Interim positron emission tomography in diffuse large B-cell lymphoma

Abstract

Rationale: In diffuse large B-cell lymphoma, early assessment of treatment response by 18-fluorodeoxyglucose positron emission tomography (PET) may trigger treatment modification. Reliable identification of good and poor responders is important. We compared three competing methods of interim PET evaluation. Methods: Images from 449 patients participating in the ‘Positron Emission Tomography-Guided Therapy of Aggressive Non-Hodgkin Lymphomas’ trial were re-analyzed by applying the visual Deauville score and the standardized uptake value (SUV)-based qPET and ΔSUV_max scales to interim PET scans performed after two cycles of chemotherapy. qPET relates residual lymphoma 18-fluorodeoxyglucose uptake to physiological liver uptake, converting the ordinal Deauville scale into a continuous scale and permitting a direct comparison with the continuous ΔSUV_max scale, which is based on SUV_max changes between baseline and interim scans. Positive and negative predictive values were calculated for progression-free survival. Results: Using established thresholds to distinguish between good and poor responders (visual Deauville score 1-3 vs. 4-5; ΔSUV_max >66% vs. ΔSUV_max ≤66%), the positive predictive value was significantly lower with Deauville than ΔSUV_max (38.4% versus 56.6%; P = 0.03). qPET and ΔSUV_max were strongly correlated on the log scale (Pearson’s r=0.75). When plotted along corresponding percentiles, the positive predictive value curves for qPET and ΔSUV_max were superimposable, with low values up to the 85th percentile and a steep rise thereafter. The recommended threshold of 66% SUV_max reduction for the identification of poor responders was equivalent to qPET=2.26 corresponding to score 5 on the visual Deauville scale. The negative predictive value curves were also superimposable, but remained flat between 80% and 70%. Conclusion: Continuous scales are better suited for interim PET-based outcome prediction than the ordinal Deauville scale. qPET and ΔSUV_max essentially carry the same information. The proportion of poor risk patients identified is less than 15%.