Staging by Positron Emission Tomography Predicts Survival in Patients With Non-small Cell Lung Cancer^*

Materials and Methods

All patients undergoing PET scanning for the evaluation of thoracic malignancy between July 1992 and November 1997 at the Wake Forest University Baptist Medical Center were identified through the computerized data bank of the PET center. Indications for PET, patient age, and testing date were recorded at the time of scanning. Specific a priori criteria were not used for the selection of patients for evaluation by PET. All scans were ordered by treating and evaluating physicians in efforts to diagnose or stage possible thoracic malignancy.

Imaging Procedures

In each patient, PET images of the thorax and the upper abdomen (liver and adrenal glands) were obtained with an ECAT 951 scanner (CTI; Knoxville, TN), with an in-plane resolution of approximately 6 mm. Patients underwent routine fasting for FDG studies. Using the protocol of our PET center, a set of 31 contiguous sections over a longitudinal axis of 10.4 cm was obtained per acquisition. Three contiguous acquisitions were obtained to encompass the thorax. Transmission scans were performed for attenuation correction before the injection of 10 mCi of FDG. PET images were obtained 45 min after FDG injection. Skin marks placed during the transmission scan were used to reposition correctly for the emission scan and ensure correct alignment for attenuation correction. A total acquisition time of 30 min (10 min per bed position) was used for both the transmission and emission scans. Emission data were reconstructed into transaxial, sagittal, and coronal planes for review. In addition, a rotating three-dimensional reconstruction image was generated and reviewed as a closed-loop video. A focus of activity was called positive by the interpreting radiologist based on the following: 1) the activity was judged to be substantially increased compared with normal tissue, and 2) the activity was visible in at least two adjacent tomographic sections in each of the three tomographic planes. Standardized uptake value was not used in this definition of PET positivity. The original, prospectively recorded PET scan reports, independent of CT or histologic data, were forwarded to the ordering physicians and used for staging, using the International Staging System.

Radiologists’ original prospectively recorded interpretations of CT scans of the chest were used for clinical staging. A GE 9800 Advantage scanner (General Electric; Milwaukee, WI) at the Wake Forest University Baptist Medical Center radiology department was the primary machine used for these examinations. Using a standardized lung cancer staging protocol, 7.0-mm slices at a pitch of 1.5:1 with IV contrast (when renal status allowed) were made from the lung apices to the top of the aortic arch and from the mid-cardiac region to the iliac crest. Three-millimeter slices at a pitch 1:1 were obtained from the top of the aortic arch through the mid-cardiac region (6 cm below the carina) for mediastinal evaluation. Independent of PET scan results, CT scans and reports were evaluated and staged according to the new 1997 International Staging Guidelines by three pulmonary physicians (T.W.M., D.P.D., R.C.). Staging for individual cases was agreed on by consensus. Lymph nodes were deemed positive if they exceeded 1.0 cm in short-axis diameter. When multiple pulmonary nodules were present on CT images and interpreted by the chest radiologist to represent possible metastatic disease, these were staged as T4 (within the same lobe as the primary) or metastatic (located within a different lobe as the primary) disease. Pleural effusions were considered to represent T4 disease. Distant lesions (adrenal glands or liver) were considered M1 disease if distant spread of tumor was suggested by the interpreting radiologist.

Surgical Staging

Independent of imaging results, each patient’s surgical stage was determined by review of all histologic data obtained from a computerized data management system maintained in the pathology department and from review of medical records. Only patients with definitive histologic staging were used in statistical analysis for the determination of the performance characteristics of both PET and CT.

Survival Data

Independent of imaging results or histologic findings, patients’ outcomes were assessed. Mortality information was obtained through the computerized database of the Cancer Center Tumor Registry of Wake Forest University Comprehensive Cancer Center. If follow-up information was not available, the patients or their families were contacted by telephone to obtain further information. Survival was defined as the time from PET until death or the last date of contact.

