PET/CT-biomarkers enable risk stratification of patients with relapsed/refractory diffuse large B-cell lymphoma enrolled in the LOTIS-2 clinical trial

Abstract

Purpose:

Significant progress has occurred in developing quantitative PET/CT biomarkers in diffuse large B-cell lymphoma (DLBCL). Total metabolic tumor volume (MTV) is the most extensively studied, enabling assessment of FDG-avid tumor burden associated with outcomes. However, prior studies evaluated the outcome of cytotoxic chemotherapy or chimeric antigen receptor T-cell without data on recently approved FDA agents. Therefore, we aimed to assess the prognosis of PET/CT-biomarkers in patients treated with loncastuximab tesirine.

Patients and Methods:

We centrally reviewed screening PET/CT scans of patients with relapsed/refractory DLBCL enrolled in the LOTIS-2 (NCT03589469) study. MTV was obtained by computing individual volumes using the SUV ≥4.0 threshold. Other PET/CT metrics, clinical factors, and the International Metabolic Prognostic Index (IMPI) were evaluated. Logistic regression was used to assess the association between biomarkers and treatment response. Cox regression was used to determine the effect of biomarkers on time-to-event outcomes. We estimated biomarker prediction as continuous and binary variables defined by cutpoints.

Results:

Across 138 patients included in this study, MTV with a cutpoint of 96ml was the biomarker associated with the highest predictive performance in univariable and multivariable models to predict failure to achieve complete metabolic response (odds ratio =5.42; P= 0.002), progression-free survival (hazard ratio (HR)= 2.68; P= 0.002), and overall survival (HR= 3.09; P<.0001). IMPI demonstrated an appropriate performance; however, not better than MTV alone.

Conclusions:

Pretreatment MTV demonstrated robust risk-stratification, with those patients demonstrating high-MTV achieving lower responses and survival to loncastuximab tesirine in relapsed/refractory DLBCL.

Introduction

Diffuse large B-cell lymphoma (DLBCL) represents the most common non-Hodgkin lymphoma (1). Patients with DLBCL are commonly treated with anthracycline-based regimens, achieving long-term remission in more than 60% of the cases (2,3). However, many patients subsequently relapse or do not respond to frontline therapies, demonstrating divergent outcomes. In patients with relapsed/refractory DLBCL, treatment options include chimeric antigen receptor (CAR) T-cell, platinum-based chemotherapy followed by autologous stem cell transplant (ASCT) and tafasitamab with lenalidomide depending on the response to first- and second-line therapy, time to relapse, and fitness for cellular approach (4–8). In the third line, two antibody-drug conjugates are presently available. Loncastuximab tesirine comprises a humanized anti-CD19 monoclonal antibody stochastically conjugated to a pyrrolobenzodiazepine dimer cytotoxin SG3199 capable of achieving an overall response rate (ORR) of 48.3% with complete response (CR) rate of 24.1% (9,10). Polatuzumab vedotin is another antibody-drug conjugate targeting CD79b to deliver a microtubule polymerization inhibitor and, in combination with bendamustine and rituximab, achieves an ORR of 41.5% with a CR rate of 38.7% (11). More recently, several CD20xCD3 bispecific antibodies demonstrated remarkable efficacy in single-arm clinical trials with epcoritamab and glofitamab receiving regulatory approval, further expanding the treatment landscape of relapsed/refractory (rel/ref) DLBCL (12–14). However, no predictive biomarkers are presently available to guide treatment selection or optimal sequence, remaining an unmet need in the field.

Over the last years, quantitative PET/CT-biomarkers have emerged as important factors for predicting treatment response and survival in DLBCL. Total metabolic tumor volume (MTV) has been the most studied among them. MTV enables assessment of FDG-avid total nodal and extranodal tumor burden associated with individualized estimates of survival in untreated and rel/ref patients (15–17). Furthermore, the recently proposed International Metabolic Prognostic Index (IMPI), including MTV with clinical factors such as stage and age, represents a significant advance to broadly integrate PET/CT-biomarkers in lymphoma clinical research (18). Despite considerable progress evaluating PET/CT-biomarkers in DLBCL, all these studies evaluated response and survival to cytotoxic chemotherapy or CAR T-cell without data on the role of these biomarkers with recently FDA-approved novel agents. Therefore, we aimed to assess the value of PET/CT-biomarkers to predict response and time-to-event outcomes to loncastuximab tesirine in patients enrolled in the LOTIS-2 trial (NCT03589469) with the goal of identify which patients will benefit the most from this agent.

