Interim PET Analysis in First-Line Therapy of Multiple Myeloma: Prognostic Value of ΔSUVmax in the FDG-Avid Patients of the IMAJEM Study Free

The last decade has seen the growing use of PET using ¹⁸F-fluorodeoxyglucose (FDG-PET) for the staging and assessment of therapy in multiple myeloma (1, 2). Several studies have shown the prognostic value of this imaging technique. At baseline, the number of focal lesions (FL; refs. 3, 4), the maximum standardized uptake value (SUVmax; ref. 4), and volume-based parameters (5) have been reported to be associated with shorter progression-free survival (PFS) in multiple myeloma patients. Similarly, FDG-PET normalization before maintenance has been demonstrated to favorably affect PFS (3, 4, 6), despite different criteria of normalization among studies. For instance, in the French IMAJEM (IMAgerie du JEune Myélome) study, a visual assessment using liver background as cutoff was used to distinguish PET positivity and negativity (6). A significantly improved PFS was reported in patients, whose disease became PET-negative before maintenance (6), whereas after only three courses of lenalidomide, bortezomib, and dexamethasone (RVD) as initial therapy, the PFS was improved among patients with a negative PET, but the improvement was not statistically significant. Quantitative measurements may improve the prognostic value of interim PET analysis, as demonstrated in lymphoma (1, 7–9). The purpose of this study was to assess the prognostic value of interim FDG-PET performed after three courses of chemotherapy analyzed using decrease in SUVmax in the FDG-avid multiple myeloma patient population included in the French prospective multicenter IMAJEM study.

The present study is based on the IMAJEM trial, which is part of the IFM/DFCI2009 study (ClinicalTrial.gov identifier: NCT01309334). The aims, inclusion, and exclusion criteria have previously been reported (10). The IFM/DFCI clinical trial is aimed at evaluating the combination of RVD versus RVD plus autologous stem-cell transplantation, followed by lenalidomide maintenance in patients with de novo multiple myeloma, younger than 65 years of age. A subgroup of 134 multiple myeloma patients were enrolled into the prospective IMAJEM trial with the primary aim of comparing the detection rate of bone lesions using MRI and FDG-PET at diagnosis, after three cycles of RVD and before maintenance. (6) All patients signed a written-informed consent form. The IMAJEM study was locally approved by the institutional ethics committee (University Hospital, Nantes, France) and in accordance with the precepts of the Helsinki declaration.

For this analysis, FDG-PET images performed at diagnosis and following three cycles of RVD (interim FDG-PET) were considered. Each of the 18 centers involved in this study applied their own acquisition procedure for patient imaging. Briefly, all patients fasted for at least 4 hours before the examinations. The blood glucose level required prior to FDG administration was set to be ≤150 mg/dL. Whole-body imaging was performed between 60 and 80 minutes after injection of FDG (from 3 to 7 MBq/kg), and the same procedure was repeated for each of the three PET/CT examinations required in this protocol. FDG-PET data from the 18 centers were centrally collected and analyzed. For the present analysis, only patients with FDG-avid bone marrow FL, defined as uptake higher than liver background, were considered. In these patients with FL, diffuse bone marrow involvement (BMI) and extramedullary disease (EMD) were also described (6).

For baseline and interim FDG-PET images, the most intense bone marrow FL (not necessarily the same at baseline and at interim analysis) was considered for SUV measurement, and the percentage difference of SUVmax (ΔSUVmax) between baseline and interim FDG-PET was calculated as previously described (11). In cases where the lesions disappeared after three cycles of RVD, the SUVmax was set at 1. Patients with a new lesion at 3 months (suspected progressive disease) were excluded from this analysis.

In addition to SUVmax, a visual analysis using the standard Deauville 5-point scale (DS; ref. 12) on the most intense residual lesion was done during the interim FDG-PET analysis, as proposed in the IMPeTUS Italian criteria (13), and patients were classified as DS 1–3 or DS 4–5.

For the purpose of comparison, the ΔSUVmax between the most intense lesion (either FL, BMI, or EMD) at baseline and after three cycles of RVD (not necessarily identical) was also considered. For this analysis, and in order to minimize false-positive results due to a BMI signal after chemotherapy, we chose to disregard patients with the highest SUVmax regarding BMI after three cycles of RVD when their baseline BMI had been considered negative.

Because SUVmax values may harbor significant bias and because of potential variance due to the multicentric nature of this study (14), a normalization of the SUVmax was performed against the liver SUVmax (ratio between FL SUVmax and liver SUVmax), as already successfully done in interim PET analyses in patients with Hodgkin lymphoma (15). The percentage difference of this normalized SUVmax was depicted as ΔrSUVmax in our study and assessed using the same population as for ΔSUVmax. A similar analysis was also performed considering ΔrSUVmax between the most intense lesion (either FL, BMI, or EMD) at baseline and after three cycles of RVD as described previously.