Statistical Methods

A patient’s primary disease was staged as 0 to IV independently by pathologic diagnosis, PET, and CT. Patients were considered negative for disease by pathologic examination, PET, or CT if disease was absent by each method. Otherwise the patients were considered positive for disease. Test performance characteristics for PET and CT were calculated using the pathologic diagnosis as the “gold standard.” Agreement of PET and CT with pathologic diagnosis was calculated as the percentage of times the test stage agreed exactly with the pathologic stage. κ statistics were calculated to quantify the degree of agreement after correcting for chance agreement.¹³ Survival estimates were obtained using the Kaplan-Meier method. Log-rank tests were used to assess the univariate significance of pathologic stage, PET stage, CT stage, and the various other covariates. Cox’s proportional hazards regression model was used to assess which variables were jointly predictive of survival.

Results

Patient Demographics

One hundred fifty-two thoracic PET scans were performed for the staging of possible NSCLC during the 5-year period of review. The median age (± SD) of patients was 64 ± 9 years. A tissue diagnosis of NSCLC was made in 113 of 152 patients (74%). In patients undergoing PET, definitive surgical (pathologic) staging was obtained in 86 patients (57%), of whom 75 patients (49%) had CT scanning reports available for review. In patients undergoing PET scanning more than once, only data from the initial study corresponding to the timing of the chest CT were used.

False-Positives and False-Negatives

Despite having PET scans interpreted as negative for cancer, four patients were found to have malignancy. Their cancer types included cystadenocarcinoma (n = 1), carcinoid (n = 1), and bronchioalveolar cell carcinoma (n = 2). Four patients had FDG uptake compatible with malignancy by PET, but histology confirmed nonmalignant conditions: foreign body-type granulomatous disease, a necrotic mass without malignancy, a benign infectious pseudotumor, and necrotizing aspergillosis.

Tumor Staging Results, Performance Characteristics of Imaging, and Clinical Correlations

PET and pathologic data were available for 86 patients, and the relationship of PET stage to pathologic stage is summarized in Table 1 . Histologic documentation of NSCLC was present in 74 patients (86%) and absent in 12 patients (14%). PET results were normal in 8 of the 12 patients without cancer (specificity, 67%) and were abnormal in 70 of the 74 patients with cancer (sensitivity, 95%). The positive and negative predictive values of PET were 70 of 74 (95%) and 8 of 12 (67%), respectively. The overall accuracy of PET for the detection of malignancy was 91%. PET scan stage correlated with pathologic stage in 48 patients (56%). PET scan stage was discordant with pathologic stage in 38 patients (44%) and was most often related to understaging by PET in 23 patients (27%). Staging agreed exactly or was off by one level for 71 of 86 patients (83%). The estimate of κ for PET and pathologic stages was 0.41 ± 0.07.

CT and pathologic data were available for 75 patients, and their relationships are shown in Table 2 . Histologic documentation of NSCLC was present in 64 patients (85%) and absent in 11 patients (15%). Of the 11 patients without cancer, CT scans were normal in 3 (specificity, 27%). Of the 64 patients with cancer, CT scans were abnormal in all (sensitivity, 100%). The positive and negative predictive values of CT were 64 of 72 (89%) and 3 of 3 (100%), respectively. The overall accuracy of CT for the detection of malignancy was 89%. CT scan stage correlated with pathologic stage in only 24 patients (32%). CT estimates of tumor staging were discordant in 51 patients (68%) and were most often related to overstaging by CT (31 patients, 41%). Staging agreed exactly or was off by one level for 41 of 75 patients (55%). The estimate of κ for CT and pathologic stages was 0.11 ± 0.07.

To determine the potential clinical implications of imaging results, patients were reclassified into surgical or nonsurgical groups according to stage as determined by histologic, PET, and CT diagnosis. (At our institution, stage III patients are not generally regarded as candidates for resection for cure.) On the basis of PET scanning, nine patients (10%) might have been considered as nonsurgical candidates (stage III or IV) when, in fact, a lesser stage was identified at surgery. Eleven patients (13%) were classified as surgical candidates (stage I or II) with more advanced disease being diagnosed histologically. CT scanning inaccurately categorized 24 patients (32%) as nonsurgical and 14 patients (19%) as surgical candidates when true status was that of a lesser or greater stage, respectively.

