Mining medical records for early signs of cancer

A team of 11 led by Professor Henk van Weert

Denmark, The Netherlands and UK

 Medical practitioners, epidemiologists, informaticians, psychologists and neuroscientists

 5 years

 

Deep Learning Grand Challenge

The challenge: Detect cancer earlier by interrogating medical and non-medical data sets using machine and deep-learning.

Earlier cancer diagnosis can save lives. This award would allow us to uncover early, until now unknown, clues in large medical, lifestyle and public health databases that could help identify early signs of cancer. We will use new information technology and machine learning as well as examine social, ethical and legal aspects of using this technology. High impact is expected, especially for those cancers which are often diagnosed too late.

Professor Henk van Weert, Principal Investigator

Background

While overall cancer survival is improving, many patients are diagnosed at a late stage of disease when the prognosis is poor and treatment options are more severe. Survival dramatically improves when the disease is diagnosed at the earliest stages. However, for some cancers, very few of the early signs and symptoms are known, making it incredibly difficult for doctors to spot and treat it early.

But could medical records and other non-medical databases hold vital clues that indicate early signs of cancer? Professor Henk van Weert and his team plan to use machine-learning technology to mine electronic health records and public databases for clues that could help GPs identify patients that are showing early signs of cancer.

The Research

A project of this ambition requires collaboration on an international scale, with expertise from multiple disciplines. Professor van Weert’s team brings together an exceptional group of GPs, cancer risk experts, record-mining specialists, statisticians and artificial intelligence experts from the Netherlands, Denmark and the UK.

The team plans to analyse the medical records of over 2 million patients in the Netherlands and Scotland, not only for coded, but also for textual data. They will interrogate this data to identify clues that could indicate if someone might be showing the early signs of one of 10 different types of cancer. These clues could be hidden in notes taken about the patient’s previous medical conditions, changes in their behaviour, or the GP’s knowledge of the family. Not stopping there, the team will also look beyond the medical world and explore public health and government databases on lifestyle and social environment. Analysing these datasets alongside medical records will offer unprecedented insight into the early signs of cancer.

Combined, this data will feed into a self-learning computer algorithm that will be validated using data in Scotland, Denmark and the Netherlands to ensure it is applicable to other data sets and healthcare systems. The team’s eventual aim is to integrate the algorithm into the electronic medical record programme that GPs use every day, helping them decide whether to refer a patient for further investigation. If successful, this approach could be expanded to other cancer types and other medical conditions in the future.

Importantly, the team are consulting cancer patients in all three countries to help inform and guide the project and, crucially, explore the ethical considerations of using personal data. Taking patients’ cancer diagnosis experiences and perspectives into account is vital to ensure the project delivers relevant and realistic outputs for future patients.

Impact 

Early detection doesn’t just have the potential to save more lives, but could help more people avoid the most aggressive cancer treatments in favour of kinder alternatives. By using the latest artificial intelligence technology, this research could revolutionise the first step in many patients’ cancer journey, helping GPs spot the early signs of the disease and getting patients the treatment they need, faster.

This team is looking to combine vast amounts of complicated data, interestingly from both medical and non-medical sets. They will be looking for patterns that can predict early signs of cancer or cancer susceptibility. This could be revolutionary for the UK healthcare system and importantly the team are keen to make sure the algorithm they produce is replicable and helpful for GPs diagnosing cancer across the world.

Professor Ed Harlow, Grand Challenge Advisory Panel 

The Team

 

Professor Henk van Weert

Grand Challenge Shortlisted Team Principal Investigator
Professor of General Medical Practice

Country: The Netherlands
Organisation: Academic Medical Center, Amsterdam
Discipline: General medical practice

 

Professor Ameen Abu-Hanna

Professor of Medical Informatics

Country: The Netherlands
Organisation: Academic Medical Center, Amsterdam
Discipline: Medical informatics​​

 

Professor Marjolein Berger

Professor of General Practice

Country: The Netherlands
Organisation: University Medical Center Groningen
Discipline: General medical practice

 

Dr Christine Campbell

Reader in Psychology

Country: UK
Organisation: University of Edinburgh
Discipline: Lecturer
 

 

Professor Niek de Wit

Professor in General Practice

Country: The Netherlands
Organisation: University Medical Center Utrecht
Discipline: General medical practice

 

Dr Claire Grover

Senior Research Fellow in Informatics

Country: UK
Organisation: University of Edinburgh
Discipline: Informatics

 

Dr Kim Mouridsen

Associate Professor of Neuroinformatics

Country: Denmark
Organisation: Aarhus University
Discipline: Neuroscience and medical informatics

 

Dr Pauline Slottje

Senior Researcher in General Practice and Elderly Care Medicine

Country: The Netherlands
Organisation: VU University Medical Center
Discipline: Epidemiology

 

Professor Carla van Gils

Professor of Clinical Epidemiology of Cancer

Country: The Netherlands
Organisation: University Medical Center Utrecht
Discipline: Epidemiology

 

Professor Peter Vedsted

Professor of General Practice

Country: Denmark
Organisation: Aarhus University
Discipline: General medical practice

 

Professor David Weller

James Mackenzie Professor of General Practice

Country: UK
Organisation: University of Edinburgh
Discipline: General medical practice