See editorial on pages 1129–31, this issue.
Total lesion glycolysis in positron emission tomography is a better predictor of outcome than the International Prognostic Index for patients with diffuse large B cell lymphoma†
Article first published online: 4 DEC 2012
Copyright © 2012 American Cancer Society
Volume 119, Issue 6, pages 1195–1202, 15 March 2013
How to Cite
Kim, T. M., Paeng, J. C., Chun, I. K., Keam, B., Jeon, Y. K., Lee, S.-H., Kim, D.-W., Lee, D. S., Kim, C. W., Chung, J.-K., Kim, I. H. and Heo, D. S. (2013), Total lesion glycolysis in positron emission tomography is a better predictor of outcome than the International Prognostic Index for patients with diffuse large B cell lymphoma. Cancer, 119: 1195–1202. doi: 10.1002/cncr.27855
- Issue published online: 4 MAR 2013
- Article first published online: 4 DEC 2012
- Manuscript Accepted: 7 JUN 2012
- Manuscript Revised: 2 MAY 2012
- Manuscript Received: 18 MAR 2012
- diffuse large B cell lymphoma;
- positron emission tomography;
- total lesion glycolysis;
- Ann Arbor stage;
- International Prognostic Index
This study was undertaken to evaluate the prognostic value of quantitative metabolic parameters in [18F]2-fluoro-2-deoxyglucose (FDG)-positron emission tomography (PET) for diffuse large B cell lymphoma (DLBCL).
A total of 140 DLBCL patients underwent FDG-PET scans before rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisolone (R-CHOP) chemotherapy. The maximal standardized uptake value (SUVmax) and total lesion glycolysis (TLG) were calculated, with the margin thresholds as 25%, 50%, and 75% of SUVmax of all lesions. Treatment outcomes were compared between groups according to metabolic parameters and the International Prognostic Index (IPI).
After a median follow-up of 28.5 months (range, 5-81 months), the 2-year progression-free survival (PFS) and overall survival (OS) were 83% and 87%, respectively. Among metabolic parameters, TLG at the threshold of 50% (TLG50) was significantly associated with treatment outcomes. High TLG50 values (>415.5) were associated with reduced survivals compared with low TLG50 values (≤415.5) (2-year PFS of 73% versus 92%, P = .007; and 2-year OS of 81% versus 93%, P = .031). High IPI score (≥3) significantly reduced OS (2-year OS of 79% versus 90%, P = .049). Ann Arbor stage III/IV adversely affected PFS (P = .013). However, high IPI score and Ann Arbor stage of III/V did not significantly shorten PFS (P = .200) and OS (P = .921), respectively. High TLG50 values independently predicted survivals by multivariate analysis (hazard ratio = 4.4; 95% confidence interval = 1.5-13.1; P = .008 for PFS and hazard ratio = 3.1; 95% confidence interval = 1.0-9.6; P = .049 for OS).
Combined assessment of volume and metabolism (ie, TLG) is predictive of survivals in DLBCL patients who are treated with R-CHOP. Cancer 2013. © 2012 American Cancer Society.
