Metabolic Tumor Volume in Lymphoma: Hype or Hope?

Follicular lymphoma (FL) is a heterogeneous disease with a spectrum of genetic, biologic, laboratory, and clinical features that determine the need for, and the variable response to, first-line therapy and, ultimately, patient outcome.^1–4 As in other lymphomas, [¹⁸F]fluorodeoxyglucose (FDG) –positron emission tomography (PET) and computed tomography (CT) imaging are widely used for staging and response assessment in FL. The intensity of FDG uptake—usually measured by standardized uptake values (SUVs)—varies widely among patients with FL, from barely detectable (corresponding to SUV numbers in the low single digits) to very intense (corresponding to high SUV numbers greater than 10). At the extreme points (very low uptake or very high uptake), the SUV may provide clinically relevant information. Low uptake is generally associated with low proliferation and less aggressive disease, while the opposite pertains to very high uptake, which may also herald transformation from indolent to aggressive lymphoma.⁵

Beyond SUV, other metrics for analysis of tumor FDG uptake include the metabolic tumor volume (MTV)⁶ and an index called total lesion glycolysis (TLG).⁷ MTV measures the volume of FDG-avid disease, for which three-dimensional regions of interest are drawn (typically by autosegmentation using computer software, with some manual adjustment by the reader) around individual lesions. Volumes for all lesions are then added to derive the total body metabolic tumor volume. TLG additionally considers the intensity of FDG uptake in each disease site (TLG = MTV × mean SUV within the lesion).

Historically, the interest in quantifying volumes of metabolically active disease was driven by three major intentions: to derive data that could be used for lesional dosimetry, to estimate patient prognosis, and to quantify the response to therapy. For many years, efforts to derive these quantitative indices remained an academic exercise that relied on homegrown software packages and manual contouring of each individual tumor site.^8–10 This only changed when computer algorithms and user-friendly, commercially available software packages became available for clinical research. Accordingly, the number of publications investigating the technical features and prognostic value of MTV has grown exponentially since the turn of the century, from fewer than 20 papers per year to more than 200 papers per year in 2015. Whereas most initial studies focused on solid tumors with regionally confined disease, the improvements in algorithms and automation in recent years have enabled the evaluation of total-body metabolic tumor volumes (TMTVs), even in patients with widespread systemic disease, including patients with lymphoma.^11–18

In the article accompanying this editorial, Meignan et al¹⁹ report on the prognostic value of TMTV in FL. The authors retrospectively compiled data from three clinical trials^20–22 and analyzed data from 185 patients. Using a threshold of 41% of maximum, the median TMTV in this population was 297 cm³. The authors used three different approaches to define the optimal cutoff for TMTV as a predictor of survival: X-tile analysis,²³ receiver operating curve (ROC) analysis, and restricted cubic spline. Of these, X-tile is the primary reliable source of cut-point definition. This method uses a training set and a validation data set, improving the robustness of the analysis. Splines are useful for modeling the relationship between TMTV as a continuous variable and survival time, but their contribution to optimal cut-point definition is minimal. The authors also seem to have used ROC analysis with survival as a binary end point, ignoring the follow-up time. This would be inappropriate.²⁴ By X-tile analysis, the authors derived the TMTV (510 cm³) that provided the best combined sensitivity and specificity for predicting progression-free survival (PFS). The 2-year PFS was 58% in patients with TMTV > 510 cm³, and 87% in patients with TMTV < 510 cm³. Interestingly, high-baseline TMTV and a persistently positive FDG-PET scan after induction therapy were both equally independent prognostic factors. Of course, baseline TMTV has the advantage of providing this prognostic information at the outset of treatment.

Shortcomings of the study by Meignan et al¹⁹ largely relate to the retrospective nature of the project. For instance, the vast majority of patients (1,634 of 1,819) from the three clinical trials were not eligible for analysis, presumably because baseline staging PETs had not been performed or were not available for electronic volumetric analysis. Patients were treated with three different drug regimens and 16 patients remained on 2-year maintenance therapy with rituximab. PET scans were obtained on scanners of different generations from three different vendors, which may potentially affect the calculation of SUV and, therefore, also the calculation of TMTV. Uptake times for FDG were not standardized; although only patients with scans obtained less than 90 minutes after FDG injection were included, the range or even median of uptake times was not provided. Because tumor FDG uptake, in general, increases with time after injection, lack of standardization of uptake times can lead to variable FDG SUV that is not due to differences in tumor biology but simply technical reasons.

