Opening up all medical records would be a huge enabler for science; for both individual studies and the amount of researchers with access to data. But medical data naturally needs to be kept private.

Differential Privacy [1] is a robust framework to tackle this problem in the area of statistical disclosure control in a manner that automatically adapts to the query and available data. We'd like to bring this powerful framework into practice and help make medical data more accessible for medical researchers while at the same time giving stronger privacy protections than what's currently possible in data-sharing agreements.

The platform developed by Nedap Healthcare supports the daily work of many healthcare workers dealing with millions of patient-records in the Netherlands. Our datasets contain everything from a medical diagnosis, prescribed medication, allergies up until the bill that is sent to the insurer and everything in between.

Differential privacy has become more accessible with the release of open-source implementations [3], but there are still gaps with the medical research practice that we'd like to close. We're looking for a bright studentwho is interested in bridging this gap. You will study the fundamental mathematical operations of differential privacy. Analyse the impact of generated noise in a dataset and its effect on the validity of medical statistical tests. Some of our key questions are:

To what extent can medical statistical queries be executed in the context of differential privacy?

How does the adaptive (Gaussian/Laplacian) noise translate to the probabilities obtained via a statistical test (e.g. t-test, chi^2-test)?

Which DP methods scale to larger datasets and what are important aspects to consider in terms of data structures, pre-processing and data quality?

What is the proper epsilon (anonymity)-tradeoff for various queries in the medical domain? How can we assist users in this trade-off in an intuitive manner? [2]

Key references

1. Dwork, Cynthia. "Differential privacy: A survey of results." International conference on theory and applications of models of computation. Springer, Berlin, Heidelberg, 2008.
2. Dwork, Cynthia, Nitin Kohli, and Deirdre Mulligan. "Differential privacy in practice: Expose your epsilons!." Journal of Privacy and Confidentiality 2019.
3. SmartNoise - https://github.com/opendp/smartnoise-samples/blob/master/data/README.md

Our offer

As an intern or graduate student at Nedap, you can work both remotely and at our cool campus in Groenlo. Of course, this is done in consultation with your team and the possibilities depend a bit on your work, but the choice is yours.

You also get plenty of room to "take the lead" in your assignment. Personal leadership and taking responsibility are key and as far as we are concerned, that also applies to students! You will work with enthusiastic professionals who have a lot of knowledge, experience, and expertise from whom you can learn a lot.

To thank you for your efforts, every six months we organize a 'Nedap Studentday' specially for you. It is a super fun day with all the other interns, graduates and working students at Nedap. 

We also offer you a decent monthly internship or graduation allowance. For a fulltime internship or graduation assignment, you'll receive between €325,- and €650,- per month. And if both parties like it, we may even offer you a job. Many students have already stayed on after their assignment!