Fluorodeoxyglucose Uptake of Primary Non-Small Cell Lung Cancer at Positron Emission Tomography: New Contrary Data on Prognostic Role

Primary lung cancer remains the leading cause of cancer death in the United States, with an estimated 174,470 new patients diagnosed and 162,460 deaths expected in 2006 (1). Non-small cell lung cancer (NSCLC) accounts for 80% of these new cases and includes the following histologic types: adenocarcinoma, squamous cell carcinoma, large cell carcinoma, and mixed histologies.

Tumor stage is used to describe disease extent and is established by imaging with computed tomography (CT) and positron emission tomography (PET) as well as by surgical procedures. To date, stage is the strongest prognostic factor for NSCLC (2) and the most important variable in selecting appropriate treatment for these patients. However, survival odds by stage (2) clearly underscore that each surgico-pathologic stage is a heterogeneous population containing individuals at much higher risk of death than others in the same stage. In particular, one third of NSCLC patients present with early-stage disease (stages I, II, and IIIA-N₁: T₃N₁M₀; ref. 3), which permits primary surgical resection with curative intent. However, over one half of these early-stage resectable patients suffer a tumor recurrence and die of their cancer despite a complete and presumably curative resection (2). Clearly, for these patients, a strategy based on surgery alone is insufficient. Further risk stratification and prognostication is necessary to optimize patient care. This would be particularly relevant for this group of early-stage patients as recent clinical trials have shown that the administration of chemotherapy after NSCLC resection confers a survival benefit (4–6). However, due to the morbidity and cost associated with chemotherapy administration, it should be reserved for those patients at risk of recurrence. Identification of these patients will require improved prognostic assessment tools. Although biological markers have been sought for this purpose, none of clinical utility has been identified yet.

PET imaging with fluorodeoxyglucose (FDG) is now widely available as an accurate staging tool for NSCLC (7, 8). Furthermore, FDG uptake in the primary NSCLC tumor as measured by PET has been proposed as a prognostic measure (9–17). In six retrospective and three prospective studies of NSCLC patients imaged with FDG PET, a statistically significant association was found between overall survival and primary tumor FDG uptake (9–17). However, these studies varied greatly in their methodology, including such factors as disease stages included, tumor staging accuracy, FDG uptake measures, and analyse enformed. In particular, these studies included a wide range of primary tumor sizes without accounting for partial volume effects present when imaging small lung lesions in a PET scanner (18, 19).

The primary goal of this prospective study was to evaluate the prognostic significance of primary NSCLC FDG uptake in a carefully staged population while correcting for partial volume effects. Our ultimate goal was to determine whether FDG uptake had prognostic significance for surgically resectable patients, the group for which identification of a worse outcome could direct the administration of adjuvant chemotherapy.

Inclusion and exclusion criteria. This study was conducted under University of Washington Human Subjects Division approval. All patients were clinically diagnosed with NSCLC and had been referred to the thoracic surgery clinics at the University of Washington Medical Center or the Veterans Affairs Puget Sound Health Care System between February 1998 and August 2004. After a preliminary CT scan of the thorax, potentially resectable NSCLC cases were referred for ¹⁸F-FDG PET scan and approached for the study. If histology of the lesion was later confirmed to be other than NSCLC, or if histology could not be confirmed, the patient was excluded from the study. Patients with a history of prior cancer were allowed to participate in the study if they had been disease free for at least 5 years. All patients had to be able to tolerate surgical resection if determined to be resectable. Patients who underwent wedge resection or segmentectomy and were later found to have local recurrence were excluded from the study due to the non-curative resection. Patients were additionally excluded if they had a history of type I diabetes, weight over 350 pounds, inability to give informed consent, or were either pregnant or breast-feeding. The primary NSCLC lesion had to be >1 cm on the mediastinal windows of the chest CT to allow accurate partial volume correction of FDG uptake measures. With these criteria, 208 patients were included in the study and followed standard care management as previously described (18). Among the 208 participants, the subgroup of surgically staged and resected patients was further analyzed to evaluate the prognostic significance of FDG uptake in that particular group. Patients included in this resection-only subgroup met the following characteristics: no administration of any therapy for NSCLC before PET scan, no surgical staging before PET scan, and no administration of induction (neoadjuvant) or adjuvant therapy as part of care. With the application of these rigorous study inclusion criteria, 103 patients were enrolled in the resection-only subgroup.