Diffuse large B cell lymphoma (DLBCL) is a biologically heterogeneous subtype of non-Hodgkin's lymphoma (NHL)1 that is curable in approximately 60% to 80% of patients in the rituximab era.2, 3 Because of the biological heterogeneity of aggressive NHL, the International Prognostic Index (IPI) has been used to identify risk groups before treatment.4 The distinct subtypes of germinal center B cell–like (GCB) and activated B cell–like DLBCL were also identified as molecular predictors of survival.5, 6 Although the revised IPI7 or the absolute lymphocyte count/revised IPI model8 was proposed as a survival predictor, the standard IPI remains significant in the rituximab era.9
[18F]2-Fluoro-2-deoxyglucose (FDG)-positron emission tomography (PET) is widely used for pretreatment staging, restaging, response monitoring, evaluation of aggressiveness, and prognostication in malignant lymphoma.10-13 High FDG uptake or hypermetabolism of glucose is a surrogate marker for aggressive biology in NHL. In FDG-PET, a standardized uptake value (SUV) is usually used as a quantitative parameter of metabolism, and maximal SUV (SUVmax) of a tumor lesion has been proven to be a survival predictor in several solid tumors.14-16 Similarly, a high SUVmax on pretreatment PET is associated with poor outcomes in DLBCL patients who received rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisolone (R-CHOP) chemotherapy.17 However, because SUVmax reflects the metabolic activity of the most aggressive cells and thus lacks prognostic significance, total lesion glycolysis (TLG), calculated by mean SUV (SUVmean) × metabolic tumor volume (MTV), has been suggested as a quantitative parameter for FDG-PET.18 TLG in addition to SUVmax predicted response to radioimmunotherapy in NHL.19 However, the quantitative parameters of FDG-PET have not been comparatively investigated as prognostic factors in untreated DLBCL.
In this study, we evaluated the prognostic significance of quantitative metabolic parameters in pretreatment FDG-PET for DLBCL patients treated with R-CHOP.
MATERIALS AND METHODS
A total of 140 patients newly diagnosed with DLBCL at Seoul National University Hospital, Seoul, Korea, between September 2004 and July 2010 met the following inclusion criteria: 1) pathologically confirmed de novo DLBCL according to World Health Organization criteria1, 20 by specialized hematopathologists (Y.K.J. and C.W.K.); 2) first-line therapy with R-CHOP; 3) FDG-PET/computed tomography (CT) before R-CHOP; and 4) no central nervous system involvement by lymphoma. The staging work-up included complete blood count, blood chemistry, chest radiograph, CT scans of the neck, chest, and abdomen, FDG-PET/CT, and bone marrow examination. Clinical and pathologic factors were reviewed and retrieved such as age at diagnosis, sex, bulky disease (defined as tumors ≥ 10 cm in diameter), Eastern Cooperative Oncology Group (ECOG) performance score (PS), lactate dehydrogenase (LDH) levels, Ann Arbor stage, number of extranodal sites, IPI,4 revised IPI,7 DLBCL subtype by Hans' algorithm (GCB versus non-GCB),21 and Ki-67 index. This retrospective study was approved by the Institutional Review Board at Seoul National University Hospital (H-0707-016-212).
Treatments and Response Evaluation
Standard R-CHOP given in 3-week cycles consisted of intravenous rituximab at 375 mg/m2, cyclophosphamide at 750 mg/m2, doxorubicin at 50 mg/m2, and vincristine at 2 mg on day 1, and oral prednisolone at 100 mg on days 1 through 5. Patients received 6 to 8 cycles of R-CHOP, and radiotherapy at a dose of 36 Gy was delivered to primary bulky disease. Nonbulky stage I/II disease was treated with 6 cycles of R-CHOP or 3 to 4 cycles of R-CHOP followed by involved-field radiotherapy. Response was assessed on the basis of the modified International Workshop criteria, and complete remission (CR) was briefly defined as the complete disappearance of all detectable clinical evidence of disease and disease-related symptoms in the presence of posttreatment residual mass of any size with negativity upon PET scan.12
FDG-PET and Image Analysis
In all patients, FDG-PET images were acquired using dedicated PET/CT scanners (Gemini, Philips; and Biograph 40, Siemens) according to our standard imaging protocol, which is consistent with a major guideline for standard oncologic PET imaging.22 Briefly, after fasting for at least 6 hours, blood glucose level was checked to be less than 120 mg/dL. Patients were injected with FDG (0.14 mCi/kg), and image acquisition was started 60 minutes later. A CT scan was performed for attenuation correction and lesion localization, and an emission scan was performed from the skull base to the proximal thigh. PET images were reconstructed using iterative algorithms. PET/CT images were reviewed by experienced nuclear medicine physicians (J.C.P. with 11 years, D.S.L. with 17 years, and J.-K.C. with 17 years of PET experience), and all visually discernible hypermetabolic lesions were selected for analysis, excluding false-positive lesions with the aid of CT images.
