Contrast enhancement is a prognostic factor in IDH1/2 mutant, but not in wild-type WHO grade II/III glioma as confirmed by machine learning. (January 2019)

• Record Type: Journal Article
• Authors: Suchorska, Bogdana; Schüller, Ulrich; Biczok, Annamaria; Lenski, Markus; Albert, Nathalie L.; Giese, Armin; Kreth, Friedrich-Wilhelm; Ertl-Wagner, Birgit; Tonn, Jörg-Christian; Ingrisch, Michael
• Is Part Of: European journal of cancer. Volume 107(2019)
• Page Start: 15
• Page End: 27
• Publication Date: 2019-01
• Subjects: IDH mutation -- Glioma -- Contrast enhancement -- Co-del 1p/19q -- Machine learning; Cancer -- Periodicals; Neoplasms -- Periodicals; Cancer -- Périodiques; Cancer; Tumors; Electronic journals; Periodicals; Electronic journals; 616.994
• DOI: 10.1016/j.ejca.2018.10.019
• Languages: English
• ISSNs: 0959-8049

Abstract

Abstract: Background: Mutation of the isocitrate dehydrogenase ( IDH ) gene and co-deletion on chromosome 1p/19q is becoming increasingly relevant for the evaluation of clinical outcome in glioma. Among the imaging parameters, contrast enhancement (CE) in WHO II/III glioma has been reported to indicate poor outcome in the past. We aimed at reassessing the prognostic value of CE in these tumours within the framework of molecular markers using a machine learning approach (random survival forests [RSF]) as well as conventional Cox regression modelling. Methods: 301 patients with WHO grade II (n = 181) or grade III glioma (n = 120) were stratified according to their molecular profile. Pre-operative magnetic resonance imaging (MRI) was reviewed and volumetric analyses of CE and T2 volumes were performed followed by conventional univariate and multivariate Cox analyses. Furthermore, the dataset was split into discovery and validation datasets, and RSFs were trained on the discovery dataset to predict the individual risk of each patient. Concordance indices for Cox and RSF models were determined and the variable importance of explanatory variables was assessed using the minimal-depth concept. Results: In IDH mut tumours only, both conventional Cox regression modelling and RSF analyses showed that CE on initial MRI is a prognostic factor for survival with dependence on volume ( p < 0.05). In contrast, presence of CE on initial MRI was not associated with outcome in IDH wt tumours. Conclusions: In patients with diffuse IDH wt gliomas WHO grade II/III, CE is not associated with survival, whereas in tumours with an IDH mutation, presence of CE on initial MRI is linked to inferior survival. Highlights: Random survival forest (RSF) method is based on a combination of decision trees. RSF identify high-risk glioma by stratifying variables according to their importance. Contrast enhancement on initial MRI is linked to poor outcome in IDH mutant glioma. No association between contrast enhancement and outcome in IDH wt tumours.