Data protection and participant level data access models

Horizon-MSCA-2022-DN-01 Project: 101120360 Data protection and participant level data access models Fabian Prasser Recap 21.10.2025 Patient level data access models 2 Privacy •Privacy in general: “Someone's right to keep their personal matters and relationships secret” •Information privacy: Right to control personal information and how it is used, processed and collected 21.10.2025 Patient level data access models 3 EU Charter of Fundamental Rights Article 7: Respect for private and family life Everyone has the right to respect for his or her private and family life, home and communications. Source: Cambridge dictionary Data protection •“Data protection is about protecting any information relating to an identified or identifiable natural (living) person” •Originates from the right to privacy 21.10.2025 Patient level data access models 4 Article 8: Protection of personal data 1. Everyone has the right to the protection of personal data concerning him or her. 2. Such data must be processed fairly for specified purposes and on the basis of the consent of the person concerned or some other legitimate basis laid down by law. EU Charter of Fundamental Rights Source: https://edps.europa.eu/data-protection/data-protection_en General Data Protection Regulation EU law on data protection and privacy in the European Union and the European Economic Area, that came into effect in 2018 Primary objectives •Protect individuals' privacy and personal data •Harmonize data privacy laws across Europe •Empower individuals over their personal data Principles: lawfulness, fairness, and transparency, purpose limitation, data minimization, accuracy, storage limitation, confidentiality, integrity, and accountability Rights of Individuals: right to access, right to be forgotten (erasure), right to data portability, right to restriction of processing, and more 21.10.2025 Patient level data access models 5 Anonymity: An easy way out? 21.10.2025 Patient level data access models 6 “Magic” Personal data Anonymous data The GDPR (and other laws) doesn’t apply to anonymous data… Does it work? Source: Damien Desfontaines Not so easy… 21.10.2025 Patient level data access models 7 Usefulness Privacy Bad science Breaking the law No risk, all data No risk,no data Max. risk, all data Good approach Source: Barth-Jones, Brüsseler Datenschutz-Symposium, 2016 Introduction: Interactive Scenario •Modification of query results •Originally primary focus of differential privacy •Each query lowers degree of privacy protection (sequential composition) •Example: perturbation of results of count queries Source: Security Methods for Statistical Databases by Karen Goodwin Sanitization Sanitization 21.10.2025 8Patient level data access models Introduction: Non-Interactive Scenario Source: Security Methods for Statistical Databases by Karen Goodwin •Modification of primary data •One-size fits all approach •Preservation of utility can be challenging •Examples: Anonymization, synthetization Sanitization 21.10.2025 9Patient level data access models Introduction to differential privacy 21.10.2025 Patient level data access models 16 Differential Privacy: Formal Definition •Formal definition: •The probability for obtaining a specific result does not change significantly if D1is used instead of D2 •Smaller values of the privacy budget ε result in stronger degrees of privacy protection 21.10.2025 17Patient level data access models Differential Privacy: Basic Mechanisms Randomized response Respondent throws a coin •If coin comes up heads, respondent gives the true answer •If it comes up tails, respondent throws coin again and •Answers „yes“ if it comes up heads •Answers „no“ if it comes up tails Provides ln(3)-differential privacy (given that the probability for each side of the coin coming up is 0.5) 21.10.2025 18Patient level data access models Differential Privacy: Basic Mechanisms Addition of random noise (when outputs are numerical): −Amount of noise depends on the sensitivity −Add appropriately scaled Laplacian noise Exponential mechanism (when outputs are categorical): −Potential outputs r are ranked using a real-valued score function s −Output is chosen with probability proportional to Source:https://en.wikipedia.org/wiki/Laplace_distribution # /media/File:Laplace_pdf_mod.svg Accessed: 28.03.2019 21.10.2025 19Patient level data access models Differential Privacy: Properties •Properties which can be used to construct more complex algorithms out of methods which satisfy ε-differential privacy •Let M, M1, …, Mn satisfy ε, ε1, …, εn-differential privacy, respectively −Sequential composition: −Parallel composition: −Robustness with respect to post-processing: 21.10.2025 20Patient level data access models Thank you for your attention! Questions? Patient level data access models