Pith. sign in

REVIEW 4 major objections 6 minor 42 references

An Algorithmic Pipeline for GDPR-Compliant Healthcare Data Anonymisation: Moving Toward Standardisation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proposes a three-stage algorithmic pipeline that classifies healthcare dataset columns by re-identification risk and measures anonymisation utility, so GDPR-compliant sharing can be standardised.

desk verdict Useful open-source integration of existing anonymisation components, but the GDPR-compliance claim outruns the evidence because the pipeline leaves sex and similar low-cardinality attributes unmodified. read the letter →

arxiv 2506.02942 v1 pith:5CO3AZLF submitted 2025-06-03 cs.CR

Hamza Khan , Lore Menten , Liesbet M. Peeters This is my paper
classification cs.CR
keywords GDPRanonymisationquasi-identifiersre-identificationriskk-anonymitynon-uniformentropyhealthcaredataopen-sourcepipeline
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the vague GDPR definitions of quasi-identifiers and sensitive attributes can be turned into a concrete, repeatable computer workflow. It presents a three-stage anonymisation pipeline: identify which columns carry re-identification risk, de-identify those columns, and evaluate the result with standard privacy and utility metrics. The authors argue that because the pipeline's code is public and its thresholds are explicit, different health-data teams can produce comparable, auditable anonymisation decisions. If the pipeline works as claimed, healthcare data sharing that currently depends on subjective interpretation could become standardised, with privacy and utility reported in the same way across datasets.

What carries the argument

The mechanism that carries the argument is the g-distinct re-identification risk rate and the α/β thresholding rule applied to it. For each column, every value's uniqueness is measured, the column-level risk is the sum of these g-distinct values, and the α/β comparison converts the number into a category: sensitive, quasi-identifier, or non-sensitive. This is what transforms GDPR's qualitative language into an algorithmic decision. The second mechanism is the QID-dimension search: the pipeline tests subsets of the identified QIDs and selects the smallest one satisfying k-anonymity ≥ 2, ℓ-diversity ≥ 2, and t-closeness ≤ 0.8 while maximising NUE, which ties the identification stage to measurable privacy and utility outcomes.

What would settle it

Take a public microdata file with a known re-identification attack (for example, a hospital discharge table that has been linked to voter records), run the pipeline, and check whether columns it labels non-sensitive—such as sex—can combine with public external tables to re-identify individuals; any successful match would show the classification step is not sufficient on its own.

Watch

Extended reading notes

Core claim

The central claim is that a single algorithmic pipeline can perform the identification step that GDPR leaves open: assigning each column to sensitive, quasi-identifying, or non-sensitive on the basis of a computed re-identification risk rate. That rate comes from g-distinct uniqueness, i.e., how many distinct values a column has and how rare each is; columns above a chosen α threshold are labelled sensitive, those between α and β are quasi-identifiers, and those below β are non-sensitive. After de-identification, the pipeline measures privacy with k-anonymity, ℓ-diversity, and t-closeness, and utility with non-uniform entropy (NUE), searching for the smallest QID set that keeps privacy above chosen floors. On two mock real-world datasets (500 and 1000 rows), k-anonymity rose from 1 to 4 and from 1 to 110 respectively while NUE stayed near 69 percent, which the authors take to show that the workflow can raise privacy without disproportionate information loss.

Load-bearing premise

The load-bearing premise is that a column's re-identification risk computed from g-distinct uniqueness, together with user-chosen α and β thresholds, is sufficient to classify columns as sensitive, quasi-identifying, or non-sensitive in a way that protects privacy; if that classification is wrong for easily known low-cardinality attributes, the pipeline's GDPR-compliance claim falls.

