Pith. sign in

REVIEW 3 major objections 6 minor 53 references

A text-to-tabular approach to generate synthetic patient data using LLMs

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Text prompts alone can generate realistic synthetic patient cohorts.

desk verdict Frozen-LLM text-to-tabular SDG is a genuinely new zero-shot framing with thorough evaluation, but the in-context example leaks target-cohort aggregates, so the 'no original data' claim is only partially tested. read the letter →

arxiv 2412.05153 v2 pith:MI5D6E77 submitted 2024-12-06 cs.LG

Margaux Tornqvist , Jean-Daniel Zucker , Tristan Fauvel , Nicolas Lambert , Mathilde Berthelot , Antoine Movschin This is my paper · ORCID
classification cs.LG
keywords syntheticdatagenerationlargelanguagemodelstabularpatientzero-shotin-contextlearningclinicalcorrelationsprivacy
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 proposes generating synthetic tabular patient data with a large language model that never sees the original dataset. Instead, the model receives a text description of the desired database, detailed column definitions, and one fictitious patient example built from published average values, and it produces whole tables of patients. The authors test this on Parkinson's and Alzheimer's cohorts and report that the generated patients preserve the shape of the real distributions and, notably, the correlations between clinical variables, though less accurately than models trained on the real data. The payoff would be a low-skill, zero-shot way to fabricate realistic patient cohorts for pilot studies, pipeline development, and teaching, without the privacy hurdles of sharing real records.

What carries the argument

The carrying mechanism is the structured prompt, assembled from three parts: prior knowledge describing the medical database and disease, instructions giving column names, types, and category encodings plus the requested row count and JSON format, and one fictitious example patient constructed from published average values of the same cohort. The prompt is sent to GPT-3.5 or GPT-4 at temperature 1; because LLMs truncate long repetitive outputs, the desired cohort is generated as many small tables of ten rows, with column order randomly permuted at each generation. The single example patient is the load-bearing context that anchors the model to the target population's typical values.

What would settle it

Run the method on a rare or private disease cohort with no published average values in the prompt, then compare the correlation matrix and marginal distributions of the synthetic cohort against the real cohort; the claim fails if strongly coupled clinical variables show near-zero correlation in the synthetic data while the real data show strong correlation.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a frozen, off-the-shelf LLM prompted only with a database description and a single prototypical patient row generates synthetic cohorts whose univariate distributions and pairwise clinical correlations are broadly faithful to real restricted-access cohorts. Compared with CTGAN, TVAE, and a Gaussian copula trained on the original training splits, the LLM's fidelity scores are lower overall, but in several correlation metrics it is competitive with or better than CTGAN, and a classifier trained on the synthetic data reaches test F1 scores comparable to one trained on real data. Because the original patient-level data is never input to the model, the method carries a structural privacy guarantee that data-dependent generators do not.

Load-bearing premise

The method assumes that one fictitious patient assembled from published average values of the target cohort, plus the model's pretrained biomedical knowledge, is enough to recreate the joint distribution of the real database; for a population with no published averages or a different distribution, the anchor no longer holds.

Editorial extensions

If this is right

  • Synthetic cohorts can be produced before any real data is in hand, so ML pipelines can be built and tested in parallel with data-access approvals.
  • Because no patient-level record is ever passed to the model, the generated data is structurally private rather than relying on differential-privacy machinery.
  • The method transfers to a second disease with a different variable set, suggesting the prompt template generalizes across well-documented conditions.
  • In the train-synthetic-test-real evaluation, classifiers trained on the synthetic Parkinson's data scored as well as classifiers trained on real data, indicating the synthetic cohort can stand in for real data in at least some prediction tasks.
  • The ablation shows that upgrading the model and adding one example patient are the prompt changes that matter most for distribution-shape fidelity.

Reading between the lines

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

  • We infer that the method's fidelity will degrade sharply for rare diseases or populations with no published summary statistics, because the single anchor example then cannot represent the target distribution.
  • We infer that the same prompt structure could be extended to longitudinal or multimodal outputs, since the LLM is already conditioned on text and could emit sequences of records or nested JSON.
  • We infer that a testable next step is to use a private benchmark dataset whose aggregate statistics are withheld, prompting only with disease knowledge, to measure how much of the fidelity is carried by the anchor example rather than by the model's pretrained knowledge.
  • We infer that the Wasserstein-distance gaps on imaging variables will persist unless the prompt provides external calibration values for those measures, since the model tends to produce prototypical values with low variance.
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

