Baran Atli, Maximilian Wartha, Matthias Gmeiner, Martin Aichholzer, Serge Weis, Dominik Vorhauer, Helga Wagner, Tanja Hauer, Carlo Serra, Petra Böhm, Michael Sonnberger, Nico Stroh-Holly, Stefan Aspalter, Kathrin Aufschnaiter, Tobias Rossmann, Maria Gollwitzer, Jawed Ziai, Annette Leibetseder, Josef Pichler, Wolfgang Thomae, Christian Dorfer, Andreas Gruber, Harald Stefanits, Philip Rauch
CE at diagnosis predicted transformation risk and timing. It remained predictive after GTR removed the enhancing tissue, a postoperative role not previously recognised. The signal was present in each integrated tumour type and survived adjustment for grade, IDH status and EOR, may reflect intrinsic tumour biology and could refine postoperative risk estimation. GTR was the only modifiable determinant. These are the first interval-censored estimates of MT timing.
PURPOSE: Malignant transformation (MT) is the principal driver of mortality in lower-grade glioma (LGG), yet its preoperative determinants and timing remain poorly defined. We quantified imaging elements of MT and its timing, correcting for bias in dating transformation.
METHODS: We evaluated contrast enhancement (CE), tumour volume and 99 radiomic features in 155 patients with WHO grade 2-3 LGG, radiologically low-grade at diagnosis with at most minimal enhancement (77 transformations; median surveillance among non-transformers 4.8 years). Because MT is ascertainable only between scans, timing was modelled as interval-censored across four complementary models, with extent of resection (EOR) time-varying.
RESULTS: CE at diagnosis was the strongest predictor of MT, exceeding tumour volume and a co-modelled 99-feature radiomic panel (Bayesian HR 22.1, 95% CrI 10.5-53.2); conditioned on transformation-free survival to six months, HR 5.65 (95% CI 2.23-14.31). No radiomic feature survived shrinkage or native interval-censoring. Initial enhancement predicted post-surgical transformation independently of grade, IDH status and EOR, even after gross-total resection (GTR), itself associated with a two-thirds lower hazard (HR 0.32, 95% CI 0.13-0.77). Interval censoring proved material: conventional dating overstated the hazard four-fold.
CONCLUSION: CE at diagnosis predicted transformation risk and timing. It remained predictive after GTR removed the enhancing tissue, a postoperative role not previously recognised. The signal was present in each integrated tumour type and survived adjustment for grade, IDH status and EOR, may reflect intrinsic tumour biology and could refine postoperative risk estimation. GTR was the only modifiable determinant. These are the first interval-censored estimates of MT timing.