Editorial extensions

If this is right

  • Health-data teams can replace manual, judgement-based QID/SA labelling with the open-source workflow, making anonymisation decisions auditable and repeatable.
  • Using NUE as a single utility number lets different anonymised datasets be compared on a common scale.
  • The observed results imply that larger datasets can reach much higher k-anonymity with about the same utility as smaller ones, so dataset size is a first-order factor in how much de-identification is needed.
  • Because the code ships with mock datasets mimicking real-world data, organisations can practise and test anonymisation strategies before working with actual patient data.
  • With small extensions, the QID-dimension stage can compare two different de-identification approaches applied to the same QID, helping users pick the one that retains more utility.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A purely data-driven threshold can label low-cardinality columns such as sex as non-sensitive even though an adversary can easily know them; the paper's own results show this, so a safe deployment would need to let users force such columns into the QID set, which the authors suggest but do not implement.
  • The α and β thresholds remain user-selected rather than derived from a target re-identification risk, so the pipeline standardises the procedure but not the most consequential privacy choice.
  • NUE is one supported utility measure, but organisations with a specific analysis in mind may still need task-specific metrics, since NUE summarises distributional similarity rather than downstream analytical validity.
  • A natural testable extension would be to calibrate α and β against actual record-linkage attacks on public microdata, turning the threshold choice from a user preference into an empirically grounded parameter.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes a three-stage pipeline for anonymising healthcare real-world data: identification of quasi-identifiers (QIDs) and sensitive attributes (SAs) via g-distinct re-identification risk with user-set alpha/beta thresholds; de-identification via suppression, masking, and generalisation; and evaluation of the QID dimension using k-anonymity, l-diversity, t-closeness, and non-uniform entropy (NUE). The pipeline is implemented in Python and tested on two mock datasets of 500 and 1000 rows. The authors report improved k-anonymity (from 1 to 4 and from 1 to 110) with NUE around 69%, and claim this constitutes a GDPR-compliant, reproducible, and standardisable approach to healthcare data anonymisation.

Significance. If the pipeline worked as claimed, it would be a useful open-source contribution: the code and mock datasets are publicly available, the reporting is transparent, and the systematic literature review adds context. However, the central GDPR-compliance claim is not currently supported. Low-cardinality attributes that an adversary can plausibly know are classified as non-sensitive and excluded from the QID set, so the reported privacy metrics overstate protection. In addition, the defined optimal-dimension criterion is internally inconsistent because it asks to maximise NUE, which is defined as information loss. The paper is therefore better understood as a proof-of-concept workflow than as a validated standard. Its strengths include machine-checkable code, reproducible mock data, and a candid Discussion that acknowledges several of these limitations.