3 major / 6 minor

Summary. The paper proposes a text-to-tabular method for generating synthetic patient data with frozen large language models (GPT-3.5 and GPT-4). The method builds a prompt containing prior knowledge about the disease and database, detailed data specifications and formatting instructions, and optionally a single fictitious example patient, and then samples synthetic tables from the LLM. The authors evaluate the approach on PPMI Parkinson's disease data and ADNI Alzheimer's disease data, comparing it against CTGAN, TVAE, and a Gaussian copula model across fidelity, privacy, and utility metrics, with five data splits and five synthetic datasets each, plus an ablation study on prompt components. The central claim is that the method does not require access to the original patient-level data but only a description of the desired database, enabling zero-shot generation in low-resource settings.

Significance. If the central claim holds, the method offers a low-cost, privacy-preserving route to synthetic clinical cohorts for well-characterized diseases, which is genuinely useful for early pipeline development and educational purposes. The paper has several concrete strengths: the evaluation framework is thorough, with multiple splits, multiple synthetic datasets, and a broad metric suite covering fidelity, privacy, and utility; the prompts and code are provided openly; the ablation study isolates key prompt components; and the comparison against three established baselines is informative. The main weakness is that the evaluation setting is narrower than the abstract's headline claim: the single in-context example is derived from published averages of the same PPMI and ADNI studies that serve as ground truth, so the method as tested already receives cohort-specific aggregate information. The paper is transparent about its limitations in the Discussion, but the abstract and introduction currently overstate the scope of what is demonstrated.

major comments (3)
  1. [Section III.C.2 and Tables I, II, V] The central claim that the method requires only a description of the desired database is not fully supported by the evaluated configuration. Section III.C.2 states that the single context example was created using average values of the selected variables extracted from publications on PPMI and ADNI, which are the same studies whose restricted-access data are used as ground truth in Tables I and II. For PPMI, the example comes from the PPMI2020 datacut while the evaluation uses PPMI2024, but both are part of the same study; for ADNI, the example comes from ADNI-based publications. The ablation in Table V quantifies the impact: for GPT-3.5, adding the example row raises Column Shape from 0.64 to 0.70 and ContingencySimilarity from 0.79 to 0.90. This shows that the example injects substantial cohort-specific information about marginal and joint distributions, so the method as evaluated is not purely text-to-tabular in the zero-shot sense claimed. I would ask the authors to either (a) test a configuration in which the example is not derived from the target cohort, such as an example constructed from a different disease or from expert knowledge independent of PPMI/ADNI, or (b) revise the abstract and introduction to state that the method uses a textual description plus target-cohort aggregate statistics when such statistics are available.
  2. [Section V (Discussion)] The Discussion explicitly acknowledges that the approach was only tested on well-documented diseases and reference databases, that rare diseases are untested, and that applying the method without access to the original patient data may require a validation framework using aggregated data from the literature or medical experts. This is an appropriate limitation, but it directly qualifies the contribution stated in the abstract, which says the method requires 'only a description of the desired database' with no such caveat. The paper should bring this scope condition into the abstract and Section I, either as a stated assumption or as an explicit limitation of the current evaluation. This is not an internal inconsistency, but the current framing makes the headline claim stronger than the evidence supports.
  3. [Section IV.C and Table V] The ablation narrative is not fully supported by the numbers in Table V. The 'Best parameters' row, which combines GPT-4 with the example row, reports ContingencySimilarity 0.75, lower than the reference GPT-3.5 no-example value of 0.79 and substantially lower than the GPT-3.5 example-row value of 0.90. The text says that combining the elements that improved fidelity produced the best distribution shapes despite a slight deterioration in correlations, but for ContingencySimilarity the deterioration relative to the reference is not slight; it is the lowest value among all rows except the sampling-1-by-1 condition. The paper should clarify whether this is considered acceptable, and should report the full metric-by-metric comparison against the reference configuration in the text, not only in the table.
minor comments (6)
  1. [Tables I and II] The metric names contain typos: 'Wassersein' should be 'Wasserstein' and 'Jenssen-Shannon' should be 'Jensen-Shannon'.
  2. [Section III.D.1] The text writes 'Gaussain copula'; this should be 'Gaussian copula'. In addition, the baseline abbreviation 'V AE' should be written 'TVAE' consistently.
  3. [Table II] The WD row uses a comma as a thousands separator ('4,618±932') while all other numbers in the paper use periods; the formatting should be unified.
  4. [Table V] The header 'Experience' should read 'Experiment'. The 'Sampling 1 by 1' row should state explicitly whether n=1 is used for every generated table or only for the final iteration after generating m tables of size n.
  5. [Tables I and II] The first column label 'SDG Model on #PPMI2024 Dtrain (Ref.)' is misleading because that column is not a synthetic-data model but the reference metric computed on the real training data; the caption should state this explicitly.
  6. [Section IV.A] The sentence about 'left and right putamen variables for PPMI and the ventricles and intracranial volumes for ADNI' refers to imaging variables, but this is only implicit; the text should state that these are the imaging variables shown in the supplementary distribution plots.

