SECURITY AND DATA NORMALIZATION COLLABORATION sharps.org Discussion by Mark Frisse and Carl Gunter
The Emerging Clusters Four groups of activities have emerged Policy rules and modules Audit of EHR access Encryption and trusted base Telemedicine Discussion today: areas of possible overlap with SHARPN Discussion of Audit Toolkit Some general discussion questions 3
Introduction Audit is important for EHRs Heavy reliance on accountability Critical trust with patients Current techniques are too ad hoc and reactive Need audit that is meaningful To do this: develop audit techniques that are more portable based on standardized logs Extensible Medical Open Audit Toolkit (EMOAT) Carl Gunter, David Liebovitz, Brad Malin, Sanjay Mehrotra together with staff and students 4
Background and Related Work Standards: DICOM, RFC 3881, IHE ATNA, NHIN Audit Log Requests, HL7 PASS Audit Analysis Community-based Anomaly Detection (CADS) Patient Flow-based Anomaly Detection (PFADS) Process Experience-based Access Management Role-Up Algorithm for role engineering Reporting support for HHS Rule 45 CFR Part 164 5
Towards Standardized Log Analysis Mapped attributes are ones that have a standard semantics Mapping type is a pair consisting of required mapped attributes and optional mapped attributes Application is compatible if it has the required attributes, may be able to use the others too Three focus areas so far The Matrix Role Hierarchies Role Mapping 6
Analytic Techniques for Scoring CADS: Create a social network from joint access to a record. Use k th nearest neighbor to form communities. Look for outliers and their neighbors. PFADS: Form a graph from observed transitions between record accesses by users grouped in classes. Rare transitions are considered outliers. 8
Analytic Techniques for Role Engineering and Reporting Role Engineering Role-up: Train a naïve Bayes classifier on actions of roles over an audit set. Use this to predict roles from actions. Choose a parameter to balance specificity and accuracy. Reporting Two views: operations and patient Scoring to aid operations Role mapping to aid patients 9
EMOAT Part 1 of 2 10
EMOAT Part 2 of 2 11
CADS Comparison 12
Reporting Application 13
Future Work on EBAM/EMOAT Continued tuning and testing of the algorithms Addressing the scalability and flexibility of EMOAT Extensions to HIE: Communication between Cerner and Epic systems within NMH, Illinois controlled substance system Extending the Matrix: JHU, AthenaHealth, eClinicalWorks, and GE Centricity. Audit worshop? Coordination with SHARPN? 14
the policy “cloud” Denise Vanderbilt JHU, NWU Helen Policy Synthesis Group Frisse (lead) Denise (co-lead) McCarthy (analyst) VU ISIS Stanford CMU Selected Research Activities Policy prioritization and synthesis common use cases & syntax Laws & Policies RTI Summaries Dartmouth Summaries Developer- readable representation Formal representation backward links U Of IL NWU JHU
Potential discussion SHARPn & privacy, consent Granularity (e.g., PCAST). Roles – access, consent, and encryption Test beds – who & for what purpose Thinking through the continuum from “top secret” to “information altruists” Software tools 16
Further Discussion How does SHARPn view the general issue of privacy, consent, and the granularity issues (e.g., PCAST). To what extent are SHARPn investigators working through formalizing roles and other issues? How do people view the continuum from "top secret" to "research for selective reasons" to "total information altruism." How are people thinking about these issues? What approaches seem most apropos? 17