Cancer registries as fuel and engine for data driven rapid learning health systems, an interview with dr. Xander Verbeek

Verbeek: ‘The mission of our organization is to reduce the impact of cancer with insights from data in the Netherlands Cancer Registry. This is a population based registry with pivotal data on disease, care and outcomes along the patient journey, which are abstracted from various distributed sources across care settings. It exists since 1989 and currently contains data on 2.7M patients. Worldwide it is one of the most renowned cancer registrations in terms of extend and quality of data.’

Cancer registries and their role in oncology

‘There are two ways to look at cancer registries. One way is as source of objective real world data. Insights from these data have a demonstrated role in reducing the impact of cancer, through epidemiologic research, but also health economics research, research on optimal organization of care, evaluation of screening programs. This research results in more than 200 scientific publications annually based on data from the Netherlands Cancer Registry (NCR). And cancer registry data are also seen as pivotal in EU beating cancer plan. So from this perspective, cancer registrations can be seen as fuel or the solar power of a learning system.

Another perspective is to regard cancer registrations as organizational entities that over the past decades have built an infrastructure that allows for abstraction and curation of real world data, for knowledge generation from that data and for dissemination of insights to make real world impact. From this perspective cancer registries can also been seen as the engine of a learning system.’

New frontiers for cancer registries considering developments in AI 

‘For sustained value creation it is essential that cancer registries embrace FAIR data principles and make sure they become more interoperable with other stakeholders in an ever more digital ecosystem. This is essential to evolve towards rapid learning health systems that can even scale beyond oncology. I think there are three key ingredients to achieve this;

To become more interoperable we need to embrace a sociotechnical systems design approach that also includes AI, next to design and implementation of standards and mainstream IT solutions alone.

Secondly, solutions such as Federated Learning will become essential in order to be able to exploit data from ever more distributed sources that are available and at the same time protect patient privacy,

And finally we need AI and Decision Support Solutions to make sure we can better exploit the huge amount of insights that come forth from that data for the benefit to patients, public health and society.’

Sociotechnical design approach and AI

‘Currently, for a significant part we depend on data managers for the abstraction of cancer registry data from patient records. We are participating in national programs to design and implement standards in electronic health records and clinical practice to optimize registration via electronic data exchange. There is a mutual benefit here for cancer registries and healthcare providers, because eliminating the need to manually copy and paste data between systems will help reduce registration burden. On the other hand cancer registries are not simply a complete 1:1 copy of data in health records. Registration of data for a major part also relies on human reasoning. Here is where AI can potentially help to support registration. For instance by applying algorithms for contextual reasoning on the validity of single data item taking into account the entire patient record. Additionally, the sociotechnical systems design approach will be needed because introducing standards, mainstream IT solutions or more advanced AI based solutions in practice should be done a way that doesn’t disturb the care process, for instance cognitive focus of care providers.’

Federated Learning

‘For the research challenges in oncology we need more data than available in any single data source. Traditionally, we would send data from those various sources to a researcher for doing the calculations. But this model raises some concern on how primarily responsible parties for curation of the data can guarantee privacy or data quality. On the other hand shielding the data from use for research to protect privacy is not desirable either, nor necessary. The model of federated learning allows us to keep data in place at the original sources. And instead of sending data to the researcher an algorithm is send to the various databases where calculations are then executed locally. And in this case only analysis results are exchanged, and combined, instead of the privacy sensitive source data itself. In the Netherlands we are collaborating on this approach with a broad community and this has been dubbed as the Personal Health Train. Were the train refers to the algorithm that travel via tracks, the learning infrastructure, to data stations. Those who are interested can have a look at the website of Vantage6.ai, which provides an overview of the infrastructure that we have built so far which we already apply in for routine epidemiological research on data from international cancer registries. The future is happening now.’

A.I. and Decision Support

‘A proven way to disseminate insights from data, is through publication in peer reviewed scientific journals or clinical practice guidelines. But this way leaves up a lot of responsibility for acting upon those insights to the stakeholders for which the insights are intended for. So complementary to publication we also need to explore and invest in AI and decision support systems that can seamlessly integrate those insights in a way that optimally fits the cognitive framework and workflow of the decision makers. This holds for clinicians, but also for policy makers, executives or politicians. For example at IKNL we apply FAIR principles to data. But we have also build a system, called Oncoguide, that can be used to model and represent knowledge that is used in practice today based on FAIR principles. For example guidelines, or indications for innovative medicines. And with this we have building block for a hybrid intelligence solution in which we combine real world FAIR data with FAIR knowledge. This way we can learn from real world data in the Netherlands Cancer Registry to improve upon the knowledge that is used to guide clinical practice already today.’

Europe’s beating cancer plan

‘In the EU beating cancer plan cancer, registries are given an important role. But the plan also refers to the disparity between cancer registries within the EU in terms of the extent and quality of data. I believe that cancer registries by working together can evolve towards both fuel and engine for this European plan. So I invite everyone who is interested to contact me.’