On a lesion basis, SUVmax and TLG values with 3 variable margin thresholds were acquired for analysis. TLG of a lesion was calculated as SUVmean × MTV, which considers both the metabolic activity and tumor burden. MTVs were measured by setting the margin thresholds as 25%, 50%, and 75% of SUVmax of each lesion, resulting in TLG25, TLG50, and TLG75, respectively. These 3 variable margin thresholds were tested to find the optimal margin threshold. In the image analysis, a spheroidal volume of interest was drawn to include the entire lesion, and metabolic tumor margins were automatically drawn by a dedicated software package (TrueD; Siemens). Afterward, the values of SUVmax, MTV, and SUVmean were measured. When lesions were bulky and conglomerated so that discrimination of a lesion from each other was unavailable, the lesions were analyzed as a cluster. Measurement of these values is automatic in the software, and thus, has a reproducibility of 100%. Finally, on a patient basis, SUVmax of a patient was defined as the highest SUVmax among those of all lesions, and TLG of a patient was defined as the sum of those of all lesions.
Between-group differences in clinical and pathologic factors were evaluated by Pearson's chi-square test or Fisher's exact test, as appropriate. Overall survival (OS) was measured from the date of diagnosis to the date of death or the last follow-up visit. Progression-free survival (PFS) was defined as the time from the date of initial treatment to the date of disease progression, death, or the last follow-up visit. Survival curves were derived by the Kaplan-Meier method.23 Comparisons between the groups were made using the log-rank test. Factors independently associated with response and survival were identified by multivariate analysis using the logistic regression model and Cox proportional hazards regression model, respectively.24 Two-sided P values < .05 were considered significant. All statistical analyses were performed using SPSS, version 19.0 (IBM Corporation, Armonk, NY).
Clinical Features and Treatment Outcomes
The clinical and pathologic characteristics of the 140 DLBCL patients are shown in Table 1. The median age was 59 years (range, 16-80 years) with a male:female ratio of 1.1:1. CR was achieved in 114 (81%) of 140 DLBCL patients, of whom 12 (11%) subsequently relapsed after frontline R-CHOP treatment. After a median follow-up of 28.5 months (range, 5-81 months), the 2-year PFS and OS in 140 DLBCL patients were 83% and 87%, respectively. Factors associated with a lower probability of achieving CR were the presence of B symptoms (hazard ratio [HR] = 3.2, 95% confidence interval [CI] = 1.3-7.9; P = .010) and ECOG PS ≥ 2 (HR = 6.5; 95% CI = 2.3-18.1; P < .001) by univariate analysis. ECOG PS ≥ 2 was an independently significant factor in multivariate analysis.
|Characteristic||No. of Patients Available||No. of Patients (%)|
|≤60 y||78 (56)|
|>60 y||62 (44)|
|Presence of B symptoms||140|
|Ann Arbor stage||140|
|2 or more||20 (14)|
|Number of extranodal sites||140|
|2 or more||28 (20)|
|Bone marrow involvement||140|
Comparison of Treatment Outcomes Based on Metabolic Parameters
Treatment outcomes were compared according to the quantitative metabolic parameters in Table 2. Median values of SUVmax, TLG25, TLG50, and TLG75 were 16.4 (range, 2.0-26.8), 817.8 (range, 9.56-2362.9), 415.5 (range, 4.74-1498.6), and 102.0 (range, 1.12-302.23), respectively, and these were used as cutoff values for group discrimination. The attainment of CR was not statistically different between the high- and low-metabolism groups, as determined by the median value. Disease progression was significantly higher in high-TLG groups than in low-TLG groups. However, SUVmax did not influence disease progression. A high TLG50 value exhibited a trend of positive correlation with death. The PFSs were inferior in the high-TLG groups compared with those in the low TLG groups, resulting in a high HR for disease progression. A high TLG50 value significantly reduced OS, unlike high SUVmax, TLG25, and TLG75 values. Clinical and pathologic factors were compared between the high- and low-TLG50 groups (Table 3). A high TLG50 was significantly associated with adverse prognostic factors including advanced stage, poor PS, elevated LDH levels, bulky disease, presence of B symptoms, high IPI score, and high Ki-67 index.
