CoChat
Sign Up
Cited by 1,542Privacy-Preserving Technologies in DataMachine Learning in HealthcareArtificial Intelligence in Healthcare and Education

Federated Learning for Healthcare Informatics

Open access
Xu, Jie, Glicksberg, Benjamin S., Su, Chang, Walker, Peter, Bian, Jiang, Wang, Fei·2021·Journal of Healthcare Informatics Research
DOI 10.1007/s41666-020-00082-4PMC PMC7659898
Open this paper in CoChat — free accountOpen PDF

Abstract

With the rapid development of computer software and hardware technologies, more and more healthcare data are becoming readily available from clinical institutions, patients, insurance companies, and pharmaceutical industries, among others. This access provides an unprecedented opportunity for data science technologies to derive data-driven insights and improve the quality of care delivery. Healthcare data, however, are usually fragmented and private making it difficult to generate robust results across populations. For example, different hospitals own the electronic health records (EHR) of different patient populations and these records are difficult to share across hospitals because of their sensitive nature. This creates a big barrier for developing effective analytical approaches that are generalizable, which need diverse, "big data." Federated learning, a mechanism of training a shared global model with a central server while keeping all the sensitive data in local institutions where the data belong, provides great promise to connect the fragmented healthcare data sources with privacy-preservation. The goal of this survey is to provide a review for federated learning technologies, particularly within the biomedical space. In particular, we summarize the general solutions to the statistical challenges, system challenges, and privacy issues in federated learning, and point out the implications and potentials in healthcare.

Integrity signals

Registry

Crossref · 10.1007/s41666-020-00082-4

Cites

117 of 117 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

30 of 45 cited sources checked · 15 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Sharing a small subset of data across clients, with uniform class distribution, can address non-IID issues.

    “Another commonly used method is globally sharing a small portion of data between all the clients [75, 111]. The shared subset is required containing a uniform distribution over classes from the central server to the clients.”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [75] Nishio T, Yonetani R (2018) Client selection for federated learning with heterogeneous resources in mobile edge. arXiv:1804.08333 source

      The Experimental Setup (p. 5) describes IID and non-IID data assignments, but not sharing a small, class-balanced subset among clients. The Results (p. 7) discuss increasing client participation or using model compression to alleviate non-IID effects. Searches for shared data and uniform class distribution found no passage addressing the claimed method. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • cites [111] Zhao Y, Li M, Lai L, Suda N, Civin D, Chandra V (2018) Federated learning with non-iid data. arXiv:1806.00582 source
  • Premisecitation

    When component distributions truly differ, pluralism can beat the ideal IID baseline.

    “When the component distributions are actually different, pluralism can outperform the “ideal” IID baseline.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [28] Eichner H, Koren T, McMahan HB, Srebro N, Talwar K (2019) Semi-cyclic stochastic gradient descent. arXiv:1904.10120 source
  • Premisecitation

    FedAvg can handle some non-IID data if all clients use the same model.

    “It is demonstrated can work with certain non independent identical distribution (non-IID) data by requiring all the clients to share the same model.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [69] McMahan B, Moore E, Ramage D, Hampson S, Arcas BA (2017) Communication-efficient learning of deep networks from decentralized data. In: Artificial intelligence and statistics, pp 1273–1282 source
  • Premisecitation

    Centralized models trained on pooled local samples minimize loss against a uniform distribution.

    “Most centralized models are trained on the aggregated training samples obtained from the samples drawn from the local clients [96, 111]. Intrinsically, the centralized model is trained to minimize the loss with respect to the uniform distribution [73]”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [96] Smith V, Chiang CK, Sanjabi M, Talwalkar AS (2017) Federated multi-task learning. In: Advances in neural information processing systems, pp 4424–4434

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [111] Zhao Y, Li M, Lai L, Suda N, Civin D, Chandra V (2018) Federated learning with non-iid data. arXiv:1806.00582 source

      Section 3.1 (p. 4) describes centralized training as optimizing population loss, but it does not say the population distribution is uniform. Sections 4.1–4.2 (p. 7) specify a uniform distribution over classes only for a separate, globally shared dataset used in the authors’ data-sharing strategy. That does not establish that centralized models generally minimize loss with respect to a uniform distribution. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [73] Mohri M, Sivek G, Suresh AT (2019) Agnostic federated learning. In: Chaudhuri K, Salakhutdinov R (eds) Proceedings of the 36th International conference on machine learning, proceedings of machine learning research, vol 97. PMLR, Long Beach, pp 4615–4625

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Premisecitation

    Convolutional neural networks trained with FedAvg can lose substantial performance because of weight divergence.