Tumor size. Primary tumor size was calculated by averaging all three dimensions of the lesion as measured on the mediastinal windows of the chest CT.

PET imaging. All PET imaging was done using a dedicated whole-body PET tomograph (PET Advance, General Electric Medical Systems). All patients were instructed to fast for 12 h before tracer administration. A blood sample was obtained to screen for abnormally high plasma glucose levels. Twenty minutes before tracer administration, patients received 1 mg of i.v. lorazepam to help decrease artifactual uptake of ¹⁸F-FDG in the neck and upper thorax that could otherwise compromise PET image interpretation. ¹⁸F-FDG (259–407 MBq, 7–11 mCi) was infused over a period of 2 min with a Harvard pump (Harvard; ref. 18). After allowing patients to rest for 45 min, they were placed supine in the scanner with the thorax fitting within two contiguous 15-cm-wide tomograph fields of view. As previously described (18), imaging started with a 15-min emission scan over the primary tumor so as to control for the time dependence of the tumor SUV. The primary tumor SUV was, therefore, measured for all patients over the 45- to 60-min interval after FDG injection. The remainder of the imaging consisted of a 10-min long emission scan over the rest of the thorax, a 10-min scan over the abdomen, and 5-min scans over the pelvis and neck. Fifteen-minute-long transmission scans were performed over each of the two thoracic fields of view and over the abdomen.

All images were collected in two-dimensional mode with scatter septae in place. Real-time random correction was applied using a delayed coincidence window. The deconvolution-based scatter correction algorithm supplied by the manufacturer was also applied. Raw PET data were reconstructed with a 12-mm Hanning filter, 55-cm image diameter, and 128 × 128 array size, as per the standard filtered back-projection on the PET Advance system. Emission, attenuation-corrected, and transmission scans were reconstructed and reviewed on a dedicated workstation by the same experienced reviewer (H.V.) with the diagnostic chest CT made available (18). The primary tumor maximum pixel-standardized uptake value (maxSUV) was determined by drawing a region of interest over the primary tumor on the attenuation-corrected images as previously described (19). The SUV is defined as: SUV = C_(μCi/mL) × W_(kg) / ID_(mCi), where C is the radiotracer concentration in a voxel of tissue (μCi/mL), W is the patient weight (kg), and ID is the injected tracer dose (mCi). The maxSUV is the maximum pixel value within the entire primary tumor.

The partial volume corrected maxSUV (PVCmaxSUV) was also evaluated for each primary tumor, accounting for the normal background uptake of lung parenchyma, as previously described (19). FDG uptake in normal lung was measured as the average SUV within a large region of interest located at the same axial level as the primary tumor but away from it, the chest wall, and the mediastinum.

The PVCmaxSUV is defined as: PVCmaxSUV = background SUV + [(measured maxSUV − background SUV) / RC], where RC represents the recovery coefficient for a lesion of a given tumor size (18). With the reconstruction variables used, for lesions of diameter < 2.8 cm, background ¹⁸F-FDG uptake contributes to the measured tumor uptake because the recovery coefficient is <1 as previously described (19).

Other definitions of the SUV based on lean body mass (LBM) and body surface area (BSA) were also investigated for prognostic value. Additional glucose correction of the SUV was also explored. These were investigated using previously reported formulas (19). LBM was defined according to the formulas by Hume (20): for males, LBM (kg) = 0.32810 × body mass (kg) + 0.33929 × height (cm) − 29.5336; for females, LBM (kg) = 0.29569 × body mass (kg) + 0.41813 × height (cm) − 43.2933. The following additional measurements of FDG uptake were tested: (a) plasma glucose-corrected PVCmaxSUV (glucose-corrected PVCmaxSUV); (b) plasma glucose-corrected and partial volume-corrected maxSUV defined with LBM (glucose-corrected LBM-PVCmaxSUV) or BSA (glucose-corrected BSA-PVCmaxSUV).