PFS was the period from the end of the three cycles of RVD to the date of documented progressive disease, death from any cause, or last follow-up. Association with PFS was examined using a univariate Cox regression model and considering quantitative parameters (SUVmax, ΔSUVmax, or ΔrSUVmax) as dichotomized variables. Because PFS was significantly prolonged in the transplantation group of the IMAJEM population (ref. 6; median, 50 months vs. 36 months, respectively), the analysis was also conducted adjusted for treatment arms. Age, hemoglobin, the revised International Staging System (R-ISS), and biochemical response after three cycles of RVD according to International Myeloma Working Group (IMWG) criteria (7, 16) were also considered in this univariate analysis. A correction for multiple testing was subsequently performed using the Benjamini–Hochberg approach. Parameters demonstrating significant association with PFS at univariate level were then evaluated using a stepwise multivariate Cox analysis including no more than four variables given the number of events observed (20 progressions) and the rule of thumb (at least 5 to 10 events per variable). The R-ISS and biochemical response were also included in the multivariate model. The optimal ΔSUVmax threshold was derived from the maximally selected rank statistics (17) from the “maxstat” R package and used to compare high- and low-risk groups for quantitative parameters by choosing the cutoff point associated with the most significant relation with outcome. The HR and associated P values were derived for these dichotomized quantitative parameters. The survival curves were drawn using the Kaplan–Meier method and the difference evaluated with the exact log-rank test. A value of P < 0.05 was considered statistically significant. All tests were conducted with R version 3.4.1 (R Foundation).

Among the 134 patients enrolled in the IMAJEM study, 71 patients with lesions more intense than liver background were considered for this analysis (Supplementary Fig. S1). Patient demographic and clinical characteristics were similar to those of the entire IMAJEM population (6). Thirty-eight patients were treated in arm A and 33 in arm B. Our cohort included 71 patients with FL (100%), 46 patients (65%) with diffuse BMI, and 6 patients (8%) with EMD at baseline. Median baseline FL SUVmax was 6.4 (ranging from 3.5 to 27.1) with 65 (92%) of the 71 FDG-avid multiple myeloma patients presenting a SUVmax value higher than 4.2. On interim FDG-PET analysis, 45 patients (64%) still presented with FL, 25 (35%) patients had BMI, and 2 patients (2%) had EMD. Median residual FL SUVmax was 1.7 (ranging from 1.0 to 15.0).

The characteristics of the studied parameters are summarized in Table 1. The median follow-up time was 21.5 months. Twelve progressions occurred in arm A, while 8 occurred in arm B.

The univariate Cox regression analysis for survival outcomes considering each parameter is shown in Table 2. The only significant parameter for PFS was ΔSUVmax (P = 0.003 after multiple tests correction) calculated using the most intense FL. Two examples are shown in Supplementary Fig. S2. It is interesting to note that interim visual analysis using DS 1–3 versus 4–5, interim SUVmax, and known prognostic factors, such as R-ISS and biochemical response [very good partial response (VGPR) or better], after three cycles of RVD in the same population did not reach significance in the univariate analysis (Table 2).

The Kaplan–Meier analysis for ΔSUVmax resulted in a median PFS of 22.6 months for those patients in the high-risk group (ΔSUVmax > −25%). The median PFS has not been reached in the low-risk group (ΔSUVmax ≤ −25%) with an HR of 4.41 (P < 0.001; Fig. 1). Results were similar when adjusted for treatment arm (P = 0.005).

In the multivariate analysis, only ΔSUVmax [P = 0.001; HR = 5.56; 95% confidence interval (CI), 1.96–15.81] and biochemical response (VGPR or better) after three cycles of RVD (P = 0.025; HR = 0.29; 95% CI, 0.1–0.85) appeared as independent prognostic factors with a more discriminative HR for ΔSUVmax.

Incorporation of BMI and EMD in the analysis did not alter the results (Table 3) except for only ΔSUVmax being retained in the multivariate analysis (P = 0.023; HR = 3.41; 95% CI, 1.18–9.84).

A similar evaluation was conducted using liver normalization of SUVmax (Table 3). The survival curves for ΔrSUVmax exhibited a similar trend with a good stratification between the high- and low-risk groups (HR = 3.26; 95% CI, 1.22–8.73; P = 0.007) and a median PFS of 22.6 months versus not reached, respectively. Again, a similar trend was reported when taking into account BMI, EMD, and normalizing by the liver SUVmax (Table 3).

This study demonstrates the added prognostic value of a quantitative measurement of early treatment-induced changes in SUVmax after three treatment cycles compared with visual analysis in patients with FDG-avid multiple myeloma. To our knowledge, this article is the first to report the relevance of SUVmax variation in multiple myeloma patients treated homogeneously in a prospective clinical trial in the setting of front-line intensive therapy, including proteasome inhibitors and immunomodulatory drugs. It is difficult to compare our results with previously published studies due to differences in therapies and PET interpretation criteria applied. Usmani and colleagues reported a significant prognostic value of FDG-PET after 7 days of an induction therapy using two cycles of VTD-PACE (bortezomib, thalidomide, dexamethasone; 4-day continuous infusions of cisplatin, doxorubicin, cyclophosphamide, etoposide; Total Therapy 3 protocol, P = 0.0003). The prognostic PET biomarker in the study by Usmani and colleagues was the persistence of more than three FL (18). In contrast, the FDG-PET results after three courses of RVD in the IMAJEM study affected PFS without reaching significance; however, the PET criterion considered was normalization (6).