It seems intuitive that patients with high tumor burden should be at higher risk for treatment failure and shorter survival than those with low tumor burden. Current staging systems are built on this paradigm, and prognostic scoring systems use clinical and biochemical surrogates (eg, lactate dehydrogenase levels) to account for overall tumor burden. Both the term “tumor burden” and the idea that this is associated with prognosis in patients with lymphoma date back at least 30 years.²⁵ Several prior studies investigated the prognostic utility of TMTV or TLG in lymphoma (Table 1). Potential shortcomings in most of these studies relate to a lack of standardization with regard to PET scanning and treatment regimens. In nearly all of these studies, so-called optimal cutoff values were derived from retrospective ROC analysis, the limitations of which were previously discussed. More appropriate and sophisticated statistical tests do exist^23,27,28 but are rarely used, thus limiting the value of many publications.

Table 1.

Studies on the Prognostic Value of MTV in Lymphoma

			Tumor Volume Parameters
Study	Type of Lymphoma	Patients (No.)	Median SUV	Threshold (%)	Median MTV (cm3)	Range (cm3)*	Predictors of PFS	Determination of MTV Cutoff
Kanoun et al15	HL	59	NR	41	117	4-1,611	MTV 225 cm3 yields	ROC analysis†, no validation sample
							4-year PFS 85% v 42%
Sasanelli et al17	DLBCL	114	NR	41	313	4-2,650	MTV 550 cm3 yields	ROC analysis, no validation sample
							3-year PFS 77% v 60%
Adams et al11	DLBCL	73	22.0	40	272	6-2,454	Neither MTV nor TLG predicted outcome	N/A
Mikhaeel et al16	DLBCL	147	27.2	41	595	2-7,360	MTV 396 cm3 yields	ROC analysis, no validation sample
							5-year PFS 92% v 42%
							Best predictive model combines MTV with i-PET Deauville score
Cottereau et al26	DLBCL	81	18	41	320	IQR: 106–668	MTV 300 cm3 yields	ROC analysis, no validation sample
							5-year PFS 75% v 42%
Schöder et al18	DLBCL	65	23.4	Various‡	226	9-3,453	MTV did not predict outcome	N/A
Ceriani et al12	PMBL	103	18.8	25	406	NR	MTV 703 cm3 yields	ROC analysis, no validation sample
							5-year PFS 97% v 60%
							TLG 5,814 yields
							5-year PFS 99% v 64%
Cottereau et al13	PTCL	108	14	41	224	3-3,824	MTV 230 cm3 yields	ROC analysis, no validation sample
							2-year PFS 71% v 26%
Meignan et al19	FL 1-3a	185	10.0	41	297	IQR: 135–567	MTV 510 cm3 yields	X-tile analysis
							2-year PFS 87% v 58%

Abbreviations: DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; HL, Hodgkin lymphoma; i-PET, interim positron emission tomography; IQR, interquartile range; MTV, metabolic tumor volume; N/A, not applicable; NR, not reported; PFS, progression-free survival; PMBL, primary mediastinal B-cell lymphoma; PTCL, peripheral T-cell lymphoma; ROC, receiver operator curve; SUV, standardized uptake value; TLG, total lesion glycolysis.

^*Unless given as IQR.

^†ROC analysis is performed to derive best combined sensitivity and specificity along the ROC; study population is then dichotomized by so-called optimal threshold derived from ROC analysis.

^‡Tested various proposed thresholds, including 41%.

To place the study by Meignan et al¹⁹ further into context, it may be helpful to shed some light on the technical aspects and potential pitfalls in calculating TMTV. The authors used a threshold of 41% of maximum for PET volume auto-segmentation, which originally dates back to phantom studies conducted about two decades ago.²⁹ With this approach, the calculation of MTV is based on finding the area of highest FDG uptake in a particular disease site (eg, with SUV 12), and then calculating a three-dimensional volume that encompasses all volume elements (voxels) up to a certain threshold. That is, with a threshold of 40% of maximum, all voxels with SUV > 4.8 would be included to calculate the MTV for this particular lesion. The outer boundaries of the volume can be confined to anatomic boundaries, as seen on the corresponding CT of the PET/CT scans. However, appropriate thresholds may well depend on primary tumor SUV and on the anatomic location of a lesion. The latter determines the contrast between FDG uptake in that lesion and regional background activity in the surrounding normal tissue. This is particularly important for lesions with relatively low FDG uptake. True SUV is underestimated in smaller lesions because of partial volume effects (the exact size up to which this underestimation occurs depends on the specific PET scanner and its recovery coefficient). Even for two equally sized lesions that differ only in their maximum SUV, the metabolic volume will differ considerably when the same percent-of-maximum SUV threshold is applied. This was recognized in a recent study in primary mediastinal B-cell lymphoma in which investigators chose a threshold of 25% of maximum for measuring MTV, on the basis of their observation that this provided the best agreement between metabolic and anatomic boundaries in large mediastinal masses with high SUV.¹² In other words, choosing a higher percent-of-maximum threshold would have led to calculating smaller metabolic volumes, potentially excluding large portions of the anatomic mass or areas of central necrosis. Finally, SUV measurements, and, hence, MTV calculations on the basis of fixed percent-of-maximum SUV thresholds, also depend on many other biologic and technical factors³⁰ that need to be standardized to assure reproducibility of measurements.