major comments (4)
  1. [Methods – Identification stage; Appendix 7] The classification step labels sex and covid19_diagnosis as non-sensitive because their g-distinct rates (0.40% and 0.20% in the two datasets) fall below the beta threshold. These are exactly the kind of attributes an adversary can plausibly know and use for linking: under GDPR Recital 26 and standard statistical disclosure control they should be treated as quasi-identifiers. Because they are left unmodified, the reported k-anonymity values (4 and 110) are computed over an incomplete QID set and overstate the protection actually offered. The Discussion acknowledges this for future versions, but the abstract and conclusion still claim GDPR compliance; this is a load-bearing inconsistency that must be resolved before the central claim can be accepted.
  2. [Methods – Quasi-identifier dimension stage] NUE is defined earlier as quantifying information loss, yet the optimal QID dimension is defined as the subset that 'maximises NUE.' This is contradictory: maximising an information-loss measure selects the least useful output. In the 1000-row dataset, NUE with two de-identified QIDs is 53.61% versus 69.05% with three; under a utility-preserving criterion the two-QID solution would be preferable, yet the paper selects three. Either the criterion should be to minimise NUE (or maximise inverse NUE), or the text must explain why a higher NUE is desirable. As written, the selection rule cannot justify the reported 'optimal' choices.
  3. [Methods – Identification stage; Results – Identification stage] The alpha and beta thresholds are set by the user (25%/1% for 500 rows, 10%/1% for 1000 rows) with no principled rule. Since the entire QID/SA classification, and therefore every downstream privacy and utility number, depends on these two numbers, the claim of a reproducible, standardised identification method is not yet established. The Discussion calls this a major challenge and notes that no consensus exists. The manuscript should at least provide a sensitivity analysis over thresholds and guidance for setting them; without this, the pipeline transfers the subjective GDPR interpretation to threshold selection.
  4. [Results – Identification stage; Appendix 7] The g-distinct re-identification rate for an attribute appears to be approximately the number of distinct values divided by the number of rows (e.g., sex: 2/500 = 0.40%). This metric conflates cardinality with disclosure risk: an attribute with many distinct values but no semantic link to identity would be classified as sensitive, while a binary demographic attribute that can be highly predictive is classified as non-sensitive. The paper does not justify this metric against alternative QID discovery methods. Because the classification is the foundation of the pipeline, this issue is load-bearing and needs to be addressed.
minor comments (6)
  1. [Methods – Usefulness metrics] NUE is described as 'information loss' while the inverse NUE is described as 'retained data utility'; please state explicitly whether higher NUE is better or worse and use consistent phrasing throughout.
  2. [Methods – Quasi-identifier dimension stage] The definition of optimal QID dimension is confusing: the 'smallest subset' and the 'maximises NUE' criteria can conflict; please specify the lexicographic order or a single objective function.
  3. [Results – Pipeline implementation and evaluation] In the 1000-row results, 'NUE was 53.61,%' contains a typo (a comma before the percent sign); please correct it.
  4. [Appendix 7] Please provide the exact formula for g-distinct and for the re-identification risk rate, as the reader cannot verify the reported percentages from the current description.
  5. [Discussion – Validation] The validation against the source paper reports results that were 'not fully identical' but does not specify which classifications differed; please include a short comparison table or a detailed description.
  6. [Discussion – Dataset size] The statement that 'smaller datasets inherently carry a greater re-identification risk' is not universally true; it depends on the number of distinct values and the attacker model. Please qualify this claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the identification and utility components are imported from external cited work, and the reported privacy/utility numbers are computed outputs of an explicitly defined transformation, not predictions forced by the same inputs.

full rationale

The paper's derivation chain is self-contained rather than circular. The identification stage implements the externally cited g-distinct/re-identification-risk classification from Mansour et al. [21] and Jadhav & Borkar [22], and the utility evaluation uses non-uniform entropy taken from external quality-model studies [29,32]; neither component is a self-citation, and the cited works are not authored by the present authors. The alpha and beta thresholds are user-set parameters, not fitted values, and the paper explicitly acknowledges their arbitrariness in the Discussion: 'the arbitrary selection of α and β thresholds poses a major challenge to the objectivity of the results.' The reported k-anonymity, l-diversity, t-closeness and NUE values are computed outputs of a defined de-identification and dimension-selection pipeline, not quantities that equal their inputs by construction. The one legitimate concern, also acknowledged in the Discussion, is that low-cardinality attributes such as sex and covid19_diagnosis fall below the beta threshold and are labelled non-sensitive, leaving them unmodified and potentially weakening the GDPR-compliance claim; however, this is a correctness or adequacy limitation about the QID set, not a circular derivation. There is no self-citation chain that forces the central claim, and no fitted parameter is renamed as a prediction. Therefore no significant circularity is found.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The ledger shows the central claim rests on three domain assumptions: uniqueness-based risk as a classifier, technical privacy metrics as a proxy for GDPR compliance, and NUE as the utility objective. The hand-set thresholds are free parameters that directly determine the reported results. No new entities are postulated.

