Abstract
Background
For pulmonary subsolid nodules (SSNs) in patients with extrapulmonary malignancies, it is still unclear what proportion of SSNs is transient and how we can more accurately diagnose these transient SSNs.
Purpose
To investigate the frequency of transient SSNs and their differentiating clinical and thin-section computed tomography (CT) features in patients with extrapulmonary malignancies.
Material and Methods
From January 2005 to February 2012, 78 SSNs in 63 individuals (30 men and 33 women; mean age, 55.1 years ± 15.5) with extrapulmonary malignancies were identified. Their clinical and thin-section CT characteristics were reviewed and compared between transient and persistent SSNs. Differentiating factors and their performance were also measured.
Results
Thirty-six of the 78 SSNs (46.2%) were transient. Between transient and persistent SSNs, there were significant differences in patients’ age, sex, detection mode, and the presence of eosinophilia, lesion multiplicity, lesion margin, and pleural retraction (P < 0.05). Multivariate analysis revealed that follow-up detected SSNs (adjusted odds ratio [OR], 38.88), multiple lesions (OR, 7.64), and an ill-defined nodular margin (OR, 11.93) were significant discriminators of transient SSNs (P < 0.05). Discrimination of transient SSNs was significantly better upon incorporating both clinical and thin-section CT features than using clinical features alone (P < 0.05).
Conclusion
Approximately half of the SSNs detected in patients with extrapulmonary malignancies were transient. Transient SSNs in these patients can be very accurately differentiated using their thin-section CT and clinical features.
Introduction
In oncology patients today, radiologic studies such as computed tomography (CT), magnetic resonance imaging (MRI), or 18F-FDG PET/CT are used for the estimation of the therapeutic efficacy of anti-cancer treatments and for the purposes of surveillance of these patients after treatment. In particular, chest CT has been one of the most commonly used imaging studies (1,2) as the lung is one of the most common sites of systemic metastasis in oncology patients. Morphologically, lung metastases commonly appear as pulmonary nodules or masses, and thus, when we encounter pulmonary nodules on chest CT in oncology patients, it is crucial to determine whether the detected pulmonary nodules are metastasis or not. When the detected pulmonary nodules are diagnosed as metastasis, systemic chemotherapy or metastasectomy may be recommended, or in the case of newly occurring metastasis during ongoing chemotherapy, a change in the chemotherapeutic agent or therapeutic method should be considered.
Pulmonary subsolid nodules (SSNs) on CT are defined as nodules containing ground-glass attenuation portions, which do not obliterate the underlying bronchial and vascular structures (3). These pulmonary SSNs include part-solid ground-glass nodules (GGNs) and pure GGNs (4). While pulmonary SSNs can be non-specific abnormalities representing various diseases (5–7), if persistent, pulmonary SSNs are likely to be malignant and are very often found to be primary lung cancers (8,9). However, according to a recent study in a lung cancer screening setting, more than half of the detected pulmonary SSNs, particularly in the case of part-solid GGNs were transient (10). Thus, 3-month follow-up CT is now recommended to exclude the possibility of transient SSNs when pulmonary SSNs are initially detected in the lung cancer screening setting (4).
As for SSNs in patients with extrapulmonary malignancies, to the best of our knowledge, there has been only one study dealing with the topic, in which Park et al. reported that pulmonary SSNs in patients with extrapulmonary malignancies were more likely to be primary lung cancers than metastasis (8). However, that study (8) only dealt with pathologically-confirmed persistent SSNs. Thus, in patients with an extrapulmonary malignancy, it is still unclear what proportion of pulmonary SSNs is transient and how we can accurately diagnose these transient SSNs. If SSNs initially detected in patients with extrapulmonary malignancies are transient, administration of new chemotherapy or alteration of the therapeutic method under the impression of systemic metastasis or treatment failure should be avoided. Therefore, the purpose of our study was to evaluate the frequency of transient SSNs, and to investigate the differentiating clinical and thin-section CT features of transient SSNs in patients with extrapulmonary malignancies.
Material and Methods
The institutional review board of Seoul National University Hospital approved this retrospective study, and waived the requirement for patients’ informed consent.