Circularity Check

1 steps flagged · score 5.0 of 10

The single in-context example is built from average values of the same PPMI/ADNI cohorts used as ground truth, so part of the reported zero-shot fidelity is supplied by the prompt rather than independently predicted.

  1. fitted input called prediction [Section III.C.2 (Prompt design); ablation Table V (Appendix A)]
    "In our case, we added a single example created using the average values of the selected variables extracted from publications on #PPMI2020 [35] and on the ADNI dataset [36], [37]."

    The single in-context example is built from the average values of the same PPMI and ADNI cohorts that later serve as ground truth for the fidelity evaluation (Tables I and II). The prompt therefore already contains the target column means, i.e., the first moments whose reproduction is scored by KSComplement, TVComplement, and the Column Shape score. The Table V ablation quantifies this loading: adding the example raises Column Shape from 0.64 to 0.70 and ContingencySimilarity from 0.79 to 0.90 on PPMI. Thus the evaluated configuration is not a purely external zero-shot test of a textual database description; a summary statistic of the exact target population is provided as an input.

full rationale

The paper is transparent and most of its derivation is self-contained: a frozen LLM generates rows from a text prompt, and the comparison against CTGAN, TVAE, and Gaussian Copula is a genuine external benchmark. The main circularity is confined to the evaluation of the zero-shot claim. The in-context example patient is derived from published average values of PPMI and ADNI, the same studies used as ground truth, and the ablation in Table V shows that this example contributes materially to the fidelity scores (Column Shape 0.64 to 0.70; ContingencySimilarity 0.79 to 0.90). Consequently the headline statement that the method 'does not require access to the original data, but only a description of the desired database' is only partially supported: in the evaluated configuration, the description includes aggregate statistics of the target cohorts. This is a moderate, not total, circularity: the LLM still supplies distributional and correlational structure beyond the single mean-value row, the authors openly state how the example is built, and the Discussion acknowledges the limitation when aggregated literature data or medical experts are not available.

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

The method's input includes a hand-constructed example patient built from published average values of the target PPMI and ADNI cohorts, which functions as a fitted anchor for the LLM. The paper also assumes that LLM pretraining on biomedical text supplies enough clinical knowledge to fill in a plausible joint distribution, and that the formatting instructions in the prompt are followed reliably. None of these assumptions is independently verified; the first is disclosed but conflicts with the abstract's 'only a description' framing.

free parameters (3)
  • Example patient row (average values) = published mean values from PPMI [35] and ADNI [36,37]
    Single in-context example derived from aggregate statistics of the target cohorts; ablation shows it improves Column Shape from 0.64 to 0.70.
  • LLM temperature = 1
    Set to 1; authors attribute lower variance and fewer outliers in synthetic data to this choice; not tuned, but affects fidelity.
  • Rows per generation n = 10
    Chosen because LLMs truncate long outputs; affects generation process but not central claim.
assumptions (4)
  • domain assumption LLM pretraining on biomedical corpora captures sufficient joint clinical knowledge to generate realistic values for well-documented diseases
    Paper's hypothesis in Introduction; results for PD/AD are decent but show lower variance.
  • domain assumption Published aggregate statistics of the target cohorts are usable as in-context examples without violating the 'no original data' claim
    The method uses summaries from the same PPMI and ADNI cohorts it evaluates against; the abstract frames this as 'only a description'.
  • domain assumption The encoding mappings provided in the prompt (e.g., categorical values mapped to integers) are correctly followed by the LLM
    Authors chose to put mappings in the prompt rather than preprocessing; assumes LLM obeys format specifications.
  • domain assumption The LLM does not memorize the restricted-access PPMI and ADNI patient-level data
    Authors selected restricted-access datasets to mitigate memorization [34]; not verified for GPT-4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A text-to-tabular approach to generate synthetic patient data using LLMs." pith.science (2026). https://pith.science/paper/MI5D6E77

