Data-driven Framework for Personalised Cancer Screening

Kreft er en av de største årsakene til sykdom og død over hele verden. Mer enn 33 000 nordmenn får kreft hvert år og 11 000 nordmenn døde av kreft i 2017. En stor del av disse kreftsykdommene kan forebygges. For eksempel har masseundersøkelser av livmorhalskreft, som ble etablert i Norden på° 1960-tallet, redusert antall tilfeller av kreft samt dødelighet med 80 %. Til tross for denne suksessen er det fortsatt en stor utfordring å forbedre screeningsprogrammet for å minimere overhypping screening og underbehandling, og dermed redusere utgiftene i et bredere helseperspektiv. Dagens kunnskap om sykdommen kombinert med et vell av tilgjengelig data og moderne teknologi kan tilby langt bedre personlig forebygging ved å estimere personlig optimal oppfølging. Eksisterende automatiske systemer for screening mot livmorhalskreft er imidlertid ekstremt konservative, siden de bare kan identifisere pasienter som er forsinket til rutinemessig screening, uten å gi personlige anbefalinger for oppfølging. Gjennom intelligent bruk av eksisterende registre og personlig helsedata, har DeCipher utviklet et data-drevet rammeverk for å gi en personlig kontinuerlig risikovurdering for å utvikle kreft, samt identifisere undergrupper av individer og faktorer som fører til lignende sykdomsprogresjon. Ved å identifisere disse undergruppene forventes det å redusere over- og underbehandling, gi en personlig screeningplan for hver kvinne og redusere kostnadene ved screeningprogrammer. DeCipher består av et enestående tverrfaglig forskerteam innen maskinlæring, screeningsprogrammer og epidemiologi. Resultater fra DeCipher vil bli gjort tilgjengelig for klinikere og individer.

DeCipher is a multidisciplinary venture with a clear vision: to develop novel interpretable data-driven tools for personalised cancer screening. The goal is to empower women to make better-informed health-related decisions in combination with healthcare professionals. By shifting from population-based statistical follow-ups to individually tailored screening, the project results will contribute to sustainable healthcare system that benefits both the individual patients and society at large. The theoretically grounded developed methods using temporal, sparse, and multimodal data for predictive analysis has lead to scientific impact within machine learning, computer science, and neighbouring fields. The methods are easily transferable and extendable to other data and applications. The project helped to consolidate the leadership of individual partners and catalyse a milieu for multidisciplinary collaboration between AI and healthcare domains.

Cancer is a major cause of morbidity and mortality worldwide, but a large proportion of the incidents are preventable. For example, mass-screening Nordic programs of cervical cancer have a proven strong effect for preventing cancer at the population level and have produced large amounts of individual and clinical data, centrally organised at nationwide registries. Despite this success, minimising over- and under-treatment, and, thus, reducing expenditure, remains a major challenge. Existing automatic decision support systems for cervical cancer prevention are, however, extremely conservative as they are mostly limited to identifying patients who are overdue for their next routine screening. Current knowledge about the cancer, together with a wealth of available data and modern technologies, can offer far better personalised prevention. DeCipher aims to develop a data-driven framework to provide a personalised time-varying risk assessment for cancer initiation and identify subgroups of individuals and biomarkers leading to similar disease progression. By unveiling structure hidden in the data via randomisation and probabilistic tools, we will develop novel theoretically grounded machine learning methods for analysis of temporal, sparse, and multimodal data. DeCipher consists of an excellent multidisciplinary research team from diverse fields such as machine learning, data mining, screening programs, and epidemiology. Leveraging Nordic screening programs and data, the project will enable better and more accurate cancer screening. Our foundational and algorithmic progress will also enable integration of data-driven techniques into biomedical domain, thus corresponding to the Medium-term Time Horizon objectives. Available to screening programs, clinicians, and individuals in the population, the DeCipher results will allow for improvement of individual’s preventive cancer healthcare while reducing the cost of screening programs.