Medicine: the debate over Big Data

From detecting how Zika or Ebola might spread to predicting surgical outcomes, from anticipating the side effects of drugs to the likelihood of readmission, Big Data can help revolutionise healthcare. But while Wil van der Aalst, a computer scientist at the Eindhoven University of Technology, champions digging deep into patient data, Nathan Lea of University College London’s Institute of Health Informatics sees big hurdles for this practice.

Technologist: How confident are you that large-scale analysis of patient data can help Europe’s healthcare systems?

Wil van der Aalst: Hospitals collect huge amounts of data that provide unprecedented opportunities for process mining, where the focus is on end-to-end behaviour involving patients, staff and machines. For instance, we can predict surgical outcomes in specific demographics or compare the quality of different intensive-care units.

Nathan Lea: It’s not yet clear what Big Data’s usefulness may be in specific contexts. How do you fit a square peg into a hole if you haven’t even discerned its shape? There are some very positive programmes around. For example, the European Medical Information Framework, which uses Big Data to identify subjects with Alzheimer’s disease who have not yet reached the stage of dementia. By connecting relevant studies across Europe, the project is driving large-scale research on risk factors for neurodegenerative disorders. However, such programmes generally require a lot of time and money.

T. What are the challenges for Big Data and informatics in a privacy-critical healthcare setting?

WvdA. On one hand, people do not see the complexity of the challenges; on the other hand, there is little awareness of the solutions that currently exist. Access control is one issue: we need to understand where data access is unjustified. Of course, doctors need access to information, but first we need to ensure that we protect patients; only then can we think about analysing their data. Data mining helps improve processes such as access control by asking deeper questions about outliers, but hospitals aren’t yet applying such solutions on a daily basis.

NL. From check-ups to emergencies, healthcare is based on a confidential therapeutic relationship between medical professional and patient in different contexts. Once I see good examples of meaningful intelligence, my confidence in it will grow.

T. Can you outline the most promising approaches for building a coherent new data ecosystem that protects patient privacy?

WvdA. One approach is to make use of various encryption types, such as homomorphic encryption. That means we can encrypt event data in a way that allows it to be analysed for a specific purpose, so it isn’t decrypted first and therefore can’t lead back to the individual patient.

NL. In the security sector, 100% guarantees don’t exist. The risk is re-identification of an individual and their records; you’ve got to accept that no matter how smart the encryption techniques, there will always be some error. We must view security, privacy and the challenges around Big Data as weaving a tapestry. It’s a tapestry of policy, procedure, best practice and guidance, and the manner in which they’re sewn together is most important. While projects are working together to build resources that manage vast datasets, linking records from multiple sources should be performed only at the local level. If hospitals and social-care organisations are already collecting and linking data, others should not be repeating that work. Keep the data within a context where use is already permitted, legal and ongoing.



, ,