    “The performance of convolutional neural networks trained with FedAvg algorithm can reduce significantly due to the weight divergence [111].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [111] Zhao Y, Li M, Lai L, Suda N, Civin D, Chandra V (2018) Federated learning with non-iid data. arXiv:1806.00582 source
  • May combine two claims. 19
  • Classified as probably not a substantial claim. 13

  • May combine two claims.

    Premisereasoning

    Because clients cannot all be guaranteed reliable, federated learning needs extra privacy protections for unreliable participants.

    “Privacy and Security: Additional privacy protections are needed for unreliable participating clients. It is impossible to ensure all clients are equally reliable.”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [13] Caldas S, Wu P, Li T, Konečnỳ J, McMahan HB, Smith V, Talwalkar A (2018) Leaf: a benchmark for federated settings. arXiv:1812.01097 source

      The Introduction (p. 1) says federated methods must address privacy and security concerns because personally generated data is sensitive. It does not say that participating clients are unreliable, that their reliability cannot be ensured, or that unreliability calls for additional privacy protections. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites [96] Smith V, Chiang CK, Sanjabi M, Talwalkar AS (2017) Federated multi-task learning. In: Advances in neural information processing systems, pp 4424–4434

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • The authors state this as an assumption: The number of participating clients in federated learning is large, potentially thousands or millions.

  • May combine two claims.

    Premisecitation

    Federated dropout lowers downlink communication, uplink update size, and local computation.

    “Federated dropout not only reduces the downlink communication but also reduces the size of uplink updates. Moreover, the local computational costs is correspondingly reduced since the local training procedure dealing with parameters with smaller dimensions.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [12] Caldas S, Konečny J, McMahan HB, Talwalkar A (2018) Expanding the reach of federated learning by reducing client resource requirements. arXiv:1812.07210 source

  • May combine two claims.

    Premisecitation

    Federated learning often involves many clients, and the client count can exceed the average number of training samples per client.

    “Communication Efficiency: The number of clients K is large and can be much bigger than the average number of training sample stored in the activated clients, i.e., K ≫ (n/K).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [13] Caldas S, Wu P, Li T, Konečnỳ J, McMahan HB, Smith V, Talwalkar A (2018) Leaf: a benchmark for federated settings. arXiv:1812.01097 source
    • cites [96] Smith V, Chiang CK, Sanjabi M, Talwalkar AS (2017) Federated multi-task learning. In: Advances in neural information processing systems, pp 4424–4434

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Fully decentralized federated learning removes the central server and lets clients communicate only with one-hop neighbors.

    “Therefore, some researches began to study fully decentralized framework where the central server is not required [41, 60, 85, 91]. The local clients are distributed over the graph/network where they only communicate with their one-hop neighbors.”
    Quote in the paperNot a substantial claim4 of 4 sources checked
    • cites [41] He C, Tan C, Tang H, Qiu S, Liu J (2019) Central server free federated learning over single-sided trust social networks. arXiv:1910.04956 source
    • cites [60] Lalitha A, Kilinc OC, Javidi T, Koushanfar F (2019) Peer-to-peer federated learning on graphs. arXiv:1901.11173 source
    • cites [85] Roy AG, Siddiqui S, Pölsterl S, Navab N, Wachinger C (2019) Braintorrent: a peer-to-peer environment for decentralized federated learning. arXiv:1905.06731 source

      The Introduction (p. 2) and Conclusion (p. 8) describe BrainTorrent as server-less, with clients communicating directly rather than through a central server. The one-hop-neighbor restriction is not supported as a general network rule: Section 2.2 (p. 4) says every client is directly connected to every other client, and an initiating client pings all the others. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [91] Shayan M, Fung C, Yoon CJ, Beschastnikh I (2018) Biscotti: a ledger for private and secure peer-to-peer machine learning. arXiv:1811.09904 source

      The Abstract (p. 1) describes Biscotti as a fully decentralized peer-to-peer approach, supporting the claim that it does not require a central server. The Noising protocol (p. 7) says a peer contacts selected noising peers, but our copy does not establish that clients communicate only with one-hop neighbors. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    q-FFL gives more weight to poorly performing devices, which lowers variance in the accuracy distribution.