Surgical staging. A rigorous imaging and surgical staging schema was performed for all potentially resectable cases as detailed in our prior report (18). Patients who were identified as having metastatic disease by ¹⁸F-FDG PET underwent further anatomic imaging or percutaneous biopsy to confirm stage IV disease. Patients identified as having pleural implants but no evidence of metastatic disease by ¹⁸F-FDG PET underwent a second review of their thoracic CT scan. If this review did not clearly identify pleural nodules, a confirmatory thoracoscopy was performed. Patients identified as having mediastinal disease by ¹⁸F-FDG PET in a location inaccessible by mediastinoscopy underwent thoracoscopy, mediastinotomy, and/or thoracotomy to determine nodal stage. All other patients underwent bronchoscopy and mediastinoscopy followed by thoracotomy for those without involvement of mediastinal nodes. The international staging system for lung cancer (2) was used in every case.

Pathology. All biopsy and resection specimens were reviewed by the pathology departments of the University of Washington Medical Center or the Veterans Affairs Puget Sound Health Care System to verify non-small cell histology and assess histologic subtype as well as tumor-node-metastasis status of each tumor.

Statistical methods. For all cases, overall survival time was calculated as the time interval between enrollment in the study (PET scan date) and last follow-up date. For subjects who died from lung cancer or other causes, the last follow-up date was defined as the date of death. For subjects who did not have a known date of death, December 15, 2004 was used as the last follow-up date. For those with a later communication with the study coordinator, that communication date was used as the last follow-up date. December 15, 2004 was chosen to allow for a 5-month lag before the last date the Social Security Death Index was searched (May 15, 2005). For the resection-only cases, overall survival time was calculated as the time interval between surgery date and last follow-up date. Disease-free survival time was calculated as the time interval between surgery date and date of first imaging confirmation of disease recurrence, death without evidence of recurrence, or last communication attesting to disease-free status. One patient was excluded from the disease-free survival analysis because the date of recurrence was unknown. Patients' clinical records and periodic detailed questionnaires sent to patients' physicians were used to establish these end points.

Univariable Cox's proportional hazards modeling was used to quantify the risk for death and recurrence from NSCLC for the following variables: age, stage, histology, tumor size, maxSUV, PVCmaxSUV, glucose-corrected PVCmaxSUV, glucose-corrected LBM-PVCmaxSUV, and glucose-corrected BSA-PVCmaxSUV. Subsequently, multivariable Cox's proportional hazards models were used to assess the potential independent effects of maxSUV (model 1) and PVCmaxSUV (model 2), adjusting for the effects of age, stage, histology, and tumor size, on risk for death or recurrence, in all subjects and in restricted analyses of subjects undergoing resection. From these models, risk estimates, estimated by hazard ratios (HR), and 95% confidence intervals (95% CI) were calculated. Kaplan-Meier methods were used to estimate survival probabilities. All analyses were conducted using SAS version 9.1 (SAS Institute, Inc.).

All NSCLC cases. The baseline characteristics of the 208 NSCLC cases are summarized in Table 1. Subjects with NSCLC had a mean age of 64 years, and only 9% of subjects were less than 50 years old. The majority (133, 64%) were male. Seventy-five patients (36%) had surgical stage I disease, 32 (15%) had stage II, 62 (30%) had stage III, and 39 (19%) had stage IV NSCLC. Fifty percent of patients had tumors ≤3 cm. By histology, 38% of tumors were adenocarcinomas, 27% were squamous cell carcinomas, 21% were large cell carcinomas, and 4% were bronchioloalveolar carcinomas. Eighteen subjects (9%) had NSCLC that was histologically unspecified or were of mixed NSCLC tumor types (NSCLC-NOS/mixed). Subjects were followed for a median time of 33.6 months [interquartile range (Q1-Q3), 9.1-43.5 months]; there were 90 deaths due to lung cancer during follow-up. The median follow-up time for the 118 surviving patients was 37.0 months (Q1-Q3, 17.3-57.4 months).