@misc{pith2026241205153,
  author       = {Pith},
  title        = {Pith review of: A text-to-tabular approach to generate synthetic patient data using LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MI5D6E77}},
  note         = {Machine review of arXiv:2412.05153}
}
read the original abstract

Access to large-scale high-quality healthcare databases is key to accelerate medical research and make insightful discoveries about diseases. However, access to such data is often limited by patient privacy concerns, data sharing restrictions and high costs. To overcome these limitations, synthetic patient data has emerged as an alternative. However, synthetic data generation (SDG) methods typically rely on machine learning (ML) models trained on original data, leading back to the data scarcity problem. We propose an approach to generate synthetic tabular patient data that does not require access to the original data, but only a description of the desired database. We leverage prior medical knowledge and in-context learning capabilities of large language models (LLMs) to generate realistic patient data, even in a low-resource setting. We quantitatively evaluate our approach against state-of-the-art SDG models, using fidelity, privacy, and utility metrics. Our results show that while LLMs may not match the performance of state-of-the-art models trained on the original data, they effectively generate realistic patient data with well-preserved clinical correlations. An ablation study highlights key elements of our prompt contributing to high-quality synthetic patient data generation. This approach, which is easy to use and does not require original data or advanced ML skills, is particularly valuable for quickly generating custom-designed patient data, supporting project implementation and providing educational resources.

Figures

Figures reproduced from arXiv: 2412.05153 by the authors.

Figure 1
Figure 1. Our text-to-tabular approach. Prior knowledge about the medical [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Evaluation framework of our approach against baseline SDG Models. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 29 canonical work pages

  1. [1]

    Synthetic data in health care: A narrative review,

    A. Gonzales, G. Guruswamy, and S. R. Smith, “Synthetic data in health care: A narrative review,” PLOS Digital Health , vol. 2, no. 1, p. e0000082, Jan. 2023, publisher: Public Library of Science. [Online]. Available: https://journals.plos.org/digitalhealth/article?id=10. 1371/journal.pdig.0000082

    work page 2023

  2. [2]

    Generation and evaluation of privacy preserving synthetic health data,

    A. Yale, S. Dash, R. Dutta, I. Guyon, A. Pavao, and K. P. Bennett, “Generation and evaluation of privacy preserving synthetic health data,” Neurocomputing, vol. 416, pp. 244–255, Nov. 2020. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0925231220305117

    work page 2020

  3. [3]

    Syn- thetic data generation for tabular health records: A systematic review,

    M. Hernandez, G. Epelde, A. Alberdi, R. Cilla, and D. Rankin, “Syn- thetic data generation for tabular health records: A systematic review,” Neurocomputing, vol. 493, pp. 28–45, Jul. 2022. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0925231222004349

    work page 2022

  4. [4]

    Mimicking Clinical Trials with Synthetic Acute Myeloid Leukemia Patients Using Generative Artificial Intelligence,

    J.-N. Eckardt, W. Hahn, C. R ¨ollig, S. Stasik, U. Platzbecker, C. M ¨uller- Tidow, H. Serve, C. D. Baldus, C. Schliemann, K. Sch ¨afer-Eckart, M. Hanoun, M. Kaufmann, A. Burchert, C. Thiede, J. Schetelig, M. Bornh ¨auser, M. Wolfien, and J. M. Middeke, “Mimicking Clinical Trials with Synthetic Acute Myeloid Leukemia Patients Using Generative Artificial I...

    work page 2023

  5. [5]

    The Effectiveness of Data Augmentation in Im- age Classification using Deep Learning,

    L. Perez and J. Wang, “The Effectiveness of Data Augmentation in Im- age Classification using Deep Learning,” Dec. 2017, arXiv:1712.04621 [cs]. [Online]. Available: http://arxiv.org/abs/1712.04621

    arXiv 2017

  6. [6]

    Curated LLM: Synergy of LLMs and Data Curation for tabular augmentation in low-data regimes,

    N. Seedat, N. Huynh, B. van Breugel, and M. van der Schaar, “Curated LLM: Synergy of LLMs and Data Curation for tabular augmentation in low-data regimes,” Jun. 2024, arXiv:2312.12112 [cs]. [Online]. Available: http://arxiv.org/abs/2312.12112

    arXiv 2024

  7. [7]

    Synthetic Oversampling: Theory and A Practical Approach Using LLMs to Address Data Imbalance,