free parameters (4)
  • alpha threshold (SA vs QID cutoff) = 25% (500 rows); 10% (1000 rows)
    User-selected; attributes with re-identification risk above alpha are labelled sensitive. The two datasets use different values, and the Discussion calls the selection arbitrary. Changing alpha changes the SA/QID split, e.g., edss at 10.04% in the 1000-row dataset is just above the 10% threshold.
  • beta threshold (QID vs NSA cutoff) = 1% for both datasets
    User-selected; attributes with risk below beta are labelled non-sensitive. Sex and covid19_diagnosis fall below this threshold and are treated as NSAs, which the authors acknowledge is questionable from an attacker-knowledge perspective.
  • missing value drop threshold = 85% missing
    Hand-selected cutoff; the covid19_self_isolation attribute was dropped because it exceeded this threshold (88% and 91.8% missing in the two datasets). This affects which attributes enter the pipeline.
  • privacy target thresholds for optimal QID dimension = k >= 2, l >= 2, t <= 0.8
    These targets define the optimal QID dimension. They are policy choices rather than derived values; if they were changed, the selected dimension, reported k-anonymity and NUE would change.
assumptions (3)
  • domain assumption A column's g-distinct re-identification risk rate is a sufficient signal for classifying it as sensitive, quasi-identifying, or non-sensitive.
    The identification stage computes g-distinct values and risk rates, then applies alpha/beta thresholds. The assumption that high uniqueness implies sensitivity and low uniqueness implies non-sensitivity is load-bearing; the authors note that sex and covid19_diagnosis fall below beta and are labelled non-sensitive even though attackers may know these attributes.
  • domain assumption Satisfying k-anonymity, l-diversity and t-closeness at the chosen thresholds is sufficient for GDPR-compliant anonymisation.
    The paper labels the pipeline GDPR-compliant based on these technical metrics on mock data. GDPR requires that individuals are no longer identifiable considering all means reasonably likely to be used; the paper does not perform that legal analysis.
  • domain assumption NUE is a valid general-purpose utility metric and can serve as the objective for choosing the optimal QID dimension.
    NUE is selected from the SLR as the best general-purpose metric, relying mainly on one included study. The paper's own description of NUE as information loss conflicts with the stated goal of maximizing NUE, so this assumption is internally inconsistent.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Algorithmic Pipeline for GDPR-Compliant Healthcare Data Anonymisation: Moving Toward Standardisation." pith.science (2026). https://pith.science/paper/5CO3AZLF

@misc{pith2026250602942,
  author       = {Pith},
  title        = {Pith review of: An Algorithmic Pipeline for GDPR-Compliant Healthcare Data Anonymisation: Moving Toward Standardisation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5CO3AZLF}},
  note         = {Machine review of arXiv:2506.02942}
}
read the original abstract

High-quality real-world data (RWD) is essential for healthcare but must be transformed to comply with the General Data Protection Regulation (GDPR). GDPRs broad definitions of quasi-identifiers (QIDs) and sensitive attributes (SAs) complicate implementation. We aim to standardise RWD anonymisation for GDPR compliance while preserving data utility by introducing an algorithmic method to identify QIDs and SAs and evaluate utility in anonymised datasets. We conducted a systematic literature review via ProQuest and PubMed to inform a three-stage anonymisation pipeline: identification, de-identification, and quasi-identifier dimension evaluation. The pipeline was implemented, validated, and tested on two mock RWD datasets (500 and 1000 rows). Privacy was assessed using k-anonymity, l-diversity, and t-closeness; utility was measured by non-uniform entropy (NUE). The review yielded two studies on QID/SA identification and five on utility metrics. Applying the pipeline, attributes were classified by re-identification risk using alpha and beta thresholds (25 percent/1 percent for 500 rows; 10 percent/1 percent for 1000 rows). Privacy metrics improved k-anonymity from 1 to 4 (500 rows) and 1 to 110 (1000 rows). NUE scores were 69.26 percent and 69.05 percent, respectively, indicating consistent utility despite varying privacy gains. We present a GDPR-compliant anonymisation pipeline for healthcare RWD that provides a reproducible approach to QID/SA identification and utility evaluation; publicly available code promotes standardisation, data privacy, and open science.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [2]

    Sharing Is Caring-Data Sharing Initiatives in Healthcare

    Hulsen T. Sharing Is Caring-Data Sharing Initiatives in Healthcare. Int J Environ Res Public Health. 2020;17:3046

    work page 2020

  2. [3]

