The CLARIFY digital decision support platform: An artificial intelligence tool for exploring multidimensional cancer data.

e13638 Background: Artificial intelligence has emerged as a mean of improving cancer care with the use of computer science. A digital framework that includes heterogeneous pipelines of real-world data can take advantage of AI, genomics and natural language processing to uncover insights that support decision-making in daily clinical practice. The goal of this study is to present an AI-based solution tool for cancer patients’ and identify clinical factors associated with relapse and survival, and to develop a prognostic model that identifies features associated with poor prognosis and stratifies patients by risk. Methods: This is a hospital-based retrospective registry included in CLARIFY (Cancer Long Survivor Artificial Intelligence Follow-up), a European project supported by the EU Horizon 2020 (grant agreement nº 875160), including 2275 patients diagnosed since 2008 at Medical Oncology Department at Hospital Universitario Puerta de Hierro-Majadahonda with non-small cell lung cancer (NSCLC) and 3000 breast cancer patients. The study was approved by the Ethics Committee at HUPHM (No. PI 148/15) and was carried out in accordance with the Helsinki Declaration. CLARIFY Decision Support Platform (DSP) is an AI-based solution tool which centralizes and analyzes real time anonymized clinical data from heterogeneous sources of data. It shows the clinical user information about individual patients or about the whole population and produces real-time descriptive statistics along with survival analysis (Kaplan Meier estimates and Cox regression model). Integrated data include clinical data from electronic health records from more than 5000 NSCLC patients and breast cancer patients, including more than 1,000,000 clinical notes, more than 900.000 clinical reports, data from wearable devices that produced around 1,000,000 variable values per patient, genomic data and data from quality of life questionnaires. Results: Using the DSP we obtained patients’ profiles, survival probabilities, we stratified over 2000 patients in low and high-risk profile and developed a machine-aided tumour-recurrence prediction model. This computational infrastructure was able to extract knowledge from different data sources allowing clinicians to analyse multiple factors that help stratifying patients by risk, in order to implement a personalized follow up care programme, aiming to make a significant impact in the patients’ quality of life. Conclusions: The reconstruction of the population’s risk profile was achieved and proved useful in clinical practice using AI. This DSP has potential application in clinical settings to improve risk stratification, early detection, and personalized surveillance management of cancer patients and may assist clinicians in their daily clinical practice.