|No. of Patients (%)|
|Mean ± SD||16.5 ± 7.7||2216.1 ± 3022.3||1267.1 ± 1824.4||288.0 ± 399.9|
|Attainment of CR|
|No||12 (17)||14 (20)||.664||10 (14)||16 (23)||.192||10 (14)||16 (23)||.192||11 (16)||15 (21)||.385|
|Yes||58 (83)||56 (80)||60 (86)||54 (77)||60 (86)||54 (77)||59 (84)||55 (79)|
|No||60 (86)||59 (84)||.813||64 (91)||55 (79)||.033||65 (93)||54 (77)||.009||65 (93)||5 (77)||.009|
|Yes||10 (14)||11 (16)||6 (9)||15 (21)||5 (7)||16 (23)||5 (7)||16 (23)|
|No||62 (89)||62 (89)||1.000||65 (93)||59 (84)||.276||66 (94)||58 (83)||.104||64 (91)||60 (86)||.560|
|Yes||8 (11)||8 (11)||5 (7)||11 (16)||4 (6)||12 (17)||6 (9)||10 (14)|
|HR for PFS, 95% CI||1.2, 0.5-2.7||.725||2.8, 1.1-7.1||.035||3.6. 1.3-10.0||.012||3.5, 1.3-9.5||.015|
|HR for OS, 95% CI||1.0, 0.4-2.6||.971||2.4, 0.8-6.8||.112||3.2, 1.0-10.0||.042||1.7, 0.6-4.8||.288|
|Number of Patients (%)|
|Clinical Factors||Low TLG50 (≤415.5)||High TLG50 (>415.5)||P|
|≤60 y||42 (60)||36 (51)|
|>60 y||28 (40)||34 (49)|
|Ann Arbor stage||<.001|
|I/II||49 (70)||28 (40)|
|III/IV||21 (30)||42 (60)|
|0-1||65 (93)||55 (79)|
|≥2||5 (7)||15 (21)|
|Normal||41 (61)||9 (13)|
|Elevated||26 (39)||61 (87)|
|Number of extranodal sites||.205|
|0-1||59 (84)||53 (76)|
|≥2||11 (16)||17 (24)|
|No||68 (97)||56 (80)|
|Yes||2 (3)||14 (20)|
|Presence of B symptoms||<.001|
|No||63 (90)||41 (59)|
|Yes||7 (10)||29 (41)|
|Mean ± SD||52.3 ± 25.0||63.3 ± 20.9||.008|
|GCB||10 (24)||11 (31)|
|Non-GCB||31 (76)||24 (69)|
|0-1||44 (64)||15 (21)|
|2||13 (19)||26 (37)|
|3||9 (13)||16 (23)|
|4-5||3 (4)||13 (19)|
Survival Analysis and Prediction of Survival
Ann Arbor stage III/V adversely affected PFS (2-year PFS of 73% versus 91%, P = .013; Fig. 1A) compared with Ann Arbor stage I/II, whereas it did not shorten OS (2-year OS of 88% versus 87%, P = .921; Fig. 1B). The IPI score was able to identify 4 risk groups for PFS (P = .003; Fig. 2A) and showed a trend toward risk group separation for OS (P = .065; Fig. 2B). However, the IPI score did not discriminate between high-intermediate and low-intermediate risk groups (2-year PFS 87% versus 70%; P = .195; and 2-year OS 85% versus 82%;, P = .793). Similarly, the revised IPI did not separate patients with DLBCL into 3 risk groups for OS (very good versus good versus poor, 2-year of OS 88% versus 91% versus 79%, respectively; P = .140) and for PFS (very good versus good versus poor, 2-year PFS of 95% versus 82% versus 77%, respectively; P = .268). In contrast, high TLG50 values significantly reduced PFS and OS compared with the findings for low TLG50 values (Fig. 3A,B). Representative PET/CT scans of patients with favorable and poor outcomes were drawn in Fig. 4.
