“Data is the Only Way to Meet the Future Needs of Our Patients”
Machine learning makes it possible to use high-volume data for routine clinical care and personalized medical care.
Martin Lindner | Feb 03, 2016
Peter Fasching, a gynecologist at Erlangen University Hospital in Germany, works closely with international teams to research the role that genomes play in breast cancer. Read about the challenge of transferring this knowledge to clinical practice, and the potential of smart data for personalized medicine.
High-throughput genome analyses and the increasing digitalization of imaging and patient data have multiplied the amount of medical information that is available. As things stand, the information is not being merged and analyzed to make it usable for personalized medicine.
Machine learning makes it possible to use high-volume data (big data) for routine clinical care and personalized medical care.
The first genomic tests already facilitate more accurate prognoses and personalized disease management in, for instance, breast cancer. In the future, computerized decision-support systems could intelligently link heterogeneous data sources, thus simplifying routine clinical care for physicians and turning big data into smart data.
Professor Fasching, does data make us healthy?
Fasching: Not yet, but we are on the threshold of being able to use the enormous volumes of data collected in research and clinical practice to truly make people better.
You mean, for example, using a patient’s genomic data.
Fasching: Yes. We have known the human genetic code – with its three billion DNA building blocks – since the year 2000, and we have learned much more about it since then, such as how genes are regulated and translated into proteins. So how can we now use all this data to benefit individual patients? As a comparison, a basic Microsoft Excel spreadsheet can currently contain around 16,000 columns. Even if we only wanted information on one million DNA components at our fingertips when treating a patient, it would still exceed our powers of imagination.
What is your solution?
Fasching: It will be important to identify patterns and regularities in the data with, for example, machine learning (i.e. with adaptive computer programs), and to then apply them to individual cases. That’s something that we, as physicians, need to learn – and it is already changing research.
You research the role that the genome plays in breast cancer. What are your findings?
Fasching: Here at Erlangen University Hospital, we have a database with information on around 12,000 patients who have participated in our Bavarian Breast Cancer Case-Control Study. For all of these women, we have pseudonymized data about the genome and about some of the protein patterns in the tumor cells. We are not unique in this; similar breast cancer databases exist in many countries. In fact, many of today’s research questions can only be answered through international cooperation. One recent collaboration, which involved over 200,000 patients and also included our data, showed that variants in more than 70 genes influence the risk of breast cancer.1 While we have long known about the serious effects of a few specific breast cancer genes (the most well-known ones being BRCA1 and BRCA2), we can now scan practically the entire genetic blueprint for less obvious genetic risks. The key to progress will be to correlate this completely new dimension of knowledge with what we already know about cancer.
What might this mean specifically?
Fasching: The first use for these genome-wide analyses is to determine the prognosis of the illness. We already use gene expression tests to identify breast cancer patients who, thanks to their favorable genomic activity profile, have a lower risk of relapse and therefore do not need to undergo chemotherapy after surgery. Naturally, this information is extremely important to individual patients.
Do doctors need a molecular-biological profile for each patient in order to provide this kind of personalized medicine?
Fasching: It is undoubtedly technically possible to create individual genomic profiles today. On the one hand, it’s a question of cost. Genome analysis requires a great deal of time and effort. We also have to establish how we want to manage the new data in the future – for example, where the data will be stored, who can access it, and how patients and their families will deal with all this new information. But there is another issue: it is simply an enormous challenge to interpret genome-wide analyses so that they can actually be used in clinical decision-making – i.e. to translate smart data into specific medical care. There is a gap between the knowledge that variations in over 70 areas of the genome influence the risk of breast cancer, and the question of what the doctor should do when it comes to the individual patient. We want to close this gap because we believe that the data can indeed help to personalize patient care. Using the data is the only way to better meet the needs of our patients.
Clinical Data Intelligence – An Interdisciplinary Research Project
A number of medical and artificial-intelligence research institutes in Germany are working on an interdisciplinary research project with Siemens experts in machine learning, semantic modeling, and medical image analysis. The aim is to lay the foundations for enhanced usability of high-volume health data for routine clinical care using innovative IT platforms. Aside from the pseudonymized genome data from around 12,000 breast cancer patients at Erlangen University Hospital (see interview), the project also combines databases from the Charité (Berlin) that contain information on over 4,000 long-term patients who have kidney failure or have had a kidney transplant.
With the help of machine-learning methods, the data is used to model chronological disease progression, medical decisions, and complications that arise in specific patients. The goal is to develop a computerized decision-support system that can predict complex drug interactions in individual patients, thus simplifying routine clinical care for physicians. Other teams in the interdisciplinary research project are developing processes for automatically recognizing patterns in imaging data or extracting semantic information from medical findings and reports. The project, which began in 2014, is sponsored by the German Federal Ministry for Economic Affairs as part of its Smart Data technology program.
You are collaborating with several other research institutes on a project about clinical data intelligence. It is by no means solely about genomic data. (see sidebar)
Fasching: That’s correct – for example, we also want to link imaging data with genetic analyses or the patient’s history. To do that, we need intelligent IT platforms.
A type of clinical supercomputer?
Fasching: First we have to develop software solutions for sorting the increasing volumes of clinical data so that we can understand the data better and see how they are connected. In the future, we might have computerized decision-support systems that could, for example, flag up genetic risk constellations or patterns in imaging data that might be relevant to a breast cancer patient’s prognosis, and then generate treatment suggestions for doctors. New, integrated diagnostic devices are also conceivable.
What do you envisage exactly?
Fasching: An idea that we are currently investigating is improved mammography screening. Mammograms often do not produce reliable assessments in women with dense breast tissue, and an additional ultrasound can be useful. Our vision is to equip a combined mammography and ultrasound device with software that not only increases the informative value of the mammogram, but also integrates data about a patient’s individual cancer risk. This enhanced screening would then be offered to high-risk patients, and we could provide personalized screenings. This is a simple example of how sensible data integration can drive the future of cancer medicine.
About the Author
Martin Lindner is an award-winning science writer based in Berlin, Germany. He went into journalism after completing his medical studies and a doctoral thesis on the history of medicine. His articles have appeared in many major newspapers and magazines in Germany and Switzerland.
The statements by Siemens’ customers described herein are based on results that were achieved in the customer's unique setting. Since there is no "typical" hospital and many variables exist (e.g., hospital size, case mix, level of IT adoption) there can be no guarantee that other customers will achieve the same results.