Publikationsrepositorium - Helmholtz-Zentrum Dresden-Rossendorf

Background: Visually Accessible Rembrandt (Repository for Molecular Brain Neoplasia Data) Images (VASARI) features, a controlled vocabulary aimed at establishing reproducible terminology for glioma reporting, have been in use for a decade, but there is no systematic evaluation of their performance.

Purpose: To conduct a systematic review and meta-analysis of the performance of the VASARI set for glioma assessment.

Materials and methods: MEDLINE and Web of Science were systematically searched until July 7, 2023. Original articles using VASARI to predict diagnosis, progression, and survival in patients with glioma were included. A modified Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was used for quality assessment. Meta-analysis was performed using a random effects model and forest-plot visualizations if five or more comparable studies with low or medium risk-of-bias QUADAS-2 were available.

Results: Thirty-five studies (4,348 patients; 34 retrospective, 1 prospective) were included. The overall risk-of-bias score was medium (n=33) and low (n=2). The most frequent objectives were overall survival (OS) (n=18) and isocitrate dehydrogenase (IDH) mutation (n=12) prediction. IDH mutation-predicting models combining different VASARI features rendered a pooled AUC-ROC of 0.82 (95% CI 0.76, 0.88) with considerable heterogeneity (I² = 100%). The highest pooled hazard ratios (HR) for single features predicting OS involved features 9 (multifocality; HR 1.80; 95%: CI 1.21, 2.67; I² = 53% ), 19 (ependymal invasion; HR 1.73; 95% CI: 1.45, 2.05; I² = 0%), and 23 (enhancing tumor crossing midline; HR 2.08; 95% CI: 1.35, 3.18; I² = 52%). Combined models including various inputs such as VASARI, clinical, pathological, and/or radiomics features outperformed single data type models in all relevant studies (n=17).

Conclusion: VASARI features perform well in predicting OS and IDH mutation status, and improve when combined models are used. The discriminatory power of single VASARI features is, however, variable.