Methods

Study population

LOTIS-2 was a single-arm, open-label multicenter study that enrolled 145 patients with rel/ref DLBCL after two or more lines of systemic therapy. Eligibility criteria were previously described but briefly included measurable disease defined by the 2014 Lugano Classification, Eastern Cooperative Oncology Group (ECOG) performance status of 0 to 2, and adequate organ function (9). Patients with bulky disease, defined as a tumor ≥10cm in the longest dimension, ASCT within 30 days, allogeneic stem cell transplant within 60 days, and active central nervous system lymphoma represented major exclusion criteria. The study was performed in accordance with the Declaration of Helsinki, regulatory requirements, and institutional review boards and ethics committees approved the protocol at individual sites. All patients provided written informed consent. Treatment consisted of single agent loncastuximab tesirine 0.150mg/kg every three weeks for two cycles, then 0.075mg/kg every three weeks for subsequent cycles for up to 1 year or until relapse or progression of disease, unacceptable toxicity, or death.

PET/CT analysis

In this post-hoc analysis, we centrally reviewed screening PET/CT images of patients enrolled in the LOTIS-2 trial. Regions of interest were set by manual adjustment in three planes to exclude adjacent physiologic FDG-avid structures. The maximum standardized uptake value (SUVmax) was defined as the maximum voxel intensity within the volumetric region of interest. Bone marrow involvement was included in volume measurement only if there was focal uptake. The spleen was considered involved if there was focal or diffuse uptake higher than 150% of the liver background (16,19). MTV was obtained by summing the metabolic volumes of all individual lesions using the previously reported SUV ≥4.0 threshold (18,20,21). Total lesion glycolysis (TLG) was calculated as the product of MTV and SUVmean for each individual lesion and then summed together. We also evaluated the prognosis of lesion dissemination, defined as the largest distance between two lesions that are farthest apart normalized by the body surface area (SDmax) and the distance between the bulkiest lesion and the lesion furthest away (SDmax bulk) (22). Following prior publication those with a single lesion or lesion conglomerate were excluded from SDmax and SDmax bulk analysis (23). Hermes Affinity Viewer software was implemented for these calculations. MTV values calculated by the first reader (R.A.K.) were independently confirmed by the second readers (M.N. and S.L.). All nuclear medicine readers were blinded for patient outcomes.

Statistical analysis

Descriptive statistics were used to summarize baseline characteristics and PET/CT biomarkers. One-way analysis of variance (ANOVA) was performed to compare the mean difference of the baseline PET/CT-biomarkers by loncastuximab tesirine treatment response. When the overall difference was significant, pairwise comparison was performed using Bonferroni correction. Response to treatment was analyzed with three groups: 1) complete metabolic response (CMR), 2) partial metabolic response (PMR), and 3) no metabolic response (NMR)/disease progression (PD)/not evaluable (NE) grouped due to similar overall survival (OS).

Logistic regression was used to assess the association between PET/CT biomarkers and binary treatment response (failure to achieve CMR vs. CMR (reference)). Cox regression was used to determine progression-free survival (PFS) and OS predictors. We tested if these quantitative metrics presented as continuous variables in the original scale or as log₂-transformed data, could predict treatment response, PFS, and OS. We used an outcome-oriented method to determine cutpoints (c) for risk-stratification into low (biomarker <c) and high (biomarker ≥c) risk groups. Cutpoints for PFS were determined using the Contal and O’Quigley method, which is based on the log-rank test (24). The same cutpoints were used for OS and treatment response (CMR status). For the binary treatment response (failure to achieve CMR (non-CMR) vs. CMR), the performance of logistic regression models was assessed using Akaike information criterion (AIC) statistics and the area under the receiver operating characteristic curve (AUC). For PFS and OS, the Integrated AUC (IAUC), computed as a weighted average of the AUC values at all the event times, was calculated to assess the predictive performance of Cox regression models. PFS and OS of the patients in the low-risk and high-risk groups were estimated using the Kaplan–Meier and compared using the log-rank test. Based on selected MTV multivariable models, we constructed nomograms for predicted probability of non-CMR and predicted PFS and OS rates at 6 and 12 months. We also assessed the performance of IMPI as a predictor of outcomes in this dataset, which was calculated as previously described (18). Analyses were conducted in SAS version 9.4 (SAS Inc., Cary, NC) and R version 4.0.3.

Data availability statement

Proposals requesting deidentified participant data collected for the LOTIS-2 study following publication can be sent to clinical.trials@adctherapeutics.com and will be evaluated on a case by-case basis. Derived data supporting the findings of this study should be requested to corresponding author.

