Noninvasive assessment of tumor grading and prediction of patient outcome were among the first applications of positron emission tomography (PET) in oncology (1). Since then, the relationship between tumor glycolytic activity, as measured by the metabolism of the glucose analogue fluorodeoxyglucose, and patient survival has been analyzed in a variety of malignant tumors. In non–small cell lung cancer (NSCLC), a series of studies has indicated that high tumor fluorodeoxyglucose uptake is associated with a poor prognosis, irrespective of tumor size or stage (2–6). Furthermore, studies have indicated that tumor fluorodeoxyglucose uptake is correlated with tumor cell proliferation and grading (7, 8), further supporting the concept that fluorodeoxyglucose-PET allows noninvasive assessment of the biological aggressiveness of the tumor tissue and provides prognostic information.
In this issue of Clinical Cancer Research, Vesselle et al. (9) report the results of a prospective study evaluating the relationship between tumor fluorodeoxyglucose uptake and patient survival in 208 patients with potentially resectable NSCLC. In accordance with previous studies, the authors found that “high” tumor fluorodeoxyglucose uptake [defined in this study as a standardized uptake value (SUV) of >7] was significantly correlated with patient survival, although the association was weaker than reported in previous studies (hazard ratio, 2.01). However, tumor fluorodeoxyglucose uptake was not an independent prognostic factor in a multivariate analysis, including tumor size (measured by computed tomography) and stage. Tumor stage was determined by fluorodeoxyglucose-PET and computed tomography and confirmed by histopathology.
Several statistical, methodologic, and biological factors need to be considered when trying to understand these unexpected results. The first question is whether the study was adequately powered to detect a significant correlation between tumor fluorodeoxyglucose uptake and survival. The article does not include a formal power analysis, but there is little doubt that the study had enough power to detect such a relationship in the whole group of patients because previous publications (3, 6) have unanimously reported that high tumor fluorodeoxyglucose uptake is a very strong prognostic factor (hazard ratio, >3). However, in patients within one stage group, the correlation between fluorodeoxyglucose uptake and patient survival was less strong or absent in some studies (2, 10, 11). Thus, the study by Vesselle may still have been underpowered to detect survival differences in the subgroup analyses.
It is also important to emphasize that most of the previous studies evaluating the correlation between tumor fluorodeoxyglucose uptake and patient survival were retrospective and used post hoc, data-driven definitions of “high” fluorodeoxyglucose uptake. Thus, they might easily have overestimated the prognostic significance of tumor fluorodeoxyglucose uptake because definitions of “high” and “low” fluorodeoxyglucose uptake were selected to maximize differences in patient survival. Furthermore, threshold values for a “high” tumor SUV range from 5 to 20 (2, 3, 5, 6, 12), suggesting a considerable heterogeneity of patient populations and/or data analysis. Methodologic differences include the use of average or maximum tumor fluorodeoxyglucose uptake, differences in the start of data acquisition, etc. A further concern about retrospective studies is the accuracy of tumor staging, especially in nonsurgically treated patients. This may have made stratification of patient survival by tumor stage inaccurate.
On the methodologic side, the study by Vesselle et al. (9) used partial volume-corrected SUVs in addition to raw SUVs, whereas most previous studies used only raw SUVs. Partial volume effects occur when lesions are smaller than the resolution of an imaging system. In PET, this causes the activity within a lesion to be blurred over an area larger than the actual lesion. This results in the activity within such a lesion being underestimated (13). For typical whole-body fluorodeoxyglucose-PET scans, partial volume effects become significant for lesions <2 to 3 cm, with the magnitude of the effect increasing with decreasing size of the lesion. For lesions with a diameter of <1.5 cm and a typical image resolution of 0.7 cm in clinical whole-body studies, partial volume effects become severe, resulting in a >50% underestimation of the true activity. Applied to NSCLC, this means that fluorodeoxyglucose uptake of T1 tumors (diameter, <3 cm) is systematically underestimated by PET imaging. Consequently, the SUV of a 1.5-cm T1 tumor will be significantly lower than that of a 5-cm T2 tumor, even if the true fluorodeoxyglucose uptake by these tumors is identical. Thus, partial volume effects can cause a spurious correlation between tumor fluorodeoxyglucose uptake and tumor size or stage. Because tumor size is a strong prognostic factor in NSCLC, partial volume effects would also lead to an apparent correlation between tumor fluorodeoxyglucose uptake and patient survival, even if the true fluorodeoxyglucose uptake of all studied tumors were identical.