    Microdata

    Hundepool A, Domingo-Ferrer J, Franconi L, Giessing S, Nordholt E, Spicer K, et al. Microdata. Stat Discl Control [Internet]. John Wiley & Sons, Ltd; 2012 [cited 2025 May 29]. p. 23–130. Available from: https://onlinelibrary.wiley.com/doi/abs/10.1002/9781118348239.ch3

    work page doi:10.1002/9781118348239.ch3 2012

  3. [4]

    identification

    Faculty of Medicine and Life Sciences, Hasselt University, Belgium * Both authors have contributed equally to this work. Corresponding author: liesbet.peeters@uhasselt.be Background: High-quality, real-world data (RWD) is crucial for various healthcare applications, but this data requires transformations to be shared in compliance with the General Data Pr...

    work page 2023

  4. [5]

    [cited 2025 May 29]

    Regulation - 2016/679 - EN - gdpr - EUR-Lex [Internet]. [cited 2025 May 29]. Available from: https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng

    work page 2016

  5. [6]

    Sensitive attribute privacy preservation of trajectory data publishing based on l-diversity

    Yao L, Chen Z, Hu H, Wu G, Wu B. Sensitive attribute privacy preservation of trajectory data publishing based on l-diversity. Distrib Parallel Databases. 2021;39:785–811

    work page 2021

  6. [7]

    Patient level dataset to study the effect of COVID-19 in people with Multiple Sclerosis

    Khan H, Geys L, Baneke P, Comi G, Peeters LM. Patient level dataset to study the effect of COVID-19 in people with Multiple Sclerosis. Sci Data. 2024;11:149

    work page 2024

  7. [8]

    Demystifying the likelihood of reidentification in neuroimaging data: A technical and regulatory analysis

    Jwa AS, Koyejo O, Poldrack RA. Demystifying the likelihood of reidentification in neuroimaging data: A technical and regulatory analysis. Imaging Neurosci. 2024;2:1–18

    work page 2024

  8. [9]

    Observational health research in Europe: understanding the General Data Protection Regulation and underlying debate

    van Veen E-B. Observational health research in Europe: understanding the General Data Protection Regulation and underlying debate. Eur J Cancer Oxf Engl 1990. 2018;104:70–80

    work page 1990

Show all 42 references
  1. [10]

    Disruptive and avoidable: GDPR challenges to secondary research uses of data

    Peloquin D, DiMaio M, Bierer B, Barnes M. Disruptive and avoidable: GDPR challenges to secondary research uses of data. Eur J Hum Genet. 2020;28:697–705

    2020

  2. [11]

    Are We There Yet? Understanding the Challenges Faced in Complying with the General Data Protection Regulation (GDPR)

    Sirur S, Nurse JRC, Webb H. Are We There Yet? Understanding the Challenges Faced in Complying with the General Data Protection Regulation (GDPR). Proc 2nd Int Workshop Multimed Priv Secur [Internet]. New York, NY, USA: Association for Computing Machinery; 2018 [cited 2025 May ...

    2018

  3. [12]

    Lindqvist J. New challenges to personal data processing agreements: is the GDPR fit to deal with contract, accountability and liability in a world of the Internet of Things? Int J Law Inf Technol. 2018;26:45–63

    2018

  4. [13]

    The Role of Quasi-identifiers in k-Anonymity Revisited [Internet]

    Bettini C, Wang XS, Jajodia S. The Role of Quasi-identifiers in k-Anonymity Revisited [Internet]. arXiv; 2006 [cited 2025 May 29]. Available from: http://arxiv.org/abs/cs/0611035

    2006 arXiv

  5. [14]

    The PRISMA 2020 statement: an updated guideline for reporting systematic reviews

    Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71

    2020

  6. [15]

    Parsifal

    Perform Systematic Literature Reviews [Internet]. Parsifal. [cited 2025 May 29]. Available from: https://parsif.al/

    2025

  7. [16]

    How-to conduct a systematic literature review: A quick guide for computer science research