    “Compared to AFL, Li et al. [64] proposed q-Fair Federated Learning (q-FFL), assigning higher weight to devices with poor performance, so that the distribution of accuracy in the network reduces in variance.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [64] Li T, Sanjabi M, Smith V (2019) Fair resource allocation in federated learning. arXiv:1905.10497 source

  • May combine two claims.

    Premisecitation

    Client data distributions in federated learning are highly different, so local samples do not represent the overall distribution.

    “Statistical Challenge: The data distribution among all clients differ greatly, i.e., , we have . It is such that any data points available locally are far from being a representative sample of the overall distribution, i.e., .”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [13] Caldas S, Wu P, Li T, Konečnỳ J, McMahan HB, Smith V, Talwalkar A (2018) Leaf: a benchmark for federated settings. arXiv:1812.01097 source

      The Introduction (p. 1) says devices have different, though perhaps related, data-generating distributions, supporting the claim of heterogeneous client data. It does not establish that every client's local samples are far from representative of the overall distribution; that absolute formulation goes beyond the passage. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [96] Smith V, Chiang CK, Sanjabi M, Talwalkar AS (2017) Federated multi-task learning. In: Advances in neural information processing systems, pp 4424–4434

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Differential privacy only protects users to a limited degree and can reduce prediction accuracy because it is lossy.

    “However, DP only protects users from data leakage to a certain extent and may reduce performance in prediction accuracy because it is a lossy method [18].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [18] Cheng K, Fan T, Jin Y, Liu Y, Chen T, Yang Q (2019) Secureboost: a lossless federated learning framework. arXiv:1901.08755 source

  • May combine two claims.

    Interpretationcitation

    Federated learning supports predictive modeling from diverse sources and can give clinicians more insight into earlier treatment risks and benefits.

    “Federated learning has also enabled predictive modeling based on diverse sources, which can provide clinicians with additional insights into the risks and benefits of treating patients earlier [9, 10, 90].”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [9] Boughorbel S, Jarray F, Venugopal N, Moosa S, Elhadi H, Makhlouf M (2019) Federated uncertainty-aware learning for distributed hospital ehr data. arXiv:1910.12191 source

      The Introduction (p. 1) says federated learning enables predictive models trained jointly from different EHR sources. The Clinical Application section (p. 2) describes predicting preterm-birth risk three months before delivery, but does not say the model gives clinicians insight into the risks and benefits of treating patients earlier. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [10] Brisimi TS, Chen R, Mela T, Olshevsky A, Paschalidis IC, Shi W (2018) Federated learning of predictive models from federated electronic health records. Int J Med Inform 112:59–67 source

      The Abstract (Objective) and Conclusions describe a federated model that uses distributed EHR data to predict future hospitalizations. The Introduction (Health application) links prediction and early prevention, but the paper does not assess the risks and benefits of treating patients earlier or show that its model gives clinicians insight into that trade-off. Page numbers are unavailable in our copy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [90] Sharma P, Shamout FE, Clifton DA (2019) Preserving patient privacy while training a predictive model of in-hospital mortality. arXiv:1912.00354 source

      The Introduction (p. 1) describes federated training of an in-hospital mortality model using data distributed across hospitals, and the Summary (p. 4) discusses the potential value of larger, more diverse datasets. The paper does not say that its model gives clinicians insight into the risks and benefits of treating patients earlier. Its evaluation used one database distributed among virtual workers, rather than data collected from multiple hospitals (Dataset, p. 3). (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Federated learning can connect EHR data across medical institutions, letting them share experiences rather than data while preserving privacy.

    “Federated learning is a viable method to connect EHR data from medical institutions, allowing them to share their experiences, and not their data, with a guarantee of privacy [9, 25, 34, 45, 65, 82].”
    Quote in the paperSubstantial claim5 of 6 sources checked
    • cites [9] Boughorbel S, Jarray F, Venugopal N, Moosa S, Elhadi H, Makhlouf M (2019) Federated uncertainty-aware learning for distributed hospital ehr data. arXiv:1910.12191 source