Results

Baseline PET/CT biomarkers

Of 145 patients enrolled in LOTIS-2, we retrieved and reviewed 138 (95%) screening scans (Supplementary Figure S1). Baseline characteristics at the time of study enrollment included most patients were male (n= 82; 59.4%), ≥60 years (n= 89; 64.5%), presented with elevated serum LDH (n= 94; 68.1%), good ECOG performance status (0–1) (n= 132; 95.7%), and advanced-stage disease (III-IV) (n= 105; 76.1%). The best response to loncastuximab tesirine in 138 included patients was as follows: 23.9% CMR (n= 33), 24.6% PMR (n= 34) and 51.5% NMR/PMD/NE (n= 71) (Supplementary Table S1 and S2). Across all patients, the median SUVmax was 24.4 (interquartile range (IQR) 16.2 to 31.6). The median MTV, TLG, SDmax (n= 106), and SDmax bulk (n= 106) were 157 ml (IQR 33.8 to 463.3), 1538.5 (IQR 236.5 to 4199), 202.3 (IQR 111.6 to 311.1), and 181.1 (IQR 105.2 to 315), respectively. The median IMPI was 1.3 (IQR 0.9 to 1.8) (Supplementary Table S3). MTV reproducibility among readers was high, with Pearson correlation coefficient of 0.997 (Supplementary Figure S2A). MTV and TLG were highly correlated with one another (Pearson correlation of 0.964; Supplementary Figure S2B), and MTV was a slightly better predictor of outcomes in univariable and multivariable models. Thus, we reported only multivariable models, including MTV. We observed similar results between SDmax and SDmax bulk (Pearson correlation= 0.961; P<.0001) (Supplementary Figure S2C).

Association with response to treatment

We first assessed the association of different PET/CT-biomarkers with treatment response. Patients achieving CMR had significantly lower SUVmax (mean= 19.2, SD= 10.6), _log2MTV (mean= 4.8, SD= 3.1), _log2TLG (mean= 7.5, SD= 3.6), and IMPI (mean= 1.0, SD= 0.5) compared to those achieving PMR or NMR/PMD/NE (P<0.010). Neither SDmax nor SDmax bulk demonstrated a significant association with response (Supplementary Table S4). The duration of CMR by MTV and other baseline PET/CT variables categorized as low vs. high groups is depicted in Supplementary Table S5.

We then fit logistic regression models assessing the association between PET/CT biomarkers and non-CMR versus CMR (reference) (Table 1). Univariable logistic regression models evaluated independent PET/CT-biomarkers, clinical factors, and IMPI as continuous and dichotomous variables using cutpoints determined for PFS. We observed good model prediction of several biomarkers including _log2MTV (1-unit increase in log₂-scale odds ratio (OR)= 1.45, P<.0001; AUC= 0.770), _log2TLG (1-unit increase in log₂-scale OR= 1.40, P<.0001; AUC= 0.765), IMPI (1-unit increase OR= 3.26, P= 0.003; AUC= 0.696), and SUVmax (1-unit increase OR= 1.06, P= 0.003; AUC= 0.671). Neither SDmax (1-unit increase OR= 1.00, P= 0.073; AUC= 0.640) nor SDmax bulk (1-unit increase OR= 1.00, P= 0.147; AUC= 0.611) predicted failure to achieve CMR. For the dichotomized biomarkers, similarly MTV with a cutpoint of 96ml was the variable associated with the highest classification performance (OR= 7.13, P<.0001; AUC= 0.726), followed by IMPI (OR= 7.31, P<.0001; AUC= 0.719), TLG (OR= 6.52, P<.0001; AUC= 0.717), SUVmax (OR= 2.92, P=0.009; AUC= 0.624), SDmax (OR= 2.16, P=0.122; AUC=0.588), and SDmax bulk (OR= 1.79, P=0.230; AUC=0.568). Elevated serum LDH was the clinical variable associated with non-CMR outcome (OR= 6.24, P<.0001; AUC= 0.709). Advanced-stage disease did not significantly predict non-CMR (OR= 1.27, P= 0.604). Interestingly, in this dataset, age ≥60 (vs. <60 (reference)) was associated with a 75% lower likelihood of non-CMR (OR= 0.25, P= 0.008; AUC= 0.634). In addition, comparing high- vs. low-MTV groups, there was no significant difference with respect to age ≥60 (63.0% vs. 66.7%; P= 0.654); however, there were significant differences with respect to elevated serum LDH (87.7% vs. 40.4%; P<.0001) and advanced-stage (87.7% vs 60%; P= 0.0001).

Table 1.

Logistic regression models evaluating association between PET/CT metrics and failure to achieve CMR (non-CMR vs CMR (reference)) (N=138)