In the present study, we hypothesized that a semiquantitative method could potentially be more sensitive than visual analysis to evaluate early response to chemotherapy, as has been shown in lymphoma (19–22). Indeed, the ability to identify slow and fast responders is important, as early tumor shrinkage is associated with a deeper response (23). We decided to compare the standard DS validated in lymphoma to SUVmax and ΔSUVmax. Visual analysis also appeared relevant for application in multiple myeloma as suggested by the recently proposed IMPeTUS criteria (13), even if not yet prospectively validated. As previously done in lymphoma (19, 20), only multiple myeloma patients with FDG-avid lesions at baseline (uptake higher than liver background) were included in the analysis, considering a low baseline SUVmax as a drawback for ΔSUVmax calculation (24). This subgroup might represent a necessary limitation of our study; however, patients with low FDG-avid multiple myeloma may harbor a more indolent form of the disease (2). Only 6 (8%) of our patients presented a SUVmax lower than the previously reported (4) prognostic threshold of 4.2. Moreover, patients showing new lesions after three cycles of RVD suggesting progression were not included in the analysis of ΔSUVmax. Considering the specific population included in our study, a semiquantitative analysis using ΔSUVmax was superior to interim SUVmax and visual analysis based on DS criteria in predicting PFS. This strong prognostic value was observed in both treatment arms and independent of biochemical response assessed according to the IMWG criteria (7). Results have yet to be interpreted with caution due to the relatively smaller number of patients with biochemical response assessment after 3 cycles of RVD (n = 57). It was also confirmed using a normalized ΔrSUVmax allowing for a reduction in the variability of SUV measurements. Although reactive bone marrow hyperplasia induced by G-CSF treatment or due to repopulating marrow in patients with anemia is not rare after chemotherapy, the ΔSUVmax calculation remained significantly associated with PFS when BMI SUVmax was considered in the quantitative analysis in patients with BMI at baseline.

Our results are consistent with previous findings in lymphoma. Indeed, in patients with FDG-avid multiple myeloma, the ΔSUVmax method may represent the dynamic and continuous process of metabolic reduction of tumor cells during treatment more accurately than a visual scale (1, 20). Moreover, it was found to have higher interobserver reproducibility and outcome prediction performance than DS in patients with FDG-avid lesions (8, 11, 19).

This study had some inherent limitations as it explored two sets of parameters which do not necessarily reflect the same situation at baseline and after three cycles of RVD. It was retrospectively designed to address the question of the prognostic value of prospectively measured markers, which may be useful for imaging-guided therapy in combination with baseline risk factors. At the current time, prognostic parameters are identified at diagnosis, before the onset of therapy. The aim is to use these parameters to define risk-adapted therapies, or to devise trial protocols focusing on patients with specific risk factors. Of note, other prognostic parameters can now be defined at specific time points during therapy. These “dynamic” prognostic parameters, such as response after induction or imaging biomarkers such as ΔSUV, could be used in the future to change or adapt therapy in case of suboptimal response. Our study shows the importance of PET in this dynamic assessment of response, which could be complementary to the assessment of MRD within the bone marrow by next-generation flow cytometry or next-generation sequencing.

In this study, ΔSUVmax was a powerful tool to predict long-term outcomes in patients with FDG-avid multiple myeloma and superior to visual assessment. This study shows the potential added value of integrating FDG-PET alongside known prognostic factors, such as R-ISS and biochemical response, in the management of multiple myeloma. This prognostic biomarker should be confirmed in other cohorts of multiple myeloma patients.

C. Hulin reports receiving other remuneration from Celgene, Janssen, Amgen, and Takeda. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Carlier, X. Leleu, D. Caillot, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

Development of methodology: T. Carlier, B. Jamet, D. Caillot, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Bailly, T. Eugene, C. Touzeau, C. Hulin, T. Facon, X. Leleu, A. Perrot, L. Garderet, M. Macro, D. Caillot, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Bailly, T. Carlier, B. Jamet, X. Leleu, D. Caillot, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

Writing, review, and/or revision of the manuscript: C. Bailly, T. Carlier, B. Jamet, C. Hulin, X. Leleu, A. Perrot, M. Macro, D. Caillot, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Carlier

Study supervision: T. Carlier, P. Moreau, F. Kraeber-Bodéré, C. Bodet-Milin

This work was supported by PO1-155258 and P50-100707 from the French Ministry of Health, Soutien aux Techniques Innovantes Coûteuses 2010 Cancer STIC 10/03.

This work has been supported in part by grants from the French National Agency for Research, called “Investissements d'Avenir” IRON Labex n° ANR-11-LABX-0018-01 and ArronaxPlusEquipex n° ANR-11-EQPX-0004, and by grants from DHU Oncogreffe and SIRIC ILIAD (Imaging and Longitudinal Investigations to Ameliorate Decision-making in Multiple Myeloma and Breast Cancer).