Integration of [18f]FDG-PET Radiomics with Liquid Biopsy Improves Outcome Prediction in Newly Diagnosed DLBCL Patients

BACKGROUND: Baseline circulating tumor DNA (ctDNA) levels and PET/CT radiomics allow to identify DLBCL destined to early relapse, but their integration has not been extensively evaluated. METHODS: A real-world cohort of newly diagnosed DLBCL referring to our institution provided with ctDNA from plasma and with baseline [18F]FDG-PET scans has been prospectively enrolled. ctDNA levels were reported as Log10haploid genome equivalents per mL of plasma (Log10hGE/mL) and molecular clusters were identified using the LymphGen tool. Using the standardized uptake value (SUV) threshold of 4.0, the following parameters were collected: i) maximum SUV of the hottest lesion (SUVmax), ii) total metabolic tumor volume (tMTV), iii) total tumor lesion glycolysis (tTLG), and iv) Dmax. The maximally selected rank statistics was used to identify the best cutoff values in predicting progression free survival (PFS) for ctDNA levels and for each PET/CT variable. RESULTS: The cohort included 120 newly diagnosed DLBCL homogenously treated with R-CHOP. Median age was 67 years. After a median follow-up of 56 months, the 40-month PFS and overall survival (OS) were 65.1% and 75.6%. Using maximally selected rank statistics, the best cutoff for predicting PFS was 2.68 Log10hGE of ctDNA (p<0.001). The optimal PFS cutoff for SUVmax, tMTV, tTLG and Dmax were 17.66 (p=0.023), 639.2 cm3 (p<0.001), 7138.34 (p<0.001) and 39 cm (p=0.0018), respectively. Patients with ctDNA levels >2.68 Log10hGE (N=43), termed ctDNA-high, had a 40-month PFS and OS of 42.9% and 55.4% compared to 77.4% and 86.5% for ctDNA-low patients (p<0.001 and p=0.001, respectively; c-indices: 0.630 for PFS and 0.648 for OS). In multivariate analysis, each single PET/CT feature, namely tMTV (HR 2.42, p=0.007), tTLG (HR 2.42, p=0.007) and Dmax (HR 1.97, p=0.031), maintained an independent association with shorter PFS when adjusted for ctDNA levels, while SUVmax (HR 1.84, p=0.087) lost its prognostic value. Therefore, patients with at least one PET/CT value (among tMTV, tTLG or Dmax) above the respective cutoff were grouped as PET-positive patients (N=67). PET-positive patients had significantly worse outcome compared to PET-negative patients with 40-month PFS and OS of 45.4% and 62.2% compared to 90.4% and 92.3% (p<0.001 and p=0.007, respectively; c-indices: 0.685 for PFS and 0.645 for OS). By multivariate analysis, PET-positive (HR 3.84, 95% CI 1.78-8.28, p<0.001) and ctDNA-high (HR 2.04, 95% CI 1.10-3.78, p=0.024) independently predicted PFS. Relying on bcoefficients, a 2-variables prognostic score was devised and patients were grouped into three risk classes. Low-risk DLBCL (N=44) were both ctDNA-low and PET-negative and presented a 40-month PFS and OS of 93.0% and 95.3%. Intermediate-risk DLBCL (N=42) were either ctDNA-high or PET-positive and presented a 40-month PFS and OS of 61.5% and 75.4%. High-risk DLBCL (N=34) were both ctDNA high and PET-positive and presented a 40-month PFS and OS of 33.6% and 49.8% (p<0.001 and p<0.001, respectively; c-indices: 0.712 for PFS and 0.696 for OS). Moreover, BN2/ST2 patients, classified with the LymphGen tool on ctDNA, presented a very good outcome after R-CHOP and were able to predict PFS and OS in intermediate-risk and high-risk DLBCL. In multivariate analysis, BN2/ST2 clusters (HR 0.37, 95% CI 0.15-0.90, p=0.029) maintained an independent association with shorter PFS even when adjusted for PET-positive (HR 3.98, 95% CI 1.84-8.59, p<0.001) and ctDNA-high (HR 2.53, 95% CI 1.34-4.78, p=0.004). Therefore, a novel 3-variables score was devised. According to the b coefficients, 1.5 points were assigned to PET-positive, 1 point to ctDNA-high and -1 point to BN2/ST2 patients. Three different groups with unique PFS and OS were identified. Low-risk DLBCL (-1 to 0.5 points, N=51) presented a 40-month PFS and OS of 92.0% and 96.0%, intermediate-risk DLBCL (1 to 1.5 points, N=46) presented a 40-month PFS and OS of 58.0% and 75.5%, and high-risk DLBCL (2.5 points, N=23) presented a 40-month PFS and OS of 20.9% and 30.4%, respectively (both p<0.001; c-indices were 0.736 for PFS and 0.743 for OS). This model outperformed the 2-variables score based on PET/CT and ctDNA levels. CONCLUSIONS: The integration of baseline PET/CT variables (tMTV, tTLG and Dmax), ctDNA levels and molecular clusters improved the outcome prediction of newly diagnosed DLBCL compared to PET/CT and ctDNA levels alone or both combined.