Models	Unit/Cutpoint categories	OR (95%CI)	P	AIC	AUC
Univariable models
SUVmax	1-unit increase	1.06 (1.02, 1.11)	0.003	145.4	0.671
Log2 MTV	1-unit increase	1.45 (1.23, 1.71)	<.0001	131.0	0.770
Log2 TLG	1-unit increase	1.40 (1.21, 1.62)	<.0001	130.3	0.765
SDmax (N=106)	1-unit increase	1.00 (0.99, 1.01)	0.073	116.1	0.640
SDmax bulk (N=106)	1-unit increase	1.00 (0.99, 1.01)	0.147	117.5	0.611
Age	1-unit increase	0.96 (0.92, 0.99)	0.023	149.7	0.614
IMPI	1-unit increase	3.26 (1.51, 7.06)	0.003	145.5	0.696
SUVmax	≥18 vs <18	2.92 (1.30, 6.56)	0.009	149.1	0.624
MTV (N=138)	≥96 vs <96	7.13 (2.90, 17.50)	<.0001	134.4	0.726
TLG	≥926 vs <926	6.52 (2.67, 15.97)	<.0001	136.2	0.717
SDmax (N=106)	≥225 vs <225	2.16 (0.81, 5.74)	0.122	117.3	0.588
SDmax bulk (N=106)	≥206 vs <206	1.79 (0.69, 4.61)	0.230	118.3	0.568
LDH	High vs Normal	6.24 (2.68, 14.55)	<.0001	136.7	0.709
Stage	III-IV vs I-II	1.27 (0.52, 3.09)	0.604	155.6	0.522
Age	≥60 vs <60	0.25 (0.09, 0.69)	0.008	147.1	0.634
IMPI	≥1.25 vs <1.25	7.31 (2.78, 19.26)	<.0001	135.5	0.719
Multivariable models:			P	AIC	AUC	AUC change
Model 1:
MTV	≥96 vs <96	5.42 (1.82, 16.13)	0.002	125.1	0.817	0.091
LDH	High vs Normal	3.21 (1.20, 8.58)	0.020	--	--	--
Stage	III-IV vs I-II	0.69 (0.23, 2.06)	0.507	--	--	--
Age	≥60 vs <60	0.21 (0.07, 0.66)	0.008	--	--	--
Model 2:
MTV	≥96 vs <96	8.84 (2.43, 32.17)	0.001	90.5*	0.864*	0.097*
SDmax (N=106)	≥225 vs <225	0.56 (0.15, 2.15)	0.399	--	--	--
LDH	High vs Normal	3.62 (1.04, 12.64)	0.043	--	--	--
Stage	III-IV vs I-II	1.71 (0.36, 8.06)	0.497	--	--	--
Age	≥60 vs <60	0.11 (0.02, 0.55)	0.007	--	--	--
Model 3:
MTV	≥96 vs <96	5.39 (1.71, 16.96)	0.004	125.4	0.817	0.091
SUVmax	≥18 vs <18	0.77 (0.25, 2.35)	0.646	--	--	--
LDH	High vs Normal	3.50 (1.28, 9.56)	0.015	--	--	--
Age	≥60 vs <60	0.21 (0.07, 0.64)	0.007	--	--	--

OR: odds ratio, the ratio between odds of non-CMR (PMR/NMR/PMD/NE) comparing patient groups. The categories for PET/CT metric were based on cutpoints for PFS. CI: confidence interval. P: p-value testing OR=1 from Wald test.

AIC: Akaike information criterion (smaller is better model fit). AUC: Area under the receiving operating curve (ROC) (higher is better model prediction). AUC change: change in AUC between multivariable model and univariable MTV model in N=138 or N=106 (*).

Abbreviations; SUVmax, maximum standardized uptake value; MTV, metabolic tumor volume, TLG, tumor lesion glycolysis; SDmax, largest distance between two lesions that are farthest apart; IMPI, International Metabolic Prognostic Index; LDH, lactate dehydrogenase

Next, we derived multivariable models, including MTV, SDmax, SUVmax, LDH, and clinical variables (stage and age) included in the IMPI. All multivariable models, including MTV, had better performance than the univariable dichotomized MTV model (AUC change >0.09; smaller AIC) (Table 1).

Association with progression-free survival and overall survival

We then fit Cox regression models evaluating the association between PET/CT biomarkers, clinical factors, and IMPI with PFS and OS. MTV as continuous (_Log2MTV) or dichotomized (cutpoint= 96 ml) was the biomarker demonstrating better performance in univariable Cox models for PFS (Table 2) and OS (Table 3); 1-unit increase in log₂MTV for PFS (hazard ratio (HR)= 1.27, P<.0001, IAUC= 0.748), and OS (HR= 1.36, P<.0001, IAUC= 0.836); MTV ≥96 vs. <96 ml (reference) for PFS (HR= 3.29, P<.0001, IAUC= 0.689), and for OS (HR= 3.56, P<.0001, IAUC= 0.732). Similar to the results observed with response assessment, age ≥60 vs <60 (reference) (PFS HR= 0.66, P= 0.096, and OS HR= 1.05, P= 0.837), and advanced-stage III-IV vs I-II (reference) (PFS HR= 1.24, P= 0.457, and OS HR= 1.32, P= 0.268) were not significant survival predictors. SDmax was statistically significant in univariable models as continuous (1-unit increase PFS HR= 1.002, P= 0.005, IAUC= 0.621, and OS HR= 1.002, P= 0.003, IAUC= 0.645) and dichotomous factor (SDmax ≥225 vs. <225 (reference) PFS HR= 2.33, P= 0.002, IAUC= 0.616, and OS HR= 1.87, P= 0.008, IAUC= 0.608). Similar results were observed for SDmax bulk. In multivariable Cox regression analysis, MTV again emerged as the factor with better discrimination performance for PFS (Model 1 IAUC= 0.744) and OS (Model 1 IAUC= 0.773), compared to models with dichotomized MTV only (PFS IAUC= 0.689, and OS IAUC= 0.732) (Table 2 and 3).