SUVs of small lesions can be corrected for partial volume effects, if their size and shape are known (13). Vesselle et al. (9) used computed tomography measures of lesion size to estimate partial volume corrections and found that after correction, tumor SUV was no longer correlated with patient survival. Although this provides some evidence that in their patient population tumor fluorodeoxyglucose uptake was not a significant prognostic factor, some caution is required when interpreting partial volume-corrected SUVs. As correction factors are large for small tumors, even minor errors in measuring the diameter of lesions can result in significant random errors of partial volume-corrected SUVs. In addition, it is assumed that the size of the malignant tissue is accurately determined by computed tomography and that the tumor homogenously takes up fluorodeoxyglucose. However, a lesion seen on a chest computed tomography is frequently a composite of malignant tissue, necrosis, and atelectasis. In other words, partial volume correction attempts to eliminate an important bias of SUVs but at the cost of errors from the estimate and assumptions in partial volume corrections. This added noise may have limited the power of the study to find a correlation between partial volume-corrected SUVs and patient survival.
The use of fluorodeoxyglucose in detecting and assessing tumor grade is based on sound cancer biology, indicating that malignant transformation of cells is associated with increases in glucose transporters, amplification of glycolysis through elimination of the high ATP yield Kreb cycle (36 ATPs/molecule of glucose) and use of the low ATP yield of glycolysis (2 ATPs/molecule of glucose), and activation of the hexose monophosphate shunt to provide glucose for synthesis of DNA and RNA (14). However the exact relationship between the rate of glycolysis and meeting the needs of different and changing biological functions of malignant degeneration is not well understood. Thus, although it is conceptually attractive to assume that rate of proliferation of tumor cells is proportional to the overall rate of glycolysis, and therefore the metabolic rate of fluorodeoxyglucose, progressive malignant transformations of cells and intercellular networks can produce varying demands on glycolysis at different stages, although generally increasing degree of malignancy is associated with increasing glycolysis (14). In several tumor types, such as esophageal cancer (15), the intensity of tumor fluorodeoxyglucose uptake has not been found to be an independent prognostic factor.
Furthermore, the selected treatment may influence the correlation between the results of a diagnostic test and patient survival. In stage IV NSCLC treated with chemotherapy, the intensity of tumor fluorodeoxyglucose uptake has been reported to be negatively correlated with response duration and time to progression, consistent with the concept that high glucose metabolic activity reflects the biological aggressiveness of the tumor tissue. However, the intensity of tumor fluorodeoxyglucose uptake was also positively correlated with the response rate (10). In other words, metabolically active tumors tended to be more aggressive but responded better to chemotherapy. As a consequence of these two opposing trends, tumor fluorodeoxyglucose uptake was not predictive for overall survival in this study (10).
In summary, the study of Vesselle et al. indicates that the relationship between tumor fluorodeoxyglucose uptake and patient survival may be significantly “blurred” by partial volume effects. Based on these data, further studies evaluating the prognostic role of fluorodeoxyglucose-PET in unselected patient populations seem to be of limited value. To determine the clinical usefulness of fluorodeoxyglucose uptake as a predictive marker, future studies are necessary that focus on specific tumor stages and treatments, such as surgically treated stage I NSCLC, and use predefined criteria for “high” and “low” tumor SUVs. In addition, it will be important to add genotyping and phenotyping of tissues where possible to better characterize the relationship between the biology of cancer and the findings from fluorodeoxyglucose PET.