Nodal Staging Results, Performance Characteristics of Imaging, and Clinical Correlations

PET and pathologic data for nodal staging were available for 81 patients. The relationship between nodal stage determined by PET and surgical pathologic findings is shown in Table 3 . Nodal metastasis of NSCLC was present in 33 patients (41%) and absent in 48 patients (59%). PET scans were normal in 37 of the 48 patients without nodal disease (specificity, 67%) and were abnormal in 22 of the 33 patients with nodal disease (sensitivity, 77%). The positive and negative predictive values of PET were 22 of 33 (67%) and 37 of 48 (77%), respectively. The overall accuracy of PET for detecting nodal disease was 73%. Nodal stage by PET scan correlated with pathologic stage in 54 patients (67%). PET scan was discordant with histologic nodal stage in 27 patients (33%) and was most often related to understaging in 15 patients (19%).

CT and pathologic data for nodal staging were available for 72 patients. The relationship of CT nodal stage to pathologic staging results is shown in Table 4 . Nodal metastasis of NSCLC was present in 28 patients (39%) and absent in 44 patients (61%). CT scans were normal in 34 of the 44 patients without nodal disease (specificity, 77%) and were abnormal in 12 of the 28 patients with nodal disease (sensitivity, 43%). The positive and negative predictive probabilities of CT were 12 of 22 (55%) and 34 of 50 (68%), respectively. The overall accuracy of CT for the detection of nodal disease was 64%. CT scan nodal stage correlated with histologic nodal stage in 42 patients (58.3%). CT scan was discordant in 30 patients (42%) and was most often related to understaging in 19 patients (26%).

Relationship of Imaging Results to Patient Outcomes

Kaplan-Meier estimates of survival based on stage as estimated on the basis of histology, PET, and CT are shown in Figures 1 –3 , respectively. Overall extent of disease and tumor and lymph node staging, as determined by histologic examination, correlated with survival (p = 0.0001 and p = 0.0001). In addition, tumor and nodal stage determined by PET scanning were both predictive of survival (p = 0.0037 and p = 0.0023), but neither tumor nor nodal stage by CT was predictive of survival (p = 0.608 and p = 0.338). One-year and 3-year percent survival for staging by each modality are shown in Table 5 . When data were analyzed according to whether or not surgical resection of tumor occurred, PET remained predictive of survival only in persons in whom surgery was not performed. Statistical analysis demonstrated that although PET stage was univariately predictive of survival, when the histologic stage was also known, no additional predictive value was obtained.

Discussion

In patients with NSCLC, PET is not only highly sensitive for detecting and staging malignancy within the framework of the new international staging guidelines, but also serves as a predictor of long-term survival. This finding was true for both overall tumor stage as well as nodal stage, as determined by PET. Similar predictions for survival were not demonstrated when staging by CT scanning was placed in the same model.

During the past several years, the role of PET scanning in the detection and staging of thoracic and extrathoracic malignancies has increased dramatically. Initial investigations of the applications of PET in patients with bronchogenic carcinoma concentrated primarily on the evaluation of the solitary pulmonary nodule. Gupta et al⁵¹⁴ and Dewan et al⁶ reported that increased FDG uptake had a 95% accuracy in differentiating malignant from nonmalignant disease in patients with nodules < 3 cm in size and later reported that PET accuracy could be increased if patient age (> 60 years) and lesion size (> 1 cm) were incorporated into the evaluation. In 50 patients, Bury et al⁷ confirmed the high sensitivity (100%) and specificity (88%) of PET. Previous investigation at our comprehensive cancer center demonstrated that PET had a sensitivity, specificity, and accuracy of 78%, 81%, and 80%, respectively, for the diagnosis of mediastinal disease in patients without clinically evident mediastinal spread of cancer.⁸ Lewis et al¹⁰ reported that preoperative PET scanning revealed nonresectable cancers in almost 20% of patients believed to have resectable disease. Studies applying PET to the evaluation of extrathoracic metastasis of lung cancer are limited, and to our knowledge, no study has previously reported the clinical staging of cancer in these patients using this modality.¹¹¹²¹⁵ We used results obtained from PET and CT to stage cancer in patients using the new International Staging Guidelines and compared these findings with pathologic staging. PET had an overall sensitivity and specificity of 95% and 67%, respectively, for lung cancer staging.