Table 2.

Cox regression analyses assessing the effect of PET/CT metrics on PFS (N=138)

Variable	Unit/Cutpoint categories	HR (95%CI)	P	AIC	IAUC
Univariable models
SUVmax	1-unit increase	1.02 (1.01, 1.04)	0.012	587.5	0.579
Log2 MTV	1-unit increase	1.27 (1.15, 1.41)	<.0001	567.4	0.748
Log2 TLG	1-unit increase	1.23 (1.13, 1.35)	<.0001	569.6	0.731
SDmax (N=106)	1-unit increase	1.002 (1.001, 1.004)	0.005	424.2	0.621
SDmax bulk (N=106)	1-unit increase	1.002 (1.001, 1.004)	0.009	425.2	0.604
Age	1-unit increase	0.99 (0.97, 1.01)	0.196	591.8	0.594
IMPI	1-unit increase	2.42 (1.61, 3.63)	<.0001	575.9	0.688
SUVmax	≥18 vs <18	2.30 (1.30, 4.07)	0.004	584.1	0.601
MTV	≥96 vs <96	3.29 (1.94, 5.59)	<.0001	571.9	0.689
TLG	≥926 vs <926	3.10 (1.84, 5.23)	<.0001	573.6	0.678
SDmax (N=106)	≥225 vs <225	2.33 (1.36, 4.00)	0.002	421.8	0.616
SDmax bulk (N=106)	≥206 vs <206	2.19 (1.28, 3.74)	0.004	423.2	0.609
LDH	High vs normal	2.71 (1.55, 4.73)	0.0004	579.3	0.656
Stage	III-IV vs I-II	1.24 (0.71, 2.16)	0.457	592.8	0.510
Age	≥60 vs <60	0.66 (0.41, 1.08)	0.096	590.7	0.562
IMPI	≥1.25 vs <1.25	3.60 (2.14, 6.05)	<.0001	568.2	0.701
Multivariable models:			P	AIC	IAUC	IAUC change
Model 1:
MTV	≥96 vs <96	2.68 (1.44, 4.99)	0.002	571.9	0.744	0.055
LDH	High vs normal	1.85 (1.01, 3.39)	0.047	--	--	--
Stage	III-IV vs I-II	0.83 (0.45, 1.52)	0.536	--	--	--
Age	≥60 vs <60	0.79 (0.48, 1.30)	0.352	--	--	--
Model 2:
MTV	≥96 vs <96	3.63 (1.61, 8.22)	0.002	407.3*	0.780*	0.093*
SDmax (N=106)	≥225 vs <225	1.80 (0.95, 3.38)	0.069	--	--	--
LDH	High vs normal	2.01 (0.98, 4.15)	0.058	--	--	--
Stage	III-IV vs I-II	0.50 (0.20, 1.25)	0.139	--	--	--
Age	≥60 vs <60	0.78 (0.44, 1.40)	0.405	--	--	--
Model 3:
MTV	≥96 vs <96	2.46 (1.26, 4.80)	0.008	572.3	0.730	0.041
SUVmax	≥18 vs <18	1.03 (0.51, 2.10)	0.929	--	--	--
LDH	High vs normal	1.86 (1.01, 3.43)	0.046	--	--	--
Age	≥60 vs <60	0.77 (0.47, 1.26)	0.295	--	--	--

HR: Hazard ratio of progression for 1-unit increase, and for comparing groups variable ≥cutpoint versus <cutpoint (ref). The cutpoints were determined by Contal and O’Quigley method. CI: confidence interval. P: p-value testing HR=1 from Wald test.

AIC: Akaike information criterion (smaller is better model fit). IAUC: Integrated area under the ROC curve, computed as a weighted average of the AUC values at all the event times (higher is better model prediction). IAUC change: change in IAUC between multivariable and univariable model using dichotomized MTV in N=138 or N=106 (*).

Abbreviations; SUVmax, maximum standardized uptake value; MTV, metabolic tumor volume, TLG, tumor lesion glycolysis; SDmax, largest distance between two lesions that are farthest apart; IMPI, International Metabolic Prognostic Index; LDH, lactate dehydrogenase

Table 3.