Study population
From January 2005 to February 2012, 9939 oncology patients who visited the department of oncology or thoracic surgery in our hospital due to malignancies underwent 16,040 chest CT examinations for tumor staging or for follow-up surveillance. From this population, one author (WSC) retrospectively selected individuals with pulmonary SSNs identified at chest CT using the electronic medical records and radiology information systems of our hospital. Our study population was determined via the following steps. First, we selected all CT scans, of which reports included the words “GGO nodule”, “GGN”, “part-solid nodule”, “nonsolid nodule”, “ground-glass opacity nodule”, “subsolid nodule”, or “ground-glass nodule”, resulting in 99 individuals in total. Second, two radiologists (WSC and CMP) reviewed all the CT examinations and included only cases that met the following criteria: (i) SSNs larger than 5 mm and smaller than 3 cm; (ii) individuals with a thin-section CT containing SSNs; and (iii) individuals with at least one or more follow-up CT studies available allowing us to determine the temporality of the SSNs; 87 out of 99 patients met these inclusion criteria. Third, we excluded patients who had primary lung cancers; 11 patients with primary lung cancers were excluded in this step. Fourth, patients with extrapulmonary malignancies, who had SSNs detected prior to the primary malignancies or had no evidence of remaining cancer for more than 5 years prior to SSN detection, were excluded as the detected SSNs were not likely to be related to the primary malignancies; 13 patients were excluded in this step. Finally, 78 SSNs in 63 individuals (30 men and 33 women; age range, 19–85 years; mean age, 55.1 years ± 15.5) were included in this study (Fig. 1). Table 1 demonstrates the sites and pathological diagnosis of primary extrapulmonary malignancies included in the present study.
Flow chart of inclusion criteria in 9939 oncology patients. Primary extrapulmonary malignancies in the study population. Numbers in parentheses indicate number of patients.
Analysis of clinical features
One author (YSS) recorded the following clinical and laboratory data for all individuals with SSNs: (i) patient age; (ii) patient sex; (iii) smoking history (current smoker vs. non-current smoker); (iv) detection mode of the lesion (baseline detection vs. follow-up detection); and (v) history of systemic chemotherapy or antimicrobial therapy. Both ex-smokers and never smokers were designated as non-current smokers in this study. In terms of the detection mode of the lesion, we designated SSNs as baseline-detected lesions when SSNs were detected at initial CT. When SSNs newly occurred on follow-up CT, we regarded them as follow-up detected lesions. Laboratory results such as (i) white blood cell (WBC) count, (ii) blood eosinophil count, (iii) presence of blood eosinophilia, (iv) C-reactive protein (CRP) level, and (v) erythrocyte sedimentation rate (ESR) were also recorded. These laboratory results were acquired from the tests performed closest to the CT examinations showing SSNs. Among the 61 individuals, WBC count and eosinophil count were available in all patients. CRP levels were available in 23 patients and ESR in 15. We defined blood eosinophilia as an eosinophil count exceeding 500 per µL.
CT examinations
Sixty-three patients underwent 215 chest CT examinations with each having undergone at least two thin-section CT examinations (mean CT number, 3.41 ± 2.16; range, 2–12; section thickness ≤1.25 mm). One hundred and twenty-seven CT scans were performed with intravenous contrast media injection while 88 CTs were performed without intravenous contrast media. All CT examinations were performed using one of four available CT scanners: Somatom Definition, Sensation-16 (Siemens Medical Systems, Erlangen, Germany), Brilliance-64 (Phillips Medical Systems, Best, The Netherlands), and LightSpeed Ultra (GE Healthcare, Milwaukee, WI, USA). Scanning parameters for chest CT were as follows: detector collimation, 1.0–1.25 mm; beam pitch, 0.75–1.0; reconstruction thickness, 1.0–1.25 mm; reconstruction interval, 1.0–1.25 mm; rotation time, 0.4–0.5 s; tube voltage, 120 kVp; tube current, 40–120 mAs; and reconstruction kernel, sharp reconstruction algorithm. All CT scans were performed during patients’ inspiration in the supine position. The mean follow-up interval between CT examinations of transient SSNs was 85.08 ± 45.21 days (range, 33–198 days), and that of persistent SSNs was 131.36 ± 122.46 days (range, 32–549 days). The total follow-up period of transient SSNs was 91.87 ± 80.60 days (range, 32–356 days), and that of persistent SSNs was 493.43 ± 551.82 days (range, 32–1864 days).
