Lower-Grade Gliomas: Predicting DNA Methylation Subtyping and its Consequences on Survival with MR Features

Rationale and Objectives To explore associations between MR imaging features, DNA methylation subtyping, and survival in lower-grade gliomas (LGG). Materials and Methods The MR data from 170 patients generated with the Cancer Imaging Archive were reviewed. The correlation was evaluated by Fisher's Exact Test, Pearson Chi-Square and binary regression analysis. Survival analysis was conducted by using time-dependent ROC analysis and the Kaplan-Meier method (the worst prognosis subgroup). Results Identified were 9 (5.3%) M1-subtype, 18 (10.6%) M2-subtype, 48 (28.2%) M3-subtype, 31 (18.2%) M4-subtype and 64 (37.6%) M5-subtype. Patients with M4-subtype had the shortest median OS (49.3 vs. 28.4) months(p < 0.05). The time-dependent ROC for the M4-subtype was 0.83 (95% confidence interval 0.72–0.95) for survival at 12 months, 0.82 (95% confidence interval 0.70–0.94) for survival at 24 months, and 0.74 (95% confidence interval 0.62–0.86) for survival at 36 months. After uni- and multivariate analysis, a nomogram was built based on proportion contrast-enhanced (CE) tumor, extranodular growth, volume_cutoff_median, and location. For the prediction of M4-subtype, the nomogram showed good discrimination, with an area under the curve (AUC) of 0.886 (95% CI: 0.820–952) and was well calibrated. On multivariate logistic regression analysis, volume ≥60cm3 (OR: 0.200; p < 0.001; 95%CI: 0.048–0.834) was associated with M1-subtype (AUC: 0.690). Hemorrhage (OR: 5.443; p = 0.002; 95%CI: 1.844–16.069) and volume > median (OR: 3.256; p = 0.05; 95%CI: 0.992–10.686) were associated with M2-subtype (AUC: 0.733). Proportion CE tumor<=5% (OR: 3.968; P=0.002; 95%CI: 1.634-9.635) was associated with M3-subtype (AUC: 0.632). Poorly-defined (OR: 2.258; p = 0.05; 95%CI: 1.000–5.101) and volume > median (OR: 2.447; p = 0.01; 95%CI: 1.244–4.813) were associated with M5-subtype (AUC: 0.645). Decision curve analysis indicated predictions for all models were clinically useful. Conclusion This preliminary radiogenomics analysis of lower-grade gliomas demonstrated associations between MR features and DNA methylation subtyping. The shortest survival was observed in patients with M4-subtype. And we have constructed nomogram that enables more accurate predictions of M4-subtype. To explore associations between MR imaging features, DNA methylation subtyping, and survival in lower-grade gliomas (LGG). The MR data from 170 patients generated with the Cancer Imaging Archive were reviewed. The correlation was evaluated by Fisher's Exact Test, Pearson Chi-Square and binary regression analysis. Survival analysis was conducted by using time-dependent ROC analysis and the Kaplan-Meier method (the worst prognosis subgroup). Identified were 9 (5.3%) M1-subtype, 18 (10.6%) M2-subtype, 48 (28.2%) M3-subtype, 31 (18.2%) M4-subtype and 64 (37.6%) M5-subtype. Patients with M4-subtype had the shortest median OS (49.3 vs. 28.4) months(p < 0.05). The time-dependent ROC for the M4-subtype was 0.83 (95% confidence interval 0.72–0.95) for survival at 12 months, 0.82 (95% confidence interval 0.70–0.94) for survival at 24 months, and 0.74 (95% confidence interval 0.62–0.86) for survival at 36 months. After uni- and multivariate analysis, a nomogram was built based on proportion contrast-enhanced (CE) tumor, extranodular growth, volume_cutoff_median, and location. For the prediction of M4-subtype, the nomogram showed good discrimination, with an area under the curve (AUC) of 0.886 (95% CI: 0.820–952) and was well calibrated. On multivariate logistic regression analysis, volume ≥60cm3 (OR: 0.200; p < 0.001; 95%CI: 0.048–0.834) was associated with M1-subtype (AUC: 0.690). Hemorrhage (OR: 5.443; p = 0.002; 95%CI: 1.844–16.069) and volume > median (OR: 3.256; p = 0.05; 95%CI: 0.992–10.686) were associated with M2-subtype (AUC: 0.733). Proportion CE tumor<=5% (OR: 3.968; P=0.002; 95%CI: 1.634-9.635) was associated with M3-subtype (AUC: 0.632). Poorly-defined (OR: 2.258; p = 0.05; 95%CI: 1.000–5.101) and volume > median (OR: 2.447; p = 0.01; 95%CI: 1.244–4.813) were associated with M5-subtype (AUC: 0.645). Decision curve analysis indicated predictions for all models were clinically useful. This preliminary radiogenomics analysis of lower-grade gliomas demonstrated associations between MR features and DNA methylation subtyping. The shortest survival was observed in patients with M4-subtype. And we have constructed nomogram that enables more accurate predictions of M4-subtype.