Subgroup of resection-only cases (no adjuvant or neoadjuvant therapy). This subset comprised 103 patients who underwent surgical resection, including 66 (64%) with stage I, 27 (26%) with stage II, and 10 (10%) with stage III NSCLC. Their mean age was 65.5 years, and 71 patients (69%) were male. The median follow-up time from surgery was 31.8 months (Q1-Q3, 17.3-53.4 months) and 24.2 months (Q1-Q3, 12.8-48.0 months) for overall survival and disease-free survival, respectively. In this resection-only subgroup, there were 26 recurrences of disease, 20 of whom subsequently died of NSCLC.

Determination of cutoff points for maxSUV and PVCmaxSUV. Because the goal of this study was to determine whether standardized uptake values were predictive of recurrence or survival of NSCLC, we performed sensitivity analyses to determine the maxSUV and PVCmaxSUV values that, in univariable Cox regression analysis, best differentiated those who did and did not recur or survive. Integer values between 3 and 15 were evaluated as potential cutoff points, separately for maxSUV and PVCmaxSUV. Ultimately, both factors were dichotomized with a cutoff point of 7 because for each factor, this cutoff proved to be the most discriminative cutoff for prognosis (based upon log-likelihood values) when analyzed univariably among all cases. However, all maxSUV values between 6 and 12 and also 15 gave significantly discriminative log-rank probability values (P < 0.05) for survival in univariable analyses. No integer value of PVCmaxSUV gave significantly discriminative log-rank probability values, although a cutoff point of 7 was the most prognostic and was marginally discriminative (P = 0.053).

Similar to what was done in all cases, we evaluated integer values between 3 and 15 as potential cutoff points in the resected cases, separately for maxSUV and PVCmaxSUV, both for the outcomes of death and recurrence. In the evaluation of death, no maxSUV values offered significant prognostic value, with all P > 0.3. The cutoff value with the highest risk estimate was 7, with a HR of 1.21 (P = 0.7). Similarly, no PVCmaxSUV values offered significant prognostic value, and all had P > 0.3. The PVCmaxSUV cutoff value with the highest risk estimate again was 7, with a HR of 1.19 (P = 0.8). The same cutoff values were subsequently used to analyze the subgroup of resection cases with respect to risk of disease recurrence.

Survival analysis of all NSCLC cases. The overall median survival time of all 208 NSCLC cases was 43.3 months. Univariably, increasing stage, increasing tumor size, and tumors with NSCLC-NOS/mixed histology were all strongly associated with an increase risk of death from lung cancer (Table 2; Fig. 1). MaxSUV ≥7 was significantly associated (HR, 2.01; 95% CI, 1.15-3.50), whereas PVCmaxSUV ≥7 was no longer significantly associated (HR, 1.85; 95% CI, 0.98-3.48) with an increased risk of death from lung cancer (Table 2; Fig. 1). However, in separate Cox regression multivariable analyses, neither maxSUV ≥7 nor PVCmaxSUV ≥7 were significantly associated with risk of death from NSCLC after adjustment for age, stage, histology, and tumor size, and neither maxSUV nor PVCmaxSUV provided additional significant prognostic information (HR, 1.25; 95% CI, 0.64-2.44 and HR, 1.45; 95% CI, 0.71-2.98, respectively).

Survival analysis for resection-only cases. Among patients who underwent surgical resection, the overall 5-year survival estimate (based upon Kaplan-Meier methodology) was 72%, and the 2-year disease-free survival time was 76%. In separate univariable Cox regression analyses (Table 3), only surgical stage was significantly associated with an increased risk of death from lung cancer after surgery, whereas neither max SUV ≥7 nor PVCmaxSUV ≥7 was significantly associated with lung cancer death (HR, 1.21; 95% CI, 0.47-3.16 and HR, 1.19; 95% CI, 0.40-3.57, respectively). After adjusting for age, surgical stage, histology, and tumor size, in separate multivariable Cox regression analyses, neither maxSUV ≥7 nor PVCmaxSUV ≥7 was significantly associated with risk of death from lung cancer after surgery (HR, 0.68; 95% CI, 0.14-3.40 and HR, 0.64; 95% CI, 0.12-3.33, respectively).