Analysis of thin-section CT features
Two chest radiologists (SML and JMG with 4 and 19 years of experience, respectively) reviewed all thin-section CT examinations while blinded to clinical information such as age, sex, pathologic results, and laboratory data. All decisions on CT findings were established in consensus. All CT findings of SSNs were evaluated with the lung window setting (window level, −700 HU; width, 1500 HU), using a Picture Archiving and Communication System (M-view, Infinite, Seoul, Korea).
“Transient” SSNs were defined as SSNs that decreased in size or disappeared at follow-up CT either spontaneously or during/after antimicrobial therapy. We designated SSNs as “persistent” when these lesions remained stable or increased in size over follow-ups. We defined pulmonary SSNs as “indeterminate” when the SSNs decreased in size or disappeared at follow-up CT in patients who underwent systemic chemotherapy during the existence of SSNs due to the ambiguity of the true cause of the SSNs’ regression, i.e. “true” spontaneous regression or chemotherapy. Determination of whether the SSNs were transient, persistent, or indeterminate was made through the consensus of two radiologists (WSC and CMP). We considered nodules to have significantly decreased in size when they showed a decrease in size of at least 20% compared with the previous CT showing the SSNs (10).
Evaluated thin-section CT features of each SSN were as follows: (i) lesion location, (ii) lesion type (pure GGN vs. part-solid nodule), (iii) lesion size, (iv) solid portion size, (v) solid proportion, (vi) lesion shape (round vs. non-round), (vii) lesion margin (well-defined vs. ill-defined), (viii) lesion border (spiculated vs. non-spiculated), (ix) lesion multiplicity, (x) presence of bubble-lucency, and (xi) presence of pleural retraction.
To determine lesion size and solid portion size, one chest radiologist (SML) recorded the longest diameter of the lesion as well as the internal solid portion. The solid proportion of the lesion was then calculated by dividing the solid portion size by the lesion size.
Statistical analysis
To investigate the differentiating factors of transient from persistent SSNs, logistic regression analysis was performed. Indeterminate SSNs were excluded from statistical analysis due to ambiguity of the true cause of the SSNs’ regression. In the case of multiple SSNs in one patient, we randomly selected one lesion among the multiple SSNs as estimation of a predictive model proved very difficult for multiple SSNs due to the high degree of correlation among them. After univariate logistic regression analysis for both categorical and continuous variables, subsequent multivariate analysis was performed to evaluate independent differentiating factors of transient SSNs from persistent SSNs. Variables with a P value <0.2 at univariate analysis were used as input variables for multivariate analysis. In our study, penalized maximum likelihood estimation was used instead of classical multivariate logistic regression analysis due to the complete separation phenomenon in the presence of pleural retraction and bubble lucency (11). The C statistic was conducted to measure the performance of multiple logistic regression models in discriminating transient from persistent SSNs. The C statistic equals the area under a receiver-operating characteristic (ROC) curve when the response is binary (12). Independent predictors of clinical features alone, thin-section CT features alone and of a combination of both clinical and thin-section CT features were used as input data for the three separate logistic regression models. Thereafter, we compared the differentiating performances of these logistic regression models. A C statistic of 0.5 indicated no ability to discriminate, while a value of 1.0 indicated perfect discrimination (13).
All statistical analyses were performed using SAS statistical software (version 9.2; SAS institute, Cary, NC, USA). A P value of <0.05 was considered to indicate a significant difference.
Results
Thirty-six of the 78 SSNs (46.2%) in 22 individuals were transient and 40 of the 78 SSNs (51.3%) in 39 individuals were persistent. Two SSNs (2.6%) in two individuals who underwent systemic chemotherapy during the existence of SSNs were determined to be indeterminate as one SSN disappeared and the other SSN substantially decreased in size at follow-up CT and we were not able to safely state whether the decrease or disappearance of the nodules were caused by chemotherapy or their natural courses.