The clinical factors associated with reduced PFS in univariate analysis were high TLG25 (HR = 2.8, 95% CI = 1.1-7.1; P = .035), high TLG50 (HR = 3.6, 95% CI = 1.3-10.0; P = .012), and high TLG75 values (HR = 3.5, 95% CI = 1.3-9.5; P = .015). Multivariate analysis indicated that high TLG50 values (HR = 3.6, 95% CI = 1.3-10.0; P = .012) were independently significant for reduced PFS. The IPI score (0-2 versus ≥ 3) did not predict PFS (HR = 1.8; 95% CI = 0.7-4.3; P = .208).
Regarding OS, the factors associated with poor survival were high TLG50 values (HR = 3.3, 95% CI = 1.0-10.0; P = .042) and high IPI score (≥ 3) (HR = 2.6, 95% CI = 1.0-6.9; P = .059). High TLG50 values remained independently significant in the multivariate analysis (HR = 3.1, 95% CI = 1.0-9.6; P = .049). In addition, high TLG50 retained a predictive capacity for PFS (HR = 2.8, 95% CI = 1.0-8.0; P = .047) and OS (HR = 3.6, 95% CI = 1.1-11.5; P = .031), regardless of Ann Arbor stage.
Our study demonstrated that metabolic tumor burden expressed as TLG could predict survivals in DLBCL patients treated with R-CHOP. It also demonstrated that TLG50 is the optimal technical measurement for TLG. High TLG50 values were associated with adverse prognostic factors, which resulted in reduced survivals compared with the findings for low TLG50 values. In addition, high TLG50 values were independently predictive of reduced PFS and OS. Although Ann Arbor stage and IPI score did not predict survivals, the study population was under-represented for patients with Ann Arbor stage III/IV and high IPI score.
In FDG-PET, which evaluates the glucose metabolism of malignant lesions, SUVmax has been the most widely studied metabolic parameter because it is the most reproducible parameter and can reflect the metabolic activity of the most aggressive cell component. However, SUVmax has limited prognostic value because it lacks information regarding the tumor burden, another important factor for prognosis. To complement such a limitation, TLG, which is calculated as a product of MTV and mean SUV within the volume, was suggested as a quantitative parameter for FDG-PET.18 Thus, TLG can reflect both the metabolic activity and tumor burden.
Thus far, the pretreatment TLG or change in TLG has been evaluated as a prognostic factor for survival or treatment response in various solid tumors.15, 25-29 High TLG values were predictive of survival in small cell lung cancer,28 esophageal cancer,29 and osteosarcoma.15 In addition, increases in TLG after chemotherapy were associated with a shorter PFS in osteosarcoma.15 Regarding the treatment response, volume-based parameters (MTV and TLG) were independent predictors for tumor progression in malignant pleural mesothelioma,27 whereas changes in TLG did not predict tumor response in breast cancer with bone metastases25 and soft-tissue sarcomas.26 The cumulative TLG of pretreatment FDG-PET scans accurately predicted response to 90Y-based radioimmunotherapy in various types of NHL.19 TLG also reflected the dose-response relationship in a Daudi-inoculated mice model treated with cyclophosphamide.30 However, the prognostic significance of TLG at diagnosis has not been verified in malignant lymphoma, including DLBCL.