    R. Nakada, Y . Xu, L. Li, and L. Zhang, “Synthetic Oversampling: Theory and A Practical Approach Using LLMs to Address Data Imbalance,” Jun. 2024, arXiv:2406.03628 [cs, stat]. [Online]. Available: http://arxiv.org/abs/2406.03628

    arXiv 2024

  8. [8]

    T1dCteGui: A User- Friendly Clinical Trial Enrichment Tool to Optimize T1D Prevention Studies by Leveraging AI/ML Based Synthetic Patient Population,

    M. Pauley, N. Henscheid, S. E. David, S. R. Karpen, K. Romero, J. T. Podichetty, and t. T. . D. Consortium (T1DC), “T1dCteGui: A User- Friendly Clinical Trial Enrichment Tool to Optimize T1D Prevention Studies by Leveraging AI/ML Based Synthetic Patient Population,” Clinical Pharmacology & Therapeutics , vol. 114, no. 3, pp. 704–711, 2023, eprint: https:/...

    work page doi:10.1002/cpt.2976 2023

Show all 53 references
  1. [9]

    Knowledge-based mechanistic modeling accurately predicts disease progression with gefitinib in EGFR-mutant lung adenocarcinoma,

    A. L’Hostis, J.-L. Palgen, A. Perrillat-Mercerot, E. Peyronnet, E. Jacob, J. Bosley, M. Duruisseaux, R. Toueg, L. Lef `evre, R. Kahoul, N. Ceres, and C. Monteiro, “Knowledge-based mechanistic modeling accurately predicts disease progression with gefitinib in EGFR-mutant lung a...

    2023

  2. [10]

    Digital Twin Generators for Disease Modeling,

    N. Alam, J. Basilico, D. Bertolini, S. C. Chetty, H. D’Angelo, R. Douglas, C. K. Fisher, F. Fuller, M. Gomes, R. Gupta, A. Lang, A. Loukianov, R. Mak-McCully, C. Murray, H. Pham, S. Qiao, E. Ryapolova-Webb, A. Smith, D. Theoharatos, A. Tolwani, E. W. Tramel, A. Vidovszky, J. V...

    2024 arXiv

  3. [11]

    Language Models are Realistic Tabular Data Generators,

    V . Borisov, K. Seßler, T. Leemann, M. Pawelczyk, and G. Kasneci, “Language Models are Realistic Tabular Data Generators,” Apr. 2023, arXiv:2210.06280 [cs]. [Online]. Available: http://arxiv.org/abs/2210. 06280

    2023 arXiv

  4. [12]

    Deep Neural Networks and Tabular Data: A Survey,

    V . Borisov, T. Leemann, K. Seßler, J. Haug, M. Pawelczyk, and G. Kasneci, “Deep Neural Networks and Tabular Data: A Survey,” IEEE Transactions on Neural Networks and Learning Systems , vol. 35, no. 6, pp. 7499–7519, Jun. 2024, arXiv:2110.01889 [cs]. [Online]. Available: http:...

    2024 arXiv

  5. [13]

    Using Bayesian Networks to Create Synthetic Data,

    J. Young, P. Graham, and R. Penny, “Using Bayesian Networks to Create Synthetic Data,” Journal of Official Statistics, vol. 25, pp. 549–567, Dec. 2009

    2009

  6. [14]

    PrivBayes: Private Data Release via Bayesian Networks,

    J. Zhang, G. Cormode, C. M. Procopiuc, D. Srivastava, and X. Xiao, “PrivBayes: Private Data Release via Bayesian Networks,” Proceedings of the 2014 ACM SIGMOD International Conference on Management of Data , p. 1423–1434, 2014. [Online]. Available: https://doi.org/10.1145/2588...

    2014

  7. [15]

    The Synthetic Data Vault,

    N. Patki, R. Wedge, and K. Veeramachaneni, “The Synthetic Data Vault,” in 2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA). Montreal, QC, Canada: IEEE, Oct. 2016, pp. 399–

    2016

  8. [16]

    Copula Flows for Synthetic Data Generation,

    S. Kamthe, S. Assefa, and M. Deisenroth, “Copula Flows for Synthetic Data Generation,” Jan. 2021, arXiv:2101.00598 [cs, stat]. [Online]. Available: http://arxiv.org/abs/2101.00598

    2021 arXiv

  9. [17]

    Generative Adversarial Networks,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative Adversarial Networks,” Jun. 2014, arXiv:1406.2661 [cs, stat]. [Online]. Available: http://arxiv.org/abs/1406.2661