Disease-free survival analysis for resection-only cases. In univariable analyses, increasing surgical stage and tumor size were significantly associated with an increased risk of recurrence (Table 4), but neither maxSUV ≥7 nor PVCmaxSUV ≥7 were significantly associated with disease-free survival in resection-only cases of NSCLC (HR, 0.91; 95% CI, 0.40-2.05 and HR, 0.69; 95% CI, 0.29-1.65, respectively). After adjusting for age, surgical stage, histology, and tumor size, both maxSUV ≥7 and PVCmaxSUV ≥7 remained insignificant in multivariable analysis (HR, 0.34; 95% CI, 0.08-1.50 and HR, 0.31; 95% CI, 0.07-1.37, respectively).

Glucose-corrected SUV. Three glucose-adjusted partial volume-corrected (PVC) SUV definitions were additionally evaluated as potential prognostic indices of risk for death or recurrence from NSCLC, including glucose- and PVC-corrected maxSUV, glucose- and PVC-corrected maxSUV normalized to LBM, and glucose- and PVC-corrected maxSUV normalized to BSA. As with maxSUV and PVCmaxSUV, sensitivity analyses were performed to determine the best cutoff for each glucose-adjusted maxSUV, which differentiated those who did and did not subsequently die from NSCLC. In separate univariable Cox regression analyses of all cases, glucose-corrected PVCmaxSUV ≥7 (HR, 1.84; 95% CI, 1.04-3.26), glucose-corrected LBM-PVCmaxSUV ≥8 (HR, 1.69; 95% CI, 1.10-2.58), and glucose-corrected BSA-PVCmaxSUV ≥1,600 (HR, 2.10; 95% CI, 1.08-4.10) were all significantly associated with risk of death from NSCLC (data not shown). However, in multivariable Cox regression analyses, after adjusting for age, stage, histology, and tumor size, neither glucose-corrected PVCmaxSUV ≥7 (HR, 1.41; 95% CI, 0.73-2.72), glucose-corrected LBM-PVCmaxSUV ≥8 (HR, 1.10; 95% CI, 0.68-1.76), nor glucose-corrected BSA-PVCmaxSUV ≥1,600 (HR, 1.65; 95% CI, 0.77-3.52) remained significantly associated with risk of death from NSCLC.

In analyses of the resection-only subgroup, neither risk of death nor risk of recurrence after surgery was significantly associated with glucose-corrected PVCmaxSUV, glucose-corrected LBM-PVCmaxSUV, or glucose-corrected BSA-PVCmaxSUV in univariable or multivariable analyses (data not shown).

Our study clearly confirms the prognostic significance of tumor stage both in univariable and multivariable analyses for all cases as well as for the subgroup of resection-only cases. This is in keeping with published literature (2). Tumor histology did not have significant prognostic value with respect to risk of death in the entire cohort or for the subset of resected tumors. Tumor size > 5 cm, increased tumor stage, and age >70 years were found to be prognostic of risk of death for all cases in multivariable analysis.

One of the strengths of our study was the thorough staging that all subjects underwent. In particular, all of the resected cases were surgically staged in addition to FDG PET staging. Because tumor stage is a strong prognostic factor in NSCLC, any study evaluating potential prognostic measurements depends on thorough staging. Our multivariable statistical analyses included every variable found to be prognostic in univariable analysis.

When considering all cases, our study found a significant association between risk of death and tumor maxSUV in univariable analysis. However, after accounting for known prognostic factors, the prognostic value of maxSUV was no longer significant. This effect was observed in two separate ways. First, when maxSUV was analyzed in a multivariable model including tumor size, stage, age, and histology, the presence of these factors reduced the prognostic power of maxSUV to below the level of significance (Table 2). This effect is likely due to the fact that, as we have shown previously, maxSUV is positively correlated with both cancer stage (18) and tumor size (19). Second, when PVC was applied to maxSUV, correcting for the dependence of maxSUV on tumor size, the prognostic value of PVCmaxSUV was no longer significant in univariable or multivariable analysis (Table 2).

