Researchers have developed a new multi-detector evaluation protocol to assess the effectiveness of structure-preserving de-identification techniques in preserving Protected Health Information (PHI) detectability. The study found that replacing PHI with realistic surrogates, such as changing a name from 'Anna S.' to 'Maria S.', resulted in a statistically insignificant drop in PHI detector recall, moving from 76.1% to 74.9%. This minimal loss was attributed to issues with malformed or out-of-distribution surrogates rather than a degradation of the detectors themselves. The findings suggest that well-formed surrogate substitution does not significantly impair the ability of downstream PHI detectors to function. AI
IMPACT This research provides a robust method for evaluating de-identification techniques, crucial for maintaining data privacy in AI applications that process sensitive health information.
RANK_REASON The cluster is based on an academic paper published on arXiv detailing a new evaluation protocol for de-identification methods. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Cedars-Sinai Medical Center
- Chicago
- de-identification
- Equivalence Testing for Psychological Research: A Tutorial
- PHI detectors
- Phi Llm
- Surrogate Substitution
- Tost
- Vidant Medical Center
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