    2014 arXiv

  10. [18]

    Modeling Tabular data using Conditional GAN,

    L. Xu, M. Skoularidou, A. Cuesta-Infante, and K. Veeramachaneni, “Modeling Tabular data using Conditional GAN,” Oct. 2019, arXiv:1907.00503 [cs, stat]. [Online]. Available: http://arxiv.org/abs/ 1907.00503

    2019 arXiv

  11. [19]

    CTAB- GAN: Effective Table Data Synthesizing,

    Z. Zhao, A. Kunar, H. Van der Scheer, R. Birke, and L. Y . Chen, “CTAB- GAN: Effective Table Data Synthesizing,” May 2021, arXiv:2102.08369 [cs]. [Online]. Available: http://arxiv.org/abs/2102.08369

    2021 arXiv

  12. [20]

    Differentially Private Generative Adversarial Network,

    L. Xie, K. Lin, S. Wang, F. Wang, and J. Zhou, “Differentially Private Generative Adversarial Network,” Feb. 2018, arXiv:1802.06739 [cs, stat]. [Online]. Available: http://arxiv.org/abs/1802.06739

    2018 arXiv

  13. [21]

    PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees,

    J. Jordon, J. Yoon, and M. v. d. Schaar, “PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees,” in International Conference on Learning , Sep. 2018. [Online]. Available: https: //openreview.net/forum?id=S1zk9iRqF7

    2018

  14. [22]

    CTAB-GAN+: Enhancing Tabular Data Synthesis,

    Z. Zhao, A. Kunar, R. Birke, and L. Y . Chen, “CTAB-GAN+: Enhancing Tabular Data Synthesis,” Frontiers in Big Data , vol. 6, 2022, publisher: arXiv Version Number: 1. [Online]. Available: https://arxiv.org/abs/2204.00401

    2022 arXiv

  15. [23]

    Structured Denoising Diffusion Models in Discrete State- Spaces,

    J. Austin, D. D. Johnson, J. Ho, D. Tarlow, and R. van den Berg, “Structured Denoising Diffusion Models in Discrete State- Spaces,” in Advances in Neural Information Processing Systems , vol. 34. Curran Associates, Inc., 2021, pp. 17 981–17 993. [Online]. Available: https://pr...

    2021

  16. [24]

    TabDDPM: Modelling Tabular Data with Diffusion Models,

    A. Kotelnikov, D. Baranchuk, I. Rubachev, and A. Babenko, “TabDDPM: Modelling Tabular Data with Diffusion Models,” Sep. 2022, arXiv:2209.15421 [cs]. [Online]. Available: http://arxiv.org/abs/ 2209.15421

    2022 arXiv

  17. [25]

    Generating and Imputing Tabular Data via Diffusion and Flow-based Gradient-Boosted Trees,

    A. Jolicoeur-Martineau, K. Fatras, and T. Kachman, “Generating and Imputing Tabular Data via Diffusion and Flow-based Gradient-Boosted Trees,” in Proceedings of The 27th International Conference on Artificial Intelligence and Statistics . PMLR, Apr. 2024, pp. 1288– 1296, iSSN:...

    2024

  18. [26]

    REaLTabFormer: Generating Realistic Relational and Tabular Data using Transformers,

    A. V . Solatorio and O. Dupriez, “REaLTabFormer: Generating Realistic Relational and Tabular Data using Transformers,” 2023, publisher: arXiv Version Number: 1. [Online]. Available: https://arxiv.org/abs/2302.02041

    2023 arXiv

  19. [27]

    TabuLa: Harnessing Language Models for Tabular Data Synthesis,

    Z. Zhao, R. Birke, and L. Chen, “TabuLa: Harnessing Language Models for Tabular Data Synthesis,” 2023, publisher: arXiv Version Number: 1. [Online]. Available: https://arxiv.org/abs/2310.12746

    2023 arXiv

  20. [28]

    Differentially Private Tabular Data Synthesis using Large Language Models,

    T. V . Tran and L. Xiong, “Differentially Private Tabular Data Synthesis using Large Language Models,” Jun. 2024, arXiv:2406.01457 [cs]. [Online]. Available: http://arxiv.org/abs/2406.01457

    2024 arXiv

  21. [29]

    MALLM-GAN: Multi-Agent Large Language Model as Generative Adversarial Network for Synthesizing Tabular Data,

    Y . Ling, X. Jiang, and Y . Kim, “MALLM-GAN: Multi-Agent Large Language Model as Generative Adversarial Network for Synthesizing Tabular Data,” Jun. 2024, arXiv:2406.10521 [cs]. [Online]. Available: http://arxiv.org/abs/2406.10521

    2024 arXiv

  22. [30]

    Are LLMs Naturally Good at Synthetic Tabular Data Generation?