Twenty of the 63 (31.7%) individuals underwent systemic chemotherapy during the existence of the SSNs and thirty of the 63 (47.6%) individuals received antimicrobial agent during the existence of the SSNs. Among 21 SSNs in 20 individuals who underwent systemic chemotherapy19 SSNs in 18 individuals were persistent at follow-up CT, while two SSNs in two individuals disappeared (n = 1) or decreased (n = 1) in size during or after systemic chemotherapy, and were designated as indeterminate. There were no patients who changed or ceased chemotherapy during the existence of SSNs. Among 35 SSNs in 30 individuals who underwent antimicrobial therapy, 25 SSNs in 24 individuals were persistent at follow up CT, while 10 SSNs in six individuals disappeared (n = 9) or decreased (n = 1) in size during or after antimicrobial therapy and were designated as transient SSNs.
Among the 36 transient SSNs, 31 (86.1%) SSNs completely resolved and five SSNs (13.9%) decreased in size by more than 50%, of which all size changes were greater than 4 mm in diameter on follow-up CT. The mean follow-up interval for confirmation of transient SSNs was 91.87 ± 80.60 days (range, 32–356 days). In our study, 23 SSNs in 15 patients were initially detected at follow-up detection, and 22 of those SSNs (95%) were transient. The mean interval between follow-up CT showing new SSNs and previous CT was 216.17 ± 250.25 days (range, 32–1027 days) and the median size of newly occurring transient SSNs at follow-up CT was 12.0 mm (range, 6.0–19.0 mm).
Of the 40 persistent SSNs, 23 lesions in 23 patients were pathologically confirmed via wedge resection (n = 1), segmentectomy (n = 1), and lobectomy (n = 21), and their pathological diagnosis was as follows: adenocarcinoma (n = 16), adenocarcinoma in situ (n = 3), atypical adenomatous hyperplasia (n = 1), adenosquamous carcinoma (n = 1), squamous cell carcinoma (n = 1), and non-specific inflammation (n = 1). Seventeen persistent SSNs in 16 patients that were not pathologically confirmed had been followed-up with regular chest CT examinations. The mean total follow-up period for these persistent SSNs was 898.61 ± 574.77 days (range, 297–1846 days). Among these 17 SSNs, 16 persistent SSNs remained unchanged during the follow-up period, while one lesion showed an increase in the entire nodule size and internal solid portion size during 733 days.
Clinical features
Clinical features of patients with transient and persistent SSNs.
Except where indicated, data are numbers of patients.
Data are means ± standard deviations.
SSN, subsolid nodule.
Thin-section CT findings
In terms of thin-section CT findings, there were no significant differences in lesion location, lesion type, solid proportion or lesion shape between transient and persistent SSNs. Transient SSNs showed ill-defined margins, and non-spiculated borders more frequently, and were more likely to be multiple lesions (P < 0.05) (Figs. 2 and 3). Pleural retraction and bubble lucency were present only in patients with persistent SSNs (P < 0.04) (Fig. 4). Table 3 demonstrates the thin-section CT features of transient and persistent SSNs.
Transient subsolid nodule in a 45-year-old man with esophageal cancer. (a) Thin-section CT examination shows a 16-mm SSN (arrow) with an ill-defined margin in the left lower lobe of the lung. This patient had multiple SSNs in the bilateral lung (not shown). (b) At follow-up CT performed 1 month after, the previous SSN completely disappeared without any anti-chemotherapeutic agent administration. Transient SSN detected on follow-up CT in a 24-year-old man with embryonal cell carcinoma. (a) Thin-section chest CT scan shows a 13-mm peripheral SSN (arrow) in the right lower lobe of the lung. (b) On previous CT taken 3 months ago, there was no such nodule in the right lower lobe. (c) At follow-up CT performed 2 months later, the size of the SSN (arrow) markedly decreased. (d) At follow-up CT performed 3 months later, this lesion finally disappeared. During the follow-up period, this patient did not receive any anti-chemotherapeutic agents. A persistent subsolid nodule in a 47-year-old woman with breast cancer. (a) Chest CT scan shows a 14-mm solitary SSN (arrow) with pleural retraction (arrowhead) in the right lower lobe of the lung. (b) At follow-up CT taken 6 months later, this SSN (arrow) with pleural retraction (arrowhead) remains unchanged. The patient underwent right lower lobectomy and this lesion was confirmed as pulmonary adenocarcinoma. Thin-section CT findings of transient and persistent SSNs. Except where indicated, data are numbers of SSNs. Data are median (range, minimum–maximun). Calculated using penalized maximum likelihood.
