ESTRO 36 Abstract Book

S102 ESTRO 36 2017 _______________________________________________________________________________________________

Making accurate prediction for these outcomes requires data on many patients, preferably from many institutions using different treatment protocols. Also these data needs to be of high quality, rich in its content and will come from many data domains (clinical, imaging, omics, society, quantified self). Integration of data across these domains and across hospitals worldwide and analysis of such complex datasets is the topic of this seminar. With lung cancer as the running example, known prognostic and predictive factors for outcomes after radiotherapy will be summarized to show the domains from which data needs to be integrated and the challenges associated with each domain. The problem of dimensionality reduction, risk of over-fitting and need for robust validation will be detailed. Then the need and barriers for integration data across vertical data partitions (e.g. departments in a hospital having a ‘slice’ of data for a give patient) and across horizontal partitions (e.g. each hospital has a patient cohort with data) will be discussed. Finally, some example implementations and first results of efforts to integrate and analyse these complex data will be shown. Radiation oncology research requires use of appropriate novel trial designs and robust statistical strategies to ensure reliable decision making. Phase I/II drug trials typically focus on identifying the maximum tolerated dose of a novel therapy. In the context of radiation oncology, this concept is not so clear cut. Various aims of phase I/II trials investigating radiotherapy-novel agent combinations include identifying optimum scheduling, determining the optimum biologic dose, and obtaining preliminary estimates of efficacy. Difficulties in attributing toxicities and evaluating longer-term toxicity also pose problems in designing early phase radiation oncology trials. Opportunities exist to apply novel clinical trial designs to the setting of phase I and II trials in radiation oncology, to overcome some of these difficulties, and to also enable investigation of a number of novel agent-radiotherapy combinations together. Several practical designs addressing this will be discussed in this session, as well as those taking into account known biomarkers in the evaluation of efficacy. SP-0204 Decision support systems and shared decision making M.A. Gambacorta 1 , G. Chiloiro 1 , C. Masciocchi 1 , N. Dinapoli 1 , F. Cellini 1 , A. Re 1 , V. Valentini 1 1 Università Cattolica del Sacro Cuore -Policlinico A. Gemelli, Radiation Oncology, Rome, Italy The personalized approach to cancer patient requires a shared decision which derives from the contribution of every single specialist involved in the cure of any single person affected by cancer. The tumor behaviour and patient outcomes are related to several factors. These factors and their interactions are sometimes poorly known by the doctors who will take the therapeutic decision for the patients. Different branches of medicine have developed their own lines of research which are sometimes difficult to be interpreted, difficult to be integrated with classical clinical factors and for these reasons, difficult to be applied in clinical practice. In the clinical prediction and decision making process, results provided by these researches are rarely included, whereas clinicians usually use few clinical and imaging data for understanding tumor behaviour, predicting patients' outcomes and thus for choosing the the most suitable treatment. In the last years, in the literature, SP-0203 Innovative clinical trial designs for Personalised Radiation Oncology S. Brown 1 1 University of Leeds, Institute of Clinical Trials Research, Leeds, United Kingdom

several studies based on the analysis of large data-base started to appear. These type of studies uses an advanced statistic aiming to find the connections between different covariates in predicting outcomes. The clinical decision is usually based on general guidelines which extrapolate information from randomized clinical trial (RCT). Moreover independent factors, derived from several RCT, are used by the Oncologists to make his prevision on tumor behaviour and consequently to choose the „right treatment“ for a specific patient. RCT enclose patients with characteristics chosen beforehand and usually the innovative information are never or rarely included. This lead to a potential miss of several information that could refine prediction and thus promote personalized treatments and to an erroneous outcomes prediction that can lead to un-appropriate treatment decision for a specific patient. Integrative data analysis has the potential to correlate data of different origins (genetic, radiology, clinic,...) with patient’s outcomes and to create a consistent dataset useful to obtain a trustful analysis for the Decision Support System (DSS). The DSS can easily be applied in clinical practice helping the Oncologist to merge several information that otherwise would be excluded in the process of decision making. The possibility to predict the outcome for a certain patient in combination with a specific treatment with more accuracy, will lead to better identification of risk groups and thus better treatment decisions in individual patients, but it will also stimulate research focused on specific risk groups which try to find new treatment options or other combinations of treatment options for these subgroups. These treatments will be more personalized, which will not only save patients from unnecessary toxicity and inconvenience, but will also facilitate the choice of the most appropriate treatment . The resulting predictive models, based on patient features, enable a more patient specific selection from the treatment options menu and a possibility to share decisions with patients based on an objective evaluation of risks and benefits. Finally, considering the important role that predictive models could play in the clinical practice, clinicians must be aware of the limits of these prediction models. They need to be internally validated taking into account the quality of the collected data. An external validation of models is also essential to support general applicability of the prediction model. Therefore structural collaboration between different groups is crucial to generate enough anonymized large databases from patients included or not in clinical trials. SP-0205 What evidence is needed to assess cost- effectiveness of new technology and how can we get it (easily)? M. Johannesma 1 1 Health Insurance Company CZ, Department Innovation & Advice, Tilburg, The Netherlands The incidence of cancer is increasing worldwide. By this, but also due to the rapid diffusion of new technologies, there is a continues rise in health care costs. For decision- making, primary studies of costs of health technology are gaining importance and cost-effectiveness analysis are increasingly used to estimate the incremental health gain for an incremental use of resources. Economic evaluation in health care, the comparative analysis of alternative interventions in terms of both their costs and consequences, are performed with the aim to determine the value for money of new treatments compared to the prevailing standard. In their aim to support decision making in health care, they represent one aspect within Symposium: Safety and clinical and cost effectiveness of multi-modality IGRT and ART