---
title: "Blockchain-Enforced Federated Ensemble for Privacy-Preserving In-Hospital Mortality Prediction"
id: "medrxiv-18-a-blockchain-enabled-federated-ensemble-framework-for-privacy-preserving"
canonical_url: "https://medichelpline.com/clinical-feed/medrxiv-18-a-blockchain-enabled-federated-ensemble-framework-for-privacy-preserving"
content_type: "clinical_feed_article"
specialty: "Critical Care"
source_name: "medRxiv (Clinical Preprints)"
source_url: "https://www.medrxiv.org/content/10.64898/2026.11.19.26368466v1?rss=1"
published_at: "2026-11-20T12:00:00.000Z"
evidence_level: "Verified Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Blockchain-Enforced Federated Ensemble for Privacy-Preserving In-Hospital Mortality Prediction
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/medrxiv-18-a-blockchain-enabled-federated-ensemble-framework-for-privacy-preserving
- **Specialty:** [Critical Care](https://medichelpline.com/clinical-feed/critical-care.md)
- **Primary Source:** medRxiv (Clinical Preprints)
- **Source URL:** [Original Journal Publication](https://www.medrxiv.org/content/10.64898/2026.11.19.26368466v1?rss=1)
- **Published At:** 2026-11-20T12:00:00.000Z
- **Evidence Rating:** Verified Feed
## Executive GIST (TL;DR)
- The study simulated a federated learning protocol, named the **Blockchain-Enabled Federated Ensemble (BFE)**, to build a shared in-hospital **mortality** prediction tool while admitting only peer-validated models. - Hospitals committed SHA-256 digests of locally trained **XGBoost** and random-forest models to a Solidity smart contract; peers posted digest-bound validation AUROCs and the contract admitted models with a median peer AUROC ≥ 0.70. - Admitted models were pooled using weighted logit averaging (BFE-PW) or stacking with federated recalibration (BFE-Stack); the study team implemented federated recalibration variants including a median-of-sites recalibration. - Data came from MIMIC-IV v3.1: 425,215 admissions (186,103 patients) with 2.35% in-hospital mortality; predictors used the first 24 hours of admission. - The simulation contrasted BFE with centralised training, local-only models, FedAvg, FedProx, median and Multi-Krum FedAvg, and federated logistic regression across seven adversarial attacks targeting training, attestation, recalibration and model delivery. - Under gross label flipping, the contract excluded poisoned models in every random seed; FedAvg lost 29% of its AUPRC under the same attack, while Multi-Krum FedAvg also resisted exclusion. - A targeted flip of female deaths passed the attestation gate and reduced sensitivity for female deaths from 32.8% to 27.5%; attacks that preserved validation discrimination were not detected by the contract. - Poisoned recalibration could catastrophically invert BFE-PW (AUROC 0.098) unless a median-of-sites recalibration was applied (AUROC 0.902). - With large-hospital weighting, BFE-PW matched FedAvg in AUROC (0.902); BFE-Stack matched FedProx (0.907 and 0.906). Both trailed local-only XGBoost (0.910) and centralised XGBoost (0.915) and showed lower AUPRC; in a 5% data regime BFE-PW outperformed local models (0.866 vs 0.844). - Pooled calibration masked a two-fold under-prediction at one site (observed-to-expected ratio 2.13); applying a local intercept corrected that site’s calibration (0.93). - Blockchain transaction costs were quantified: one aggregation required 37 transactions and ~3.3 million gas and rose roughly quadratically with consortium size. - Authors conclude contract-enforced peer attestation provides auditable exclusion of grossly poisoned or substituted models at modest accuracy cost but cannot detect attacks that preserve validation discrimination; the approach is most applicable to consortia of smaller hospitals needing verifiable model provenance.