    S. Xu, C.-T. Lee, M. Sharma, R. B. Yousuf, N. Muralidhar, and N. Ramakrishnan, “Are LLMs Naturally Good at Synthetic Tabular Data Generation?” Jun. 2024, arXiv:2406.14541 [cs]. [Online]. Available: http://arxiv.org/abs/2406.14541

    2024 arXiv

  23. [31]

    Creating virtual patients using large language models: scalable, global, and low cost,

    D. A. Cook, “Creating virtual patients using large language models: scalable, global, and low cost,” Medical Teacher , Jul. 2024, publisher: Taylor & Francis. [Online]. Available: https: //www.tandfonline.com/doi/abs/10.1080/0142159X.2024.2376879

    2024

  24. [32]

    Parkinson’s disease symptoms: The patient’s perspective,

    M. Politis, K. Wu, S. Molloy, P. G. Bain, K. R. Chaudhuri, and P. Piccini, “Parkinson’s disease symptoms: The patient’s perspective,” Movement Disorders , vol. 25, no. 11, pp. 1646–1651, 2010, eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mds.23135. [Online]. Availab...

    2010 doi

  25. [33]

    Alzheimer Disease,

    R. J. Castellani, R. K. Rolston, and M. A. Smith, “Alzheimer Disease,” Disease-a-month : DM , vol. 56, no. 9, p. 484, Sep. 2010. [Online]. Available: https://pmc.ncbi.nlm.nih.gov/articles/PMC2941917/

    2010

  26. [34]

    Elephants Never Forget: Testing Language Models for Memorization of Tabular Data,

    S. Bordt, H. Nori, and R. Caruana, “Elephants Never Forget: Testing Language Models for Memorization of Tabular Data,” Mar. 2024, arXiv:2403.06644 [cs]. [Online]. Available: http://arxiv.org/abs/2403. 06644

    2024 arXiv

  27. [35]

    The Parkinson’s progression markers initiative (PPMI) – establishing a PD biomarker cohort,

    K. Marek, S. Chowdhury, A. Siderowf, S. Lasch, C. S. Coffey, C. Caspell-Garcia, T. Simuni, D. Jennings, C. M. Tanner, J. Q. Trojanowski, L. M. Shaw, J. Seibyl, N. Schuff, A. Singleton, K. Kieburtz, A. W. Toga, B. Mollenhauer, D. Galasko, L. M. Chahine, D. Weintraub, T. Foroud,...

    2018

  28. [36]

    A Comparison of Accelerated and Non-accelerated MRI Scans for Brain V olume and Boundary Shift Integral Measures of V olume Change: Evidence from the ADNI Dataset,

    E. N. Manning, K. K. Leung, J. M. Nicholas, I. B. Malone, M. J. Cardoso, J. M. Schott, N. C. Fox, and J. Barnes, “A Comparison of Accelerated and Non-accelerated MRI Scans for Brain V olume and Boundary Shift Integral Measures of V olume Change: Evidence from the ADNI Dataset,...

    2017

  29. [37]

    Identification of mild cognitive impairment subtypes predicting conversion to Alzheimer’s disease using multimodal data,

    M. Kikuchi, K. Kobayashi, S. Itoh, K. Kasuga, A. Miyashita, T. Ikeuchi, E. Yumoto, Y . Kosaka, Y . Fushimi, T. Takeda, S. Manabe, S. Hattori, A. D. N. Initiative, A. Nakaya, K. Kamijo, and Y . Matsumura, “Identification of mild cognitive impairment subtypes predicting conversi...

