TL;DR
This paper introduces CAMRA, a novel protocol for auditing access to electronic medical records (EMRs) that preserves patient confidentiality. It enables multiple organizations to collaborate on audits without revealing sensitive patient identifiers, addressing a critical privacy challenge in healthcare.
What Problem Does This Paper Address?
Accessing a patient's medical record promptly is crucial for effective healthcare. However, to facilitate this, healthcare organizations often grant broad access privileges to Electronic Medical Record (EMR) systems. This can lead to unauthorized access and violations of patient privacy. EMR access logs are used to investigate these potential breaches, but they often lack sufficient information. To conduct a thorough audit, data from disparate systems (like HR databases) must be merged. This merging process, however, can inadvertently expose sensitive patient and care provider information, creating a privacy paradox: the audit needed to protect privacy can itself compromise it. The paper aims to solve this by allowing for secure, privacy-preserving record linkage across organizational boundaries for audit purposes.
Methods
The core of the CAMRA protocol is a privacy-preserving technique for linking data from different organizations without revealing sensitive identifiers. It leverages a commutative cryptography approach, specifically a quasi-commutative encryption method. This allows organizations to encrypt their data using their own keys, and then encrypt it again using another organization's keys. The key innovation is that the order of encryption doesn't matter, and crucially, the protocol is designed to be extended to an asymmetric keyed cryptosystem, allowing for the recovery of original identifiers using specific key pairs.
The protocol involves a semi-trusted third party that facilitates the record linkage. Organizations encrypt their datasets and send them to this third party. The third party performs record linkage on these encrypted datasets. The results are then sent back to the EMR auditor, who can then decrypt the information to identify potential violations. The protocol also incorporates a "blinding" mechanism to prevent information leaks during the encryption and decryption processes, ensuring that neither the third party nor other participating organizations can deduce sensitive information.
Datasets
Not discussed in the paper. The paper focuses on a protocol and its theoretical underpinnings, using hypothetical examples like Vanderbilt University Medical Center (VUMC) and its HR department for illustration.
Key Insights and Ideas
- Privacy-Preserving Audits are Possible: The paper demonstrates that it's feasible to conduct thorough audits of EMR access logs without compromising patient confidentiality, a significant step towards balancing security and privacy.
- Commutative Cryptography as a Solution: The use of commutative cryptography is a clever way to enable secure data sharing and linkage across multiple organizations without requiring them to trust each other with their raw data or encryption keys.
- The Role of the Semi-Trusted Third Party: The protocol effectively isolates the sensitive linkage process to a third party that only sees encrypted data, minimizing the risk of direct data exposure.
- Addressing the "Break the Glass" Model's Weaknesses: While many healthcare systems rely on a "break the glass" model (broad access with harsh penalties), this paper offers a technical solution to detect misuse more effectively and with less privacy risk.
Key Results
The paper presents a protocol that is:
- Collusion-Resistant: It's designed to prevent any set of organizations, including the third party, from learning sensitive identifiers from the encrypted data.
- Scalable: The architecture can be extended to accommodate an arbitrary number of participating organizations by adding more encryption/decryption keys.
- Extensible for Differential Responses: The protocol can be augmented to provide tailored responses to each participating organization.
The paper does not present quantitative results from a deployed system but rather outlines the protocol's design and its theoretical properties.
Limitations and Future Work
- Fault Tolerance for Data Variations: The current protocol struggles with typographical errors or variations in personal information (e.g., "Jon" vs. "Jonathon") because hashes and encrypted identifiers lose their inherent similarity. The authors note that existing methods for measuring string similarity in encrypted environments do not scale for multi-party CAMRA. Future work could involve developing scalable record linkage algorithms for encrypted data to reduce false non-linkage rates.
- Third-Party Involvement: While the protocol uses a semi-trusted third party, the authors express a desire to develop confidential EMR audit models that limit or remove the involvement of third parties altogether, citing research in computational theory that suggests this is possible, though computationally burdensome in practice.
- Trust Between Organizations: The paper acknowledges that the protocol can be extended to detect malicious actions (like using faulty encryption keys) when trust between organizations is low, referencing prior work. However, the core protocol assumes a certain level of cooperation.
Summary updated