One of the most important findings of this study was that PET scan staging was, like histologic stage, predictive of long-term survival. This relationship, true for both overall tumor stage as well as lymph node stage, could not be documented for staging on the basis of chest CT. Although this finding may seem intuitive, to our knowledge it has not been reported previously and has major clinical implications that, necessarily, will vary with institutional philosophies regarding the management of patients with stage III disease and the evolving role of mutimodality therapy. We believe that certain individuals, in whom advanced disease is not detected during PET imaging, have metabolically less active tumors. Such neoplasms would tend to grow more slowly and metastasize later, factors that portend longer survival. One question which arose during this investigation was whether PET might provide a greater predictor of extended long-term survival than would histologic staging. This hypothesis was evaluated using multivariate analysis. Although there was a trend toward multivariate prediction of survival by PET, this trend was not statistically significant. When both histologic stage and PET stage were known, the other modality did not predict an increased survival benefit. Our report provides clinicians with the first estimates of long-term survival on the basis of PET data.

To determine whether PET might have influenced clinical practice (if used alone for determination of surgical intervention), patients were categorized as surgical (no disease and stages I and II) and nonsurgical (stages III and IV) candidates. Using these criteria, nine patients (10%) might have been considered erroneously to be nonsurgical candidates, ie, staged as either III or IV by PET, when in fact surgical staging was that of a lesser stage. PET was more reliable than CT, which overstaged 32% of patients. In addition, 11 patients were staged by PET as surgical candidates, ie, staged as no disease or stages I or II, and surgical staging was that of a higher level. Thus, although PET demonstrates excellent accuracy in the detection and staging of malignancy, it should not be used in isolation, but rather, in combination with histologic and other clinical data when available.

Although our investigation confirms prior reports and provides new information about PET staging and survival, several limitations should be addressed. This experience at a single institution lacks the variability in patient population, physician practices, and imaging interpretation that might benefit a multicenter study. Despite the large number of patients undergoing PET, prospective interpretations of imaging procedures and histologic data were collected retrospectively, and no control group of patients managed without PET was used for comparison. It is likely that this referral population chosen for PET scanning represents a selected group, and the high diagnostic sensitivity of both CT and PET is consistent with such selection bias. Patients with more complicated clinical presentations, comorbid illnesses, or questionable tumor stage may have been chosen to undergo PET scanning to assist clinicians with decision making, influencing physician practice. Other patients, perceived to have more advanced disease by either PET scanning or CT scanning reports, may not have undergone invasive diagnostic procedures to confirm the true tumor stage.

Despite these concerns, these data suggest that PET, using metabolic activity to determine the extent of neoplastic involvement, represents a promising, noninvasive method of staging NSCLC in patients in a manner predictive of survival that has only been available previously through invasive techniques requiring histologic sampling. In certain patient populations, such as the otherwise healthy, low-risk patient with clinical stage I and II malignancy, PET may have little clinical indication. In other groups, PET provides an alternative nonintrusive method for staging disease, which also significantly predicts survival. PET may prove especially worthwhile in patients with comorbid illnesses, accentuated risks from surgery, or invasive diagnostic procedures, or in individuals with poor functional status and clinically suspected advanced disease. Prospective multicenter randomized controlled trials are necessary to address these possibilities, and to determine whether a PET-based management algorithm will contribute to better outcomes in patients with NSCLC.