    2022

  30. [38]

    [Online]

    Synthetic Data Metrics , DataCebo, Inc., 5 2024, version 0.14.1. [Online]. Available: https://docs.sdv.dev/sdmetrics/

    2024

  31. [39]

    On Wasserstein Two Sample Testing and Related Families of Nonparametric Tests,

    A. Ramdas, N. Garcia, and M. Cuturi, “On Wasserstein Two Sample Testing and Related Families of Nonparametric Tests,” Oct. 2015, arXiv:1509.02237 [math, stat]. [Online]. Available: http: //arxiv.org/abs/1509.02237

    2015 arXiv

  32. [40]

    Divergence measures based on the Shannon entropy,

    J. Lin, “Divergence measures based on the Shannon entropy,” IEEE Transactions on Information Theory , vol. 37, no. 1, pp. 145–151, Jan. 1991, conference Name: IEEE Transactions on Information Theory. [Online]. Available: https://ieeexplore.ieee.org/document/61115

    1991

  33. [41]

    Distinctive Image Features from Scale-Invariant Keypoints,

    D. G. Lowe, “Distinctive Image Features from Scale-Invariant Keypoints,” International Journal of Computer Vision , vol. 60, no. 2, pp. 91–110, Nov. 2004. [Online]. Available: https://doi.org/10.1023/B: VISI.0000029664.99615.94

    2004

  34. [42]

    Gender Differences in the Prevalence of Parkinson’s Disease,

    A. Zirra, S. C. Rao, J. Bestwick, R. Rajalingam, C. Marras, C. Blauwendraat, I. F. Mata, and A. J. Noyce, “Gender Differences in the Prevalence of Parkinson’s Disease,” Movement Disorders Clinical Practice, vol. 10, no. 1, pp. 86–93, Nov. 2022. [Online]. Available: https://www...

    2022

  35. [43]

    Quantitative Measurement of Rigidity in Parkinson’s Disease: A Systematic Review,

    M. d. R. Ferreira-S ´anchez, M. Moreno-Verd ´u, and R. Cano-de-la Cuerda, “Quantitative Measurement of Rigidity in Parkinson’s Disease: A Systematic Review,” Sensors (Basel, Switzerland) , vol. 20, no. 3, p. 880, Feb. 2020. [Online]. Available: https://www.ncbi.nlm.nih.gov/pmc...

    2020

  36. [44]

    Generation and evaluation of synthetic patient data,

    A. Goncalves, P. Ray, B. Soper, J. Stevens, L. Coyle, and A. P. Sales, “Generation and evaluation of synthetic patient data,” BMC Medical Research Methodology , vol. 20, no. 1, p. 108, Dec

  37. [45]

    How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models,

    A. M. Alaa, B. van Breugel, E. Saveliev, and M. van der Schaar, “How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models,” Jul. 2022, arXiv:2102.08921 [cs, stat]. [Online]. Available: http://arxiv.org/abs/2102.08921

    2022 arXiv

  38. [46]

    Synthetic data, real errors: how (not) to publish and use synthetic data,

    B. van Breugel, Z. Qian, and M. van der Schaar, “Synthetic data, real errors: how (not) to publish and use synthetic data,” Jul. 2023, arXiv:2305.09235 [cs]. [Online]. Available: http://arxiv.org/abs/2305. 09235 APPENDIX A A. Data

    2023 arXiv

  39. [49]

    For up-to-date information on the study, visit www.ppmi-info.org

    PPMI: Data used in the preparation of this article were obtained on February, 6th, 2024 from the Parkinson’s Progression Markers Initiative (PPMI) database (www.ppmi- info.org/access-dataspecimens/download-data), RRID:SCR 006431. For up-to-date information on the study, visit ...

    2024

  40. [50]

    As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analy- sis or writing of this report

    ADNI: Data used in preparation of this article were ob- tained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did...

    2024

  41. [51]

    Fidelity: a) Distribution plots: • Univariate distributions plots: We compared distribu- tions one by one of synthetic data against real data to examine the distribution shape and identify potential issues in the patterns learned by the SDG Model. • Joint bivariate distributio...

  42. [52]

    The Euclidean distance is computed for each point of the synthetic data and then averaged

    Privacy preservation: • Distance to closest record (DCR): Distance to closest record measures the closest distance between a point in the synthetic data and the real data. The Euclidean distance is computed for each point of the synthetic data and then averaged. As suggested b...

  43. [53]

    The evaluation process involves training the ML algorithm with synthetic data

    Utility: • BinaryAdaboostClassifier To evaluate the added value of using synthetic data for a ML binary prediction task, an AdaBoost classifier is fitted to the training data. The evaluation process involves training the ML algorithm with synthetic data. The efficacy of the ML...

  44. [410]

    Available: http://ieeexplore.ieee.org/document/7796926/

    [Online]. Available: http://ieeexplore.ieee.org/document/7796926/

  45. [2020]

    Available: https://bmcmedresmethodol.biomedcentral

    [Online]. Available: https://bmcmedresmethodol.biomedcentral. com/articles/10.1186/s12874-020-00977-1

    doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.