    Carrera-Rivera A, Ochoa W, Larrinaga F, Lasa G. How-to conduct a systematic literature review: A quick guide for computer science research. MethodsX. 2022;9:101895

    2022

  8. [17]

    STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies

    Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527

    2015

  9. [18]

    EndNote [Internet]

    The EndNote Team. EndNote [Internet]. Philadelphia, PA: Clarivate; 2013 [cited 2025 May 29]. Available from: https://support.clarivate.com/Endnote/s/article/Citing-the-EndNote-program-as-a-reference?language=en_US

    2013

  10. [19]

    Quality assessment in systematic literature reviews : A software engineering perspective

    Yang L, Zhang H, Shen H, Huang X, Zhou X, Rong G, et al. Quality assessment in systematic literature reviews : A software engineering perspective. 2020 [cited 2025 May 29]; Available from: https://acuresearchbank.acu.edu.au/item/8wv75/quality-assessment-in-systematic-literatur...

    2020

  11. [20]

    The Real-World Data Challenges Radar: A Review on the Challenges and Risks regarding the Use of Real-World Data

    Grimberg F, Asprion PM, Schneider B, Miho E, Babrak L, Habbabeh A. The Real-World Data Challenges Radar: A Review on the Challenges and Risks regarding the Use of Real-World Data. Digit Biomark. 2021;5:148–57

    2021

  12. [21]

    Anonymization of Data Sets with NULL Values

    Ciglic M, Eder J, Koncilia C. Anonymization of Data Sets with NULL Values. Trans Large-Scale Data- Knowl-Centered Syst XXIV. Germany: Springer Berlin / Heidelberg; 2016. p. 193–220

    2016

  13. [22]

    Quasi-Identifier Recognition Algorithm for Privacy Preservation of Cloud Data Based on Risk Reidentification

    Mansour HO, Siraj MM, Ghaleb FA, Saeed F, Alkhammash EH, Maarof MA. Quasi-Identifier Recognition Algorithm for Privacy Preservation of Cloud Data Based on Risk Reidentification. Wirel Commun Mob Comput. 2021;2021:7154705

    2021

  14. [23]

    Quasi-identifier recognition with echo chamber optimization-based anonymization for privacy preservation of cloud storage

    Jadhav PS, Borkar GM. Quasi-identifier recognition with echo chamber optimization-based anonymization for privacy preservation of cloud storage. Concurr Comput Pract Exp. 2024;36:e7906

    2024

  15. [24]

    De-Identification Guideline

    Krehling L. De-Identification Guideline. 2020

    2020

  16. [25]

    De-identification of personal information [Internet]

    Garfinkel SL. De-identification of personal information [Internet]. National Institute of Standards and Technology; 2015 Oct p. NIST IR 8053. Report No.: NIST IR 8053. Available from: https://nvlpubs.nist.gov/nistpubs/ir/2015/NIST.IR.8053.pdf

    2015

  17. [26]

    ACHIEVING k-ANONYMITY PRIVACY PROTECTION USING GENERALIZATION AND SUPPRESSION

    Sweeney L. ACHIEVING k-ANONYMITY PRIVACY PROTECTION USING GENERALIZATION AND SUPPRESSION. Int J Uncertain Fuzziness Knowl-Based Syst. 2002;10:571–88

    2002

  18. [27]

    k-ANONYMITY: A MODEL FOR PROTECTING PRIVACY

    Sweeney L. k-ANONYMITY: A MODEL FOR PROTECTING PRIVACY. Int J Uncertain Fuzziness Knowl-Based Syst. 2002;10:557–70

    2002

  19. [28]

    L-diversity: Privacy beyond k-anonymity

    Machanavajjhala A, Kifer D, Gehrke J, Venkitasubramaniam M. L-diversity: Privacy beyond k-anonymity. ACM Trans Knowl Discov Data. 2007;1:3-es