In our study, we assessed the prognostic value of TLG in comparison with conventional prognostic factors such as IPI and the Ann Arbor staging system. In addition, as a technical aspect, we tried to find the optimal margin threshold for TLG or MTV measurement, because no standard method currently exists for TLG or MTV. Although measurement with a relative margin threshold to SUVmax has been the most commonly used method, the optimal margin threshold has scarcely been studied. Thus, we tried several margin thresholds on a quartile basis, and as a result, TLG50 was selected as the optimal threshold. In calculation of MTV, an extremely high threshold excludes a moderately hypermetabolic component from the MTV, which consequently underestimates tumor burden. Contrarily, an excessively low threshold overestimates tumor burden. Taken together, this study showed that margin threshold of 50% is more representative of the physiology than others.
Even in comparison with other prognostic factors, TLG50 was the most significant factor for treatment outcomes in DLBCL patients treated with R-CHOP. TLG50 at diagnosis was an independent predictor for PFS as well as OS. Furthermore, high TLG50 values were significantly associated with advanced stage, elevated LDH levels, bulky disease, and high Ki-67 index, suggesting a large tumor volume. Although SUVmax independently predicted survival in DLBCL, an SUVmax cutoff value of 30 did not provide a fair balance, and SUVmax ≥ 30 did not significantly separate the PFS curve.17 In contrast, SUVmax did not predict the treatment outcomes of DLBCL patients who received R-CHOP in our analysis, probably because the cutoff of SUVmax was determined by median value rather than receiver operating characteristic curves. This implies that tumor volume and overall metabolic activity of the volume, or metabolic tumor burden, may be more important than the highest metabolic activity within the tumor in predicting treatment outcomes in DLBCL. However, the unknown results of changes in TLG after R-CHOP therapy and molecular subtypes by gene expression profiling might weaken the prognostic significance of TLG in our analysis. In addition, cross-calibration of 2 PET scanners from different manufacturers, retrospective study design, and high proportion of low-risk patients might compromise our results.
Since the introduction of IPI in the pre-rituximab era,4 it has retained predictive capacity for survival in DLBCL patients treated with R-CHOP or R-CHOP–like regimens.9 Similarly, IPI correlated with survival outcomes in this study. However, the IPI score did not discriminate between risk subgroups for survivals. Thus, it would be desirable to design a new integrative prognostic factor that includes conventional factors and metabolic parameters.
Our study indicated that metabolic tumor burden expressed as TLG is predictive of survivals, and a high TLG is significantly associated with reduced survival in DLBCL patients treated with R-CHOP. The IPI and Ann Arbor staging system do not appear to predict survivals, and IPI does not discern between risk subgroups for survival in our study. Pretreatment TLG as a metabolic prognostic factor will be a great help to identify risk subgroups and design treatment plans in DLBCL. Therefore, an aggressive treatment might be helpful to DLBCL patients with high TLG. Future efforts should be made to validate the prognostic value of TLG even in a high-risk group or to develop a new prognostic model that incorporates TLG in DLBCL in a prospective study.
This study was supported by grants from the Innovative Research Institute for Cell Therapy, Republic of Korea (A062260) and the Korea Healthcare Technology R&D Project, Ministry of Health and Welfare (A070001).
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosure.
- 1Diffuse large B-cell lymphoma, not otherwise specified. In: Swerdlow SH, Campo E, Harris NL, et al., eds. WHO Classification of Tumours of Haematopoietic and Lymphoid Tissues. 4th ed. Lyon, France: IARC; 2008: 233-237., , , .
- 18Tumor treatment response based on visual and quantitative changes in global tumor glycolysis using PET-FDG imaging. The visual response score and the change in total lesion glycolysis. Clin Positron Imaging. 1999; 2: 159-171., , , et al.
- 23Nonparametric estimation from incomplete observation. J Am Stat Assoc. 1958; 53: 457-481., .
- 24Regression models and life-table. J R Stat Soc B. 1972; 34: 187-120..