Prospective validation of the Magnetic Resonance Tumor Regression Grade (MR-TRG) and correlation with pathologic endpoints score in NRG Oncology GI002 [abstract] Abstract uri icon


  • Purpose/Objective(s): Total neoadjuvant therapy (TNT) including chemotherapy (chemo) and concurrent chemo and radiotherapy (RT) is a promising strategy for rectal cancer. Characterizing tumor regression during TNT represents an important area of research. We report a prospective validation study of the Magnetic Resonance Tumor Regression Grade (MR-TRG) using MRIs from NRG-GI002, a phase II trial of TNT for rectal cancer. Materials/Methods: This was the a priori designed imaging biomarker study of NRG-GI002 seeking to correlate the MR-TRG with the pathologic Neoadjuvant Rectal score (NAR) and pathologic complete response (pCR). 110 patients (pts) were needed with 2 MRI's, 1 pre-TNT and 1 post. Three diagnostic radiologists independently reviewed each MRI and completed a consensus read. Radiologists’ assessment of complete response (mriCR) was tested for its positive predictive value (PPV), negative predictive value (NPV), sensitivity (sen), and specificity (spec) with pCR. Chi-squared test was used for association of mriCR and pCR, and Wilcoxon-Mann-Whitney test was used for association of MR-TRG and pCR. Spearman Rho was used for association of MR-TRG/mriCR and NAR. P-values < .05 (2-sided) were considered significant. Each reader's T-stage, N-stage, extra-mural vascular invasion (EMVI), MR-TRG score and diffusion weighted imaging (DWI) assessment was evaluated for inter-reader variability by Kappa statistics. Kappa values of agreement were interpreted as: poor, < 0.0; slight, 0.0–0.2; fair, 0.2–0.4; moderate, 0.4–0.6; substantial, 0.6–0.8; and almost perfect, 0.8–1.0. All analyses were performed using a data management and decision management software. Radiologists were blinded to pathologic data. Results: A total of 126 patients, from 76 institutions, had 2 MRI's (pre-TNT and post), 3 radiologists completed 756 MRI interpretations. Of 126 patients, 29% were female (N = 37), 85% white (N = 107) and median age was 55 (24-76). Individual and consensus MR-TRG scores were highly associated with pCR (P < 0.01) and NAR (Rho = .39, P < 0.001). The addition of DWI to MR-TRG improved both sen and spec over MR-TRG alone (P = 0.035). On the consensus read for mriCR, sen = 53.3%, spec = 70.7%, PPV = 42.1%, and NPV = 79.1% compared to pCR, and was highly associated (P = 0.02). The Kappa agreement score for T-stage, N-stage, and EMVI was 0.34, 0.67, and 0.39 reflecting fair, substantial, and fair agreement respectively. The Kappa agreements score across readers were 0.26 for MR-TRG and 0.32 for DWI, reflecting fair agreement. The mriCR, had a Kappa statistic of 0.85, reflecting near perfect agreement. Conclusion: MR-TRG is significantly correlated with pathologic NAR score. Binary assessment of CR, using MR-TRG and DWI, had near perfect agreement across readers. The addition of DWI to MR-TRG improved both the sen and spec, reflecting utility of DWI in this setting. Correlation of mriCR with pCR was statistically significant. Notably, PPVs here were limited, implicating the need for stronger biomarkers when planning a national rectal organ preservation study. NCT02921256.

publication date

  • 2021