = 71 [56%] vs. n WT = 50 [38%], Note:-Combined EGFR mutation stands for mutation at least was found in two exons (18)(19)(20)(21), for example point mutation in 18 and 20 exons.
-Seldom mutation means at least two combination of the way of EGFR mutation, such as deletion combined insertion in exon 19.
-There was one patients had both deletion in exon 19 and p.L858R.
Lymph nodes. The median short axis diameter of the biggest N1 LN in EGFR mutated patients was smaller (12 mm vs. 14 mm; P = 0.023) ( Table 2). Other comparisons such as the short axis of the biggest LN in the other two levels and LN staging status were not statistically different.
Metastases. While random and diffuse, pulmonary metastases were more frequent in the EGFR mutation group (n M : 11 [8%] vs. n WT : 4 [3%], P = 0.052), this observation did not reach significance. The frequency of pleural metastases, of which 63 patients were confirmed by pathology and 17 patients were diagnosed according to clinical criteria, was higher in the EGFR mutation group (n M = 50 [36%] vs. n WT = 30 [21%], P = 0.004). With distant metastases, the overall incidence and the patients with more than five metastases were similar between the two groups. However, liver metastases were found significantly more often in EGFR mutated patients (P = 0.010) while bilateral adrenal gland metastases were found only in EGFR wt patients (P = 0.002). Interestingly, half of the adrenal metastases patients in EGFR wt group were manifested as bilateral metastases (Table 2) (S1 Table).
Univariate and multivariable logistic regression analysis. Univariate logistic analysis showed that gender, smoking history, AB, pleural tags, pleural metastasis and liver metastasis were significantly different in two groups and the P values were 0.008, 0.002, 0.033, 0.014, 0.004 and 0.013, respectively. (Table 2).  4: N1: ipsilateral peribronchial and/or ipsilateral hilar LN and intrapulmonary nodes, only included the short size of LN more than 5mm. 5: Information concerning liver and adrenal gland metastases was missing for 10 and 12 patients, respectively. #: The P value was calculated by multivariable logistic regression analysis which adjusted for age, gender and smoking history.
Adjusted for age, gender and smoking history of two groups respectively, we found that the incidence of pleural tags (P = 0.007), pleural (P = 0.004) and liver metastases (P = 0.043) are significantly higher in EGFR mutated patients. (Table 2) However, the incidence of AB was nearly to show the significant difference in two groups (P = 0.059). Further, multivariable regression analysis found that combined age, gender, smoking history, AB, pleural tags, pleural and liver metastasis together which showed the highest predictive value (AUC = 0.741) (Fig 3).

Association of the CT features with EGFR exon-19 deletion or p.L858R
After adjusting for age, gender and smoking history, multivariable logistic regression analysis showed that AB strongly associated with exon-19 deletion mutation (OR, 2.91; 95%CI: 1.25-7.79; P = 0.011), as compared to the rest of EGFR mutations. Whereas, compared with the rest of EGFR mutated patients, patients with p.L858R mutation had significant less AB (OR, 3.12; 95%CI: 1.25-7.79, P = 0.015) ( Table 3). With other imaging features, there were no significant differences were found when associated them with EGFR exon-19 deletion or p.L858R.

Discussion
This study has identified certain clinical and imaging characteristics which were correlated with EGFR mutations. According to previous reports [9,10], we found that women and nonsmokers tended to have EGFR mutation more often. To our knowledge, no study has previously been done with the objective of a comprehensive comparison of CT features of ADC patients with different EGFR mutation status, hence only single pattern such as GGO [16,17,[29][30][31][32] have been evaluated so far. In addition, there were two studies have showed that ADC patients with malignant pleural effusion (MPE) had a higher incidence of EGFR mutation [33,34].
Our study has showed that, several radiological features associated with EGFR mutation in ADC: the number of patients with AB, pleural tags, pleural and liver metastases was significantly higher if EGFR is mutated. Logistic regression analysis showed that the model composed by age, gender, smoking history, AB, pleural tag (n!1), pleural and liver metastasis have a moderate predictive value for EGFR mutation. These could enable radiologists to better understand the imaging features which correlated with EGFR mutation and to applying this understanding into clinical practice by allowing radiologists to raise clinical suspicion for EGFR mutation.
Lepidic predominant ADC were already described to show AB frequently [35], while a correlation of EGFR mutations with the lepidic pattern has also been demonstrated [20]. This might explain why EGFR mutated patients in our cohort showed more AB. Koenigkam Santos M et al. [36] found that ADC was more commonly associated with pleural tags, compared to squamous cell carcinomas. In our study, mutated ADC had more pleural tags than wild type ADC.
Several reports [33,37] found that overall survival (OS) could be prolonged in EGFR positive ADC patients with MPE undergoing EGFR-TKI therapy. Meanwhile, several retrospective studies [33,34] discovered that patients with ADC and MPE had a higher rate of EGFR mutation. A functional variant of the EGFR promoter, 216G/T (rs712829), was associated with pleural spread of ADC which is usually accompanied with MPE [38]. This might explain the significantly higher number of pleural metastases in EGFR mutated patients. The accompanied high incidence of pleural effusion might cause more compressive atelectasis (n M = 12 vs. n WT = 6) in ADC with EGFR mutation.
In contrast to the previous findings [18,[39][40][41], neither significant differences in LN staging, pulmonary, nor brain metastases were noticed, while satellite metastases were significantly more frequent and of smaller size in EGFR mutated tumors. The latter findings were probably associated with the abundant angiogenesis of EGFR pathways [42] since this important mediator supports local spread resulting in multiple metastases in the same lobe or the whole lung. This mechanism might also explain the few necrosis in EGFR mutated tumors (n M = 1; n WT = 7, S1 Table). In addition, the short axis of N1 LN in EGFR mutated tumors was smaller. A possible explanation for this maybe EGFR was expressed in almost all cells with the only exception of the mature lymphohematopoietic cells [42]. The majority of detected EGFR mutations were either p.L858R or exon-19 deletions (39 + 62 = 101, 73%, Table 3). An association of CT features and OS with p.L858R or exon-19 deletions might be different. It has been demonstrated that p.L858R were correlated with the CT feature 'invasive solid pattern' [29], had not prolonged OS after EGFR-TKI therapy [43]. While exon-19 deletion, in contrast, had a longer progression-free survival (PFS) [44] and OS [43,45] after TKI treatment. Therefore, it is useful to know the correlation of significant clinical features and CT pattern with EGFR p.L858R or exon-19 deletions. As we found exon-19 deletion mutations were significantly more frequent in women and this was consistent with the finding that EGFR mutation favoring female patients. Even excitingly, we found exon-19 deletion is correlated with a significantly greater number of AB, whereas tumors with p.L858R mutation had significantly less frequent AB (Table 3).
Besides, our study has its own limitations. This was a retrospective study which would induce patient selection bias and the case included in the study were not enough, even we have already included as many cases as we could. However, these results might serve as a basement for our further international cooperative study and all those results still need be further validated by our prospective study.

Conclusion
Clinical and CT-derived imaging characteristics are associated with activating EGFR mutations. Especially, the presence of AB, pleural tag, pleural and liver metastases may help to increase pretest probability for EGFR mutation. In addition, the presence of AB positively associated with EGFR exon-19 mutation.
Supporting information S1