      The Introduction (p. 1) says federated learning jointly trains models from different EHR sources without sharing data, and §3.2 (p. 3) says the proposed algorithm preserves data privacy by sending model updates to the data location. Those passages support the paraphrase, but not the citing sentence’s stronger claim of a privacy guarantee; the Introduction also identifies privacy as an ongoing challenge. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [25] Duan R, Boland MR, Liu Z, Liu Y, Chang HH, Xu H, Chu H, Schmid CH, Forrest CB, Holmes JH, et al. (2020) Learning from electronic health records across multiple sites: a communication-efficient and privacy-preserving distributed algorithm. J Am Med Inform Assoc 27(3):376–385 source

      The Conclusion (page not provided) says ODAL enables analysis across multiple clinical datasets without transferring patient-level data, supporting the claim that institutions can learn from EHR data without sharing those records. It defines privacy preservation in that limited sense; it does not establish the citing sentence’s broader “guarantee of privacy.” (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites [34] Gruendner J, Schwachhofer T, Sippl P, Wolf N, Erpenbeck M, Gulden C, Kapsner LA, Zierk J, Mate S, Stürzl M, et al. (2019) KETOS: Clinical decision support and machine learning as a service–A training and deployment platform based on Docker, OMOP-CDM, and FHIR Web Services. PloS one 14(10):1–16 source

      The Results section (Example model 3; page unavailable) describes training models across a federated hospital network while only aggregated statistics leave each institution. The Discussion likewise says patient data need not leave hospitals. These passages support privacy-preserving cross-institutional training, but they do not establish the claimed guarantee of privacy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [45] Huang L, Liu D (2019) Patient clustering improves efficiency of federated machine learning to predict mortality and hospital stay time using distributed electronic medical records. arXiv:1903.09296 source

      The Introduction (page unavailable) says federated learning keeps EMR data local and aggregates locally computed results, allowing institutions to learn together without sharing the records. The Conclusions (page unavailable) describe the authors’ clustering method as avoiding a privacy breach. Neither passage establishes the citing sentence’s stronger claim of a guarantee of privacy. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [65] Li Z, Roberts K, Jiang X, Long Q (2019) Distributed learning from multiple ehr databases: Contextual embedding models for medical events. J Biomed Inform 92:103138 source

      The Distributed Noise Contrastive Estimation and Privacy Protection sections describe building a model from multiple EHR databases without sharing patient data; sites share a differentially private count table instead. But the Conclusion and Discussion says the experiment used two EHR systems at one institution, not different hospitals, and calls for evaluation across hospitals. The paper describes privacy protection, not an unqualified guarantee of privacy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [82] Raja PV, Sivasankar E (2014) Modern framework for distributed healthcare data analytics based on hadoop. In: Information and communication technology-EurAsia conference. Springer, pp 348–355 source

  • May combine two claims.

    Interpretationreasoning

    Federated learning can connect separated healthcare data sources while preserving privacy.

    “Federated learning, a mechanism of training a shared global model with a central server while keeping all the sensitive data in local institutions where the data belong, provides great promise to connect the fragmented healthcare data sources with privacy-preservation.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Agnostic federated learning optimizes a centralized model for any target distribution that is a mixture of client distributions.

    “Mohri et al. [73] proposed a minimax optimization scheme, i.e., agnostic federated learning (AFL), where the centralized model is optimized for any possible target distribution formed by a mixture of the client distributions.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [73] Mohri M, Sivek G, Suresh AT (2019) Agnostic federated learning. In: Chaudhuri K, Salakhutdinov R (eds) Proceedings of the 36th International conference on machine learning, proceedings of machine learning research, vol 97. PMLR, Long Beach, pp 4615–4625

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Communication cost can be lowered by cutting the number of clients, the update size, or the number of updates.

    “Apparently, we can consider three ways to reduce the communication cost: (a) reduce the number of clients K, (b) reduce the update size, (c) reduce the number of updates U.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisereasoning

    Healthcare data are often split across institutions and kept private, which hinders producing reliable results across populations.

    “Healthcare data, however, are usually fragmented and private making it difficult to generate robust results across populations.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Federated learning is promising for analyzing healthcare data.

    “Federated learning holds great promises on healthcare data analytics.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisenone given

    Federated learning trains one shared global model through a central server while sensitive data stay at their local institutions.

    “Federated learning, a mechanism of training a shared global model with a central server while keeping all the sensitive data in local institutions where the data belong”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Without well-designed incentives, self-interested mobile or wearable devices will avoid federated learning tasks, which hinders adoption.