Results of multivariate logistic regression analysis
Results of multivariate analysis in predicting transient SSNs.
CI, confidence interval; OR, odds ratio.
Differentiating performance of logistic regression models
ROC curve analysis and C statistic were performed to determine the performance of established multivariate logistic regression models in discriminating transient from persistent SSNs using significant clinical and thin-section CT features. When only the detection mode of the lesion was used as input data as a clinical independent predictor, the AUC was 0.805 (95% CI: 0.699–0.911). When independent predictors at thin-section CT, including lesion margin and multiplicity were used as input data, the AUC was 0.870 (95% CI: 0.772–0.968). When both the clinical and thin-section CT features were used as input data, the AUC was 0.954 (95% CI: 0.893–1.000) and the sensitivity and specificity at the optimal cut-off, at which point the sensitivity and specificity are balanced and maximized, were 95.5% and 89.7%, respectively. The differentiating performance of the logistic regression model incorporating a combination of clinical and thin-section CT features was significantly higher than those using only the clinical features alone (0.954 vs. 0.805, P = 0.0025), and was also higher than that using only the thin-section CT features alone (0.954 vs. 0.870), although the latter improved performance only just missed statistical significance (P = 0.06) (Fig. 5).
Receiver-operating characteristic (ROC) curve analysis of multiple logistic regression models in discriminating transient from persistent SSNs in patients with extrapulmonary malignancies. This graph shows the ROC analysis of the two combinations of independent predictors in discriminating transient from persistent SSNs in patients with extrapulmonary malignancies. The AUC value of the combination of both clinical and CT predictors (AUC, 0.954) was significant higher than that of clinical predictors, alone (AUC, 0.805) (P < 0.05), and was notably higher than that of CT predictors, alone (AUC, 0.870), although the difference did not meet statistical significance.
Discussion
When evaluating therapeutic response in oncology patients, the evaluation of preexisting lesion change or detection of newly occurring lesions is critically important as the occurrence of new metastases or aggravation of preexisting malignancies is a constant concern in these patients. Thus, the accurate prediction of transiency of initially-detected pulmonary SSNs in these patients may help avoid incorrect disease status assessment as well as unnecessary alteration of ongoing treatment. Furthermore, it would help relieve the potential prolonged uncertainty and anxiety that oncology patients often experience.
In our study, we found that approximately half of the SSNs in patients with extrapulmonary malignancies were transient. Lee et al. reported similar results, albeit in a lung cancer screening population, stating that approximately 70% of initially-detected pulmonary part-solid nodules were transient because of inflammation or hemorrhage (10). Our results suggest that overlooking the possibility of transient SSNs could lead to mismanagement in a substantial proportion of patients with extrapulmonary malignancies. Therefore, meticulous evaluation of initially detected SSNs in these patients should be performed using a combination of the patients’ clinical as well as CT features as these transient SSNs could be very accurately differentiated using these features (AUC, 0.954).
There was some ambiguity in defining ‘transient SSNs’ in the case of individuals who underwent systemic chemotherapy during the existence of the SSNs. In our study, we categorized two SSNs in two individuals who underwent systemic chemotherapy during the existence of SSNs as “indeterminate nodules” as we were not able to safely state whether the decrease or disappearance of the nodules were caused by chemotherapy or through their natural courses. We determined that we could not exclude the possibility of the lesions being metastasis responding to the treatment or the possibility of being transient inflammations that tended to show spontaneous disappearance over follow-up. However, 10 SSNs in six individuals which disappeared or decreased in size during or after antimicrobial therapy were defined as “transient lesions”, not as “indeterminate nodules” as it is reasonable to consider that these lesions were caused by infection or inflammation. We believed that it would not be reasonable to consider such lesions as metastasis or primary lung cancers.