## Clinical Analysis & Structured Key Points
A Blockchain-Enabled Federated Ensemble Framework for Privacy-Preserving Mortality Prediction Across Heterogeneous Healthcare Systems | medRxiv Skip to main content A Blockchain-Enabled Federated Ensemble Framework for Privacy-Preserving Mortality Prediction Across Heterogeneous Healthcare Systems View ORCID Profile Mostak Ahmed , View ORCID Profile Sabera Tabassum Wahid , View ORCID Profile Golam Sarwar Khondoker , View ORCID Profile Golam Rabbi , View ORCID Profile Rony Md Ariful Islam , View ORCID Profile Md Mostafizur Rahman , View ORCID Profile Al Rafy , View ORCID Profile Sanjida Islam , View ORCID Profile Md. Masuqul Haque doi: https://doi.org/10.64898/2026.11.19.26368466 Mostak Ahmed 1 Information and Communication Technology (ICT) Cell, University of Dhaka, Dhaka 1000, Bangladesh; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Mostak Ahmed For correspondence: mostak{at}du.ac.bd Sabera Tabassum Wahid 2 Master of Health Economics, The University of Queensland, Brisbane, Queensland, Australia; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sabera Tabassum Wahid Golam Sarwar Khondoker 3 Department of Data Science and Machine Learning, State University of Bangladesh, Dhaka, Bangladesh; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Golam Sarwar Khondoker Golam Rabbi 4 Department of Artificial Intelligence and Engineering, Zhengzhou University, Zhengzhou, Henan, China; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Golam Rabbi Rony Md Ariful Islam 5 School of Information Engineering, Department of Computer Science and Technology, North China University of Water Resources and Electric Power, Zhengzhou, Henan, China; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Rony Md Ariful Islam Md Mostafizur Rahman 6 Master of Science in Computer Science programme, San Francisco Bay University, Fremont, California, USA; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Md Mostafizur Rahman Al Rafy 7 College of Engineering and Technology, Westcliff University, Irvine, California, USA; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Al Rafy Sanjida Islam 8 Department of Medicine, Unity Care Homes Plc, Eidgah, Cox's Bazar, Bangladesh; Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sanjida Islam Md. Masuqul Haque 9 Department of Chemistry, University of Rajshahi, Rajshahi 6205, Bangladesh Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Md. Masuqul Haque Abstract Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Objective: To test whether a smart contract that admits hospital-trained models to a shared mortality model only after peer hospitals verify their performance can exclude poisoned or substituted models, and at what cost. Design: Retrospective simulation study of a federated learning protocol. Methodology: In the Blockchain-Enabled Federated Ensemble (BFE), hospitals committed SHA-256 digests of locally trained XGBoost and random-forest models to a Solidity contract; peers posted digest-bound validation AUROCs, and the contract admitted models with a median peer AUROC of at least 0.70. Admitted models were pooled by weighted logit averaging (BFE-PW) or stacking (BFE-Stack) with federated recalibration. Code and drafts were produced with an AI coding assistant under the authors' supervision. Setting: Four hospitals simulated from one academic centre with age, renal case-mix and laboratory-availability skew; a data-scarce regime keeping 5-20% of each hospital's patients; and a natural partition by care era. Participants: 425,215 adult admissions (186,103 patients; 2.35% in-hospital mortality) in MIMIC-IV v3.1, with predictors from the first 24 hours. Interventions: BFE against centralised, local-only, FedAvg, FedProx, median and Multi-Krum FedAvg and federated logistic regression, under seven attacks on training, attestation, recalibration and model delivery. Main outcome measures: AUROC and AUPRC (joint patient-clustered bootstrap), calibration, attack effects, gas and transactions. Results: Under gross label flipping, FedAvg lost 29% of its AUPRC while the contract excluded the poisoned models in every seed (Multi-Krum FedAvg was equally protected); all substituted models were detected, and collusion blocked release. A targeted flip of female deaths passed the gate and lowered sensitivity for deaths in women from 32.8% to 27.5%; poisoned recalibration inverted BFE-PW (AUROC 0.098) unless a median-of-sites recalibration was used (0.902). With large hospitals, BFE-PW matched FedAvg in AUROC (0.902) and BFE-Stack matched FedProx (0.907 and 0.906), below local-only (0.910) and centralised XGBoost (0.915), with lower AUPRC; with 5% of the data, BFE-PW exceeded local models (0.866 against 0.844). Pooled calibration hid two-fold under-prediction at one hospital (observed-to-expected ratio 2.13), which a local intercept corrected (0.93). One aggregation used 37 transactions and 3.3 million gas, rising roughly quadratically with consortium size. Conclusions: Contract-enforced peer attestation gave auditable, automatic exclusion of grossly poisoned and substituted models at a small accuracy cost, but did not detect attacks that preserve validation discrimination. It is most useful for consortia of small hospitals that need a verifiable record of which models entered a shared tool. Competing Interest Statement The authors have declared no competing interest. Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: The Institutional Review Boards of Beth Israel Deaconess Medical Center (Boston, Massachusetts, USA) and the Massachusetts Institute of Technology (Cambridge, Massachusetts, USA) gave ethical approval for the collection and sharing of the de-identified MIMIC-IV database and waived the requirement for informed consent. The authors accessed MIMIC-IV v3.1 under the PhysioNet Credentialed Health Data Use Agreement; no further ethical approval was required for this secondary analysis of de-identified data. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Data Availability MIMIC-IV v3.1 is available from PhysioNet (https://doi.org/10.13026/kpb9-mt58) to credentialed users who complete the required training and sign the PhysioNet Credentialed Health Data Use Agreement. Under that agreement the authors cannot share patient-level data, derived patient-level data or model predictions. All aggregate results are reported in the article and its supplementary material. The complete analysis code, including the Solidity smart contract, is available at https://github.com/mostakphoenixsoftbd/Blockchain-Enabled-Federated-Ensemble-Framework and archived at Zenodo (https://doi.org/10.5281/zenodo.23947714). https://doi.org/10.13026/kpb9-mt58 https://github.com/mostakphoenixsoftbd/Blockchain-Enabled-Federated-Ensemble-Framework Copyright The copyright holder for this preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC 4.0 International license . Back to top Previous Next Posted November 20, 2026. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following A Blockchain-Enabled Federated Ensemble Framework for Privacy-Preserving Mortality Prediction Across Heterogeneous Healthcare Systems Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share A Blockchain-Enabled Federated Ensemble Framework for Privacy-Preserving Mortality Prediction Across Heterogeneous Healthcare Systems Mostak Ahmed , Sabera Tabassum Wahid , Golam Sarwar Khondoker , Golam Rabbi , Rony Md Ariful Islam , Md Mostafizur Rahman , Al Rafy , Sanjida Islam , Md. Masuqul Haque medRxiv 2026.11.19.26368466; doi: https://doi.org/10.64898/2026.11.19.26368466 Share This Article: Copy Citation Tools A Blockchain-Enabled Federated Ensemble Framework for Privacy-Preserving Mortality Prediction Across Heterogeneous Healthcare Systems Mostak Ahmed , Sabera Tabassum Wahid , Golam Sarwar Khondoker , Golam Rabbi , Rony Md Ariful Islam , Md Mostafizur Rahman , Al Rafy , Sanjida Islam , Md. Masuqul Haque medRxiv 2026.11.19.26368466; doi: https://doi.org/10.64898/2026.11.19.26368466 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Areas All Articles Addiction Medicine (642) Allergy and Immunology (928) Anesthesia (355) Cardiovascular Medicine (5059) Dentistry and Oral Medicine (496) Dermatology (432) Emergency Medicine (681) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1703) Epidemiology (16352) Forensic Medicine (37) Gastroenterology (1249) Genetic and Genomic Medicine (7333) Geriatric Medicine (761) Health Economics (1101) Health Informatics (5350) Health Policy (1480) Health Systems and Quality Improvement (1847) Hematology (604) HIV/AIDS (1389) Infectious Diseases (except HIV/AIDS) (16537) Intensive Care and Critical Care Medicine (1210) Medical Education (702) Medical Ethics (157) Nephrology (757) Neurology (7664) Nursing (380) Nutrition (1144) Obstetrics and Gynecology (1308) Occupational and Environmental Health (1046) Oncology (3769) Ophthalmology (1105) Orthopedics (414) Otolaryngology (474) Pain Medicine (508) Palliative Medicine (144) Pathology (727) Pediatrics (1877) Pharmacology and Therapeutics (759) Primary Care Research (793) Psychiatry and Clinical Psychology (6186) Public and Global Health (10051) Radiology and Imaging (2562) Rehabilitation Medicine and Physical Therapy (1540) Respiratory Medicine (1280) Rheumatology (666) Sexual and Reproductive Health (814) Sports Medicine (601) Surgery (809) Toxicology (109) Transplantation (321) Urology (309)
## Related Clinical Research

- [Triglyceride-Glucose Index Linked to Airflow Limitation Severity in Older Nondiabetic, Normoxemic](https://medichelpline.com/clinical-feed/pubmed-43067697.md) (DOI: 10.12659/MSM.954698)
- [Auditable Reference Standard for extracting clinical data from transesophageal echocardiography re](https://medichelpline.com/clinical-feed/medrxiv-19-building-an-auditable-reference-standard-for-clinical-information-extraction.md)
- [Personalized Prenatal Care Linked to Delivery Patterns and Neonatal Outcomes in High-Risk Pregnanc](https://medichelpline.com/clinical-feed/pubmed-43063094.md) (DOI: 10.1097/MD.0000000000051549)
- [High blood pressure in Ugandan children and adolescents with Type 1 diabetes: multicenter prevalen](https://medichelpline.com/clinical-feed/plos-one-11-high-blood-pressure-among-children-and-adolescents-with-type-1-diabetes.md)
- [Leveraging MAHA and Kennedy to Combat Measles Vaccine Hesitancy](https://medichelpline.com/clinical-feed/stat-news-0-opinion-how-kennedy-and-maha-could-help-stop-measles.md)

## Navigation
- [← Back to Critical Care Feed](https://medichelpline.com/clinical-feed/critical-care.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.