    “Without well-designed incentives, self-interested mobile or other wearable devices will be reluctant to participate in federal learning tasks, which will hinder the adoption of federated learning [53].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [53] Kang J, Xiong Z, Niyato D, Yu H, Liang YC, Kim DI (2019) Incentive design for efficient federated learning in mobile networks: A contract theory approach. arXiv:1905.07479 source

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    Federated learning aims to make isolated institutions or devices share experiences, and the resulting large medical dataset significantly improves model performance.

    “Federated tries to make isolated institutions or devices share their experiences, and the performance of machine learning model will be significantly improved by the formed large medical dataset.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Differential privacy protects individual data privacy and ensures that adding or removing a record does not substantially change an analysis outcome.

    “Differential privacy (DP) [26] is an alternative theoretical model for protecting the privacy of individual data, which has been widely applied to many areas, not only traditional algorithms, e.g., boosting [27], principal component analysis [15], support vector machine [86], but also deep learning research [1, 70]. It ensures that the addition or removal does not substantially affect the outcome”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] Dwork C, Kenthapadi K, McSherry F, Mironov I, Naor M (2006) Our data, ourselves: Privacy via distributed noise generation. In: Annual international conference on the theory and applications of cryptographic techniques. Springer, pp 486–503 source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Most health systems have problems with data clutter and efficiency.

    “However, most health systems suffer from data clutter and efficiency problems.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Federated learning enables collaborative modeling that supports efficient machine learning while ensuring privacy and legal compliance across parties or nodes.

    “This widespread adoption is due in part by the fact that it enables a collaborative modeling mechanism that allows for efficient ML all while ensuring data privacy and legal compliance between multiple parties or multiple computing nodes.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Using dirty data as samples makes analysis results worthless.

    “The analyzed results are apparently worthless when dirty data are accidentally used as samples.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Premisenone given

    Hospital medical devices have incomparable advantages over wearable devices.

    “Medical devices in hospitals have incomparable advantages over wearable devices.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Secure multi-party computation ensures that parties share nothing with each other or a third party.

    “Although SMC guarantees that none of the parties shares anything with each other or with any third party, it can not prevent an adversary from learning some individual information”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [8] Bonawitz K, Ivanov V, Kreuter B, Marcedone A, McMahan HB, Patel S, Ramage D, Segal A, Seth K (2017) Practical secure aggregation for privacy-preserving machine learning. In: Proceedings of the 2017 ACM SIGSAC conference on computer and communications security. ACM, pp 1175–1191 source
    • cites [78] Pathak M, Rane S, Raj B (2010) Multiparty differential privacy via aggregation of locally trained classifiers. In: Advances in neural information processing systems, pp 1876–1884

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationnone given

    Federated learning can connect isolated medical institutions, hospitals, or devices so they share experiences while preserving privacy.

    “Federated learning has the potential to connect all the isolated medical institutions, hospitals, or devices to make them share their experiences with privacy guarantee.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Prediction tasks in federated healthcare are currently limited and relatively simple.

    “However, the prediction task is currently restricted and relatively simple.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Data quality from multiple sources is uneven, and no uniform data standard exists.

    “The quality of data collected from multiple sources is uneven and there is no uniform data standard.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Common data models such as OMOP support wider replication analyses but do not solve the limits of joint data access.

    “The creation and utility of standardized common data models, such as OMOP [44], allow for more wide-spread replication analyses but it does not overcome the limitations of joint data access.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [44] Hripcsak G, Duke JD, Shah NH, Reich CG, Huser V, Schuemie MJ, Suchard MA, Park RW, Wong ICK, Rijnbeek PR, et al. (2015) Observational health data sciences and informatics (ohdsi): opportunities for observational researchers. Stud Health Technol Inform 216:574

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

27 of 27

  • The authors state this as an assumption.

    The number of participating clients in federated learning is large, potentially thousands or millions.

    “In federated learning, we usually assume the number of participated clients (e.g., phones, cars, clinical institutions...) is large, potentially in the thousands or millions.”

  • The authors state this as an assumption.

    There are K activated clients where data reside, each with a data distribution and a number of samples.

    “Mathematically, assume there are K activated clients where the data reside in (a client could be a mobile phone, a wearable device, or a clinical institution data warehouse, etc.). Let denote the data distribution associated with client k and nk the number of samples available from that client.”
2 of 2

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • DFL can be integrated with P2P-based communication technologies such as blockchain to build decentralized FL systems.