Cox regression analyses assessing the effect of PET/CT metrics on OS (N=138)

Variable	Unit/Cutpoint categories	HR (95%CI)	P	AIC	IAUC
Univariable models
SUVmax	1-unit increase	1.019 (1.003, 1.034)	0.016	808.6	0.605
Log2 MTV	1-unit increase	1.36 (1.24, 1.49)	<.0001	764.8	0.836
Log2 TLG	1-unit increase	1.30 (1.19, 1.41)	<.0001	769.9	0.822
SDmax (N=106)	1-unit increase	1.002 (1.001, 1.004)	0.003	585.9	0.645
SDmax bulk (N=106)	1-unit increase	1.002 (1.001, 1.004)	0.008	587.4	0.631
Age	1-unit increase	1.00 (0.99, 1.02)	0.972	814.0	0.498
IMPI	1-unit increase	3.16 (2.23, 4.47)	<.0001	774.8	0.799
SUVmax	≥18 vs <18	1.72 (1.10, 2.69)	0.017	807.9	0.613
MTV	≥96 vs <96	3.56 (2.26, 5.60)	<.0001	780.7	0.732
TLG	≥926 vs <926	3.34 (2.14, 5.21)	<.0001	783.3	0.728
SDmax (N=106)	≥225 vs <225	1.87 (1.18, 2.98)	0.008	587.1	0.608
SDmax bulk (N=106)	≥206 vs <206	1.93 (1.21, 3.07)	0.006	586.4	0.611
LDH	High vs normal	2.98 (1.81, 4.91)	<.0001	792.2	0.667
Stage	III-IV vs I-II	1.32 (0.81, 2.14)	0.268	812.7	0.542
Age	≥60 vs <60	1.05 (0.68, 1.60)	0.837	813.9	0.504
IMPI	≥1.25 vs <1.25	3.53 (2.29, 5.43)	<.0001	779.9	0.746
Multivariable models:			P	AIC	IAUC	IAUC change
Model 1:
MTV	≥96 vs <96	3.09 (1.84, 5.17)	<.0001	777.1	0.773	0.041
LDH	High vs normal	2.10 (1.24, 3.56)	0.006	--	--	--
Stage	III-IV vs I-II	0.81 (0.47, 1.38)	0.428	--	--	--
Age	≥60 vs <60	1.14 (0.74, 1.76)	0.555	--	--	--
Model 2:
MTV	≥96 vs <96	3.75 (1.90, 7.40)	0.0001	558.2	0.796	0.090*
SDmax (N=106)	≥225 vs <225	1.48 (0.88, 2.48)	0.142	--	--	--
LDH	High vs normal	2.40 (1.27, 4.55)	0.007	--	--	--
Stage	III-IV vs I-II	0.64 (0.31, 1.35)	0.242	--	--	--
Age	≥60 vs <60	0.88 (0.53, 1.45)	0.608	--	--	--
Model 3:
MTV	≥96 vs <96	3.29 (1.90, 5.67)	<.0001	776.5	0.764	0.032
SUVmax	≥18 vs <18	0.74 (0.44, 1.25)	0.266	--	--	--
LDH	High vs normal	2.21 (1.30, 3.77)	0.003	--	--	--
Age	≥60 vs <60	1.13 (0.73, 1.73)	0.585	--	--	--

HR: Hazard ratio of death for 1-unit increase, and for comparing groups variable ≥cutpoint versus <cutpoint (ref). The cutpoints for PFS were used for OS. CI: confidence interval. P: p-value testing HR=1 from Wald test.

AIC: Akaike information criterion (smaller is better model fit). IAUC: Integrated area under the ROC curve, computed as a weighted average of the AUC values at all the event times (higher is better model prediction). IAUC change: change in IAUC between multivariable and univariable model using dichotomized MTV in N=138 or N=106 (*).

Abbreviations; SUVmax, maximum standardized uptake value; MTV, metabolic tumor volume, TLG, tumor lesion glycolysis; SDmax, largest distance between two lesions that are farthest apart; IMPI, International Metabolic Prognostic Index; LDH, lactate dehydrogenase

The estimated median PFS by risk groups (low vs. high) were SUVmax (19.2 vs. 3.0 months; P= 0.003) (Figure 1A), MTV (15.9 vs. 2.8 months; P<.0001) (Figure 1B), SDmax (7.4 vs 2.7 months; P=0.002) (Figure 1C), and IMPI (15.9 vs 2.7 months; P<.0001) (Figure 1D). The estimated median OS by risk groups (low vs. high) were SUVmax (14.2 vs. 6.7 months; P= 0.016) (Figure 2A), MTV (19.2 vs. 5.4 months; P<.0001) (Figure 2B), SDmax (8.6 vs. 4.8 months; P=0.007) (Figure 2C), and IMPI (18.4 vs 5.1 months; P<.0001) (Figure 2D). As expected, patients achieving CMR to loncastuximab tesirine demonstrated longer PFS (median not estimable (NE) for CMR, 7.4 months for PMR, 1.3 months for NMR/PMD/NE; P<.0001) and OS (median NE for CMR, 10.8 months for PMR, 5.4 months for NMR/PMD/NE; P<.0001). 12-month PFS and OS rates for CMR were 81.6% (95%CI 57.5 to 92.8%) and 78% (95%CI 59.3 to 88.9%), for PMR 10.7% (95%CI 0.7 to 36.1) and 39.9% (95%CI 23 to 56.3%), and for NMR/PMD/NE 4.8% (95%CI 0.5 to 17.5%) and 18.3% (95%CI 10.1 to 28.4%), respectively (Supplementary Figure S3).

