Supporting Personalized Treatment Decisions in Head and Neck Cancer through Big Data

Head and Neck Carcinomas (HNC) are aggressive and heterogeneous tumors with a high fatality rate. Treatment may be extremely invasive and its impact on quality of life can be devastating. Neither molecular sub-classification nor prognostic models are well established in clinical practice, because both are currently inconsistent, platform- and population-dependent. This demonstrates the urgency for accurate patients' classification at diagnosis for personalized treatment decisions. This project exploited the world's largest pool of treated HNC patients (approximately 2500) from Italy, France and Germany. Since the data were collected by different centres without a common protocol many challenges had to be overcome before analysing the data together. For instance, questions regarding which measurements were equivalent across datasets, which patients could be considered in the analysis and how the final combined data should be organised were addressed in an important step called data harmonisation/integration. The harmonisation work was performed in collaboration with partners at the Polytechnic University of Madrid with the data stored at the Services for Sensitive Data (TSD/USIT/UiO). After data harmonisation, statistical models were developed by the UiO team to answer relevant clinical questions defined by our clinical partners. The data set combined existing clinically annotated genomic data which allowed the discovery of clinically relevant biomarkers. These biomarkers were tested and validated, showing associations between cancer survival and biomarkers. In one case the association between survival and the biomarker depended on the type of cancer treatment a patient received, suggesting that this biomarker may be relevant for making personalized clinical treatment decisions. We also investigated ethical and legal aspects of data-driven clinical decision making compared to current evidence-based approaches. This was an international collaborative project led by a team in Milano, Italy.

The work performed on data semantics for head and neck cancer (HNC) and data integration from different biomolecular platforms has promoted new knowledge generation and exchange between medical oncologists and bioinformaticians from the participating hospitals and experts in ontology systems from our technical partners. The HNC ontology developed in the project is anticipated to be beneficial to the wider HNC research community by establishing a standardized ontology to apply in future studies. The data annotation, quality check rules and integration processes and know-how are also being reused in the newly funded IDEA4RC Horizon Europe project. The generated data have fostered further analyses of the produced models on completely independent datasets (e.g., from the Head and Neck 5000 study in UK) as well as new research with a group in the US, to study radiosensitivity index and to reuse the data as a basis for a novel experiment on the generation and use of synthetic cohorts for in-silico studies. Lastly, the collaborations established by this project have allowed cross-country recruitment, generation of training opportunities for new staff and recruitment of young researchers and new technical staff.

Head and Neck Carcinomas (HNC) are aggressive and heterogeneous tumors with a high fatality rate. Treatment may be extremely invasive and result in highly impairing late sequelae. Many prognostic profiles and models have been discovered, but neither molecular sub-classification nor prognostic models are routinely used in clinical practice, because both are currently inconsistent, platform- and population-dependent, highlighting the need for accurate patients’ classification at diagnosis for personalized treatment decision. This project will focus on: validation of multifactorial methods combining existing clinically annotated omics datasets; investigation of ethical and legal aspects of data-driven clinical decision making vs. current evidence-based approach. We start from one of the world largest pools of treated HNC patients (approximately 2500), where the efficacy of the treatment has been recorded in varying forms together with a rich pool of omics and clinical data. In a 3 years study, we will retrospectively analyze these multi-source data using various types of classification, regression and statistical learning methods. We will (a) assess the role of omics, in addition to a currently used staging system to assist outcome of HNC; (b) produce and validate actionable prognostic and predictive models and algorithms to orient personalized treatment decisions, and integrate these into decisions support tools. Clinical endpoints are to improve patients’ stratification for disease outcome and response to treatment, to inform tailored treatment decisions and design clinical confirmatory studies based on new models to personalized medicine. Translational endpoints are to (a) validate effective signatures for HNC outcome and treatment response prediction, (b) propose treatment decision support tools, (c) test the acceptability of Big Data driven research through a small pilot study, drawing new ethical and regulatory frameworks.