    2007

  20. [29]

    t-Closeness: Privacy Beyond k-Anonymity and -Diversity

    Li N, Li T, Venkatasubramanian S, Labs T. t-Closeness: Privacy Beyond k-Anonymity and -Diversity

  21. [30]

    Privacy Protection in Social Science Research: Possibilities and Impossibilities

    Albright JJ. Privacy Protection in Social Science Research: Possibilities and Impossibilities. PS Polit Sci Polit. 2011;44:777–82

    2011

  22. [31]

    An Experimental Comparison of Quality Models for Health Data De-Identification

    Eicher J, Kuhn KA, Prasser F. An Experimental Comparison of Quality Models for Health Data De-Identification. Stud Health Technol Inform. 2017;245:704–8

    2017

  23. [32]

    Utility-driven assessment of anonymized data via clustering

    Ferrão ME, Prata P, Fazendeiro P. Utility-driven assessment of anonymized data via clustering. Sci Data. 2022;9:456

    2022

  24. [33]

    Utility-preserving transaction data anonymization with low information loss

    Loukides G, Gkoulalas-Divanis A. Utility-preserving transaction data anonymization with low information loss. Expert Syst Appl. 2012;39:9764–77

    2012

  25. [34]

    A Generic Method for Assessing the Quality of De-Identified Health Data

    Prasser F, Bild R, Kuhn KA. A Generic Method for Assessing the Quality of De-Identified Health Data. Stud Health Technol Inform. 2016;228:312–6

    2016

  26. [35]

    Finding Quasi-identifiers for K-Anonymity Model by the Set of Cut-vertex

    Yan Y, Wang W, Hao X, Zhang L. Finding Quasi-identifiers for K-Anonymity Model by the Set of Cut-vertex. 2018

    2018

  27. [36]

    Efficient Algorithms for Masking and Finding Quasi-Identifiers

    Motwani R, Xu Y. Efficient Algorithms for Masking and Finding Quasi-Identifiers

  28. [37]

    Podlesny NJ. Erkennung von Quasi-Identifikatoren zum Schutz der Privatsphäre vor Rückschlüssen in hochdimensionalen DatensätzenQuasi-identifier discovery to prevent privacy violating inferences in large high dimensional datasets [Internet]. Universität Potsdam; 2023 [cited 202...

    2023

  29. [38]

    Learning quasi-identifiers for privacy-preserving exchanges: a rough set theory approach

    Wafo Soh C, Njilla LL, Kwiat KK, Kamhoua CA. Learning quasi-identifiers for privacy-preserving exchanges: a rough set theory approach. Granul Comput. 2020;5:71–84

    2020

  30. [39]

    [cited 2025 May 29]

    ARX – Data Anonymization Tool – A comprehensive software for privacy-preserving microdata publishing [Internet]. [cited 2025 May 29]. Available from: https://arx.deidentifier.org/

    2025

  31. [40]

    Comparative Analysis of Python and Java for Beginners

    Khoirom MS, Sonia M, Laikhuram B, Laishram J, Singh TD. Comparative Analysis of Python and Java for Beginners. 2020;07

    2020

  32. [41]

    Python for Data Analytics, Scientific and Technical Applications

    Nagpal A, Gabrani G. Python for Data Analytics, Scientific and Technical Applications. 2019 Amity Int Conf Artif Intell AICAI [Internet]. 2019 [cited 2025 May 29]. p. 140–5. Available from: https://ieeexplore.ieee.org/document/8701341

    2019

  33. [42]

    Utility-preserving anonymization for health data publishing

    Lee H, Kim S, Kim JW, Chung YD. Utility-preserving anonymization for health data publishing. BMC Med Inform Decis Mak. 2017;17:104

    2017

  34. [43]

    Empowering open data sharing for social good: a privacy-aware approach

    Carvalho T, Antunes L, Costa Santos C, Moniz N. Empowering open data sharing for social good: a privacy-aware approach. Sci Data. 2025;12:248. Appendices Appendix 1: Data extraction forms Given the already present definitions entailing quasi-identifiers and sensitive attribute...

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.