In terms of the detection mode of SSNs, the proportion of transient lesions among SSNs detected at follow-up CT was 0.95. The mean interval between previous CT and follow-up CT showing newly occurring SSNs was 216.17 days and the median size of transient SSNs found at follow-up CT was 12.0 mm. Thus, the calculated mean volume doubling time of follow-up detected SSNs was 27.86 days, which is too short to be metastasis. According to Chojniak et al., the mean tumor doubling time for pulmonary metastases is 118 ± 134 days (range, 22–930 days) (14). As for second primary lung cancers, the mean interval between the first extrapulmonary malignancy and second primary lung cancer has been reported to be 46.6 ± 21.5 months (15). Pulmonary metastases which manifest as SSNs are very rare (16) and most persistent SSNs are adenocarcinomas or precancerous lesions which have a very long volume doubling time (17,18). Therefore, it could be reasonable for us to think that SSNs found at short-term follow-up CT are neither likely to be metastases nor second primary lung cancers. Thus, when SSNs are newly detected at follow-up CT in patients with extrapulmonary malignancies, the probability of it being a transient nodule can be much higher than that of metastasis or second primary lung cancer.
Among thin-section CT features, lesion multiplicity and ill-defined border were revealed to be significant differentiating factors of transient SSNs from persistent SSNs in patients with extrapulmonary malignancies. This is similar to a previous study that was designed to determine the differentiating factors of transient SSNs from persistent SSNs in a lung cancer screening population (10). We believe that radiologists will be able to use these features in confidently diagnosing transient SSNs in patients with extrapulmonary malignancy.
With respect to the clinical features, blood eosinophilia showed very high specificity but low sensitivity for discriminating transient SSNs from persistent SSNs in the lung cancer screening population in a previous study (10). However, multivariate analysis in our study revealed that the blood eosinophil count and presence of eosinophilia were not significant independent predictors of transient SSNs. This result suggests that benign diseases manifesting as transient SSNs in patients with extrapulmonary malignancies might be different from that of a healthy screening population. In the lung cancer screening population, a significant proportion of transient SSNs are thought to be closely associated with eosinophilic pneumonia (9) although peripheral blood eosinophilia does not always imply pulmonary eosinophilia (19). On the other hand, in patients with extrapulmonary malignancies, other benign diseases such as organizing pneumonia or opportunistic infection in addition to eosinophilic pneumonia may have been the cause of transient SSNs due to the use of chemotherapy agents or the cancer itself.
Using logistic regression analysis, ROC analysis, and C statistic, we found several independent predictors of transient SSNs: detection at follow-up, multiple SSNs, and an ill-defined nodular margin. A combination of these clinical and CT features showed a statistically significant increase in predicting the performance of transient SSNs compared with clinical features alone. We believe that radiologists can discriminate transient SSNs very accurately with a combination of these factors, helping to reduce the misinterpretation of disease status in patients with extrapulmonary malignancies.
Differing from a previous investigation (10), our study included a substantial proportion of postcontrast CT examinations. According to Honda et al. (20), the measured volume of pulmonary solid nodules obtained using three-dimensional volumetric software increased after administration of contrast medium and the mean postcontrast volume ratio, meaning postcontrast over precontrast volume, was just 1.065 using the standard algorithm (20), which is too small for radiologists to misdiagnose persistent nodules as transient. However, to the best of our knowledge, there have been no studies investigating the postcontrast volume ratio of SSNs. Thus, a substantial proportion of postcontrast CT examinations may be a limitation of our study.
In addition, Lee et al. (10) reported that transient SSNs were more likely to have a greater solid proportion than persistent SSNs, which was not observed in our study. Such a difference might be due to the difference in our study populations, as our study was performed in an oncologic population rather than a screening population. We believe that further studies with a large population and prospective design would be necessary to better elucidate this issue.
Our study had several limitations. First, our study was a retrospective study. Thus, there may have been the possibility of selection bias. Second, we resolved any discrepancy between observers through consensus, and did not evaluate interobserver agreement in terms of interpretation of SSNs’ CT features. Third, in terms of the transiency of SSNs detected at follow-up, the interval between CT examinations was not uniform. Further prospective studies with uniform follow-up intervals may be necessary in a larger population to confirm our observation.
In conclusion, approximately half of the SSNs in patients with extrapulmonary cancers are transient and can be accurately diagnosed with a combination of their thin-section CT features and clinical information.
Footnotes
Funding
This study was supported by the Research Grant of the Korean Foundation for Cancer Research (grant number: CB-2011-02-01).