Estimating the re-identification risk of clinical data sets
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
BACKGROUND: De-identification is a common way to protect patient privacy when disclosing clinical data for secondary purposes, such as research. One type of attack that de-identification protects against is linking the disclosed patient data with public and semi-public registries. Uniqueness is a commonly used measure of re-identification risk under this attack. If uniqueness can be measured accurately then the risk from this kind of attack can be managed. In practice, it is often not possible to measure uniqueness directly, therefore it must be estimated. METHODS: We evaluated the accuracy of uniqueness estimators on clinically relevant data sets. Four candidate estimators were identified because they were evaluated in the past and found to have good accuracy or because they were new and not evaluated comparatively before: the Zayatz estimator, slide negative binomial estimator, Pitman's estimator, and mu-argus. A Monte Carlo simulation was performed to evaluate the uniqueness estimators on six clinically relevant data sets. We varied the sampling fraction and the uniqueness in the population (the value being estimated). The median relative error and inter-quartile range of the uniqueness estimates was measured across 1000 runs. RESULTS: There was no single estimator that performed well across all of the conditions. We developed a decision rule which selected between the Pitman, slide negative binomial and Zayatz estimators depending on the sampling fraction and the difference between estimates. This decision rule had the best consistent median relative error across multiple conditions and data sets. CONCLUSION: This study identified an accurate decision rule that can be used by health privacy researchers and disclosure control professionals to estimate uniqueness in clinical data sets. The decision rule provides a reliable way to measure re-identification risk.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.013 | 0.181 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.018 | 0.070 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it