We have previously described PVC and its importance (19). Partial volume effects result from the limited reconstructed resolution of a PET scanner and artificially lower measured values of tumor uptake when tumor size is less than approximately 2 to 3 times the reconstructed resolution of the PET image, roughly 3 cm for most PET scanners. Therefore, not correcting for such effects may result in falsely attributing a biological significance to lower FDG uptake values found in small lesions. Consequently, the known prognostic significance of tumor size (21–23) is reflected as prognostic significance of tumor FDG uptake. Our PVC algorithm assumes that tumors are spherical and surrounded by uniform background activity. All of the tumors we studied were potentially resectable and, as such, were surrounded by normal lung or located at the pulmonary hilum. This ensured that the tumors were entirely or mostly surrounded by normal lung activity at PET. Furthermore, PVC applies only to tumors <3 cm in diameter, which are in general very close to spherical, unlike many larger tumors.

Among the patients who underwent complete and presumably curative resection (resection-only subgroup), neither the maxSUV nor the PVCmaxSUV predicted the risk of death from NSCLC after surgery or the risk of tumor recurrence, either in univariable or multivariable analysis (Tables 3 and 4). In many analyses, we even observed HR <1.0 for the high-uptake group. This lack of a consistent direction in the relationship between maxSUV and poor outcome strongly suggests that there is no prognostic effect of maxSUV in patients undergoing resection.

These conclusions for all patients and for the resection-only subgroup remain unchanged for the wide range of SUV definitions that we evaluated.

Our results are in contrast with those published in the literature. It should be noted that our cohort is unique in consisting of prospectively recruited patients who have undergone CT scanning and initial thoracic surgery evaluation and who are deemed potentially resectable by the surgeon before PET. To assess whether there may have been systematic differences in our study population compared with others', we compared our subjects' ages, stages, tumor sizes, treatment methods, follow-up, and survival times with those reported in previous studies (9–17). For all measurements, our subjects fell within the ranges reported previously. In terms of our PET methods, the differences in our study were (a) a 12-h period of fasting before PET, whereas other studies required only 4 to 6 h of fasting; (b) a 15-min emission over the primary tumor, longer than that used by most other investigators; and (c) a standard, 45-min uptake period between FDG injection and the start of the emission scan over the field of view containing the primary tumor. This uptake period was, therefore, centered on 52.5 min after injection, slightly shorter than most investigators and equivalent to others (9, 17). Our strict control of the uptake period and starting the scan over the primary tumor were chosen to provide less variability in our data, compared with other studies where the primary tumor could have been in the second, third, or fourth field of view, leading to different uptake periods depending on tumor location.

Potentially more significant were differences in data analysis. Among the three prospective and six retrospective studies reporting a significant prognostic value for tumor SUV (9–17), none used PVC to correct SUV values for tumor size, although ≥40% of tumors were <3 cm in all four studies reporting such data (9, 10, 13, 17). Furthermore, only two studies included tumor size in their multivariable analyses (9, 13), a necessary step to account for the dependence of SUV on size in small tumors. Finally, in the majority of studies (9–12, 14, 17), the use of data-driven optimal cutoff points to determine “high” versus “low” SUV values in the evaluation of FDG SUV as a prognostic factor resulted in an overestimation of the true significance of FDG SUV in predicting recurrence or survival (24). This has resulted in a wide range of SUV cutoff values that have been determined to have prognostic value, from as low as 5 in two studies (11, 17) to as high as 20 in another (12), adding uncertainty to the potential clinical usefulness of these study findings.