Several other methods for autosegmentation of PET volumes exist (eg, threshold-based, gradient-based, statistical, and texture-based methods). All have specific advantages and disadvantages, and none may be completely accurate in measuring tumor volumes in all organs and settings. However, one might argue that accuracy (ie, our ability to derive an MTV that reflects the true viable tumor tissue volume in cm³) is perhaps less important than reproducibility (ie, our ability to arrive at the same volume regardless of the equipment and software used for image acquisition, reconstruction, and analysis) as long as the outcome of our measurements yields prognostic information. For instance, Meignan et al¹⁹ used two different software programs for calculating MTV (PET-VCAR [GE Healthcare, Little Chalfont, UK] and Imagys [Keosys, Saint-Herblain, France]). To their credit, the investigators established the reproducibility of measurements in a random subset of 20% of scans with reasonable results.

Meignan et al¹⁹ pooled data from three clinical trials conducted across many countries and continents. Given the retrospective nature of the study, we must assume that the PET scanners were not cross-calibrated to ensure that the same SUV was measured when the same lesion was assessed on different scanners. Thus, while the authors can assure us that the TMTV calculations were reasonably reproducible when performed by different observers using two distinct software packages, it is unknowable whether the same TMTV would have been calculated if the same patient, under the same biologic conditions, would have undergone PET imaging on two different scanners. Finally, modern PET scanners using the time-of-flight technique (with or without point-spread function reconstruction) generally provide better detection of smaller FDG-avid lesions and often yield higher SUV numbers than scanners of older generations, in particular for smaller lesions.^31,32 Although application of appropriate image reconstruction and filtering techniques can help to standardize SUV measurements across participating sites in multicenter trials,³³ this obviously could not be done in the study by Meignan et al in view of its retrospective nature.

Is it too early to declare victory and call TMTV a new prognostic biomarker in lymphoma? Probably yes. As a single parameter, TMTV remains far from perfect for prognostication. In the Meignan et al study,¹⁹ added prognostic information was derived when TMTV was used in combination with the Follicular Lymphoma International Prognostic Index score. In other studies, a combination of TMTV with treatment response assessed on interim PET imaging led to better segregation of prognostic groups.¹⁶ Quantitatively, the situation also seems less clear; all proposed so-called optimal TMTV cutoffs (Table 1) were derived retrospectively, varied from study to study, and will require prospective validation. Importantly, cutoffs are only indicators of probability (in this case, the probability of poor PFS) and always reflect a trade-off between sensitivity and specificity. Cutoffs may also vary depending on the specific patient population, the range of tumor volumes in this population, and, possibly, the drug regimen used. The latter seems particularly important because biomarkers may potentially lose their prognostic power when standard treatment is changed to a more aggressive regimen,³⁴ or when novel targeted therapies are administered. Going forward, we should better understand the biologic reasons for why TMTV is associated with patient outcome and ask how these data could be put to good clinical use. Does a large disseminated tumor volume make a patient less likely to respond to a certain drug regimen and dose? Is this related to suboptimal drug concentration at each tumor cell? Is it not surprising that, currently, patients with a TMTV of 300 cm³ often receive exactly the same treatment (regimen and dose) as do patients with a TMTV of 3,000 cm³? Do patients with larger TMTVs simply need more of the same treatment, or would it be reasonable to offer these patients more aggressive therapies while sparing patients with lesser tumor volumes (and better prognosis) the adverse effects associated with such treatments? On the contrary, should patients with smaller TMTVs receive less chemotherapy?

On other fronts, it is becoming increasingly clear how the tumor microenvironment contributes to the prognosis and drug response in patients with FL,³⁵ and a “clinicogenetic” prognostic score incorporating a seven-gene signature has been proposed for better risk stratification.⁴ In the end, the most meaningful and actionable information will probably come from a combination of clinical, imaging, and biologic factors rather than from any single parameter alone. It will take some time until all of these novel imaging and biologic features are confirmed, tests become standardized and widely available, and results can be integrated into the routine armamentarium of the hematologic oncologist. In the meantime, we need hypothesis-testing studies that apply imaging features and biologic signatures for better risk stratification and treatment selection in patients with lymphoma. Despite their shortcomings, the data provided by Meignan et al¹⁹ move us one step along in this direction.