Article Text

Download PDFPDF

#225 Integrating ESMO guidelines with an immunological enhanced endometrial cancer risk classification model
  1. Valentina Bruno1,
  2. Martina Betti1,
  3. Lorenzo D’Ambrosio1,
  4. Alice Massacci1,
  5. Alessandro Buda2,
  6. Benito Chiofalo1,
  7. Giulia Piaggio1,
  8. Gennaro Ciliberto1,
  9. Paola Nisticò1,
  10. Matteo Pallocca1 and
  11. Enrico Vizza1
  1. 1IRCCS Regina Elena National Cancer Institute, Rome, Italy
  2. 2San Gerardo Hospital, University of Milano-Bicocca, Monza, Italy


Introduction/Background Endometrial cancer (EC) treatments are related to known prognostic factors included in the risk classes defined by the ESMO-ESGO-ESTRO consensus conference,together with the biomolecular classification. However,these prognostic factors are not sufficient to predict recurrence rate at early stages. The integration of immune signatures to existing molecular based models has not been extensively evaluated,nor mentioned in the guidelines. The aim of this study is to improve clinical risk prediction models by integrating existing guidelines with new -omic immunological predictive features extracted from TCGA-UCEC dataset.

Abstract #225 Figure 1

(A) Performances with and without novel features. TP: Percentage of correctly predicted high-risk profiles. TN: Percentage of misclassified high-risk profiles. Accuracy: Balanced between TP and TN. (B) Predicted survival curves for the two groups of interest (Relapse vs. No Relapse). (C) Feature importance and effect on DFS. Higher values are associated with a stronger effect on predictions. Higher segment lengths are associated with lower consistency of the feature importance across trees in the Random Forest. Green lines are associated with higher DFS, red lines are associated with a lower DFS, and gray lines are associated to a non-univocal effect. (D) Representation of all features of the model grouped by category; features in bold are novel.

Methodology By deconvolution tools,we estimated the relative abundances of five main immune populations in public data and then applied feature selection methods to generate a machine learning (ML)-based model for disease-free survival probability prediction. We have also further investigated factors that may ease the re-stratification of early-stage cases which do experience relapse regardless of their low-risk profiles, trough deconvolution and differential expression analysis.

Results We first obtain a ML-based model that can predict recurrence with a higher accuracy than guidelines parameters by introducing the immune framework, so far neglected by EC guidelines. Furthermore,we obtain an immune-based ultra-stratification in early stages population: to summarize, ’hot tumors’ EC subtype tends not to relapse,and among recurrences ‘cold tumors’ EC subtype has a worst prognosis than ‘ultra-hot tumors’ in terms of OS.

Conclusion In conclusion, we introduce a ML-based model to improve EC recurrence risk prediction, by integrating well-established risk class prognostic factors with new –omic immunological features, so far neglected by the guidelines. Furthermore, we identify novel endometrial cancer immunological profiles that enable an ultra-stratification of early-stage cases, discriminating those patients that experience relapse despite being assigned to the low-risk class. In endometrial cancer framework, our model can predict recurrence with a higher accuracy than guidelines parameters, opening up to precision oncology approaches in terms of prognosis, decision making- treatment and follow-up.

Disclosures None

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.