Ahuja et al. retrospectively studied the relationship between tumor FDG SUV and overall survival for 155 stage I to IV NSCLC patients and concluded that SUV >10 was associated with shorter survival (9). However, these authors also noted that this result was not consistent across tumor sizes. Specifically, when this study tested an interaction between tumor FDG SUV and tumor size in a multivariable analysis model, SUV >10 lost its prognostic significance for lesions <3 cm, the group in which tumor size and uncorrected SUV are correlated. In further agreement with our findings, the survival plot for their subset of stage I/II (resectable) tumors does not show prognostic significance for high versus low FDG uptake.

In a study of 125 stage I to IIIB NSCLC patients undergoing a variety of treatments, Vansteenkiste et al. showed that for the group of 91 stage I to IIIA patients who underwent surgical resection, lesions <3 cm in diameter had a 2-year survival of 86% if SUV <7 and 60% survival if SUV >7 (10). However, all lesions >3 cm had a SUV >7 and a 2-year survival of 43% (univariable analysis). These results underscore the effect of size on tumor SUV and survival, but the reported multivariable analysis did not include lesion size as a variable. No multivariable analysis of the prognostic significance of FDG uptake was reported for their 91 resected cases, the group of greatest interest from a clinical management perspective.

In a retrospective study of stage I to IIIB NSCLC patients separated into surgical and radiotherapy groups, Sasaki et al. found prognostic significance at maxSUV >5 for overall survival and disease-free survival (11). However, tumor stage was included in multivariable analysis as independent T and N descriptors rather than as a complete numerical stage, which is the most established predictor of prognosis in NSCLC (2). Tumor size, a strong prognostic factor in their univariable analysis, was not included in multivariable analysis. T stage should not be considered a sufficient replacement for tumor size, as these two measurements are only loosely associated.

Dhital et al. did not include tumor size or stage in their multivariable analysis of 77 patients (12). Downey et al. found no prognostic significance for FDG uptake in lesions <3 cm in their study of 100 patients (13). Among a total of 73 patients, Jeong et al. found prognostic significance for uncorrected maxSUV in adenocarcinomas but not in squamous cell cancers (14). These three series included stages I to IIIB or IV without specifically addressing the issue of recurrence risk in resected early-stage disease.

In a small study of 51 inoperable lung cancer patients treated with high-dose radiotherapy Borst et al. reported pretreatment lesion maxSUV as an important prognostic factor for survival and treatment response (15). However, their multivariable analysis did not take into account chemotherapy, tumor volume, or lymph node status, all found to be prognostic in their univariable analyses.

In the largest study to date, Cerfolio et al. (16) evaluated 315 patients in a retrospective database review and found that maxSUV ≥10 was associated with a significant reduction in disease-free survival in patients with stage IB and stage II NSCLC. However, tumor size was not taken into account in univariable or multivariable analyses. Multivariable analysis included stage as a dichotomized variable (stages I/II versus III/IV).

Only one survival study provided raw SUV data. In 57 stage I to IIIB tumors consisting mostly of adenocarcinomas, Higashi et al. reported a survival advantage for patients with FDG SUV <5 (17). However, when we applied our PVC method (25) to this data, the tumor FDG uptake was no longer a statistically significant predictor of survival, dropping from P < 0.001 to P = 0.08. This conclusion remained the same, regardless of whether normal lung background SUV was estimated at 0.25 or 0 for the PVC process. Although we applied correction coefficients generated for our GE Advance PET scanner to data acquired with another scanner, and the definition of SUV used by Higashi et al. (average of pixels >90% of maxSUV) differs slightly from our SUV definition (maxSUV), this re-analysis underscores the effect of not accounting for partial volume effects in survival studies.

In conclusion, considering the findings of our study and potential limitations of previous investigations, there is not sufficient evidence to suggest that FDG uptake in a primary NSCLC provides prognostic information beyond the prognosis conferred by tumor size and stage. Therefore, care should be exercised in not making treatment decisions based on the level of metabolic activity of the primary tumor at FDG PET. In particular, such measurements should not be used to identify patients at higher risk of recurrence after surgical resection and who could potentially benefit from adjuvant chemotherapy. A new approach is needed to identify these more aggressive early stage tumors. FDG PET remains an extremely accurate tool for staging lung cancer but treatment selection should not be altered on the basis of primary tumor FDG uptake.