    “Due to the contemporary features, DFL can be integrated with P2Pbased communication technologies such as blockchain [37] to build decentralized FL systems.”

    From Federated Learning for Internet of Things: A Comprehensive Survey (2021) · cites this paper as [37]

    The “Peer-to-Peer Learning” section (page unavailable) describes fully decentralized federated learning in which clients communicate with neighboring clients without a central server. It does not identify blockchain as a communication technology for such a system or discuss integrating it with decentralized federated learning. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • FL enables healthcare records located across different institutions to be connected without revealing personal information.

    “FL enables healthcare records that are located across different institutions to be connected without revealing personal information [11].”

    From A comprehensive survey on federated learning techniques for healthcare informatics (2023) · cites this paper as [11]

  • Xu et al. [11] surveyed papers focusing on FL in the biomedical area to provide a review.

    “Xu et al. [11] surveyed the papers that focus FL in the biomedical area to provide a review.”

    From A systematic review on federated learning in medical image analysis (2023) · cites this paper as [11]

  • Federated Learning addresses privacy and data integrity concerns in healthcare AI.

    “By enabling multiple institutions to collaborate on AI model development without exchanging patient data, FL addresses the most pressing concerns in healthcare AI: privacy and data integrity.”

    From Balancing Privacy and Progress: A Review of Privacy Challenges, Systemic Oversight, and Patient Perceptions in AI-Driven Healthcare (2024) · cites this paper as [19]

  • Federated Learning enables multiple institutions to collaborate on AI model development without exchanging patient data.

    “By enabling multiple institutions to collaborate on AI model development without exchanging patient data, FL addresses the most pressing concerns in healthcare AI: privacy and data integrity.”

    From Balancing Privacy and Progress: A Review of Privacy Challenges, Systemic Oversight, and Patient Perceptions in AI-Driven Healthcare (2024) · cites this paper as [19]

Cites

Flagged first, then by citations

13 of the 130 references couldn’t be matched to a known paper.

Research works printed but not linked: 5

  1. Kawa D, Punyani S, Nayak P, Karkera A, Jyotinagar V (2019) Credit risk assessment from combined bank records using federated learning. International Research Journal of Engineering and Technology (IRJET) 6(4):1355–1358
  2. Mohri M, Sivek G, Suresh AT (2019) Agnostic federated learning. In: Chaudhuri K, Salakhutdinov R (eds) Proceedings of the 36th International conference on machine learning, proceedings of machine learning research, vol 97. PMLR, Long Beach, pp 4615–4625
  3. Pathak M, Rane S, Raj B (2010) Multiparty differential privacy via aggregation of locally trained classifiers. In: Advances in neural information processing systems, pp 1876–1884
  4. Rehak D, Dodds P, Lannom L (2005) A model and infrastructure for federated learning content repositories. In: Interoperability of web-based educational systems workshop, vol 143. Citeseer
  5. Smith V, Chiang CK, Sanjabi M, Talwalkar AS (2017) Federated multi-task learning. In: Advances in neural information processing systems, pp 4424–4434

Cited by

Sorted by citations

Related

Authors
Xu, J., Glicksberg, B. S., Su, C., Walker, P., Bian, J., & Wang, F.
Affiliations
  • Weill Cornell Medicine (US)
  • Icahn School of Medicine at Mount Sinai (US)
  • United States Department of Defense (US)
  • University of Florida (US)
Corresponding author
Wang, Fei
ORCID
Venue
Journal of Healthcare Informatics Research
Publisher
Springer Science and Business Media LLC
Published
2020-11-12
Volume
5
Issue
1
Pages
1-19
ISSN
  • 2509-4971 (ISSN-L)
  • 2509-498X
DOI
10.1007/s41666-020-00082-4
PMC
PMC7659898
OpenAlex
W3100779497
Cited by
1,542
References
130
Citations by year
2026: 239 · 2025: 366 · 2024: 314 · 2023: 278 · 2022: 204 · 2021: 111 · 2020: 29 · 2019: 1
FWCI
61.09
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Healthcare Informatics Research (publishedVersion)
  • PubMed Central (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
  • National Science Foundation (1750326)
  • National Institute on Aging (P30AG028740)
  • Office of Naval Research (N00014-18-1-2585, N00014-18, N00014)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
federated learning, healthcare data privacy, electronic health records, data fragmentation, privacy-preserving machine learning, biomedical applications

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-09-28.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-28.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review