Figure 1.

Kaplan-Meier curves for progression-free survival by dichotomized PET/CT biomarkers. SUVmax (A), MTV (B), SDmax (C), and IMPI (D)

Figure 2.

Kaplan-Meier curves for overall survival by dichotomized PET/CT biomarkers. SUVmax (A), MTV (B), SDmax (C), and IMPI (D)

Nomograms for response to treatment and survival

Based on the nomogram using multivariable logistic model 1 from Table 1, the predicted probability of non-CMR among patients with elevated LDH, advanced-stage, and age ≥60 was 0.587 for patients with low-MTV (total point= 69) and 0.885 for patients with high-MTV (total point= 169). The predicted probability of non-CMR was 0.706 for patients with normal LDH, advanced-stage, age ≥60, and high-MTV (total point= 100) (Figure 3A). We then constructed nomograms for survival. Based on nomogram using multivariable Cox model 1 from Table 2, the predicted PFS rate at six months among patients with elevated LDH, advanced-stage, age ≥60 was 0.648 for patients with low-MTV (nomogram total point= 62) and 0.312 for patients with high-MTV (total point= 162). It was 0.533 for patients with normal LDH, advanced-stage, age ≥60, and high-MTV (total point= 100) (Figure 3B). Based on the nomogram using multivariable model 1 (Table 3), the corresponding predicted OS rates at six months were 0.774 (total point= 78), 0.453 (total point= 178), and 0.687 (total point= 112), respectively (Figure 3C).

Figure 3.

Nomograms for non-complete metabolic response (A), progression-free survival (B), and overall survival (C) based on MTV multivariable models

Discussion

In this post-hoc clinical trial analysis, we evaluated the performance of PET/CT biomarkers to predict response, PFS, and OS to loncastuximab tesirine. As a continuous and dichotomous variable, MTV obtained from pretreatment PET/CT predicted outcome and was the biomarker associated with the highest discriminatory power.

IMPI represents a significant advance for implementing MTV in lymphoma research. Using a linear spline model, investigators constructed IMPI using MTV with age as continuous variables and individual stage as I to IV outperforming the IPI to predict 3-year survival in patients with newly diagnosed DLBCL treated with R-CHOP (rituximab, cyclophosphamide, adriamycin, vincristine, and prednisone) (18). Most recently, two studies evaluated the role of IMPI in DLBCL. In a post-hoc analysis of a risk-adapted frontline immunochemotherapy program (n= 166), the IMPI exhibited limited discrimination with slightly overestimated event rates but demonstrated good calibration for high-risk patients (25). Caveats of this analysis are that the authors used an MTV threshold of 41% of SUVmax instead of SUV ≥4.0 used in the development of IMPI and an augmentation with R-ICE (rituximab, ifosfamide, carboplatin, and etoposide) after R-CHOP which may have influenced the result differences between studies. Subsequently, a small study (n= 39) evaluated IMPI performance in patients with rel/ref DLBCL treated with CAR T-cell, observing a correlation with PFS but not with duration of response or OS (26). IMPI demonstrated an appropriate performance in our dataset, underscoring the potential of this model also to predict outcomes in the rel/ref setting. However, we did not observe a better discrimination of IMPI over MTV alone. In our study, age ≥60 years and advanced-stage did not emerge as significant factors associated with outcomes across univariable Cox analyses. This lack of association with outcome may explain the less accurate performance of IMPI in our dataset. LOTIS-2 trial enrolled heavily pretreated patients with rel/ref disease, commonly associated with worse outcomes than an untreated population. Furthermore, age ≥60 years and advanced-stage disease are integral parts of the International Prognosis Index (IPI), underscoring the importance of more sophisticated biomarkers capable of better capturing tumor burden and predicting survival with newer therapies.

In patients with rel/ref DLBCL, MTV has been evaluated before ASCT and CAR T-cell but not to recently approved FDA agents (15,17,19,27–29). In patients selected for ASCT, studies have demonstrated the prognostic power of MTV to risk-stratify patients with divergent survival before platinum-based salvage chemotherapy (17) and ASCT (27). Similarly, high-MTV, defined by values between 80 and 147.5ml (15,29), was an independent prognostic factor associated with shorter survival in patients treated with CAR T-cell. Investigators from the ZUMA-7 trial presented in abstract form superior event-free survival (EFS) with axi-cel (HR=0.423) compared to standard-of-care in patients with low and high-MTV (30). Although axi-cel demonstrated better outcomes independent of the MTV subgroup, those with high-MTV demonstrated shorter survival (HR= 1.441), underscoring a higher risk for treatment failure in this population. High-MTV is associated with other poor prognosis factors, including elevated serum LDH and β−2 microglobulin, advanced-stage disease, poor ECOG performance status, and bulky disease (16,31,32). However, the prognosis implications of MTV are independent of the classical definitions of tumor burden, including stage and bulky disease. The biology related to patients with high-MTV remains largely unknown, but in patients with rel/ref DLBCL, the presence of circulatory inflammatory cytokines such as IL-6, IL-15, and TNF-α and intratumoral interferon signature genes and activated macrophages are features observed in this population (28,33) suggesting there may be a pathogenic role of the tumor microenvironment.

Different SUV thresholds and methodologies have been proposed to calculate MTV and determine appropriate cutpoints in DLBCL (20,34,35). In the present study, we decided to focus our analysis using an MTV threshold of SUV ≥4.0 following prior studies demonstrating better performance and appropriate interobserver correlation (21,36). Therefore, comparing our results with studies employing a 41% SUVmax threshold would not be prudent as tumor volumes will differ; for example, earlier studies evaluating MTV in patients treated with CAR T-cell (15,19,29). We evaluated the predictive and prognosis power of PET/CT biomarkers as continuous and dichotomous variables. This is the first study evaluating these metrics in patients homogeneously and prospectively treated with loncastuximab tesirine. Therefore, there is currently no external dataset to validate the cutpoints derived from our study, representing a limitation of our analysis. Nevertheless, analyzing the data as continuous factors addresses several concerns by avoiding loss of valuable data, better observing a genuine relationship with outcome, and improving the ability to directly compare results between different datasets with similar methodology (37,38). SDmax is an emerging tumor dissemination biomarker with growing evidence demonstrating its independent value to predict outcomes in DLBCL (23,39). We observed a limited discriminatory power of this biomarker in the rel/ref setting to predict response but there was a significant correlation of SDmax with survival, albeit less powerful than MTV or IMPI. SDmax bulk showed similar associations as SDmax and did not improve upon its discriminatory power in this cohort.

We also constructed nomograms based on multivariable data to reduce statistical predictive models into a single numerical estimate of the probability of achieving a CMR and time-to-event outcome (40). Across nomograms, MTV possesses the highest effect and is converted to 100 points. It is important to underscore that nomograms rank the importance of an effect in predicting the outcome only in the context of the other covariates currently in the model. Furthermore, outcome predictions come with associated confidence intervals, which should be considered when interpreting nomogram results.

Currently, no biomarker guides treatment selection in patients with rel/ref DLBCL. Patients are selected for second-line therapy based on response to frontline therapy, time to relapse, comorbidities, and access to cellular products (41,42). A similar situation occurs in the third line setting with an increasing number of agents recently approved. The main reason for empiric treatment selection instead of biomarker driven is the absence of predictive biomarkers integrated into treatment-decision models capable of guiding the treating physician. MTV can assist in selecting patients that will benefit the most from specific programs. It will be necessary for future clinical trials to regularly determine MTV values, which will likely require unique cutpoints for each therapeutic agent (34). Circulating tumor DNA (ctDNA) further enhances prognostication, and may complement MTV in developing better risk-stratification associated with subsequent molecular disease monitoring (43). Unfortunately, ctDNA was not available to be incorporated into our models, but future studies should regularly assess prognostication of imaging and circulatory biomarkers aiming to leverage into clinical decision-making platforms.

In conclusion, our results confirm the value of PET/CT-biomarkers as prognostic factors in DLBCL. MTV is the metric with better predictive and prognostic performance to identify patients at risk for treatment failure in the LOTIS-2 cohort. Rel/ref DLBCL with high-MTV remains a population with poor outcomes, and additional studies are needed to identify optimal therapy for these patients.

Translational relevance.

Quantitative biomarkers extracted from routine PET/CT have gained attention to predict outcomes in diffuse large B-cell lymphoma (DLBCL). Among all PET/CT biomarkers, metabolic tumor volume (MTV) demonstrated robust risk-stratification independent of methodology and treatment line. Over the last years, significant advances in drug development for DLBCL have occurred with multiple available agents. However, no biomarker guiding treatment selection with recently FDA-approved agents exists. Here, we evaluate for the first time the predictive power of PET/CT biomarkers for outcomes in a clinical trial cohort of patients treated with loncastuximab tesirine. On multivariable analyses, we found MTV derived from baseline scans was the biomarker with better performance to predict response and survival. Our data suggest that patients with relapsed/refractory DLBCL demonstrating high-MTV experience lower response rates to loncastuximab tesirine.